xref: /openbmc/linux/arch/x86/kvm/x86.c (revision f5132b01386b5a67f1ff673bb2b96a507a3f7e41)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * derived from drivers/kvm/kvm_main.c
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  * Copyright (C) 2008 Qumranet, Inc.
8  * Copyright IBM Corporation, 2008
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Amit Shah    <amit.shah@qumranet.com>
15  *   Ben-Ami Yassour <benami@il.ibm.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  */
21 
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
34 #include <linux/fs.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <trace/events/kvm.h>
50 
51 #define CREATE_TRACE_POINTS
52 #include "trace.h"
53 
54 #include <asm/debugreg.h>
55 #include <asm/msr.h>
56 #include <asm/desc.h>
57 #include <asm/mtrr.h>
58 #include <asm/mce.h>
59 #include <asm/i387.h>
60 #include <asm/xcr.h>
61 #include <asm/pvclock.h>
62 #include <asm/div64.h>
63 
64 #define MAX_IO_MSRS 256
65 #define KVM_MAX_MCE_BANKS 32
66 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
67 
68 #define emul_to_vcpu(ctxt) \
69 	container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
70 
71 /* EFER defaults:
72  * - enable syscall per default because its emulated by KVM
73  * - enable LME and LMA per default on 64 bit KVM
74  */
75 #ifdef CONFIG_X86_64
76 static
77 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
78 #else
79 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
80 #endif
81 
82 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
83 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
84 
85 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
86 static void process_nmi(struct kvm_vcpu *vcpu);
87 
88 struct kvm_x86_ops *kvm_x86_ops;
89 EXPORT_SYMBOL_GPL(kvm_x86_ops);
90 
91 int ignore_msrs = 0;
92 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
93 
94 bool kvm_has_tsc_control;
95 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
96 u32  kvm_max_guest_tsc_khz;
97 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
98 
99 #define KVM_NR_SHARED_MSRS 16
100 
101 struct kvm_shared_msrs_global {
102 	int nr;
103 	u32 msrs[KVM_NR_SHARED_MSRS];
104 };
105 
106 struct kvm_shared_msrs {
107 	struct user_return_notifier urn;
108 	bool registered;
109 	struct kvm_shared_msr_values {
110 		u64 host;
111 		u64 curr;
112 	} values[KVM_NR_SHARED_MSRS];
113 };
114 
115 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
116 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
117 
118 struct kvm_stats_debugfs_item debugfs_entries[] = {
119 	{ "pf_fixed", VCPU_STAT(pf_fixed) },
120 	{ "pf_guest", VCPU_STAT(pf_guest) },
121 	{ "tlb_flush", VCPU_STAT(tlb_flush) },
122 	{ "invlpg", VCPU_STAT(invlpg) },
123 	{ "exits", VCPU_STAT(exits) },
124 	{ "io_exits", VCPU_STAT(io_exits) },
125 	{ "mmio_exits", VCPU_STAT(mmio_exits) },
126 	{ "signal_exits", VCPU_STAT(signal_exits) },
127 	{ "irq_window", VCPU_STAT(irq_window_exits) },
128 	{ "nmi_window", VCPU_STAT(nmi_window_exits) },
129 	{ "halt_exits", VCPU_STAT(halt_exits) },
130 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
131 	{ "hypercalls", VCPU_STAT(hypercalls) },
132 	{ "request_irq", VCPU_STAT(request_irq_exits) },
133 	{ "irq_exits", VCPU_STAT(irq_exits) },
134 	{ "host_state_reload", VCPU_STAT(host_state_reload) },
135 	{ "efer_reload", VCPU_STAT(efer_reload) },
136 	{ "fpu_reload", VCPU_STAT(fpu_reload) },
137 	{ "insn_emulation", VCPU_STAT(insn_emulation) },
138 	{ "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
139 	{ "irq_injections", VCPU_STAT(irq_injections) },
140 	{ "nmi_injections", VCPU_STAT(nmi_injections) },
141 	{ "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
142 	{ "mmu_pte_write", VM_STAT(mmu_pte_write) },
143 	{ "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
144 	{ "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
145 	{ "mmu_flooded", VM_STAT(mmu_flooded) },
146 	{ "mmu_recycled", VM_STAT(mmu_recycled) },
147 	{ "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
148 	{ "mmu_unsync", VM_STAT(mmu_unsync) },
149 	{ "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
150 	{ "largepages", VM_STAT(lpages) },
151 	{ NULL }
152 };
153 
154 u64 __read_mostly host_xcr0;
155 
156 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
157 
158 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
159 {
160 	int i;
161 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
162 		vcpu->arch.apf.gfns[i] = ~0;
163 }
164 
165 static void kvm_on_user_return(struct user_return_notifier *urn)
166 {
167 	unsigned slot;
168 	struct kvm_shared_msrs *locals
169 		= container_of(urn, struct kvm_shared_msrs, urn);
170 	struct kvm_shared_msr_values *values;
171 
172 	for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
173 		values = &locals->values[slot];
174 		if (values->host != values->curr) {
175 			wrmsrl(shared_msrs_global.msrs[slot], values->host);
176 			values->curr = values->host;
177 		}
178 	}
179 	locals->registered = false;
180 	user_return_notifier_unregister(urn);
181 }
182 
183 static void shared_msr_update(unsigned slot, u32 msr)
184 {
185 	struct kvm_shared_msrs *smsr;
186 	u64 value;
187 
188 	smsr = &__get_cpu_var(shared_msrs);
189 	/* only read, and nobody should modify it at this time,
190 	 * so don't need lock */
191 	if (slot >= shared_msrs_global.nr) {
192 		printk(KERN_ERR "kvm: invalid MSR slot!");
193 		return;
194 	}
195 	rdmsrl_safe(msr, &value);
196 	smsr->values[slot].host = value;
197 	smsr->values[slot].curr = value;
198 }
199 
200 void kvm_define_shared_msr(unsigned slot, u32 msr)
201 {
202 	if (slot >= shared_msrs_global.nr)
203 		shared_msrs_global.nr = slot + 1;
204 	shared_msrs_global.msrs[slot] = msr;
205 	/* we need ensured the shared_msr_global have been updated */
206 	smp_wmb();
207 }
208 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
209 
210 static void kvm_shared_msr_cpu_online(void)
211 {
212 	unsigned i;
213 
214 	for (i = 0; i < shared_msrs_global.nr; ++i)
215 		shared_msr_update(i, shared_msrs_global.msrs[i]);
216 }
217 
218 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
219 {
220 	struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
221 
222 	if (((value ^ smsr->values[slot].curr) & mask) == 0)
223 		return;
224 	smsr->values[slot].curr = value;
225 	wrmsrl(shared_msrs_global.msrs[slot], value);
226 	if (!smsr->registered) {
227 		smsr->urn.on_user_return = kvm_on_user_return;
228 		user_return_notifier_register(&smsr->urn);
229 		smsr->registered = true;
230 	}
231 }
232 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
233 
234 static void drop_user_return_notifiers(void *ignore)
235 {
236 	struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
237 
238 	if (smsr->registered)
239 		kvm_on_user_return(&smsr->urn);
240 }
241 
242 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
243 {
244 	if (irqchip_in_kernel(vcpu->kvm))
245 		return vcpu->arch.apic_base;
246 	else
247 		return vcpu->arch.apic_base;
248 }
249 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
250 
251 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
252 {
253 	/* TODO: reserve bits check */
254 	if (irqchip_in_kernel(vcpu->kvm))
255 		kvm_lapic_set_base(vcpu, data);
256 	else
257 		vcpu->arch.apic_base = data;
258 }
259 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
260 
261 #define EXCPT_BENIGN		0
262 #define EXCPT_CONTRIBUTORY	1
263 #define EXCPT_PF		2
264 
265 static int exception_class(int vector)
266 {
267 	switch (vector) {
268 	case PF_VECTOR:
269 		return EXCPT_PF;
270 	case DE_VECTOR:
271 	case TS_VECTOR:
272 	case NP_VECTOR:
273 	case SS_VECTOR:
274 	case GP_VECTOR:
275 		return EXCPT_CONTRIBUTORY;
276 	default:
277 		break;
278 	}
279 	return EXCPT_BENIGN;
280 }
281 
282 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
283 		unsigned nr, bool has_error, u32 error_code,
284 		bool reinject)
285 {
286 	u32 prev_nr;
287 	int class1, class2;
288 
289 	kvm_make_request(KVM_REQ_EVENT, vcpu);
290 
291 	if (!vcpu->arch.exception.pending) {
292 	queue:
293 		vcpu->arch.exception.pending = true;
294 		vcpu->arch.exception.has_error_code = has_error;
295 		vcpu->arch.exception.nr = nr;
296 		vcpu->arch.exception.error_code = error_code;
297 		vcpu->arch.exception.reinject = reinject;
298 		return;
299 	}
300 
301 	/* to check exception */
302 	prev_nr = vcpu->arch.exception.nr;
303 	if (prev_nr == DF_VECTOR) {
304 		/* triple fault -> shutdown */
305 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
306 		return;
307 	}
308 	class1 = exception_class(prev_nr);
309 	class2 = exception_class(nr);
310 	if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
311 		|| (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
312 		/* generate double fault per SDM Table 5-5 */
313 		vcpu->arch.exception.pending = true;
314 		vcpu->arch.exception.has_error_code = true;
315 		vcpu->arch.exception.nr = DF_VECTOR;
316 		vcpu->arch.exception.error_code = 0;
317 	} else
318 		/* replace previous exception with a new one in a hope
319 		   that instruction re-execution will regenerate lost
320 		   exception */
321 		goto queue;
322 }
323 
324 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
325 {
326 	kvm_multiple_exception(vcpu, nr, false, 0, false);
327 }
328 EXPORT_SYMBOL_GPL(kvm_queue_exception);
329 
330 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
331 {
332 	kvm_multiple_exception(vcpu, nr, false, 0, true);
333 }
334 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
335 
336 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
337 {
338 	if (err)
339 		kvm_inject_gp(vcpu, 0);
340 	else
341 		kvm_x86_ops->skip_emulated_instruction(vcpu);
342 }
343 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
344 
345 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
346 {
347 	++vcpu->stat.pf_guest;
348 	vcpu->arch.cr2 = fault->address;
349 	kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
350 }
351 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
352 
353 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
354 {
355 	if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
356 		vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
357 	else
358 		vcpu->arch.mmu.inject_page_fault(vcpu, fault);
359 }
360 
361 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
362 {
363 	atomic_inc(&vcpu->arch.nmi_queued);
364 	kvm_make_request(KVM_REQ_NMI, vcpu);
365 }
366 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
367 
368 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
369 {
370 	kvm_multiple_exception(vcpu, nr, true, error_code, false);
371 }
372 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
373 
374 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
375 {
376 	kvm_multiple_exception(vcpu, nr, true, error_code, true);
377 }
378 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
379 
380 /*
381  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
382  * a #GP and return false.
383  */
384 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
385 {
386 	if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
387 		return true;
388 	kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
389 	return false;
390 }
391 EXPORT_SYMBOL_GPL(kvm_require_cpl);
392 
393 /*
394  * This function will be used to read from the physical memory of the currently
395  * running guest. The difference to kvm_read_guest_page is that this function
396  * can read from guest physical or from the guest's guest physical memory.
397  */
398 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
399 			    gfn_t ngfn, void *data, int offset, int len,
400 			    u32 access)
401 {
402 	gfn_t real_gfn;
403 	gpa_t ngpa;
404 
405 	ngpa     = gfn_to_gpa(ngfn);
406 	real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
407 	if (real_gfn == UNMAPPED_GVA)
408 		return -EFAULT;
409 
410 	real_gfn = gpa_to_gfn(real_gfn);
411 
412 	return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
413 }
414 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
415 
416 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
417 			       void *data, int offset, int len, u32 access)
418 {
419 	return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
420 				       data, offset, len, access);
421 }
422 
423 /*
424  * Load the pae pdptrs.  Return true is they are all valid.
425  */
426 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
427 {
428 	gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
429 	unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
430 	int i;
431 	int ret;
432 	u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
433 
434 	ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
435 				      offset * sizeof(u64), sizeof(pdpte),
436 				      PFERR_USER_MASK|PFERR_WRITE_MASK);
437 	if (ret < 0) {
438 		ret = 0;
439 		goto out;
440 	}
441 	for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
442 		if (is_present_gpte(pdpte[i]) &&
443 		    (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
444 			ret = 0;
445 			goto out;
446 		}
447 	}
448 	ret = 1;
449 
450 	memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
451 	__set_bit(VCPU_EXREG_PDPTR,
452 		  (unsigned long *)&vcpu->arch.regs_avail);
453 	__set_bit(VCPU_EXREG_PDPTR,
454 		  (unsigned long *)&vcpu->arch.regs_dirty);
455 out:
456 
457 	return ret;
458 }
459 EXPORT_SYMBOL_GPL(load_pdptrs);
460 
461 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
462 {
463 	u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
464 	bool changed = true;
465 	int offset;
466 	gfn_t gfn;
467 	int r;
468 
469 	if (is_long_mode(vcpu) || !is_pae(vcpu))
470 		return false;
471 
472 	if (!test_bit(VCPU_EXREG_PDPTR,
473 		      (unsigned long *)&vcpu->arch.regs_avail))
474 		return true;
475 
476 	gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
477 	offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
478 	r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
479 				       PFERR_USER_MASK | PFERR_WRITE_MASK);
480 	if (r < 0)
481 		goto out;
482 	changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
483 out:
484 
485 	return changed;
486 }
487 
488 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
489 {
490 	unsigned long old_cr0 = kvm_read_cr0(vcpu);
491 	unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
492 				    X86_CR0_CD | X86_CR0_NW;
493 
494 	cr0 |= X86_CR0_ET;
495 
496 #ifdef CONFIG_X86_64
497 	if (cr0 & 0xffffffff00000000UL)
498 		return 1;
499 #endif
500 
501 	cr0 &= ~CR0_RESERVED_BITS;
502 
503 	if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
504 		return 1;
505 
506 	if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
507 		return 1;
508 
509 	if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
510 #ifdef CONFIG_X86_64
511 		if ((vcpu->arch.efer & EFER_LME)) {
512 			int cs_db, cs_l;
513 
514 			if (!is_pae(vcpu))
515 				return 1;
516 			kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
517 			if (cs_l)
518 				return 1;
519 		} else
520 #endif
521 		if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
522 						 kvm_read_cr3(vcpu)))
523 			return 1;
524 	}
525 
526 	kvm_x86_ops->set_cr0(vcpu, cr0);
527 
528 	if ((cr0 ^ old_cr0) & X86_CR0_PG) {
529 		kvm_clear_async_pf_completion_queue(vcpu);
530 		kvm_async_pf_hash_reset(vcpu);
531 	}
532 
533 	if ((cr0 ^ old_cr0) & update_bits)
534 		kvm_mmu_reset_context(vcpu);
535 	return 0;
536 }
537 EXPORT_SYMBOL_GPL(kvm_set_cr0);
538 
539 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
540 {
541 	(void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
542 }
543 EXPORT_SYMBOL_GPL(kvm_lmsw);
544 
545 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
546 {
547 	u64 xcr0;
548 
549 	/* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
550 	if (index != XCR_XFEATURE_ENABLED_MASK)
551 		return 1;
552 	xcr0 = xcr;
553 	if (kvm_x86_ops->get_cpl(vcpu) != 0)
554 		return 1;
555 	if (!(xcr0 & XSTATE_FP))
556 		return 1;
557 	if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
558 		return 1;
559 	if (xcr0 & ~host_xcr0)
560 		return 1;
561 	vcpu->arch.xcr0 = xcr0;
562 	vcpu->guest_xcr0_loaded = 0;
563 	return 0;
564 }
565 
566 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
567 {
568 	if (__kvm_set_xcr(vcpu, index, xcr)) {
569 		kvm_inject_gp(vcpu, 0);
570 		return 1;
571 	}
572 	return 0;
573 }
574 EXPORT_SYMBOL_GPL(kvm_set_xcr);
575 
576 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
577 {
578 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
579 	unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
580 				   X86_CR4_PAE | X86_CR4_SMEP;
581 	if (cr4 & CR4_RESERVED_BITS)
582 		return 1;
583 
584 	if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
585 		return 1;
586 
587 	if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
588 		return 1;
589 
590 	if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
591 		return 1;
592 
593 	if (is_long_mode(vcpu)) {
594 		if (!(cr4 & X86_CR4_PAE))
595 			return 1;
596 	} else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
597 		   && ((cr4 ^ old_cr4) & pdptr_bits)
598 		   && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
599 				   kvm_read_cr3(vcpu)))
600 		return 1;
601 
602 	if (kvm_x86_ops->set_cr4(vcpu, cr4))
603 		return 1;
604 
605 	if ((cr4 ^ old_cr4) & pdptr_bits)
606 		kvm_mmu_reset_context(vcpu);
607 
608 	if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
609 		kvm_update_cpuid(vcpu);
610 
611 	return 0;
612 }
613 EXPORT_SYMBOL_GPL(kvm_set_cr4);
614 
615 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
616 {
617 	if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
618 		kvm_mmu_sync_roots(vcpu);
619 		kvm_mmu_flush_tlb(vcpu);
620 		return 0;
621 	}
622 
623 	if (is_long_mode(vcpu)) {
624 		if (cr3 & CR3_L_MODE_RESERVED_BITS)
625 			return 1;
626 	} else {
627 		if (is_pae(vcpu)) {
628 			if (cr3 & CR3_PAE_RESERVED_BITS)
629 				return 1;
630 			if (is_paging(vcpu) &&
631 			    !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
632 				return 1;
633 		}
634 		/*
635 		 * We don't check reserved bits in nonpae mode, because
636 		 * this isn't enforced, and VMware depends on this.
637 		 */
638 	}
639 
640 	/*
641 	 * Does the new cr3 value map to physical memory? (Note, we
642 	 * catch an invalid cr3 even in real-mode, because it would
643 	 * cause trouble later on when we turn on paging anyway.)
644 	 *
645 	 * A real CPU would silently accept an invalid cr3 and would
646 	 * attempt to use it - with largely undefined (and often hard
647 	 * to debug) behavior on the guest side.
648 	 */
649 	if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
650 		return 1;
651 	vcpu->arch.cr3 = cr3;
652 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
653 	vcpu->arch.mmu.new_cr3(vcpu);
654 	return 0;
655 }
656 EXPORT_SYMBOL_GPL(kvm_set_cr3);
657 
658 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
659 {
660 	if (cr8 & CR8_RESERVED_BITS)
661 		return 1;
662 	if (irqchip_in_kernel(vcpu->kvm))
663 		kvm_lapic_set_tpr(vcpu, cr8);
664 	else
665 		vcpu->arch.cr8 = cr8;
666 	return 0;
667 }
668 EXPORT_SYMBOL_GPL(kvm_set_cr8);
669 
670 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
671 {
672 	if (irqchip_in_kernel(vcpu->kvm))
673 		return kvm_lapic_get_cr8(vcpu);
674 	else
675 		return vcpu->arch.cr8;
676 }
677 EXPORT_SYMBOL_GPL(kvm_get_cr8);
678 
679 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
680 {
681 	switch (dr) {
682 	case 0 ... 3:
683 		vcpu->arch.db[dr] = val;
684 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
685 			vcpu->arch.eff_db[dr] = val;
686 		break;
687 	case 4:
688 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
689 			return 1; /* #UD */
690 		/* fall through */
691 	case 6:
692 		if (val & 0xffffffff00000000ULL)
693 			return -1; /* #GP */
694 		vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
695 		break;
696 	case 5:
697 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
698 			return 1; /* #UD */
699 		/* fall through */
700 	default: /* 7 */
701 		if (val & 0xffffffff00000000ULL)
702 			return -1; /* #GP */
703 		vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
704 		if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
705 			kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
706 			vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
707 		}
708 		break;
709 	}
710 
711 	return 0;
712 }
713 
714 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
715 {
716 	int res;
717 
718 	res = __kvm_set_dr(vcpu, dr, val);
719 	if (res > 0)
720 		kvm_queue_exception(vcpu, UD_VECTOR);
721 	else if (res < 0)
722 		kvm_inject_gp(vcpu, 0);
723 
724 	return res;
725 }
726 EXPORT_SYMBOL_GPL(kvm_set_dr);
727 
728 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
729 {
730 	switch (dr) {
731 	case 0 ... 3:
732 		*val = vcpu->arch.db[dr];
733 		break;
734 	case 4:
735 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
736 			return 1;
737 		/* fall through */
738 	case 6:
739 		*val = vcpu->arch.dr6;
740 		break;
741 	case 5:
742 		if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
743 			return 1;
744 		/* fall through */
745 	default: /* 7 */
746 		*val = vcpu->arch.dr7;
747 		break;
748 	}
749 
750 	return 0;
751 }
752 
753 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
754 {
755 	if (_kvm_get_dr(vcpu, dr, val)) {
756 		kvm_queue_exception(vcpu, UD_VECTOR);
757 		return 1;
758 	}
759 	return 0;
760 }
761 EXPORT_SYMBOL_GPL(kvm_get_dr);
762 
763 /*
764  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
765  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
766  *
767  * This list is modified at module load time to reflect the
768  * capabilities of the host cpu. This capabilities test skips MSRs that are
769  * kvm-specific. Those are put in the beginning of the list.
770  */
771 
772 #define KVM_SAVE_MSRS_BEGIN	9
773 static u32 msrs_to_save[] = {
774 	MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
775 	MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
776 	HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
777 	HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
778 	MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
779 	MSR_STAR,
780 #ifdef CONFIG_X86_64
781 	MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
782 #endif
783 	MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
784 };
785 
786 static unsigned num_msrs_to_save;
787 
788 static u32 emulated_msrs[] = {
789 	MSR_IA32_TSCDEADLINE,
790 	MSR_IA32_MISC_ENABLE,
791 	MSR_IA32_MCG_STATUS,
792 	MSR_IA32_MCG_CTL,
793 };
794 
795 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
796 {
797 	u64 old_efer = vcpu->arch.efer;
798 
799 	if (efer & efer_reserved_bits)
800 		return 1;
801 
802 	if (is_paging(vcpu)
803 	    && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
804 		return 1;
805 
806 	if (efer & EFER_FFXSR) {
807 		struct kvm_cpuid_entry2 *feat;
808 
809 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
810 		if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
811 			return 1;
812 	}
813 
814 	if (efer & EFER_SVME) {
815 		struct kvm_cpuid_entry2 *feat;
816 
817 		feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
818 		if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
819 			return 1;
820 	}
821 
822 	efer &= ~EFER_LMA;
823 	efer |= vcpu->arch.efer & EFER_LMA;
824 
825 	kvm_x86_ops->set_efer(vcpu, efer);
826 
827 	vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
828 
829 	/* Update reserved bits */
830 	if ((efer ^ old_efer) & EFER_NX)
831 		kvm_mmu_reset_context(vcpu);
832 
833 	return 0;
834 }
835 
836 void kvm_enable_efer_bits(u64 mask)
837 {
838        efer_reserved_bits &= ~mask;
839 }
840 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
841 
842 
843 /*
844  * Writes msr value into into the appropriate "register".
845  * Returns 0 on success, non-0 otherwise.
846  * Assumes vcpu_load() was already called.
847  */
848 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
849 {
850 	return kvm_x86_ops->set_msr(vcpu, msr_index, data);
851 }
852 
853 /*
854  * Adapt set_msr() to msr_io()'s calling convention
855  */
856 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
857 {
858 	return kvm_set_msr(vcpu, index, *data);
859 }
860 
861 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
862 {
863 	int version;
864 	int r;
865 	struct pvclock_wall_clock wc;
866 	struct timespec boot;
867 
868 	if (!wall_clock)
869 		return;
870 
871 	r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
872 	if (r)
873 		return;
874 
875 	if (version & 1)
876 		++version;  /* first time write, random junk */
877 
878 	++version;
879 
880 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
881 
882 	/*
883 	 * The guest calculates current wall clock time by adding
884 	 * system time (updated by kvm_guest_time_update below) to the
885 	 * wall clock specified here.  guest system time equals host
886 	 * system time for us, thus we must fill in host boot time here.
887 	 */
888 	getboottime(&boot);
889 
890 	wc.sec = boot.tv_sec;
891 	wc.nsec = boot.tv_nsec;
892 	wc.version = version;
893 
894 	kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
895 
896 	version++;
897 	kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
898 }
899 
900 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
901 {
902 	uint32_t quotient, remainder;
903 
904 	/* Don't try to replace with do_div(), this one calculates
905 	 * "(dividend << 32) / divisor" */
906 	__asm__ ( "divl %4"
907 		  : "=a" (quotient), "=d" (remainder)
908 		  : "0" (0), "1" (dividend), "r" (divisor) );
909 	return quotient;
910 }
911 
912 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
913 			       s8 *pshift, u32 *pmultiplier)
914 {
915 	uint64_t scaled64;
916 	int32_t  shift = 0;
917 	uint64_t tps64;
918 	uint32_t tps32;
919 
920 	tps64 = base_khz * 1000LL;
921 	scaled64 = scaled_khz * 1000LL;
922 	while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
923 		tps64 >>= 1;
924 		shift--;
925 	}
926 
927 	tps32 = (uint32_t)tps64;
928 	while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
929 		if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
930 			scaled64 >>= 1;
931 		else
932 			tps32 <<= 1;
933 		shift++;
934 	}
935 
936 	*pshift = shift;
937 	*pmultiplier = div_frac(scaled64, tps32);
938 
939 	pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
940 		 __func__, base_khz, scaled_khz, shift, *pmultiplier);
941 }
942 
943 static inline u64 get_kernel_ns(void)
944 {
945 	struct timespec ts;
946 
947 	WARN_ON(preemptible());
948 	ktime_get_ts(&ts);
949 	monotonic_to_bootbased(&ts);
950 	return timespec_to_ns(&ts);
951 }
952 
953 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
954 unsigned long max_tsc_khz;
955 
956 static inline int kvm_tsc_changes_freq(void)
957 {
958 	int cpu = get_cpu();
959 	int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
960 		  cpufreq_quick_get(cpu) != 0;
961 	put_cpu();
962 	return ret;
963 }
964 
965 u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
966 {
967 	if (vcpu->arch.virtual_tsc_khz)
968 		return vcpu->arch.virtual_tsc_khz;
969 	else
970 		return __this_cpu_read(cpu_tsc_khz);
971 }
972 
973 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
974 {
975 	u64 ret;
976 
977 	WARN_ON(preemptible());
978 	if (kvm_tsc_changes_freq())
979 		printk_once(KERN_WARNING
980 		 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
981 	ret = nsec * vcpu_tsc_khz(vcpu);
982 	do_div(ret, USEC_PER_SEC);
983 	return ret;
984 }
985 
986 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
987 {
988 	/* Compute a scale to convert nanoseconds in TSC cycles */
989 	kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
990 			   &vcpu->arch.tsc_catchup_shift,
991 			   &vcpu->arch.tsc_catchup_mult);
992 }
993 
994 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
995 {
996 	u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
997 				      vcpu->arch.tsc_catchup_mult,
998 				      vcpu->arch.tsc_catchup_shift);
999 	tsc += vcpu->arch.last_tsc_write;
1000 	return tsc;
1001 }
1002 
1003 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1004 {
1005 	struct kvm *kvm = vcpu->kvm;
1006 	u64 offset, ns, elapsed;
1007 	unsigned long flags;
1008 	s64 sdiff;
1009 
1010 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1011 	offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1012 	ns = get_kernel_ns();
1013 	elapsed = ns - kvm->arch.last_tsc_nsec;
1014 	sdiff = data - kvm->arch.last_tsc_write;
1015 	if (sdiff < 0)
1016 		sdiff = -sdiff;
1017 
1018 	/*
1019 	 * Special case: close write to TSC within 5 seconds of
1020 	 * another CPU is interpreted as an attempt to synchronize
1021 	 * The 5 seconds is to accommodate host load / swapping as
1022 	 * well as any reset of TSC during the boot process.
1023 	 *
1024 	 * In that case, for a reliable TSC, we can match TSC offsets,
1025 	 * or make a best guest using elapsed value.
1026 	 */
1027 	if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1028 	    elapsed < 5ULL * NSEC_PER_SEC) {
1029 		if (!check_tsc_unstable()) {
1030 			offset = kvm->arch.last_tsc_offset;
1031 			pr_debug("kvm: matched tsc offset for %llu\n", data);
1032 		} else {
1033 			u64 delta = nsec_to_cycles(vcpu, elapsed);
1034 			offset += delta;
1035 			pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1036 		}
1037 		ns = kvm->arch.last_tsc_nsec;
1038 	}
1039 	kvm->arch.last_tsc_nsec = ns;
1040 	kvm->arch.last_tsc_write = data;
1041 	kvm->arch.last_tsc_offset = offset;
1042 	kvm_x86_ops->write_tsc_offset(vcpu, offset);
1043 	raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1044 
1045 	/* Reset of TSC must disable overshoot protection below */
1046 	vcpu->arch.hv_clock.tsc_timestamp = 0;
1047 	vcpu->arch.last_tsc_write = data;
1048 	vcpu->arch.last_tsc_nsec = ns;
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1051 
1052 static int kvm_guest_time_update(struct kvm_vcpu *v)
1053 {
1054 	unsigned long flags;
1055 	struct kvm_vcpu_arch *vcpu = &v->arch;
1056 	void *shared_kaddr;
1057 	unsigned long this_tsc_khz;
1058 	s64 kernel_ns, max_kernel_ns;
1059 	u64 tsc_timestamp;
1060 
1061 	/* Keep irq disabled to prevent changes to the clock */
1062 	local_irq_save(flags);
1063 	tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
1064 	kernel_ns = get_kernel_ns();
1065 	this_tsc_khz = vcpu_tsc_khz(v);
1066 	if (unlikely(this_tsc_khz == 0)) {
1067 		local_irq_restore(flags);
1068 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1069 		return 1;
1070 	}
1071 
1072 	/*
1073 	 * We may have to catch up the TSC to match elapsed wall clock
1074 	 * time for two reasons, even if kvmclock is used.
1075 	 *   1) CPU could have been running below the maximum TSC rate
1076 	 *   2) Broken TSC compensation resets the base at each VCPU
1077 	 *      entry to avoid unknown leaps of TSC even when running
1078 	 *      again on the same CPU.  This may cause apparent elapsed
1079 	 *      time to disappear, and the guest to stand still or run
1080 	 *	very slowly.
1081 	 */
1082 	if (vcpu->tsc_catchup) {
1083 		u64 tsc = compute_guest_tsc(v, kernel_ns);
1084 		if (tsc > tsc_timestamp) {
1085 			kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1086 			tsc_timestamp = tsc;
1087 		}
1088 	}
1089 
1090 	local_irq_restore(flags);
1091 
1092 	if (!vcpu->time_page)
1093 		return 0;
1094 
1095 	/*
1096 	 * Time as measured by the TSC may go backwards when resetting the base
1097 	 * tsc_timestamp.  The reason for this is that the TSC resolution is
1098 	 * higher than the resolution of the other clock scales.  Thus, many
1099 	 * possible measurments of the TSC correspond to one measurement of any
1100 	 * other clock, and so a spread of values is possible.  This is not a
1101 	 * problem for the computation of the nanosecond clock; with TSC rates
1102 	 * around 1GHZ, there can only be a few cycles which correspond to one
1103 	 * nanosecond value, and any path through this code will inevitably
1104 	 * take longer than that.  However, with the kernel_ns value itself,
1105 	 * the precision may be much lower, down to HZ granularity.  If the
1106 	 * first sampling of TSC against kernel_ns ends in the low part of the
1107 	 * range, and the second in the high end of the range, we can get:
1108 	 *
1109 	 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1110 	 *
1111 	 * As the sampling errors potentially range in the thousands of cycles,
1112 	 * it is possible such a time value has already been observed by the
1113 	 * guest.  To protect against this, we must compute the system time as
1114 	 * observed by the guest and ensure the new system time is greater.
1115 	 */
1116 	max_kernel_ns = 0;
1117 	if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1118 		max_kernel_ns = vcpu->last_guest_tsc -
1119 				vcpu->hv_clock.tsc_timestamp;
1120 		max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1121 				    vcpu->hv_clock.tsc_to_system_mul,
1122 				    vcpu->hv_clock.tsc_shift);
1123 		max_kernel_ns += vcpu->last_kernel_ns;
1124 	}
1125 
1126 	if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1127 		kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1128 				   &vcpu->hv_clock.tsc_shift,
1129 				   &vcpu->hv_clock.tsc_to_system_mul);
1130 		vcpu->hw_tsc_khz = this_tsc_khz;
1131 	}
1132 
1133 	if (max_kernel_ns > kernel_ns)
1134 		kernel_ns = max_kernel_ns;
1135 
1136 	/* With all the info we got, fill in the values */
1137 	vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1138 	vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1139 	vcpu->last_kernel_ns = kernel_ns;
1140 	vcpu->last_guest_tsc = tsc_timestamp;
1141 	vcpu->hv_clock.flags = 0;
1142 
1143 	/*
1144 	 * The interface expects us to write an even number signaling that the
1145 	 * update is finished. Since the guest won't see the intermediate
1146 	 * state, we just increase by 2 at the end.
1147 	 */
1148 	vcpu->hv_clock.version += 2;
1149 
1150 	shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1151 
1152 	memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1153 	       sizeof(vcpu->hv_clock));
1154 
1155 	kunmap_atomic(shared_kaddr, KM_USER0);
1156 
1157 	mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1158 	return 0;
1159 }
1160 
1161 static bool msr_mtrr_valid(unsigned msr)
1162 {
1163 	switch (msr) {
1164 	case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1165 	case MSR_MTRRfix64K_00000:
1166 	case MSR_MTRRfix16K_80000:
1167 	case MSR_MTRRfix16K_A0000:
1168 	case MSR_MTRRfix4K_C0000:
1169 	case MSR_MTRRfix4K_C8000:
1170 	case MSR_MTRRfix4K_D0000:
1171 	case MSR_MTRRfix4K_D8000:
1172 	case MSR_MTRRfix4K_E0000:
1173 	case MSR_MTRRfix4K_E8000:
1174 	case MSR_MTRRfix4K_F0000:
1175 	case MSR_MTRRfix4K_F8000:
1176 	case MSR_MTRRdefType:
1177 	case MSR_IA32_CR_PAT:
1178 		return true;
1179 	case 0x2f8:
1180 		return true;
1181 	}
1182 	return false;
1183 }
1184 
1185 static bool valid_pat_type(unsigned t)
1186 {
1187 	return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1188 }
1189 
1190 static bool valid_mtrr_type(unsigned t)
1191 {
1192 	return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1193 }
1194 
1195 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1196 {
1197 	int i;
1198 
1199 	if (!msr_mtrr_valid(msr))
1200 		return false;
1201 
1202 	if (msr == MSR_IA32_CR_PAT) {
1203 		for (i = 0; i < 8; i++)
1204 			if (!valid_pat_type((data >> (i * 8)) & 0xff))
1205 				return false;
1206 		return true;
1207 	} else if (msr == MSR_MTRRdefType) {
1208 		if (data & ~0xcff)
1209 			return false;
1210 		return valid_mtrr_type(data & 0xff);
1211 	} else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1212 		for (i = 0; i < 8 ; i++)
1213 			if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1214 				return false;
1215 		return true;
1216 	}
1217 
1218 	/* variable MTRRs */
1219 	return valid_mtrr_type(data & 0xff);
1220 }
1221 
1222 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1223 {
1224 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1225 
1226 	if (!mtrr_valid(vcpu, msr, data))
1227 		return 1;
1228 
1229 	if (msr == MSR_MTRRdefType) {
1230 		vcpu->arch.mtrr_state.def_type = data;
1231 		vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1232 	} else if (msr == MSR_MTRRfix64K_00000)
1233 		p[0] = data;
1234 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1235 		p[1 + msr - MSR_MTRRfix16K_80000] = data;
1236 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1237 		p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1238 	else if (msr == MSR_IA32_CR_PAT)
1239 		vcpu->arch.pat = data;
1240 	else {	/* Variable MTRRs */
1241 		int idx, is_mtrr_mask;
1242 		u64 *pt;
1243 
1244 		idx = (msr - 0x200) / 2;
1245 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1246 		if (!is_mtrr_mask)
1247 			pt =
1248 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1249 		else
1250 			pt =
1251 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1252 		*pt = data;
1253 	}
1254 
1255 	kvm_mmu_reset_context(vcpu);
1256 	return 0;
1257 }
1258 
1259 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1260 {
1261 	u64 mcg_cap = vcpu->arch.mcg_cap;
1262 	unsigned bank_num = mcg_cap & 0xff;
1263 
1264 	switch (msr) {
1265 	case MSR_IA32_MCG_STATUS:
1266 		vcpu->arch.mcg_status = data;
1267 		break;
1268 	case MSR_IA32_MCG_CTL:
1269 		if (!(mcg_cap & MCG_CTL_P))
1270 			return 1;
1271 		if (data != 0 && data != ~(u64)0)
1272 			return -1;
1273 		vcpu->arch.mcg_ctl = data;
1274 		break;
1275 	default:
1276 		if (msr >= MSR_IA32_MC0_CTL &&
1277 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1278 			u32 offset = msr - MSR_IA32_MC0_CTL;
1279 			/* only 0 or all 1s can be written to IA32_MCi_CTL
1280 			 * some Linux kernels though clear bit 10 in bank 4 to
1281 			 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1282 			 * this to avoid an uncatched #GP in the guest
1283 			 */
1284 			if ((offset & 0x3) == 0 &&
1285 			    data != 0 && (data | (1 << 10)) != ~(u64)0)
1286 				return -1;
1287 			vcpu->arch.mce_banks[offset] = data;
1288 			break;
1289 		}
1290 		return 1;
1291 	}
1292 	return 0;
1293 }
1294 
1295 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1296 {
1297 	struct kvm *kvm = vcpu->kvm;
1298 	int lm = is_long_mode(vcpu);
1299 	u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1300 		: (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1301 	u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1302 		: kvm->arch.xen_hvm_config.blob_size_32;
1303 	u32 page_num = data & ~PAGE_MASK;
1304 	u64 page_addr = data & PAGE_MASK;
1305 	u8 *page;
1306 	int r;
1307 
1308 	r = -E2BIG;
1309 	if (page_num >= blob_size)
1310 		goto out;
1311 	r = -ENOMEM;
1312 	page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1313 	if (IS_ERR(page)) {
1314 		r = PTR_ERR(page);
1315 		goto out;
1316 	}
1317 	if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1318 		goto out_free;
1319 	r = 0;
1320 out_free:
1321 	kfree(page);
1322 out:
1323 	return r;
1324 }
1325 
1326 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1327 {
1328 	return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1329 }
1330 
1331 static bool kvm_hv_msr_partition_wide(u32 msr)
1332 {
1333 	bool r = false;
1334 	switch (msr) {
1335 	case HV_X64_MSR_GUEST_OS_ID:
1336 	case HV_X64_MSR_HYPERCALL:
1337 		r = true;
1338 		break;
1339 	}
1340 
1341 	return r;
1342 }
1343 
1344 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1345 {
1346 	struct kvm *kvm = vcpu->kvm;
1347 
1348 	switch (msr) {
1349 	case HV_X64_MSR_GUEST_OS_ID:
1350 		kvm->arch.hv_guest_os_id = data;
1351 		/* setting guest os id to zero disables hypercall page */
1352 		if (!kvm->arch.hv_guest_os_id)
1353 			kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1354 		break;
1355 	case HV_X64_MSR_HYPERCALL: {
1356 		u64 gfn;
1357 		unsigned long addr;
1358 		u8 instructions[4];
1359 
1360 		/* if guest os id is not set hypercall should remain disabled */
1361 		if (!kvm->arch.hv_guest_os_id)
1362 			break;
1363 		if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1364 			kvm->arch.hv_hypercall = data;
1365 			break;
1366 		}
1367 		gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1368 		addr = gfn_to_hva(kvm, gfn);
1369 		if (kvm_is_error_hva(addr))
1370 			return 1;
1371 		kvm_x86_ops->patch_hypercall(vcpu, instructions);
1372 		((unsigned char *)instructions)[3] = 0xc3; /* ret */
1373 		if (__copy_to_user((void __user *)addr, instructions, 4))
1374 			return 1;
1375 		kvm->arch.hv_hypercall = data;
1376 		break;
1377 	}
1378 	default:
1379 		pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1380 			  "data 0x%llx\n", msr, data);
1381 		return 1;
1382 	}
1383 	return 0;
1384 }
1385 
1386 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1387 {
1388 	switch (msr) {
1389 	case HV_X64_MSR_APIC_ASSIST_PAGE: {
1390 		unsigned long addr;
1391 
1392 		if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1393 			vcpu->arch.hv_vapic = data;
1394 			break;
1395 		}
1396 		addr = gfn_to_hva(vcpu->kvm, data >>
1397 				  HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1398 		if (kvm_is_error_hva(addr))
1399 			return 1;
1400 		if (__clear_user((void __user *)addr, PAGE_SIZE))
1401 			return 1;
1402 		vcpu->arch.hv_vapic = data;
1403 		break;
1404 	}
1405 	case HV_X64_MSR_EOI:
1406 		return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1407 	case HV_X64_MSR_ICR:
1408 		return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1409 	case HV_X64_MSR_TPR:
1410 		return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1411 	default:
1412 		pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1413 			  "data 0x%llx\n", msr, data);
1414 		return 1;
1415 	}
1416 
1417 	return 0;
1418 }
1419 
1420 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1421 {
1422 	gpa_t gpa = data & ~0x3f;
1423 
1424 	/* Bits 2:5 are resrved, Should be zero */
1425 	if (data & 0x3c)
1426 		return 1;
1427 
1428 	vcpu->arch.apf.msr_val = data;
1429 
1430 	if (!(data & KVM_ASYNC_PF_ENABLED)) {
1431 		kvm_clear_async_pf_completion_queue(vcpu);
1432 		kvm_async_pf_hash_reset(vcpu);
1433 		return 0;
1434 	}
1435 
1436 	if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1437 		return 1;
1438 
1439 	vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1440 	kvm_async_pf_wakeup_all(vcpu);
1441 	return 0;
1442 }
1443 
1444 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1445 {
1446 	if (vcpu->arch.time_page) {
1447 		kvm_release_page_dirty(vcpu->arch.time_page);
1448 		vcpu->arch.time_page = NULL;
1449 	}
1450 }
1451 
1452 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1453 {
1454 	u64 delta;
1455 
1456 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1457 		return;
1458 
1459 	delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1460 	vcpu->arch.st.last_steal = current->sched_info.run_delay;
1461 	vcpu->arch.st.accum_steal = delta;
1462 }
1463 
1464 static void record_steal_time(struct kvm_vcpu *vcpu)
1465 {
1466 	if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1467 		return;
1468 
1469 	if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1470 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1471 		return;
1472 
1473 	vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1474 	vcpu->arch.st.steal.version += 2;
1475 	vcpu->arch.st.accum_steal = 0;
1476 
1477 	kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1478 		&vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1479 }
1480 
1481 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1482 {
1483 	switch (msr) {
1484 	case MSR_EFER:
1485 		return set_efer(vcpu, data);
1486 	case MSR_K7_HWCR:
1487 		data &= ~(u64)0x40;	/* ignore flush filter disable */
1488 		data &= ~(u64)0x100;	/* ignore ignne emulation enable */
1489 		if (data != 0) {
1490 			pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1491 				data);
1492 			return 1;
1493 		}
1494 		break;
1495 	case MSR_FAM10H_MMIO_CONF_BASE:
1496 		if (data != 0) {
1497 			pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1498 				"0x%llx\n", data);
1499 			return 1;
1500 		}
1501 		break;
1502 	case MSR_AMD64_NB_CFG:
1503 		break;
1504 	case MSR_IA32_DEBUGCTLMSR:
1505 		if (!data) {
1506 			/* We support the non-activated case already */
1507 			break;
1508 		} else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1509 			/* Values other than LBR and BTF are vendor-specific,
1510 			   thus reserved and should throw a #GP */
1511 			return 1;
1512 		}
1513 		pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1514 			__func__, data);
1515 		break;
1516 	case MSR_IA32_UCODE_REV:
1517 	case MSR_IA32_UCODE_WRITE:
1518 	case MSR_VM_HSAVE_PA:
1519 	case MSR_AMD64_PATCH_LOADER:
1520 		break;
1521 	case 0x200 ... 0x2ff:
1522 		return set_msr_mtrr(vcpu, msr, data);
1523 	case MSR_IA32_APICBASE:
1524 		kvm_set_apic_base(vcpu, data);
1525 		break;
1526 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1527 		return kvm_x2apic_msr_write(vcpu, msr, data);
1528 	case MSR_IA32_TSCDEADLINE:
1529 		kvm_set_lapic_tscdeadline_msr(vcpu, data);
1530 		break;
1531 	case MSR_IA32_MISC_ENABLE:
1532 		vcpu->arch.ia32_misc_enable_msr = data;
1533 		break;
1534 	case MSR_KVM_WALL_CLOCK_NEW:
1535 	case MSR_KVM_WALL_CLOCK:
1536 		vcpu->kvm->arch.wall_clock = data;
1537 		kvm_write_wall_clock(vcpu->kvm, data);
1538 		break;
1539 	case MSR_KVM_SYSTEM_TIME_NEW:
1540 	case MSR_KVM_SYSTEM_TIME: {
1541 		kvmclock_reset(vcpu);
1542 
1543 		vcpu->arch.time = data;
1544 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1545 
1546 		/* we verify if the enable bit is set... */
1547 		if (!(data & 1))
1548 			break;
1549 
1550 		/* ...but clean it before doing the actual write */
1551 		vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1552 
1553 		vcpu->arch.time_page =
1554 				gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1555 
1556 		if (is_error_page(vcpu->arch.time_page)) {
1557 			kvm_release_page_clean(vcpu->arch.time_page);
1558 			vcpu->arch.time_page = NULL;
1559 		}
1560 		break;
1561 	}
1562 	case MSR_KVM_ASYNC_PF_EN:
1563 		if (kvm_pv_enable_async_pf(vcpu, data))
1564 			return 1;
1565 		break;
1566 	case MSR_KVM_STEAL_TIME:
1567 
1568 		if (unlikely(!sched_info_on()))
1569 			return 1;
1570 
1571 		if (data & KVM_STEAL_RESERVED_MASK)
1572 			return 1;
1573 
1574 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1575 							data & KVM_STEAL_VALID_BITS))
1576 			return 1;
1577 
1578 		vcpu->arch.st.msr_val = data;
1579 
1580 		if (!(data & KVM_MSR_ENABLED))
1581 			break;
1582 
1583 		vcpu->arch.st.last_steal = current->sched_info.run_delay;
1584 
1585 		preempt_disable();
1586 		accumulate_steal_time(vcpu);
1587 		preempt_enable();
1588 
1589 		kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1590 
1591 		break;
1592 
1593 	case MSR_IA32_MCG_CTL:
1594 	case MSR_IA32_MCG_STATUS:
1595 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1596 		return set_msr_mce(vcpu, msr, data);
1597 
1598 	/* Performance counters are not protected by a CPUID bit,
1599 	 * so we should check all of them in the generic path for the sake of
1600 	 * cross vendor migration.
1601 	 * Writing a zero into the event select MSRs disables them,
1602 	 * which we perfectly emulate ;-). Any other value should be at least
1603 	 * reported, some guests depend on them.
1604 	 */
1605 	case MSR_K7_EVNTSEL0:
1606 	case MSR_K7_EVNTSEL1:
1607 	case MSR_K7_EVNTSEL2:
1608 	case MSR_K7_EVNTSEL3:
1609 		if (data != 0)
1610 			pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1611 				"0x%x data 0x%llx\n", msr, data);
1612 		break;
1613 	/* at least RHEL 4 unconditionally writes to the perfctr registers,
1614 	 * so we ignore writes to make it happy.
1615 	 */
1616 	case MSR_K7_PERFCTR0:
1617 	case MSR_K7_PERFCTR1:
1618 	case MSR_K7_PERFCTR2:
1619 	case MSR_K7_PERFCTR3:
1620 		pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1621 			"0x%x data 0x%llx\n", msr, data);
1622 		break;
1623 	case MSR_K7_CLK_CTL:
1624 		/*
1625 		 * Ignore all writes to this no longer documented MSR.
1626 		 * Writes are only relevant for old K7 processors,
1627 		 * all pre-dating SVM, but a recommended workaround from
1628 		 * AMD for these chips. It is possible to speicify the
1629 		 * affected processor models on the command line, hence
1630 		 * the need to ignore the workaround.
1631 		 */
1632 		break;
1633 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1634 		if (kvm_hv_msr_partition_wide(msr)) {
1635 			int r;
1636 			mutex_lock(&vcpu->kvm->lock);
1637 			r = set_msr_hyperv_pw(vcpu, msr, data);
1638 			mutex_unlock(&vcpu->kvm->lock);
1639 			return r;
1640 		} else
1641 			return set_msr_hyperv(vcpu, msr, data);
1642 		break;
1643 	case MSR_IA32_BBL_CR_CTL3:
1644 		/* Drop writes to this legacy MSR -- see rdmsr
1645 		 * counterpart for further detail.
1646 		 */
1647 		pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1648 		break;
1649 	default:
1650 		if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1651 			return xen_hvm_config(vcpu, data);
1652 		if (kvm_pmu_msr(vcpu, msr))
1653 			return kvm_pmu_set_msr(vcpu, msr, data);
1654 		if (!ignore_msrs) {
1655 			pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1656 				msr, data);
1657 			return 1;
1658 		} else {
1659 			pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1660 				msr, data);
1661 			break;
1662 		}
1663 	}
1664 	return 0;
1665 }
1666 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1667 
1668 
1669 /*
1670  * Reads an msr value (of 'msr_index') into 'pdata'.
1671  * Returns 0 on success, non-0 otherwise.
1672  * Assumes vcpu_load() was already called.
1673  */
1674 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1675 {
1676 	return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1677 }
1678 
1679 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1680 {
1681 	u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1682 
1683 	if (!msr_mtrr_valid(msr))
1684 		return 1;
1685 
1686 	if (msr == MSR_MTRRdefType)
1687 		*pdata = vcpu->arch.mtrr_state.def_type +
1688 			 (vcpu->arch.mtrr_state.enabled << 10);
1689 	else if (msr == MSR_MTRRfix64K_00000)
1690 		*pdata = p[0];
1691 	else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1692 		*pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1693 	else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1694 		*pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1695 	else if (msr == MSR_IA32_CR_PAT)
1696 		*pdata = vcpu->arch.pat;
1697 	else {	/* Variable MTRRs */
1698 		int idx, is_mtrr_mask;
1699 		u64 *pt;
1700 
1701 		idx = (msr - 0x200) / 2;
1702 		is_mtrr_mask = msr - 0x200 - 2 * idx;
1703 		if (!is_mtrr_mask)
1704 			pt =
1705 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1706 		else
1707 			pt =
1708 			  (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1709 		*pdata = *pt;
1710 	}
1711 
1712 	return 0;
1713 }
1714 
1715 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1716 {
1717 	u64 data;
1718 	u64 mcg_cap = vcpu->arch.mcg_cap;
1719 	unsigned bank_num = mcg_cap & 0xff;
1720 
1721 	switch (msr) {
1722 	case MSR_IA32_P5_MC_ADDR:
1723 	case MSR_IA32_P5_MC_TYPE:
1724 		data = 0;
1725 		break;
1726 	case MSR_IA32_MCG_CAP:
1727 		data = vcpu->arch.mcg_cap;
1728 		break;
1729 	case MSR_IA32_MCG_CTL:
1730 		if (!(mcg_cap & MCG_CTL_P))
1731 			return 1;
1732 		data = vcpu->arch.mcg_ctl;
1733 		break;
1734 	case MSR_IA32_MCG_STATUS:
1735 		data = vcpu->arch.mcg_status;
1736 		break;
1737 	default:
1738 		if (msr >= MSR_IA32_MC0_CTL &&
1739 		    msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1740 			u32 offset = msr - MSR_IA32_MC0_CTL;
1741 			data = vcpu->arch.mce_banks[offset];
1742 			break;
1743 		}
1744 		return 1;
1745 	}
1746 	*pdata = data;
1747 	return 0;
1748 }
1749 
1750 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1751 {
1752 	u64 data = 0;
1753 	struct kvm *kvm = vcpu->kvm;
1754 
1755 	switch (msr) {
1756 	case HV_X64_MSR_GUEST_OS_ID:
1757 		data = kvm->arch.hv_guest_os_id;
1758 		break;
1759 	case HV_X64_MSR_HYPERCALL:
1760 		data = kvm->arch.hv_hypercall;
1761 		break;
1762 	default:
1763 		pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1764 		return 1;
1765 	}
1766 
1767 	*pdata = data;
1768 	return 0;
1769 }
1770 
1771 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1772 {
1773 	u64 data = 0;
1774 
1775 	switch (msr) {
1776 	case HV_X64_MSR_VP_INDEX: {
1777 		int r;
1778 		struct kvm_vcpu *v;
1779 		kvm_for_each_vcpu(r, v, vcpu->kvm)
1780 			if (v == vcpu)
1781 				data = r;
1782 		break;
1783 	}
1784 	case HV_X64_MSR_EOI:
1785 		return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1786 	case HV_X64_MSR_ICR:
1787 		return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1788 	case HV_X64_MSR_TPR:
1789 		return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1790 	case HV_X64_MSR_APIC_ASSIST_PAGE:
1791 		data = vcpu->arch.hv_vapic;
1792 		break;
1793 	default:
1794 		pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1795 		return 1;
1796 	}
1797 	*pdata = data;
1798 	return 0;
1799 }
1800 
1801 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1802 {
1803 	u64 data;
1804 
1805 	switch (msr) {
1806 	case MSR_IA32_PLATFORM_ID:
1807 	case MSR_IA32_EBL_CR_POWERON:
1808 	case MSR_IA32_DEBUGCTLMSR:
1809 	case MSR_IA32_LASTBRANCHFROMIP:
1810 	case MSR_IA32_LASTBRANCHTOIP:
1811 	case MSR_IA32_LASTINTFROMIP:
1812 	case MSR_IA32_LASTINTTOIP:
1813 	case MSR_K8_SYSCFG:
1814 	case MSR_K7_HWCR:
1815 	case MSR_VM_HSAVE_PA:
1816 	case MSR_K7_EVNTSEL0:
1817 	case MSR_K7_PERFCTR0:
1818 	case MSR_K8_INT_PENDING_MSG:
1819 	case MSR_AMD64_NB_CFG:
1820 	case MSR_FAM10H_MMIO_CONF_BASE:
1821 		data = 0;
1822 		break;
1823 	case MSR_IA32_UCODE_REV:
1824 		data = 0x100000000ULL;
1825 		break;
1826 	case MSR_MTRRcap:
1827 		data = 0x500 | KVM_NR_VAR_MTRR;
1828 		break;
1829 	case 0x200 ... 0x2ff:
1830 		return get_msr_mtrr(vcpu, msr, pdata);
1831 	case 0xcd: /* fsb frequency */
1832 		data = 3;
1833 		break;
1834 		/*
1835 		 * MSR_EBC_FREQUENCY_ID
1836 		 * Conservative value valid for even the basic CPU models.
1837 		 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1838 		 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1839 		 * and 266MHz for model 3, or 4. Set Core Clock
1840 		 * Frequency to System Bus Frequency Ratio to 1 (bits
1841 		 * 31:24) even though these are only valid for CPU
1842 		 * models > 2, however guests may end up dividing or
1843 		 * multiplying by zero otherwise.
1844 		 */
1845 	case MSR_EBC_FREQUENCY_ID:
1846 		data = 1 << 24;
1847 		break;
1848 	case MSR_IA32_APICBASE:
1849 		data = kvm_get_apic_base(vcpu);
1850 		break;
1851 	case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1852 		return kvm_x2apic_msr_read(vcpu, msr, pdata);
1853 		break;
1854 	case MSR_IA32_TSCDEADLINE:
1855 		data = kvm_get_lapic_tscdeadline_msr(vcpu);
1856 		break;
1857 	case MSR_IA32_MISC_ENABLE:
1858 		data = vcpu->arch.ia32_misc_enable_msr;
1859 		break;
1860 	case MSR_IA32_PERF_STATUS:
1861 		/* TSC increment by tick */
1862 		data = 1000ULL;
1863 		/* CPU multiplier */
1864 		data |= (((uint64_t)4ULL) << 40);
1865 		break;
1866 	case MSR_EFER:
1867 		data = vcpu->arch.efer;
1868 		break;
1869 	case MSR_KVM_WALL_CLOCK:
1870 	case MSR_KVM_WALL_CLOCK_NEW:
1871 		data = vcpu->kvm->arch.wall_clock;
1872 		break;
1873 	case MSR_KVM_SYSTEM_TIME:
1874 	case MSR_KVM_SYSTEM_TIME_NEW:
1875 		data = vcpu->arch.time;
1876 		break;
1877 	case MSR_KVM_ASYNC_PF_EN:
1878 		data = vcpu->arch.apf.msr_val;
1879 		break;
1880 	case MSR_KVM_STEAL_TIME:
1881 		data = vcpu->arch.st.msr_val;
1882 		break;
1883 	case MSR_IA32_P5_MC_ADDR:
1884 	case MSR_IA32_P5_MC_TYPE:
1885 	case MSR_IA32_MCG_CAP:
1886 	case MSR_IA32_MCG_CTL:
1887 	case MSR_IA32_MCG_STATUS:
1888 	case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1889 		return get_msr_mce(vcpu, msr, pdata);
1890 	case MSR_K7_CLK_CTL:
1891 		/*
1892 		 * Provide expected ramp-up count for K7. All other
1893 		 * are set to zero, indicating minimum divisors for
1894 		 * every field.
1895 		 *
1896 		 * This prevents guest kernels on AMD host with CPU
1897 		 * type 6, model 8 and higher from exploding due to
1898 		 * the rdmsr failing.
1899 		 */
1900 		data = 0x20000000;
1901 		break;
1902 	case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1903 		if (kvm_hv_msr_partition_wide(msr)) {
1904 			int r;
1905 			mutex_lock(&vcpu->kvm->lock);
1906 			r = get_msr_hyperv_pw(vcpu, msr, pdata);
1907 			mutex_unlock(&vcpu->kvm->lock);
1908 			return r;
1909 		} else
1910 			return get_msr_hyperv(vcpu, msr, pdata);
1911 		break;
1912 	case MSR_IA32_BBL_CR_CTL3:
1913 		/* This legacy MSR exists but isn't fully documented in current
1914 		 * silicon.  It is however accessed by winxp in very narrow
1915 		 * scenarios where it sets bit #19, itself documented as
1916 		 * a "reserved" bit.  Best effort attempt to source coherent
1917 		 * read data here should the balance of the register be
1918 		 * interpreted by the guest:
1919 		 *
1920 		 * L2 cache control register 3: 64GB range, 256KB size,
1921 		 * enabled, latency 0x1, configured
1922 		 */
1923 		data = 0xbe702111;
1924 		break;
1925 	default:
1926 		if (kvm_pmu_msr(vcpu, msr))
1927 			return kvm_pmu_get_msr(vcpu, msr, pdata);
1928 		if (!ignore_msrs) {
1929 			pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1930 			return 1;
1931 		} else {
1932 			pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1933 			data = 0;
1934 		}
1935 		break;
1936 	}
1937 	*pdata = data;
1938 	return 0;
1939 }
1940 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1941 
1942 /*
1943  * Read or write a bunch of msrs. All parameters are kernel addresses.
1944  *
1945  * @return number of msrs set successfully.
1946  */
1947 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1948 		    struct kvm_msr_entry *entries,
1949 		    int (*do_msr)(struct kvm_vcpu *vcpu,
1950 				  unsigned index, u64 *data))
1951 {
1952 	int i, idx;
1953 
1954 	idx = srcu_read_lock(&vcpu->kvm->srcu);
1955 	for (i = 0; i < msrs->nmsrs; ++i)
1956 		if (do_msr(vcpu, entries[i].index, &entries[i].data))
1957 			break;
1958 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
1959 
1960 	return i;
1961 }
1962 
1963 /*
1964  * Read or write a bunch of msrs. Parameters are user addresses.
1965  *
1966  * @return number of msrs set successfully.
1967  */
1968 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1969 		  int (*do_msr)(struct kvm_vcpu *vcpu,
1970 				unsigned index, u64 *data),
1971 		  int writeback)
1972 {
1973 	struct kvm_msrs msrs;
1974 	struct kvm_msr_entry *entries;
1975 	int r, n;
1976 	unsigned size;
1977 
1978 	r = -EFAULT;
1979 	if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1980 		goto out;
1981 
1982 	r = -E2BIG;
1983 	if (msrs.nmsrs >= MAX_IO_MSRS)
1984 		goto out;
1985 
1986 	size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1987 	entries = memdup_user(user_msrs->entries, size);
1988 	if (IS_ERR(entries)) {
1989 		r = PTR_ERR(entries);
1990 		goto out;
1991 	}
1992 
1993 	r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1994 	if (r < 0)
1995 		goto out_free;
1996 
1997 	r = -EFAULT;
1998 	if (writeback && copy_to_user(user_msrs->entries, entries, size))
1999 		goto out_free;
2000 
2001 	r = n;
2002 
2003 out_free:
2004 	kfree(entries);
2005 out:
2006 	return r;
2007 }
2008 
2009 int kvm_dev_ioctl_check_extension(long ext)
2010 {
2011 	int r;
2012 
2013 	switch (ext) {
2014 	case KVM_CAP_IRQCHIP:
2015 	case KVM_CAP_HLT:
2016 	case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2017 	case KVM_CAP_SET_TSS_ADDR:
2018 	case KVM_CAP_EXT_CPUID:
2019 	case KVM_CAP_CLOCKSOURCE:
2020 	case KVM_CAP_PIT:
2021 	case KVM_CAP_NOP_IO_DELAY:
2022 	case KVM_CAP_MP_STATE:
2023 	case KVM_CAP_SYNC_MMU:
2024 	case KVM_CAP_USER_NMI:
2025 	case KVM_CAP_REINJECT_CONTROL:
2026 	case KVM_CAP_IRQ_INJECT_STATUS:
2027 	case KVM_CAP_ASSIGN_DEV_IRQ:
2028 	case KVM_CAP_IRQFD:
2029 	case KVM_CAP_IOEVENTFD:
2030 	case KVM_CAP_PIT2:
2031 	case KVM_CAP_PIT_STATE2:
2032 	case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2033 	case KVM_CAP_XEN_HVM:
2034 	case KVM_CAP_ADJUST_CLOCK:
2035 	case KVM_CAP_VCPU_EVENTS:
2036 	case KVM_CAP_HYPERV:
2037 	case KVM_CAP_HYPERV_VAPIC:
2038 	case KVM_CAP_HYPERV_SPIN:
2039 	case KVM_CAP_PCI_SEGMENT:
2040 	case KVM_CAP_DEBUGREGS:
2041 	case KVM_CAP_X86_ROBUST_SINGLESTEP:
2042 	case KVM_CAP_XSAVE:
2043 	case KVM_CAP_ASYNC_PF:
2044 	case KVM_CAP_GET_TSC_KHZ:
2045 		r = 1;
2046 		break;
2047 	case KVM_CAP_COALESCED_MMIO:
2048 		r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2049 		break;
2050 	case KVM_CAP_VAPIC:
2051 		r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2052 		break;
2053 	case KVM_CAP_NR_VCPUS:
2054 		r = KVM_SOFT_MAX_VCPUS;
2055 		break;
2056 	case KVM_CAP_MAX_VCPUS:
2057 		r = KVM_MAX_VCPUS;
2058 		break;
2059 	case KVM_CAP_NR_MEMSLOTS:
2060 		r = KVM_MEMORY_SLOTS;
2061 		break;
2062 	case KVM_CAP_PV_MMU:	/* obsolete */
2063 		r = 0;
2064 		break;
2065 	case KVM_CAP_IOMMU:
2066 		r = iommu_present(&pci_bus_type);
2067 		break;
2068 	case KVM_CAP_MCE:
2069 		r = KVM_MAX_MCE_BANKS;
2070 		break;
2071 	case KVM_CAP_XCRS:
2072 		r = cpu_has_xsave;
2073 		break;
2074 	case KVM_CAP_TSC_CONTROL:
2075 		r = kvm_has_tsc_control;
2076 		break;
2077 	case KVM_CAP_TSC_DEADLINE_TIMER:
2078 		r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2079 		break;
2080 	default:
2081 		r = 0;
2082 		break;
2083 	}
2084 	return r;
2085 
2086 }
2087 
2088 long kvm_arch_dev_ioctl(struct file *filp,
2089 			unsigned int ioctl, unsigned long arg)
2090 {
2091 	void __user *argp = (void __user *)arg;
2092 	long r;
2093 
2094 	switch (ioctl) {
2095 	case KVM_GET_MSR_INDEX_LIST: {
2096 		struct kvm_msr_list __user *user_msr_list = argp;
2097 		struct kvm_msr_list msr_list;
2098 		unsigned n;
2099 
2100 		r = -EFAULT;
2101 		if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2102 			goto out;
2103 		n = msr_list.nmsrs;
2104 		msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2105 		if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2106 			goto out;
2107 		r = -E2BIG;
2108 		if (n < msr_list.nmsrs)
2109 			goto out;
2110 		r = -EFAULT;
2111 		if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2112 				 num_msrs_to_save * sizeof(u32)))
2113 			goto out;
2114 		if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2115 				 &emulated_msrs,
2116 				 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2117 			goto out;
2118 		r = 0;
2119 		break;
2120 	}
2121 	case KVM_GET_SUPPORTED_CPUID: {
2122 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2123 		struct kvm_cpuid2 cpuid;
2124 
2125 		r = -EFAULT;
2126 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2127 			goto out;
2128 		r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2129 						      cpuid_arg->entries);
2130 		if (r)
2131 			goto out;
2132 
2133 		r = -EFAULT;
2134 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2135 			goto out;
2136 		r = 0;
2137 		break;
2138 	}
2139 	case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2140 		u64 mce_cap;
2141 
2142 		mce_cap = KVM_MCE_CAP_SUPPORTED;
2143 		r = -EFAULT;
2144 		if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2145 			goto out;
2146 		r = 0;
2147 		break;
2148 	}
2149 	default:
2150 		r = -EINVAL;
2151 	}
2152 out:
2153 	return r;
2154 }
2155 
2156 static void wbinvd_ipi(void *garbage)
2157 {
2158 	wbinvd();
2159 }
2160 
2161 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2162 {
2163 	return vcpu->kvm->arch.iommu_domain &&
2164 		!(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2165 }
2166 
2167 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2168 {
2169 	/* Address WBINVD may be executed by guest */
2170 	if (need_emulate_wbinvd(vcpu)) {
2171 		if (kvm_x86_ops->has_wbinvd_exit())
2172 			cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2173 		else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2174 			smp_call_function_single(vcpu->cpu,
2175 					wbinvd_ipi, NULL, 1);
2176 	}
2177 
2178 	kvm_x86_ops->vcpu_load(vcpu, cpu);
2179 	if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2180 		/* Make sure TSC doesn't go backwards */
2181 		s64 tsc_delta;
2182 		u64 tsc;
2183 
2184 		tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2185 		tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2186 			     tsc - vcpu->arch.last_guest_tsc;
2187 
2188 		if (tsc_delta < 0)
2189 			mark_tsc_unstable("KVM discovered backwards TSC");
2190 		if (check_tsc_unstable()) {
2191 			kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2192 			vcpu->arch.tsc_catchup = 1;
2193 		}
2194 		kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2195 		if (vcpu->cpu != cpu)
2196 			kvm_migrate_timers(vcpu);
2197 		vcpu->cpu = cpu;
2198 	}
2199 
2200 	accumulate_steal_time(vcpu);
2201 	kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2202 }
2203 
2204 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2205 {
2206 	kvm_x86_ops->vcpu_put(vcpu);
2207 	kvm_put_guest_fpu(vcpu);
2208 	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2209 }
2210 
2211 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2212 				    struct kvm_lapic_state *s)
2213 {
2214 	memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2215 
2216 	return 0;
2217 }
2218 
2219 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2220 				    struct kvm_lapic_state *s)
2221 {
2222 	memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2223 	kvm_apic_post_state_restore(vcpu);
2224 	update_cr8_intercept(vcpu);
2225 
2226 	return 0;
2227 }
2228 
2229 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2230 				    struct kvm_interrupt *irq)
2231 {
2232 	if (irq->irq < 0 || irq->irq >= 256)
2233 		return -EINVAL;
2234 	if (irqchip_in_kernel(vcpu->kvm))
2235 		return -ENXIO;
2236 
2237 	kvm_queue_interrupt(vcpu, irq->irq, false);
2238 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2239 
2240 	return 0;
2241 }
2242 
2243 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2244 {
2245 	kvm_inject_nmi(vcpu);
2246 
2247 	return 0;
2248 }
2249 
2250 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2251 					   struct kvm_tpr_access_ctl *tac)
2252 {
2253 	if (tac->flags)
2254 		return -EINVAL;
2255 	vcpu->arch.tpr_access_reporting = !!tac->enabled;
2256 	return 0;
2257 }
2258 
2259 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2260 					u64 mcg_cap)
2261 {
2262 	int r;
2263 	unsigned bank_num = mcg_cap & 0xff, bank;
2264 
2265 	r = -EINVAL;
2266 	if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2267 		goto out;
2268 	if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2269 		goto out;
2270 	r = 0;
2271 	vcpu->arch.mcg_cap = mcg_cap;
2272 	/* Init IA32_MCG_CTL to all 1s */
2273 	if (mcg_cap & MCG_CTL_P)
2274 		vcpu->arch.mcg_ctl = ~(u64)0;
2275 	/* Init IA32_MCi_CTL to all 1s */
2276 	for (bank = 0; bank < bank_num; bank++)
2277 		vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2278 out:
2279 	return r;
2280 }
2281 
2282 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2283 				      struct kvm_x86_mce *mce)
2284 {
2285 	u64 mcg_cap = vcpu->arch.mcg_cap;
2286 	unsigned bank_num = mcg_cap & 0xff;
2287 	u64 *banks = vcpu->arch.mce_banks;
2288 
2289 	if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2290 		return -EINVAL;
2291 	/*
2292 	 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2293 	 * reporting is disabled
2294 	 */
2295 	if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2296 	    vcpu->arch.mcg_ctl != ~(u64)0)
2297 		return 0;
2298 	banks += 4 * mce->bank;
2299 	/*
2300 	 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2301 	 * reporting is disabled for the bank
2302 	 */
2303 	if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2304 		return 0;
2305 	if (mce->status & MCI_STATUS_UC) {
2306 		if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2307 		    !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2308 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2309 			return 0;
2310 		}
2311 		if (banks[1] & MCI_STATUS_VAL)
2312 			mce->status |= MCI_STATUS_OVER;
2313 		banks[2] = mce->addr;
2314 		banks[3] = mce->misc;
2315 		vcpu->arch.mcg_status = mce->mcg_status;
2316 		banks[1] = mce->status;
2317 		kvm_queue_exception(vcpu, MC_VECTOR);
2318 	} else if (!(banks[1] & MCI_STATUS_VAL)
2319 		   || !(banks[1] & MCI_STATUS_UC)) {
2320 		if (banks[1] & MCI_STATUS_VAL)
2321 			mce->status |= MCI_STATUS_OVER;
2322 		banks[2] = mce->addr;
2323 		banks[3] = mce->misc;
2324 		banks[1] = mce->status;
2325 	} else
2326 		banks[1] |= MCI_STATUS_OVER;
2327 	return 0;
2328 }
2329 
2330 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2331 					       struct kvm_vcpu_events *events)
2332 {
2333 	process_nmi(vcpu);
2334 	events->exception.injected =
2335 		vcpu->arch.exception.pending &&
2336 		!kvm_exception_is_soft(vcpu->arch.exception.nr);
2337 	events->exception.nr = vcpu->arch.exception.nr;
2338 	events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2339 	events->exception.pad = 0;
2340 	events->exception.error_code = vcpu->arch.exception.error_code;
2341 
2342 	events->interrupt.injected =
2343 		vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2344 	events->interrupt.nr = vcpu->arch.interrupt.nr;
2345 	events->interrupt.soft = 0;
2346 	events->interrupt.shadow =
2347 		kvm_x86_ops->get_interrupt_shadow(vcpu,
2348 			KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2349 
2350 	events->nmi.injected = vcpu->arch.nmi_injected;
2351 	events->nmi.pending = vcpu->arch.nmi_pending != 0;
2352 	events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2353 	events->nmi.pad = 0;
2354 
2355 	events->sipi_vector = vcpu->arch.sipi_vector;
2356 
2357 	events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2358 			 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2359 			 | KVM_VCPUEVENT_VALID_SHADOW);
2360 	memset(&events->reserved, 0, sizeof(events->reserved));
2361 }
2362 
2363 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2364 					      struct kvm_vcpu_events *events)
2365 {
2366 	if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2367 			      | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2368 			      | KVM_VCPUEVENT_VALID_SHADOW))
2369 		return -EINVAL;
2370 
2371 	process_nmi(vcpu);
2372 	vcpu->arch.exception.pending = events->exception.injected;
2373 	vcpu->arch.exception.nr = events->exception.nr;
2374 	vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2375 	vcpu->arch.exception.error_code = events->exception.error_code;
2376 
2377 	vcpu->arch.interrupt.pending = events->interrupt.injected;
2378 	vcpu->arch.interrupt.nr = events->interrupt.nr;
2379 	vcpu->arch.interrupt.soft = events->interrupt.soft;
2380 	if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2381 		kvm_x86_ops->set_interrupt_shadow(vcpu,
2382 						  events->interrupt.shadow);
2383 
2384 	vcpu->arch.nmi_injected = events->nmi.injected;
2385 	if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2386 		vcpu->arch.nmi_pending = events->nmi.pending;
2387 	kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2388 
2389 	if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2390 		vcpu->arch.sipi_vector = events->sipi_vector;
2391 
2392 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2393 
2394 	return 0;
2395 }
2396 
2397 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2398 					     struct kvm_debugregs *dbgregs)
2399 {
2400 	memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2401 	dbgregs->dr6 = vcpu->arch.dr6;
2402 	dbgregs->dr7 = vcpu->arch.dr7;
2403 	dbgregs->flags = 0;
2404 	memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2405 }
2406 
2407 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2408 					    struct kvm_debugregs *dbgregs)
2409 {
2410 	if (dbgregs->flags)
2411 		return -EINVAL;
2412 
2413 	memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2414 	vcpu->arch.dr6 = dbgregs->dr6;
2415 	vcpu->arch.dr7 = dbgregs->dr7;
2416 
2417 	return 0;
2418 }
2419 
2420 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2421 					 struct kvm_xsave *guest_xsave)
2422 {
2423 	if (cpu_has_xsave)
2424 		memcpy(guest_xsave->region,
2425 			&vcpu->arch.guest_fpu.state->xsave,
2426 			xstate_size);
2427 	else {
2428 		memcpy(guest_xsave->region,
2429 			&vcpu->arch.guest_fpu.state->fxsave,
2430 			sizeof(struct i387_fxsave_struct));
2431 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2432 			XSTATE_FPSSE;
2433 	}
2434 }
2435 
2436 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2437 					struct kvm_xsave *guest_xsave)
2438 {
2439 	u64 xstate_bv =
2440 		*(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2441 
2442 	if (cpu_has_xsave)
2443 		memcpy(&vcpu->arch.guest_fpu.state->xsave,
2444 			guest_xsave->region, xstate_size);
2445 	else {
2446 		if (xstate_bv & ~XSTATE_FPSSE)
2447 			return -EINVAL;
2448 		memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2449 			guest_xsave->region, sizeof(struct i387_fxsave_struct));
2450 	}
2451 	return 0;
2452 }
2453 
2454 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2455 					struct kvm_xcrs *guest_xcrs)
2456 {
2457 	if (!cpu_has_xsave) {
2458 		guest_xcrs->nr_xcrs = 0;
2459 		return;
2460 	}
2461 
2462 	guest_xcrs->nr_xcrs = 1;
2463 	guest_xcrs->flags = 0;
2464 	guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2465 	guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2466 }
2467 
2468 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2469 				       struct kvm_xcrs *guest_xcrs)
2470 {
2471 	int i, r = 0;
2472 
2473 	if (!cpu_has_xsave)
2474 		return -EINVAL;
2475 
2476 	if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2477 		return -EINVAL;
2478 
2479 	for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2480 		/* Only support XCR0 currently */
2481 		if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2482 			r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2483 				guest_xcrs->xcrs[0].value);
2484 			break;
2485 		}
2486 	if (r)
2487 		r = -EINVAL;
2488 	return r;
2489 }
2490 
2491 long kvm_arch_vcpu_ioctl(struct file *filp,
2492 			 unsigned int ioctl, unsigned long arg)
2493 {
2494 	struct kvm_vcpu *vcpu = filp->private_data;
2495 	void __user *argp = (void __user *)arg;
2496 	int r;
2497 	union {
2498 		struct kvm_lapic_state *lapic;
2499 		struct kvm_xsave *xsave;
2500 		struct kvm_xcrs *xcrs;
2501 		void *buffer;
2502 	} u;
2503 
2504 	u.buffer = NULL;
2505 	switch (ioctl) {
2506 	case KVM_GET_LAPIC: {
2507 		r = -EINVAL;
2508 		if (!vcpu->arch.apic)
2509 			goto out;
2510 		u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2511 
2512 		r = -ENOMEM;
2513 		if (!u.lapic)
2514 			goto out;
2515 		r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2516 		if (r)
2517 			goto out;
2518 		r = -EFAULT;
2519 		if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2520 			goto out;
2521 		r = 0;
2522 		break;
2523 	}
2524 	case KVM_SET_LAPIC: {
2525 		r = -EINVAL;
2526 		if (!vcpu->arch.apic)
2527 			goto out;
2528 		u.lapic = memdup_user(argp, sizeof(*u.lapic));
2529 		if (IS_ERR(u.lapic)) {
2530 			r = PTR_ERR(u.lapic);
2531 			goto out;
2532 		}
2533 
2534 		r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2535 		if (r)
2536 			goto out;
2537 		r = 0;
2538 		break;
2539 	}
2540 	case KVM_INTERRUPT: {
2541 		struct kvm_interrupt irq;
2542 
2543 		r = -EFAULT;
2544 		if (copy_from_user(&irq, argp, sizeof irq))
2545 			goto out;
2546 		r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2547 		if (r)
2548 			goto out;
2549 		r = 0;
2550 		break;
2551 	}
2552 	case KVM_NMI: {
2553 		r = kvm_vcpu_ioctl_nmi(vcpu);
2554 		if (r)
2555 			goto out;
2556 		r = 0;
2557 		break;
2558 	}
2559 	case KVM_SET_CPUID: {
2560 		struct kvm_cpuid __user *cpuid_arg = argp;
2561 		struct kvm_cpuid cpuid;
2562 
2563 		r = -EFAULT;
2564 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2565 			goto out;
2566 		r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2567 		if (r)
2568 			goto out;
2569 		break;
2570 	}
2571 	case KVM_SET_CPUID2: {
2572 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2573 		struct kvm_cpuid2 cpuid;
2574 
2575 		r = -EFAULT;
2576 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2577 			goto out;
2578 		r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2579 					      cpuid_arg->entries);
2580 		if (r)
2581 			goto out;
2582 		break;
2583 	}
2584 	case KVM_GET_CPUID2: {
2585 		struct kvm_cpuid2 __user *cpuid_arg = argp;
2586 		struct kvm_cpuid2 cpuid;
2587 
2588 		r = -EFAULT;
2589 		if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2590 			goto out;
2591 		r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2592 					      cpuid_arg->entries);
2593 		if (r)
2594 			goto out;
2595 		r = -EFAULT;
2596 		if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2597 			goto out;
2598 		r = 0;
2599 		break;
2600 	}
2601 	case KVM_GET_MSRS:
2602 		r = msr_io(vcpu, argp, kvm_get_msr, 1);
2603 		break;
2604 	case KVM_SET_MSRS:
2605 		r = msr_io(vcpu, argp, do_set_msr, 0);
2606 		break;
2607 	case KVM_TPR_ACCESS_REPORTING: {
2608 		struct kvm_tpr_access_ctl tac;
2609 
2610 		r = -EFAULT;
2611 		if (copy_from_user(&tac, argp, sizeof tac))
2612 			goto out;
2613 		r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2614 		if (r)
2615 			goto out;
2616 		r = -EFAULT;
2617 		if (copy_to_user(argp, &tac, sizeof tac))
2618 			goto out;
2619 		r = 0;
2620 		break;
2621 	};
2622 	case KVM_SET_VAPIC_ADDR: {
2623 		struct kvm_vapic_addr va;
2624 
2625 		r = -EINVAL;
2626 		if (!irqchip_in_kernel(vcpu->kvm))
2627 			goto out;
2628 		r = -EFAULT;
2629 		if (copy_from_user(&va, argp, sizeof va))
2630 			goto out;
2631 		r = 0;
2632 		kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2633 		break;
2634 	}
2635 	case KVM_X86_SETUP_MCE: {
2636 		u64 mcg_cap;
2637 
2638 		r = -EFAULT;
2639 		if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2640 			goto out;
2641 		r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2642 		break;
2643 	}
2644 	case KVM_X86_SET_MCE: {
2645 		struct kvm_x86_mce mce;
2646 
2647 		r = -EFAULT;
2648 		if (copy_from_user(&mce, argp, sizeof mce))
2649 			goto out;
2650 		r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2651 		break;
2652 	}
2653 	case KVM_GET_VCPU_EVENTS: {
2654 		struct kvm_vcpu_events events;
2655 
2656 		kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2657 
2658 		r = -EFAULT;
2659 		if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2660 			break;
2661 		r = 0;
2662 		break;
2663 	}
2664 	case KVM_SET_VCPU_EVENTS: {
2665 		struct kvm_vcpu_events events;
2666 
2667 		r = -EFAULT;
2668 		if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2669 			break;
2670 
2671 		r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2672 		break;
2673 	}
2674 	case KVM_GET_DEBUGREGS: {
2675 		struct kvm_debugregs dbgregs;
2676 
2677 		kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2678 
2679 		r = -EFAULT;
2680 		if (copy_to_user(argp, &dbgregs,
2681 				 sizeof(struct kvm_debugregs)))
2682 			break;
2683 		r = 0;
2684 		break;
2685 	}
2686 	case KVM_SET_DEBUGREGS: {
2687 		struct kvm_debugregs dbgregs;
2688 
2689 		r = -EFAULT;
2690 		if (copy_from_user(&dbgregs, argp,
2691 				   sizeof(struct kvm_debugregs)))
2692 			break;
2693 
2694 		r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2695 		break;
2696 	}
2697 	case KVM_GET_XSAVE: {
2698 		u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2699 		r = -ENOMEM;
2700 		if (!u.xsave)
2701 			break;
2702 
2703 		kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2704 
2705 		r = -EFAULT;
2706 		if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2707 			break;
2708 		r = 0;
2709 		break;
2710 	}
2711 	case KVM_SET_XSAVE: {
2712 		u.xsave = memdup_user(argp, sizeof(*u.xsave));
2713 		if (IS_ERR(u.xsave)) {
2714 			r = PTR_ERR(u.xsave);
2715 			goto out;
2716 		}
2717 
2718 		r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2719 		break;
2720 	}
2721 	case KVM_GET_XCRS: {
2722 		u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2723 		r = -ENOMEM;
2724 		if (!u.xcrs)
2725 			break;
2726 
2727 		kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2728 
2729 		r = -EFAULT;
2730 		if (copy_to_user(argp, u.xcrs,
2731 				 sizeof(struct kvm_xcrs)))
2732 			break;
2733 		r = 0;
2734 		break;
2735 	}
2736 	case KVM_SET_XCRS: {
2737 		u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
2738 		if (IS_ERR(u.xcrs)) {
2739 			r = PTR_ERR(u.xcrs);
2740 			goto out;
2741 		}
2742 
2743 		r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2744 		break;
2745 	}
2746 	case KVM_SET_TSC_KHZ: {
2747 		u32 user_tsc_khz;
2748 
2749 		r = -EINVAL;
2750 		if (!kvm_has_tsc_control)
2751 			break;
2752 
2753 		user_tsc_khz = (u32)arg;
2754 
2755 		if (user_tsc_khz >= kvm_max_guest_tsc_khz)
2756 			goto out;
2757 
2758 		kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
2759 
2760 		r = 0;
2761 		goto out;
2762 	}
2763 	case KVM_GET_TSC_KHZ: {
2764 		r = -EIO;
2765 		if (check_tsc_unstable())
2766 			goto out;
2767 
2768 		r = vcpu_tsc_khz(vcpu);
2769 
2770 		goto out;
2771 	}
2772 	default:
2773 		r = -EINVAL;
2774 	}
2775 out:
2776 	kfree(u.buffer);
2777 	return r;
2778 }
2779 
2780 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2781 {
2782 	int ret;
2783 
2784 	if (addr > (unsigned int)(-3 * PAGE_SIZE))
2785 		return -1;
2786 	ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2787 	return ret;
2788 }
2789 
2790 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2791 					      u64 ident_addr)
2792 {
2793 	kvm->arch.ept_identity_map_addr = ident_addr;
2794 	return 0;
2795 }
2796 
2797 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2798 					  u32 kvm_nr_mmu_pages)
2799 {
2800 	if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2801 		return -EINVAL;
2802 
2803 	mutex_lock(&kvm->slots_lock);
2804 	spin_lock(&kvm->mmu_lock);
2805 
2806 	kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2807 	kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2808 
2809 	spin_unlock(&kvm->mmu_lock);
2810 	mutex_unlock(&kvm->slots_lock);
2811 	return 0;
2812 }
2813 
2814 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2815 {
2816 	return kvm->arch.n_max_mmu_pages;
2817 }
2818 
2819 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2820 {
2821 	int r;
2822 
2823 	r = 0;
2824 	switch (chip->chip_id) {
2825 	case KVM_IRQCHIP_PIC_MASTER:
2826 		memcpy(&chip->chip.pic,
2827 			&pic_irqchip(kvm)->pics[0],
2828 			sizeof(struct kvm_pic_state));
2829 		break;
2830 	case KVM_IRQCHIP_PIC_SLAVE:
2831 		memcpy(&chip->chip.pic,
2832 			&pic_irqchip(kvm)->pics[1],
2833 			sizeof(struct kvm_pic_state));
2834 		break;
2835 	case KVM_IRQCHIP_IOAPIC:
2836 		r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2837 		break;
2838 	default:
2839 		r = -EINVAL;
2840 		break;
2841 	}
2842 	return r;
2843 }
2844 
2845 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2846 {
2847 	int r;
2848 
2849 	r = 0;
2850 	switch (chip->chip_id) {
2851 	case KVM_IRQCHIP_PIC_MASTER:
2852 		spin_lock(&pic_irqchip(kvm)->lock);
2853 		memcpy(&pic_irqchip(kvm)->pics[0],
2854 			&chip->chip.pic,
2855 			sizeof(struct kvm_pic_state));
2856 		spin_unlock(&pic_irqchip(kvm)->lock);
2857 		break;
2858 	case KVM_IRQCHIP_PIC_SLAVE:
2859 		spin_lock(&pic_irqchip(kvm)->lock);
2860 		memcpy(&pic_irqchip(kvm)->pics[1],
2861 			&chip->chip.pic,
2862 			sizeof(struct kvm_pic_state));
2863 		spin_unlock(&pic_irqchip(kvm)->lock);
2864 		break;
2865 	case KVM_IRQCHIP_IOAPIC:
2866 		r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2867 		break;
2868 	default:
2869 		r = -EINVAL;
2870 		break;
2871 	}
2872 	kvm_pic_update_irq(pic_irqchip(kvm));
2873 	return r;
2874 }
2875 
2876 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2877 {
2878 	int r = 0;
2879 
2880 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2881 	memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2882 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2883 	return r;
2884 }
2885 
2886 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2887 {
2888 	int r = 0;
2889 
2890 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2891 	memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2892 	kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2893 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2894 	return r;
2895 }
2896 
2897 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2898 {
2899 	int r = 0;
2900 
2901 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2902 	memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2903 		sizeof(ps->channels));
2904 	ps->flags = kvm->arch.vpit->pit_state.flags;
2905 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2906 	memset(&ps->reserved, 0, sizeof(ps->reserved));
2907 	return r;
2908 }
2909 
2910 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2911 {
2912 	int r = 0, start = 0;
2913 	u32 prev_legacy, cur_legacy;
2914 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2915 	prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2916 	cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2917 	if (!prev_legacy && cur_legacy)
2918 		start = 1;
2919 	memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2920 	       sizeof(kvm->arch.vpit->pit_state.channels));
2921 	kvm->arch.vpit->pit_state.flags = ps->flags;
2922 	kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2923 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2924 	return r;
2925 }
2926 
2927 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2928 				 struct kvm_reinject_control *control)
2929 {
2930 	if (!kvm->arch.vpit)
2931 		return -ENXIO;
2932 	mutex_lock(&kvm->arch.vpit->pit_state.lock);
2933 	kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2934 	mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2935 	return 0;
2936 }
2937 
2938 /**
2939  * write_protect_slot - write protect a slot for dirty logging
2940  * @kvm: the kvm instance
2941  * @memslot: the slot we protect
2942  * @dirty_bitmap: the bitmap indicating which pages are dirty
2943  * @nr_dirty_pages: the number of dirty pages
2944  *
2945  * We have two ways to find all sptes to protect:
2946  * 1. Use kvm_mmu_slot_remove_write_access() which walks all shadow pages and
2947  *    checks ones that have a spte mapping a page in the slot.
2948  * 2. Use kvm_mmu_rmap_write_protect() for each gfn found in the bitmap.
2949  *
2950  * Generally speaking, if there are not so many dirty pages compared to the
2951  * number of shadow pages, we should use the latter.
2952  *
2953  * Note that letting others write into a page marked dirty in the old bitmap
2954  * by using the remaining tlb entry is not a problem.  That page will become
2955  * write protected again when we flush the tlb and then be reported dirty to
2956  * the user space by copying the old bitmap.
2957  */
2958 static void write_protect_slot(struct kvm *kvm,
2959 			       struct kvm_memory_slot *memslot,
2960 			       unsigned long *dirty_bitmap,
2961 			       unsigned long nr_dirty_pages)
2962 {
2963 	/* Not many dirty pages compared to # of shadow pages. */
2964 	if (nr_dirty_pages < kvm->arch.n_used_mmu_pages) {
2965 		unsigned long gfn_offset;
2966 
2967 		for_each_set_bit(gfn_offset, dirty_bitmap, memslot->npages) {
2968 			unsigned long gfn = memslot->base_gfn + gfn_offset;
2969 
2970 			spin_lock(&kvm->mmu_lock);
2971 			kvm_mmu_rmap_write_protect(kvm, gfn, memslot);
2972 			spin_unlock(&kvm->mmu_lock);
2973 		}
2974 		kvm_flush_remote_tlbs(kvm);
2975 	} else {
2976 		spin_lock(&kvm->mmu_lock);
2977 		kvm_mmu_slot_remove_write_access(kvm, memslot->id);
2978 		spin_unlock(&kvm->mmu_lock);
2979 	}
2980 }
2981 
2982 /*
2983  * Get (and clear) the dirty memory log for a memory slot.
2984  */
2985 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2986 				      struct kvm_dirty_log *log)
2987 {
2988 	int r;
2989 	struct kvm_memory_slot *memslot;
2990 	unsigned long n, nr_dirty_pages;
2991 
2992 	mutex_lock(&kvm->slots_lock);
2993 
2994 	r = -EINVAL;
2995 	if (log->slot >= KVM_MEMORY_SLOTS)
2996 		goto out;
2997 
2998 	memslot = id_to_memslot(kvm->memslots, log->slot);
2999 	r = -ENOENT;
3000 	if (!memslot->dirty_bitmap)
3001 		goto out;
3002 
3003 	n = kvm_dirty_bitmap_bytes(memslot);
3004 	nr_dirty_pages = memslot->nr_dirty_pages;
3005 
3006 	/* If nothing is dirty, don't bother messing with page tables. */
3007 	if (nr_dirty_pages) {
3008 		struct kvm_memslots *slots, *old_slots;
3009 		unsigned long *dirty_bitmap, *dirty_bitmap_head;
3010 
3011 		dirty_bitmap = memslot->dirty_bitmap;
3012 		dirty_bitmap_head = memslot->dirty_bitmap_head;
3013 		if (dirty_bitmap == dirty_bitmap_head)
3014 			dirty_bitmap_head += n / sizeof(long);
3015 		memset(dirty_bitmap_head, 0, n);
3016 
3017 		r = -ENOMEM;
3018 		slots = kmemdup(kvm->memslots, sizeof(*kvm->memslots), GFP_KERNEL);
3019 		if (!slots)
3020 			goto out;
3021 
3022 		memslot = id_to_memslot(slots, log->slot);
3023 		memslot->nr_dirty_pages = 0;
3024 		memslot->dirty_bitmap = dirty_bitmap_head;
3025 		update_memslots(slots, NULL);
3026 
3027 		old_slots = kvm->memslots;
3028 		rcu_assign_pointer(kvm->memslots, slots);
3029 		synchronize_srcu_expedited(&kvm->srcu);
3030 		kfree(old_slots);
3031 
3032 		write_protect_slot(kvm, memslot, dirty_bitmap, nr_dirty_pages);
3033 
3034 		r = -EFAULT;
3035 		if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3036 			goto out;
3037 	} else {
3038 		r = -EFAULT;
3039 		if (clear_user(log->dirty_bitmap, n))
3040 			goto out;
3041 	}
3042 
3043 	r = 0;
3044 out:
3045 	mutex_unlock(&kvm->slots_lock);
3046 	return r;
3047 }
3048 
3049 long kvm_arch_vm_ioctl(struct file *filp,
3050 		       unsigned int ioctl, unsigned long arg)
3051 {
3052 	struct kvm *kvm = filp->private_data;
3053 	void __user *argp = (void __user *)arg;
3054 	int r = -ENOTTY;
3055 	/*
3056 	 * This union makes it completely explicit to gcc-3.x
3057 	 * that these two variables' stack usage should be
3058 	 * combined, not added together.
3059 	 */
3060 	union {
3061 		struct kvm_pit_state ps;
3062 		struct kvm_pit_state2 ps2;
3063 		struct kvm_pit_config pit_config;
3064 	} u;
3065 
3066 	switch (ioctl) {
3067 	case KVM_SET_TSS_ADDR:
3068 		r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3069 		if (r < 0)
3070 			goto out;
3071 		break;
3072 	case KVM_SET_IDENTITY_MAP_ADDR: {
3073 		u64 ident_addr;
3074 
3075 		r = -EFAULT;
3076 		if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3077 			goto out;
3078 		r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3079 		if (r < 0)
3080 			goto out;
3081 		break;
3082 	}
3083 	case KVM_SET_NR_MMU_PAGES:
3084 		r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3085 		if (r)
3086 			goto out;
3087 		break;
3088 	case KVM_GET_NR_MMU_PAGES:
3089 		r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3090 		break;
3091 	case KVM_CREATE_IRQCHIP: {
3092 		struct kvm_pic *vpic;
3093 
3094 		mutex_lock(&kvm->lock);
3095 		r = -EEXIST;
3096 		if (kvm->arch.vpic)
3097 			goto create_irqchip_unlock;
3098 		r = -ENOMEM;
3099 		vpic = kvm_create_pic(kvm);
3100 		if (vpic) {
3101 			r = kvm_ioapic_init(kvm);
3102 			if (r) {
3103 				mutex_lock(&kvm->slots_lock);
3104 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3105 							  &vpic->dev_master);
3106 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3107 							  &vpic->dev_slave);
3108 				kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3109 							  &vpic->dev_eclr);
3110 				mutex_unlock(&kvm->slots_lock);
3111 				kfree(vpic);
3112 				goto create_irqchip_unlock;
3113 			}
3114 		} else
3115 			goto create_irqchip_unlock;
3116 		smp_wmb();
3117 		kvm->arch.vpic = vpic;
3118 		smp_wmb();
3119 		r = kvm_setup_default_irq_routing(kvm);
3120 		if (r) {
3121 			mutex_lock(&kvm->slots_lock);
3122 			mutex_lock(&kvm->irq_lock);
3123 			kvm_ioapic_destroy(kvm);
3124 			kvm_destroy_pic(kvm);
3125 			mutex_unlock(&kvm->irq_lock);
3126 			mutex_unlock(&kvm->slots_lock);
3127 		}
3128 	create_irqchip_unlock:
3129 		mutex_unlock(&kvm->lock);
3130 		break;
3131 	}
3132 	case KVM_CREATE_PIT:
3133 		u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3134 		goto create_pit;
3135 	case KVM_CREATE_PIT2:
3136 		r = -EFAULT;
3137 		if (copy_from_user(&u.pit_config, argp,
3138 				   sizeof(struct kvm_pit_config)))
3139 			goto out;
3140 	create_pit:
3141 		mutex_lock(&kvm->slots_lock);
3142 		r = -EEXIST;
3143 		if (kvm->arch.vpit)
3144 			goto create_pit_unlock;
3145 		r = -ENOMEM;
3146 		kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3147 		if (kvm->arch.vpit)
3148 			r = 0;
3149 	create_pit_unlock:
3150 		mutex_unlock(&kvm->slots_lock);
3151 		break;
3152 	case KVM_IRQ_LINE_STATUS:
3153 	case KVM_IRQ_LINE: {
3154 		struct kvm_irq_level irq_event;
3155 
3156 		r = -EFAULT;
3157 		if (copy_from_user(&irq_event, argp, sizeof irq_event))
3158 			goto out;
3159 		r = -ENXIO;
3160 		if (irqchip_in_kernel(kvm)) {
3161 			__s32 status;
3162 			status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3163 					irq_event.irq, irq_event.level);
3164 			if (ioctl == KVM_IRQ_LINE_STATUS) {
3165 				r = -EFAULT;
3166 				irq_event.status = status;
3167 				if (copy_to_user(argp, &irq_event,
3168 							sizeof irq_event))
3169 					goto out;
3170 			}
3171 			r = 0;
3172 		}
3173 		break;
3174 	}
3175 	case KVM_GET_IRQCHIP: {
3176 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3177 		struct kvm_irqchip *chip;
3178 
3179 		chip = memdup_user(argp, sizeof(*chip));
3180 		if (IS_ERR(chip)) {
3181 			r = PTR_ERR(chip);
3182 			goto out;
3183 		}
3184 
3185 		r = -ENXIO;
3186 		if (!irqchip_in_kernel(kvm))
3187 			goto get_irqchip_out;
3188 		r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3189 		if (r)
3190 			goto get_irqchip_out;
3191 		r = -EFAULT;
3192 		if (copy_to_user(argp, chip, sizeof *chip))
3193 			goto get_irqchip_out;
3194 		r = 0;
3195 	get_irqchip_out:
3196 		kfree(chip);
3197 		if (r)
3198 			goto out;
3199 		break;
3200 	}
3201 	case KVM_SET_IRQCHIP: {
3202 		/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3203 		struct kvm_irqchip *chip;
3204 
3205 		chip = memdup_user(argp, sizeof(*chip));
3206 		if (IS_ERR(chip)) {
3207 			r = PTR_ERR(chip);
3208 			goto out;
3209 		}
3210 
3211 		r = -ENXIO;
3212 		if (!irqchip_in_kernel(kvm))
3213 			goto set_irqchip_out;
3214 		r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3215 		if (r)
3216 			goto set_irqchip_out;
3217 		r = 0;
3218 	set_irqchip_out:
3219 		kfree(chip);
3220 		if (r)
3221 			goto out;
3222 		break;
3223 	}
3224 	case KVM_GET_PIT: {
3225 		r = -EFAULT;
3226 		if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3227 			goto out;
3228 		r = -ENXIO;
3229 		if (!kvm->arch.vpit)
3230 			goto out;
3231 		r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3232 		if (r)
3233 			goto out;
3234 		r = -EFAULT;
3235 		if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3236 			goto out;
3237 		r = 0;
3238 		break;
3239 	}
3240 	case KVM_SET_PIT: {
3241 		r = -EFAULT;
3242 		if (copy_from_user(&u.ps, argp, sizeof u.ps))
3243 			goto out;
3244 		r = -ENXIO;
3245 		if (!kvm->arch.vpit)
3246 			goto out;
3247 		r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3248 		if (r)
3249 			goto out;
3250 		r = 0;
3251 		break;
3252 	}
3253 	case KVM_GET_PIT2: {
3254 		r = -ENXIO;
3255 		if (!kvm->arch.vpit)
3256 			goto out;
3257 		r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3258 		if (r)
3259 			goto out;
3260 		r = -EFAULT;
3261 		if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3262 			goto out;
3263 		r = 0;
3264 		break;
3265 	}
3266 	case KVM_SET_PIT2: {
3267 		r = -EFAULT;
3268 		if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3269 			goto out;
3270 		r = -ENXIO;
3271 		if (!kvm->arch.vpit)
3272 			goto out;
3273 		r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3274 		if (r)
3275 			goto out;
3276 		r = 0;
3277 		break;
3278 	}
3279 	case KVM_REINJECT_CONTROL: {
3280 		struct kvm_reinject_control control;
3281 		r =  -EFAULT;
3282 		if (copy_from_user(&control, argp, sizeof(control)))
3283 			goto out;
3284 		r = kvm_vm_ioctl_reinject(kvm, &control);
3285 		if (r)
3286 			goto out;
3287 		r = 0;
3288 		break;
3289 	}
3290 	case KVM_XEN_HVM_CONFIG: {
3291 		r = -EFAULT;
3292 		if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3293 				   sizeof(struct kvm_xen_hvm_config)))
3294 			goto out;
3295 		r = -EINVAL;
3296 		if (kvm->arch.xen_hvm_config.flags)
3297 			goto out;
3298 		r = 0;
3299 		break;
3300 	}
3301 	case KVM_SET_CLOCK: {
3302 		struct kvm_clock_data user_ns;
3303 		u64 now_ns;
3304 		s64 delta;
3305 
3306 		r = -EFAULT;
3307 		if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3308 			goto out;
3309 
3310 		r = -EINVAL;
3311 		if (user_ns.flags)
3312 			goto out;
3313 
3314 		r = 0;
3315 		local_irq_disable();
3316 		now_ns = get_kernel_ns();
3317 		delta = user_ns.clock - now_ns;
3318 		local_irq_enable();
3319 		kvm->arch.kvmclock_offset = delta;
3320 		break;
3321 	}
3322 	case KVM_GET_CLOCK: {
3323 		struct kvm_clock_data user_ns;
3324 		u64 now_ns;
3325 
3326 		local_irq_disable();
3327 		now_ns = get_kernel_ns();
3328 		user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3329 		local_irq_enable();
3330 		user_ns.flags = 0;
3331 		memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3332 
3333 		r = -EFAULT;
3334 		if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3335 			goto out;
3336 		r = 0;
3337 		break;
3338 	}
3339 
3340 	default:
3341 		;
3342 	}
3343 out:
3344 	return r;
3345 }
3346 
3347 static void kvm_init_msr_list(void)
3348 {
3349 	u32 dummy[2];
3350 	unsigned i, j;
3351 
3352 	/* skip the first msrs in the list. KVM-specific */
3353 	for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3354 		if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3355 			continue;
3356 		if (j < i)
3357 			msrs_to_save[j] = msrs_to_save[i];
3358 		j++;
3359 	}
3360 	num_msrs_to_save = j;
3361 }
3362 
3363 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3364 			   const void *v)
3365 {
3366 	int handled = 0;
3367 	int n;
3368 
3369 	do {
3370 		n = min(len, 8);
3371 		if (!(vcpu->arch.apic &&
3372 		      !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3373 		    && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3374 			break;
3375 		handled += n;
3376 		addr += n;
3377 		len -= n;
3378 		v += n;
3379 	} while (len);
3380 
3381 	return handled;
3382 }
3383 
3384 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3385 {
3386 	int handled = 0;
3387 	int n;
3388 
3389 	do {
3390 		n = min(len, 8);
3391 		if (!(vcpu->arch.apic &&
3392 		      !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3393 		    && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3394 			break;
3395 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3396 		handled += n;
3397 		addr += n;
3398 		len -= n;
3399 		v += n;
3400 	} while (len);
3401 
3402 	return handled;
3403 }
3404 
3405 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3406 			struct kvm_segment *var, int seg)
3407 {
3408 	kvm_x86_ops->set_segment(vcpu, var, seg);
3409 }
3410 
3411 void kvm_get_segment(struct kvm_vcpu *vcpu,
3412 		     struct kvm_segment *var, int seg)
3413 {
3414 	kvm_x86_ops->get_segment(vcpu, var, seg);
3415 }
3416 
3417 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3418 {
3419 	gpa_t t_gpa;
3420 	struct x86_exception exception;
3421 
3422 	BUG_ON(!mmu_is_nested(vcpu));
3423 
3424 	/* NPT walks are always user-walks */
3425 	access |= PFERR_USER_MASK;
3426 	t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3427 
3428 	return t_gpa;
3429 }
3430 
3431 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3432 			      struct x86_exception *exception)
3433 {
3434 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3435 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3436 }
3437 
3438  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3439 				struct x86_exception *exception)
3440 {
3441 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3442 	access |= PFERR_FETCH_MASK;
3443 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3444 }
3445 
3446 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3447 			       struct x86_exception *exception)
3448 {
3449 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3450 	access |= PFERR_WRITE_MASK;
3451 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3452 }
3453 
3454 /* uses this to access any guest's mapped memory without checking CPL */
3455 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3456 				struct x86_exception *exception)
3457 {
3458 	return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3459 }
3460 
3461 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3462 				      struct kvm_vcpu *vcpu, u32 access,
3463 				      struct x86_exception *exception)
3464 {
3465 	void *data = val;
3466 	int r = X86EMUL_CONTINUE;
3467 
3468 	while (bytes) {
3469 		gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3470 							    exception);
3471 		unsigned offset = addr & (PAGE_SIZE-1);
3472 		unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3473 		int ret;
3474 
3475 		if (gpa == UNMAPPED_GVA)
3476 			return X86EMUL_PROPAGATE_FAULT;
3477 		ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3478 		if (ret < 0) {
3479 			r = X86EMUL_IO_NEEDED;
3480 			goto out;
3481 		}
3482 
3483 		bytes -= toread;
3484 		data += toread;
3485 		addr += toread;
3486 	}
3487 out:
3488 	return r;
3489 }
3490 
3491 /* used for instruction fetching */
3492 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3493 				gva_t addr, void *val, unsigned int bytes,
3494 				struct x86_exception *exception)
3495 {
3496 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3497 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3498 
3499 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3500 					  access | PFERR_FETCH_MASK,
3501 					  exception);
3502 }
3503 
3504 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3505 			       gva_t addr, void *val, unsigned int bytes,
3506 			       struct x86_exception *exception)
3507 {
3508 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3509 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3510 
3511 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3512 					  exception);
3513 }
3514 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3515 
3516 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3517 				      gva_t addr, void *val, unsigned int bytes,
3518 				      struct x86_exception *exception)
3519 {
3520 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3521 	return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3522 }
3523 
3524 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3525 				       gva_t addr, void *val,
3526 				       unsigned int bytes,
3527 				       struct x86_exception *exception)
3528 {
3529 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3530 	void *data = val;
3531 	int r = X86EMUL_CONTINUE;
3532 
3533 	while (bytes) {
3534 		gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3535 							     PFERR_WRITE_MASK,
3536 							     exception);
3537 		unsigned offset = addr & (PAGE_SIZE-1);
3538 		unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3539 		int ret;
3540 
3541 		if (gpa == UNMAPPED_GVA)
3542 			return X86EMUL_PROPAGATE_FAULT;
3543 		ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3544 		if (ret < 0) {
3545 			r = X86EMUL_IO_NEEDED;
3546 			goto out;
3547 		}
3548 
3549 		bytes -= towrite;
3550 		data += towrite;
3551 		addr += towrite;
3552 	}
3553 out:
3554 	return r;
3555 }
3556 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3557 
3558 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3559 				gpa_t *gpa, struct x86_exception *exception,
3560 				bool write)
3561 {
3562 	u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3563 
3564 	if (vcpu_match_mmio_gva(vcpu, gva) &&
3565 		  check_write_user_access(vcpu, write, access,
3566 		  vcpu->arch.access)) {
3567 		*gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3568 					(gva & (PAGE_SIZE - 1));
3569 		trace_vcpu_match_mmio(gva, *gpa, write, false);
3570 		return 1;
3571 	}
3572 
3573 	if (write)
3574 		access |= PFERR_WRITE_MASK;
3575 
3576 	*gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3577 
3578 	if (*gpa == UNMAPPED_GVA)
3579 		return -1;
3580 
3581 	/* For APIC access vmexit */
3582 	if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3583 		return 1;
3584 
3585 	if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3586 		trace_vcpu_match_mmio(gva, *gpa, write, true);
3587 		return 1;
3588 	}
3589 
3590 	return 0;
3591 }
3592 
3593 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3594 			const void *val, int bytes)
3595 {
3596 	int ret;
3597 
3598 	ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3599 	if (ret < 0)
3600 		return 0;
3601 	kvm_mmu_pte_write(vcpu, gpa, val, bytes);
3602 	return 1;
3603 }
3604 
3605 struct read_write_emulator_ops {
3606 	int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
3607 				  int bytes);
3608 	int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
3609 				  void *val, int bytes);
3610 	int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3611 			       int bytes, void *val);
3612 	int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3613 				    void *val, int bytes);
3614 	bool write;
3615 };
3616 
3617 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
3618 {
3619 	if (vcpu->mmio_read_completed) {
3620 		memcpy(val, vcpu->mmio_data, bytes);
3621 		trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3622 			       vcpu->mmio_phys_addr, *(u64 *)val);
3623 		vcpu->mmio_read_completed = 0;
3624 		return 1;
3625 	}
3626 
3627 	return 0;
3628 }
3629 
3630 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3631 			void *val, int bytes)
3632 {
3633 	return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
3634 }
3635 
3636 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3637 			 void *val, int bytes)
3638 {
3639 	return emulator_write_phys(vcpu, gpa, val, bytes);
3640 }
3641 
3642 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
3643 {
3644 	trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3645 	return vcpu_mmio_write(vcpu, gpa, bytes, val);
3646 }
3647 
3648 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3649 			  void *val, int bytes)
3650 {
3651 	trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3652 	return X86EMUL_IO_NEEDED;
3653 }
3654 
3655 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3656 			   void *val, int bytes)
3657 {
3658 	memcpy(vcpu->mmio_data, val, bytes);
3659 	memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3660 	return X86EMUL_CONTINUE;
3661 }
3662 
3663 static struct read_write_emulator_ops read_emultor = {
3664 	.read_write_prepare = read_prepare,
3665 	.read_write_emulate = read_emulate,
3666 	.read_write_mmio = vcpu_mmio_read,
3667 	.read_write_exit_mmio = read_exit_mmio,
3668 };
3669 
3670 static struct read_write_emulator_ops write_emultor = {
3671 	.read_write_emulate = write_emulate,
3672 	.read_write_mmio = write_mmio,
3673 	.read_write_exit_mmio = write_exit_mmio,
3674 	.write = true,
3675 };
3676 
3677 static int emulator_read_write_onepage(unsigned long addr, void *val,
3678 				       unsigned int bytes,
3679 				       struct x86_exception *exception,
3680 				       struct kvm_vcpu *vcpu,
3681 				       struct read_write_emulator_ops *ops)
3682 {
3683 	gpa_t gpa;
3684 	int handled, ret;
3685 	bool write = ops->write;
3686 
3687 	if (ops->read_write_prepare &&
3688 		  ops->read_write_prepare(vcpu, val, bytes))
3689 		return X86EMUL_CONTINUE;
3690 
3691 	ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
3692 
3693 	if (ret < 0)
3694 		return X86EMUL_PROPAGATE_FAULT;
3695 
3696 	/* For APIC access vmexit */
3697 	if (ret)
3698 		goto mmio;
3699 
3700 	if (ops->read_write_emulate(vcpu, gpa, val, bytes))
3701 		return X86EMUL_CONTINUE;
3702 
3703 mmio:
3704 	/*
3705 	 * Is this MMIO handled locally?
3706 	 */
3707 	handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
3708 	if (handled == bytes)
3709 		return X86EMUL_CONTINUE;
3710 
3711 	gpa += handled;
3712 	bytes -= handled;
3713 	val += handled;
3714 
3715 	vcpu->mmio_needed = 1;
3716 	vcpu->run->exit_reason = KVM_EXIT_MMIO;
3717 	vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3718 	vcpu->mmio_size = bytes;
3719 	vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3720 	vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
3721 	vcpu->mmio_index = 0;
3722 
3723 	return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
3724 }
3725 
3726 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
3727 			void *val, unsigned int bytes,
3728 			struct x86_exception *exception,
3729 			struct read_write_emulator_ops *ops)
3730 {
3731 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3732 
3733 	/* Crossing a page boundary? */
3734 	if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3735 		int rc, now;
3736 
3737 		now = -addr & ~PAGE_MASK;
3738 		rc = emulator_read_write_onepage(addr, val, now, exception,
3739 						 vcpu, ops);
3740 
3741 		if (rc != X86EMUL_CONTINUE)
3742 			return rc;
3743 		addr += now;
3744 		val += now;
3745 		bytes -= now;
3746 	}
3747 
3748 	return emulator_read_write_onepage(addr, val, bytes, exception,
3749 					   vcpu, ops);
3750 }
3751 
3752 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3753 				  unsigned long addr,
3754 				  void *val,
3755 				  unsigned int bytes,
3756 				  struct x86_exception *exception)
3757 {
3758 	return emulator_read_write(ctxt, addr, val, bytes,
3759 				   exception, &read_emultor);
3760 }
3761 
3762 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3763 			    unsigned long addr,
3764 			    const void *val,
3765 			    unsigned int bytes,
3766 			    struct x86_exception *exception)
3767 {
3768 	return emulator_read_write(ctxt, addr, (void *)val, bytes,
3769 				   exception, &write_emultor);
3770 }
3771 
3772 #define CMPXCHG_TYPE(t, ptr, old, new) \
3773 	(cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3774 
3775 #ifdef CONFIG_X86_64
3776 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3777 #else
3778 #  define CMPXCHG64(ptr, old, new) \
3779 	(cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3780 #endif
3781 
3782 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
3783 				     unsigned long addr,
3784 				     const void *old,
3785 				     const void *new,
3786 				     unsigned int bytes,
3787 				     struct x86_exception *exception)
3788 {
3789 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3790 	gpa_t gpa;
3791 	struct page *page;
3792 	char *kaddr;
3793 	bool exchanged;
3794 
3795 	/* guests cmpxchg8b have to be emulated atomically */
3796 	if (bytes > 8 || (bytes & (bytes - 1)))
3797 		goto emul_write;
3798 
3799 	gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3800 
3801 	if (gpa == UNMAPPED_GVA ||
3802 	    (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3803 		goto emul_write;
3804 
3805 	if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3806 		goto emul_write;
3807 
3808 	page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3809 	if (is_error_page(page)) {
3810 		kvm_release_page_clean(page);
3811 		goto emul_write;
3812 	}
3813 
3814 	kaddr = kmap_atomic(page, KM_USER0);
3815 	kaddr += offset_in_page(gpa);
3816 	switch (bytes) {
3817 	case 1:
3818 		exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3819 		break;
3820 	case 2:
3821 		exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3822 		break;
3823 	case 4:
3824 		exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3825 		break;
3826 	case 8:
3827 		exchanged = CMPXCHG64(kaddr, old, new);
3828 		break;
3829 	default:
3830 		BUG();
3831 	}
3832 	kunmap_atomic(kaddr, KM_USER0);
3833 	kvm_release_page_dirty(page);
3834 
3835 	if (!exchanged)
3836 		return X86EMUL_CMPXCHG_FAILED;
3837 
3838 	kvm_mmu_pte_write(vcpu, gpa, new, bytes);
3839 
3840 	return X86EMUL_CONTINUE;
3841 
3842 emul_write:
3843 	printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3844 
3845 	return emulator_write_emulated(ctxt, addr, new, bytes, exception);
3846 }
3847 
3848 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3849 {
3850 	/* TODO: String I/O for in kernel device */
3851 	int r;
3852 
3853 	if (vcpu->arch.pio.in)
3854 		r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3855 				    vcpu->arch.pio.size, pd);
3856 	else
3857 		r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3858 				     vcpu->arch.pio.port, vcpu->arch.pio.size,
3859 				     pd);
3860 	return r;
3861 }
3862 
3863 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
3864 			       unsigned short port, void *val,
3865 			       unsigned int count, bool in)
3866 {
3867 	trace_kvm_pio(!in, port, size, count);
3868 
3869 	vcpu->arch.pio.port = port;
3870 	vcpu->arch.pio.in = in;
3871 	vcpu->arch.pio.count  = count;
3872 	vcpu->arch.pio.size = size;
3873 
3874 	if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3875 		vcpu->arch.pio.count = 0;
3876 		return 1;
3877 	}
3878 
3879 	vcpu->run->exit_reason = KVM_EXIT_IO;
3880 	vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3881 	vcpu->run->io.size = size;
3882 	vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3883 	vcpu->run->io.count = count;
3884 	vcpu->run->io.port = port;
3885 
3886 	return 0;
3887 }
3888 
3889 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
3890 				    int size, unsigned short port, void *val,
3891 				    unsigned int count)
3892 {
3893 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3894 	int ret;
3895 
3896 	if (vcpu->arch.pio.count)
3897 		goto data_avail;
3898 
3899 	ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
3900 	if (ret) {
3901 data_avail:
3902 		memcpy(val, vcpu->arch.pio_data, size * count);
3903 		vcpu->arch.pio.count = 0;
3904 		return 1;
3905 	}
3906 
3907 	return 0;
3908 }
3909 
3910 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
3911 				     int size, unsigned short port,
3912 				     const void *val, unsigned int count)
3913 {
3914 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3915 
3916 	memcpy(vcpu->arch.pio_data, val, size * count);
3917 	return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
3918 }
3919 
3920 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3921 {
3922 	return kvm_x86_ops->get_segment_base(vcpu, seg);
3923 }
3924 
3925 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
3926 {
3927 	kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
3928 }
3929 
3930 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
3931 {
3932 	if (!need_emulate_wbinvd(vcpu))
3933 		return X86EMUL_CONTINUE;
3934 
3935 	if (kvm_x86_ops->has_wbinvd_exit()) {
3936 		int cpu = get_cpu();
3937 
3938 		cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3939 		smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
3940 				wbinvd_ipi, NULL, 1);
3941 		put_cpu();
3942 		cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
3943 	} else
3944 		wbinvd();
3945 	return X86EMUL_CONTINUE;
3946 }
3947 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
3948 
3949 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
3950 {
3951 	kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
3952 }
3953 
3954 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3955 {
3956 	return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
3957 }
3958 
3959 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3960 {
3961 
3962 	return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
3963 }
3964 
3965 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3966 {
3967 	return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3968 }
3969 
3970 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
3971 {
3972 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3973 	unsigned long value;
3974 
3975 	switch (cr) {
3976 	case 0:
3977 		value = kvm_read_cr0(vcpu);
3978 		break;
3979 	case 2:
3980 		value = vcpu->arch.cr2;
3981 		break;
3982 	case 3:
3983 		value = kvm_read_cr3(vcpu);
3984 		break;
3985 	case 4:
3986 		value = kvm_read_cr4(vcpu);
3987 		break;
3988 	case 8:
3989 		value = kvm_get_cr8(vcpu);
3990 		break;
3991 	default:
3992 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3993 		return 0;
3994 	}
3995 
3996 	return value;
3997 }
3998 
3999 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4000 {
4001 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4002 	int res = 0;
4003 
4004 	switch (cr) {
4005 	case 0:
4006 		res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4007 		break;
4008 	case 2:
4009 		vcpu->arch.cr2 = val;
4010 		break;
4011 	case 3:
4012 		res = kvm_set_cr3(vcpu, val);
4013 		break;
4014 	case 4:
4015 		res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4016 		break;
4017 	case 8:
4018 		res = kvm_set_cr8(vcpu, val);
4019 		break;
4020 	default:
4021 		vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4022 		res = -1;
4023 	}
4024 
4025 	return res;
4026 }
4027 
4028 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4029 {
4030 	return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4031 }
4032 
4033 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4034 {
4035 	kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4036 }
4037 
4038 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4039 {
4040 	kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4041 }
4042 
4043 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4044 {
4045 	kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4046 }
4047 
4048 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4049 {
4050 	kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4051 }
4052 
4053 static unsigned long emulator_get_cached_segment_base(
4054 	struct x86_emulate_ctxt *ctxt, int seg)
4055 {
4056 	return get_segment_base(emul_to_vcpu(ctxt), seg);
4057 }
4058 
4059 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4060 				 struct desc_struct *desc, u32 *base3,
4061 				 int seg)
4062 {
4063 	struct kvm_segment var;
4064 
4065 	kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4066 	*selector = var.selector;
4067 
4068 	if (var.unusable)
4069 		return false;
4070 
4071 	if (var.g)
4072 		var.limit >>= 12;
4073 	set_desc_limit(desc, var.limit);
4074 	set_desc_base(desc, (unsigned long)var.base);
4075 #ifdef CONFIG_X86_64
4076 	if (base3)
4077 		*base3 = var.base >> 32;
4078 #endif
4079 	desc->type = var.type;
4080 	desc->s = var.s;
4081 	desc->dpl = var.dpl;
4082 	desc->p = var.present;
4083 	desc->avl = var.avl;
4084 	desc->l = var.l;
4085 	desc->d = var.db;
4086 	desc->g = var.g;
4087 
4088 	return true;
4089 }
4090 
4091 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4092 				 struct desc_struct *desc, u32 base3,
4093 				 int seg)
4094 {
4095 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4096 	struct kvm_segment var;
4097 
4098 	var.selector = selector;
4099 	var.base = get_desc_base(desc);
4100 #ifdef CONFIG_X86_64
4101 	var.base |= ((u64)base3) << 32;
4102 #endif
4103 	var.limit = get_desc_limit(desc);
4104 	if (desc->g)
4105 		var.limit = (var.limit << 12) | 0xfff;
4106 	var.type = desc->type;
4107 	var.present = desc->p;
4108 	var.dpl = desc->dpl;
4109 	var.db = desc->d;
4110 	var.s = desc->s;
4111 	var.l = desc->l;
4112 	var.g = desc->g;
4113 	var.avl = desc->avl;
4114 	var.present = desc->p;
4115 	var.unusable = !var.present;
4116 	var.padding = 0;
4117 
4118 	kvm_set_segment(vcpu, &var, seg);
4119 	return;
4120 }
4121 
4122 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4123 			    u32 msr_index, u64 *pdata)
4124 {
4125 	return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4126 }
4127 
4128 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4129 			    u32 msr_index, u64 data)
4130 {
4131 	return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4132 }
4133 
4134 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4135 {
4136 	emul_to_vcpu(ctxt)->arch.halt_request = 1;
4137 }
4138 
4139 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4140 {
4141 	preempt_disable();
4142 	kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4143 	/*
4144 	 * CR0.TS may reference the host fpu state, not the guest fpu state,
4145 	 * so it may be clear at this point.
4146 	 */
4147 	clts();
4148 }
4149 
4150 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4151 {
4152 	preempt_enable();
4153 }
4154 
4155 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4156 			      struct x86_instruction_info *info,
4157 			      enum x86_intercept_stage stage)
4158 {
4159 	return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4160 }
4161 
4162 static struct x86_emulate_ops emulate_ops = {
4163 	.read_std            = kvm_read_guest_virt_system,
4164 	.write_std           = kvm_write_guest_virt_system,
4165 	.fetch               = kvm_fetch_guest_virt,
4166 	.read_emulated       = emulator_read_emulated,
4167 	.write_emulated      = emulator_write_emulated,
4168 	.cmpxchg_emulated    = emulator_cmpxchg_emulated,
4169 	.invlpg              = emulator_invlpg,
4170 	.pio_in_emulated     = emulator_pio_in_emulated,
4171 	.pio_out_emulated    = emulator_pio_out_emulated,
4172 	.get_segment         = emulator_get_segment,
4173 	.set_segment         = emulator_set_segment,
4174 	.get_cached_segment_base = emulator_get_cached_segment_base,
4175 	.get_gdt             = emulator_get_gdt,
4176 	.get_idt	     = emulator_get_idt,
4177 	.set_gdt             = emulator_set_gdt,
4178 	.set_idt	     = emulator_set_idt,
4179 	.get_cr              = emulator_get_cr,
4180 	.set_cr              = emulator_set_cr,
4181 	.cpl                 = emulator_get_cpl,
4182 	.get_dr              = emulator_get_dr,
4183 	.set_dr              = emulator_set_dr,
4184 	.set_msr             = emulator_set_msr,
4185 	.get_msr             = emulator_get_msr,
4186 	.halt                = emulator_halt,
4187 	.wbinvd              = emulator_wbinvd,
4188 	.fix_hypercall       = emulator_fix_hypercall,
4189 	.get_fpu             = emulator_get_fpu,
4190 	.put_fpu             = emulator_put_fpu,
4191 	.intercept           = emulator_intercept,
4192 };
4193 
4194 static void cache_all_regs(struct kvm_vcpu *vcpu)
4195 {
4196 	kvm_register_read(vcpu, VCPU_REGS_RAX);
4197 	kvm_register_read(vcpu, VCPU_REGS_RSP);
4198 	kvm_register_read(vcpu, VCPU_REGS_RIP);
4199 	vcpu->arch.regs_dirty = ~0;
4200 }
4201 
4202 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4203 {
4204 	u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4205 	/*
4206 	 * an sti; sti; sequence only disable interrupts for the first
4207 	 * instruction. So, if the last instruction, be it emulated or
4208 	 * not, left the system with the INT_STI flag enabled, it
4209 	 * means that the last instruction is an sti. We should not
4210 	 * leave the flag on in this case. The same goes for mov ss
4211 	 */
4212 	if (!(int_shadow & mask))
4213 		kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4214 }
4215 
4216 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4217 {
4218 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4219 	if (ctxt->exception.vector == PF_VECTOR)
4220 		kvm_propagate_fault(vcpu, &ctxt->exception);
4221 	else if (ctxt->exception.error_code_valid)
4222 		kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4223 				      ctxt->exception.error_code);
4224 	else
4225 		kvm_queue_exception(vcpu, ctxt->exception.vector);
4226 }
4227 
4228 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4229 			      const unsigned long *regs)
4230 {
4231 	memset(&ctxt->twobyte, 0,
4232 	       (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4233 	memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4234 
4235 	ctxt->fetch.start = 0;
4236 	ctxt->fetch.end = 0;
4237 	ctxt->io_read.pos = 0;
4238 	ctxt->io_read.end = 0;
4239 	ctxt->mem_read.pos = 0;
4240 	ctxt->mem_read.end = 0;
4241 }
4242 
4243 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4244 {
4245 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4246 	int cs_db, cs_l;
4247 
4248 	/*
4249 	 * TODO: fix emulate.c to use guest_read/write_register
4250 	 * instead of direct ->regs accesses, can save hundred cycles
4251 	 * on Intel for instructions that don't read/change RSP, for
4252 	 * for example.
4253 	 */
4254 	cache_all_regs(vcpu);
4255 
4256 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4257 
4258 	ctxt->eflags = kvm_get_rflags(vcpu);
4259 	ctxt->eip = kvm_rip_read(vcpu);
4260 	ctxt->mode = (!is_protmode(vcpu))		? X86EMUL_MODE_REAL :
4261 		     (ctxt->eflags & X86_EFLAGS_VM)	? X86EMUL_MODE_VM86 :
4262 		     cs_l				? X86EMUL_MODE_PROT64 :
4263 		     cs_db				? X86EMUL_MODE_PROT32 :
4264 							  X86EMUL_MODE_PROT16;
4265 	ctxt->guest_mode = is_guest_mode(vcpu);
4266 
4267 	init_decode_cache(ctxt, vcpu->arch.regs);
4268 	vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4269 }
4270 
4271 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4272 {
4273 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4274 	int ret;
4275 
4276 	init_emulate_ctxt(vcpu);
4277 
4278 	ctxt->op_bytes = 2;
4279 	ctxt->ad_bytes = 2;
4280 	ctxt->_eip = ctxt->eip + inc_eip;
4281 	ret = emulate_int_real(ctxt, irq);
4282 
4283 	if (ret != X86EMUL_CONTINUE)
4284 		return EMULATE_FAIL;
4285 
4286 	ctxt->eip = ctxt->_eip;
4287 	memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4288 	kvm_rip_write(vcpu, ctxt->eip);
4289 	kvm_set_rflags(vcpu, ctxt->eflags);
4290 
4291 	if (irq == NMI_VECTOR)
4292 		vcpu->arch.nmi_pending = 0;
4293 	else
4294 		vcpu->arch.interrupt.pending = false;
4295 
4296 	return EMULATE_DONE;
4297 }
4298 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4299 
4300 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4301 {
4302 	int r = EMULATE_DONE;
4303 
4304 	++vcpu->stat.insn_emulation_fail;
4305 	trace_kvm_emulate_insn_failed(vcpu);
4306 	if (!is_guest_mode(vcpu)) {
4307 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4308 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4309 		vcpu->run->internal.ndata = 0;
4310 		r = EMULATE_FAIL;
4311 	}
4312 	kvm_queue_exception(vcpu, UD_VECTOR);
4313 
4314 	return r;
4315 }
4316 
4317 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4318 {
4319 	gpa_t gpa;
4320 
4321 	if (tdp_enabled)
4322 		return false;
4323 
4324 	/*
4325 	 * if emulation was due to access to shadowed page table
4326 	 * and it failed try to unshadow page and re-entetr the
4327 	 * guest to let CPU execute the instruction.
4328 	 */
4329 	if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4330 		return true;
4331 
4332 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4333 
4334 	if (gpa == UNMAPPED_GVA)
4335 		return true; /* let cpu generate fault */
4336 
4337 	if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4338 		return true;
4339 
4340 	return false;
4341 }
4342 
4343 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4344 			      unsigned long cr2,  int emulation_type)
4345 {
4346 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4347 	unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4348 
4349 	last_retry_eip = vcpu->arch.last_retry_eip;
4350 	last_retry_addr = vcpu->arch.last_retry_addr;
4351 
4352 	/*
4353 	 * If the emulation is caused by #PF and it is non-page_table
4354 	 * writing instruction, it means the VM-EXIT is caused by shadow
4355 	 * page protected, we can zap the shadow page and retry this
4356 	 * instruction directly.
4357 	 *
4358 	 * Note: if the guest uses a non-page-table modifying instruction
4359 	 * on the PDE that points to the instruction, then we will unmap
4360 	 * the instruction and go to an infinite loop. So, we cache the
4361 	 * last retried eip and the last fault address, if we meet the eip
4362 	 * and the address again, we can break out of the potential infinite
4363 	 * loop.
4364 	 */
4365 	vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4366 
4367 	if (!(emulation_type & EMULTYPE_RETRY))
4368 		return false;
4369 
4370 	if (x86_page_table_writing_insn(ctxt))
4371 		return false;
4372 
4373 	if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4374 		return false;
4375 
4376 	vcpu->arch.last_retry_eip = ctxt->eip;
4377 	vcpu->arch.last_retry_addr = cr2;
4378 
4379 	if (!vcpu->arch.mmu.direct_map)
4380 		gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4381 
4382 	kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4383 
4384 	return true;
4385 }
4386 
4387 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4388 			    unsigned long cr2,
4389 			    int emulation_type,
4390 			    void *insn,
4391 			    int insn_len)
4392 {
4393 	int r;
4394 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4395 	bool writeback = true;
4396 
4397 	kvm_clear_exception_queue(vcpu);
4398 
4399 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4400 		init_emulate_ctxt(vcpu);
4401 		ctxt->interruptibility = 0;
4402 		ctxt->have_exception = false;
4403 		ctxt->perm_ok = false;
4404 
4405 		ctxt->only_vendor_specific_insn
4406 			= emulation_type & EMULTYPE_TRAP_UD;
4407 
4408 		r = x86_decode_insn(ctxt, insn, insn_len);
4409 
4410 		trace_kvm_emulate_insn_start(vcpu);
4411 		++vcpu->stat.insn_emulation;
4412 		if (r != EMULATION_OK)  {
4413 			if (emulation_type & EMULTYPE_TRAP_UD)
4414 				return EMULATE_FAIL;
4415 			if (reexecute_instruction(vcpu, cr2))
4416 				return EMULATE_DONE;
4417 			if (emulation_type & EMULTYPE_SKIP)
4418 				return EMULATE_FAIL;
4419 			return handle_emulation_failure(vcpu);
4420 		}
4421 	}
4422 
4423 	if (emulation_type & EMULTYPE_SKIP) {
4424 		kvm_rip_write(vcpu, ctxt->_eip);
4425 		return EMULATE_DONE;
4426 	}
4427 
4428 	if (retry_instruction(ctxt, cr2, emulation_type))
4429 		return EMULATE_DONE;
4430 
4431 	/* this is needed for vmware backdoor interface to work since it
4432 	   changes registers values  during IO operation */
4433 	if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4434 		vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4435 		memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4436 	}
4437 
4438 restart:
4439 	r = x86_emulate_insn(ctxt);
4440 
4441 	if (r == EMULATION_INTERCEPTED)
4442 		return EMULATE_DONE;
4443 
4444 	if (r == EMULATION_FAILED) {
4445 		if (reexecute_instruction(vcpu, cr2))
4446 			return EMULATE_DONE;
4447 
4448 		return handle_emulation_failure(vcpu);
4449 	}
4450 
4451 	if (ctxt->have_exception) {
4452 		inject_emulated_exception(vcpu);
4453 		r = EMULATE_DONE;
4454 	} else if (vcpu->arch.pio.count) {
4455 		if (!vcpu->arch.pio.in)
4456 			vcpu->arch.pio.count = 0;
4457 		else
4458 			writeback = false;
4459 		r = EMULATE_DO_MMIO;
4460 	} else if (vcpu->mmio_needed) {
4461 		if (!vcpu->mmio_is_write)
4462 			writeback = false;
4463 		r = EMULATE_DO_MMIO;
4464 	} else if (r == EMULATION_RESTART)
4465 		goto restart;
4466 	else
4467 		r = EMULATE_DONE;
4468 
4469 	if (writeback) {
4470 		toggle_interruptibility(vcpu, ctxt->interruptibility);
4471 		kvm_set_rflags(vcpu, ctxt->eflags);
4472 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4473 		memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4474 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4475 		kvm_rip_write(vcpu, ctxt->eip);
4476 	} else
4477 		vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4478 
4479 	return r;
4480 }
4481 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4482 
4483 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4484 {
4485 	unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4486 	int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4487 					    size, port, &val, 1);
4488 	/* do not return to emulator after return from userspace */
4489 	vcpu->arch.pio.count = 0;
4490 	return ret;
4491 }
4492 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4493 
4494 static void tsc_bad(void *info)
4495 {
4496 	__this_cpu_write(cpu_tsc_khz, 0);
4497 }
4498 
4499 static void tsc_khz_changed(void *data)
4500 {
4501 	struct cpufreq_freqs *freq = data;
4502 	unsigned long khz = 0;
4503 
4504 	if (data)
4505 		khz = freq->new;
4506 	else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4507 		khz = cpufreq_quick_get(raw_smp_processor_id());
4508 	if (!khz)
4509 		khz = tsc_khz;
4510 	__this_cpu_write(cpu_tsc_khz, khz);
4511 }
4512 
4513 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4514 				     void *data)
4515 {
4516 	struct cpufreq_freqs *freq = data;
4517 	struct kvm *kvm;
4518 	struct kvm_vcpu *vcpu;
4519 	int i, send_ipi = 0;
4520 
4521 	/*
4522 	 * We allow guests to temporarily run on slowing clocks,
4523 	 * provided we notify them after, or to run on accelerating
4524 	 * clocks, provided we notify them before.  Thus time never
4525 	 * goes backwards.
4526 	 *
4527 	 * However, we have a problem.  We can't atomically update
4528 	 * the frequency of a given CPU from this function; it is
4529 	 * merely a notifier, which can be called from any CPU.
4530 	 * Changing the TSC frequency at arbitrary points in time
4531 	 * requires a recomputation of local variables related to
4532 	 * the TSC for each VCPU.  We must flag these local variables
4533 	 * to be updated and be sure the update takes place with the
4534 	 * new frequency before any guests proceed.
4535 	 *
4536 	 * Unfortunately, the combination of hotplug CPU and frequency
4537 	 * change creates an intractable locking scenario; the order
4538 	 * of when these callouts happen is undefined with respect to
4539 	 * CPU hotplug, and they can race with each other.  As such,
4540 	 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4541 	 * undefined; you can actually have a CPU frequency change take
4542 	 * place in between the computation of X and the setting of the
4543 	 * variable.  To protect against this problem, all updates of
4544 	 * the per_cpu tsc_khz variable are done in an interrupt
4545 	 * protected IPI, and all callers wishing to update the value
4546 	 * must wait for a synchronous IPI to complete (which is trivial
4547 	 * if the caller is on the CPU already).  This establishes the
4548 	 * necessary total order on variable updates.
4549 	 *
4550 	 * Note that because a guest time update may take place
4551 	 * anytime after the setting of the VCPU's request bit, the
4552 	 * correct TSC value must be set before the request.  However,
4553 	 * to ensure the update actually makes it to any guest which
4554 	 * starts running in hardware virtualization between the set
4555 	 * and the acquisition of the spinlock, we must also ping the
4556 	 * CPU after setting the request bit.
4557 	 *
4558 	 */
4559 
4560 	if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4561 		return 0;
4562 	if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4563 		return 0;
4564 
4565 	smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4566 
4567 	raw_spin_lock(&kvm_lock);
4568 	list_for_each_entry(kvm, &vm_list, vm_list) {
4569 		kvm_for_each_vcpu(i, vcpu, kvm) {
4570 			if (vcpu->cpu != freq->cpu)
4571 				continue;
4572 			kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4573 			if (vcpu->cpu != smp_processor_id())
4574 				send_ipi = 1;
4575 		}
4576 	}
4577 	raw_spin_unlock(&kvm_lock);
4578 
4579 	if (freq->old < freq->new && send_ipi) {
4580 		/*
4581 		 * We upscale the frequency.  Must make the guest
4582 		 * doesn't see old kvmclock values while running with
4583 		 * the new frequency, otherwise we risk the guest sees
4584 		 * time go backwards.
4585 		 *
4586 		 * In case we update the frequency for another cpu
4587 		 * (which might be in guest context) send an interrupt
4588 		 * to kick the cpu out of guest context.  Next time
4589 		 * guest context is entered kvmclock will be updated,
4590 		 * so the guest will not see stale values.
4591 		 */
4592 		smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4593 	}
4594 	return 0;
4595 }
4596 
4597 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4598 	.notifier_call  = kvmclock_cpufreq_notifier
4599 };
4600 
4601 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4602 					unsigned long action, void *hcpu)
4603 {
4604 	unsigned int cpu = (unsigned long)hcpu;
4605 
4606 	switch (action) {
4607 		case CPU_ONLINE:
4608 		case CPU_DOWN_FAILED:
4609 			smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4610 			break;
4611 		case CPU_DOWN_PREPARE:
4612 			smp_call_function_single(cpu, tsc_bad, NULL, 1);
4613 			break;
4614 	}
4615 	return NOTIFY_OK;
4616 }
4617 
4618 static struct notifier_block kvmclock_cpu_notifier_block = {
4619 	.notifier_call  = kvmclock_cpu_notifier,
4620 	.priority = -INT_MAX
4621 };
4622 
4623 static void kvm_timer_init(void)
4624 {
4625 	int cpu;
4626 
4627 	max_tsc_khz = tsc_khz;
4628 	register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4629 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4630 #ifdef CONFIG_CPU_FREQ
4631 		struct cpufreq_policy policy;
4632 		memset(&policy, 0, sizeof(policy));
4633 		cpu = get_cpu();
4634 		cpufreq_get_policy(&policy, cpu);
4635 		if (policy.cpuinfo.max_freq)
4636 			max_tsc_khz = policy.cpuinfo.max_freq;
4637 		put_cpu();
4638 #endif
4639 		cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4640 					  CPUFREQ_TRANSITION_NOTIFIER);
4641 	}
4642 	pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4643 	for_each_online_cpu(cpu)
4644 		smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4645 }
4646 
4647 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4648 
4649 int kvm_is_in_guest(void)
4650 {
4651 	return __this_cpu_read(current_vcpu) != NULL;
4652 }
4653 
4654 static int kvm_is_user_mode(void)
4655 {
4656 	int user_mode = 3;
4657 
4658 	if (__this_cpu_read(current_vcpu))
4659 		user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
4660 
4661 	return user_mode != 0;
4662 }
4663 
4664 static unsigned long kvm_get_guest_ip(void)
4665 {
4666 	unsigned long ip = 0;
4667 
4668 	if (__this_cpu_read(current_vcpu))
4669 		ip = kvm_rip_read(__this_cpu_read(current_vcpu));
4670 
4671 	return ip;
4672 }
4673 
4674 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4675 	.is_in_guest		= kvm_is_in_guest,
4676 	.is_user_mode		= kvm_is_user_mode,
4677 	.get_guest_ip		= kvm_get_guest_ip,
4678 };
4679 
4680 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4681 {
4682 	__this_cpu_write(current_vcpu, vcpu);
4683 }
4684 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4685 
4686 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4687 {
4688 	__this_cpu_write(current_vcpu, NULL);
4689 }
4690 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4691 
4692 static void kvm_set_mmio_spte_mask(void)
4693 {
4694 	u64 mask;
4695 	int maxphyaddr = boot_cpu_data.x86_phys_bits;
4696 
4697 	/*
4698 	 * Set the reserved bits and the present bit of an paging-structure
4699 	 * entry to generate page fault with PFER.RSV = 1.
4700 	 */
4701 	mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
4702 	mask |= 1ull;
4703 
4704 #ifdef CONFIG_X86_64
4705 	/*
4706 	 * If reserved bit is not supported, clear the present bit to disable
4707 	 * mmio page fault.
4708 	 */
4709 	if (maxphyaddr == 52)
4710 		mask &= ~1ull;
4711 #endif
4712 
4713 	kvm_mmu_set_mmio_spte_mask(mask);
4714 }
4715 
4716 int kvm_arch_init(void *opaque)
4717 {
4718 	int r;
4719 	struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4720 
4721 	if (kvm_x86_ops) {
4722 		printk(KERN_ERR "kvm: already loaded the other module\n");
4723 		r = -EEXIST;
4724 		goto out;
4725 	}
4726 
4727 	if (!ops->cpu_has_kvm_support()) {
4728 		printk(KERN_ERR "kvm: no hardware support\n");
4729 		r = -EOPNOTSUPP;
4730 		goto out;
4731 	}
4732 	if (ops->disabled_by_bios()) {
4733 		printk(KERN_ERR "kvm: disabled by bios\n");
4734 		r = -EOPNOTSUPP;
4735 		goto out;
4736 	}
4737 
4738 	r = kvm_mmu_module_init();
4739 	if (r)
4740 		goto out;
4741 
4742 	kvm_set_mmio_spte_mask();
4743 	kvm_init_msr_list();
4744 
4745 	kvm_x86_ops = ops;
4746 	kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4747 			PT_DIRTY_MASK, PT64_NX_MASK, 0);
4748 
4749 	kvm_timer_init();
4750 
4751 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
4752 
4753 	if (cpu_has_xsave)
4754 		host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4755 
4756 	return 0;
4757 
4758 out:
4759 	return r;
4760 }
4761 
4762 void kvm_arch_exit(void)
4763 {
4764 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4765 
4766 	if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4767 		cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4768 					    CPUFREQ_TRANSITION_NOTIFIER);
4769 	unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4770 	kvm_x86_ops = NULL;
4771 	kvm_mmu_module_exit();
4772 }
4773 
4774 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4775 {
4776 	++vcpu->stat.halt_exits;
4777 	if (irqchip_in_kernel(vcpu->kvm)) {
4778 		vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4779 		return 1;
4780 	} else {
4781 		vcpu->run->exit_reason = KVM_EXIT_HLT;
4782 		return 0;
4783 	}
4784 }
4785 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4786 
4787 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4788 {
4789 	u64 param, ingpa, outgpa, ret;
4790 	uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4791 	bool fast, longmode;
4792 	int cs_db, cs_l;
4793 
4794 	/*
4795 	 * hypercall generates UD from non zero cpl and real mode
4796 	 * per HYPER-V spec
4797 	 */
4798 	if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4799 		kvm_queue_exception(vcpu, UD_VECTOR);
4800 		return 0;
4801 	}
4802 
4803 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4804 	longmode = is_long_mode(vcpu) && cs_l == 1;
4805 
4806 	if (!longmode) {
4807 		param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4808 			(kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4809 		ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4810 			(kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4811 		outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4812 			(kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4813 	}
4814 #ifdef CONFIG_X86_64
4815 	else {
4816 		param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4817 		ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4818 		outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4819 	}
4820 #endif
4821 
4822 	code = param & 0xffff;
4823 	fast = (param >> 16) & 0x1;
4824 	rep_cnt = (param >> 32) & 0xfff;
4825 	rep_idx = (param >> 48) & 0xfff;
4826 
4827 	trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4828 
4829 	switch (code) {
4830 	case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4831 		kvm_vcpu_on_spin(vcpu);
4832 		break;
4833 	default:
4834 		res = HV_STATUS_INVALID_HYPERCALL_CODE;
4835 		break;
4836 	}
4837 
4838 	ret = res | (((u64)rep_done & 0xfff) << 32);
4839 	if (longmode) {
4840 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4841 	} else {
4842 		kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4843 		kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4844 	}
4845 
4846 	return 1;
4847 }
4848 
4849 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4850 {
4851 	unsigned long nr, a0, a1, a2, a3, ret;
4852 	int r = 1;
4853 
4854 	if (kvm_hv_hypercall_enabled(vcpu->kvm))
4855 		return kvm_hv_hypercall(vcpu);
4856 
4857 	nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4858 	a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4859 	a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4860 	a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4861 	a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4862 
4863 	trace_kvm_hypercall(nr, a0, a1, a2, a3);
4864 
4865 	if (!is_long_mode(vcpu)) {
4866 		nr &= 0xFFFFFFFF;
4867 		a0 &= 0xFFFFFFFF;
4868 		a1 &= 0xFFFFFFFF;
4869 		a2 &= 0xFFFFFFFF;
4870 		a3 &= 0xFFFFFFFF;
4871 	}
4872 
4873 	if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4874 		ret = -KVM_EPERM;
4875 		goto out;
4876 	}
4877 
4878 	switch (nr) {
4879 	case KVM_HC_VAPIC_POLL_IRQ:
4880 		ret = 0;
4881 		break;
4882 	default:
4883 		ret = -KVM_ENOSYS;
4884 		break;
4885 	}
4886 out:
4887 	kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4888 	++vcpu->stat.hypercalls;
4889 	return r;
4890 }
4891 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4892 
4893 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
4894 {
4895 	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4896 	char instruction[3];
4897 	unsigned long rip = kvm_rip_read(vcpu);
4898 
4899 	/*
4900 	 * Blow out the MMU to ensure that no other VCPU has an active mapping
4901 	 * to ensure that the updated hypercall appears atomically across all
4902 	 * VCPUs.
4903 	 */
4904 	kvm_mmu_zap_all(vcpu->kvm);
4905 
4906 	kvm_x86_ops->patch_hypercall(vcpu, instruction);
4907 
4908 	return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
4909 }
4910 
4911 /*
4912  * Check if userspace requested an interrupt window, and that the
4913  * interrupt window is open.
4914  *
4915  * No need to exit to userspace if we already have an interrupt queued.
4916  */
4917 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4918 {
4919 	return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4920 		vcpu->run->request_interrupt_window &&
4921 		kvm_arch_interrupt_allowed(vcpu));
4922 }
4923 
4924 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4925 {
4926 	struct kvm_run *kvm_run = vcpu->run;
4927 
4928 	kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4929 	kvm_run->cr8 = kvm_get_cr8(vcpu);
4930 	kvm_run->apic_base = kvm_get_apic_base(vcpu);
4931 	if (irqchip_in_kernel(vcpu->kvm))
4932 		kvm_run->ready_for_interrupt_injection = 1;
4933 	else
4934 		kvm_run->ready_for_interrupt_injection =
4935 			kvm_arch_interrupt_allowed(vcpu) &&
4936 			!kvm_cpu_has_interrupt(vcpu) &&
4937 			!kvm_event_needs_reinjection(vcpu);
4938 }
4939 
4940 static void vapic_enter(struct kvm_vcpu *vcpu)
4941 {
4942 	struct kvm_lapic *apic = vcpu->arch.apic;
4943 	struct page *page;
4944 
4945 	if (!apic || !apic->vapic_addr)
4946 		return;
4947 
4948 	page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4949 
4950 	vcpu->arch.apic->vapic_page = page;
4951 }
4952 
4953 static void vapic_exit(struct kvm_vcpu *vcpu)
4954 {
4955 	struct kvm_lapic *apic = vcpu->arch.apic;
4956 	int idx;
4957 
4958 	if (!apic || !apic->vapic_addr)
4959 		return;
4960 
4961 	idx = srcu_read_lock(&vcpu->kvm->srcu);
4962 	kvm_release_page_dirty(apic->vapic_page);
4963 	mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4964 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
4965 }
4966 
4967 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4968 {
4969 	int max_irr, tpr;
4970 
4971 	if (!kvm_x86_ops->update_cr8_intercept)
4972 		return;
4973 
4974 	if (!vcpu->arch.apic)
4975 		return;
4976 
4977 	if (!vcpu->arch.apic->vapic_addr)
4978 		max_irr = kvm_lapic_find_highest_irr(vcpu);
4979 	else
4980 		max_irr = -1;
4981 
4982 	if (max_irr != -1)
4983 		max_irr >>= 4;
4984 
4985 	tpr = kvm_lapic_get_cr8(vcpu);
4986 
4987 	kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4988 }
4989 
4990 static void inject_pending_event(struct kvm_vcpu *vcpu)
4991 {
4992 	/* try to reinject previous events if any */
4993 	if (vcpu->arch.exception.pending) {
4994 		trace_kvm_inj_exception(vcpu->arch.exception.nr,
4995 					vcpu->arch.exception.has_error_code,
4996 					vcpu->arch.exception.error_code);
4997 		kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4998 					  vcpu->arch.exception.has_error_code,
4999 					  vcpu->arch.exception.error_code,
5000 					  vcpu->arch.exception.reinject);
5001 		return;
5002 	}
5003 
5004 	if (vcpu->arch.nmi_injected) {
5005 		kvm_x86_ops->set_nmi(vcpu);
5006 		return;
5007 	}
5008 
5009 	if (vcpu->arch.interrupt.pending) {
5010 		kvm_x86_ops->set_irq(vcpu);
5011 		return;
5012 	}
5013 
5014 	/* try to inject new event if pending */
5015 	if (vcpu->arch.nmi_pending) {
5016 		if (kvm_x86_ops->nmi_allowed(vcpu)) {
5017 			--vcpu->arch.nmi_pending;
5018 			vcpu->arch.nmi_injected = true;
5019 			kvm_x86_ops->set_nmi(vcpu);
5020 		}
5021 	} else if (kvm_cpu_has_interrupt(vcpu)) {
5022 		if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5023 			kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5024 					    false);
5025 			kvm_x86_ops->set_irq(vcpu);
5026 		}
5027 	}
5028 }
5029 
5030 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5031 {
5032 	if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5033 			!vcpu->guest_xcr0_loaded) {
5034 		/* kvm_set_xcr() also depends on this */
5035 		xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5036 		vcpu->guest_xcr0_loaded = 1;
5037 	}
5038 }
5039 
5040 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5041 {
5042 	if (vcpu->guest_xcr0_loaded) {
5043 		if (vcpu->arch.xcr0 != host_xcr0)
5044 			xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5045 		vcpu->guest_xcr0_loaded = 0;
5046 	}
5047 }
5048 
5049 static void process_nmi(struct kvm_vcpu *vcpu)
5050 {
5051 	unsigned limit = 2;
5052 
5053 	/*
5054 	 * x86 is limited to one NMI running, and one NMI pending after it.
5055 	 * If an NMI is already in progress, limit further NMIs to just one.
5056 	 * Otherwise, allow two (and we'll inject the first one immediately).
5057 	 */
5058 	if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5059 		limit = 1;
5060 
5061 	vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5062 	vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5063 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5064 }
5065 
5066 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5067 {
5068 	int r;
5069 	bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5070 		vcpu->run->request_interrupt_window;
5071 	bool req_immediate_exit = 0;
5072 
5073 	if (vcpu->requests) {
5074 		if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5075 			kvm_mmu_unload(vcpu);
5076 		if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5077 			__kvm_migrate_timers(vcpu);
5078 		if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5079 			r = kvm_guest_time_update(vcpu);
5080 			if (unlikely(r))
5081 				goto out;
5082 		}
5083 		if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5084 			kvm_mmu_sync_roots(vcpu);
5085 		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5086 			kvm_x86_ops->tlb_flush(vcpu);
5087 		if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5088 			vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5089 			r = 0;
5090 			goto out;
5091 		}
5092 		if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5093 			vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5094 			r = 0;
5095 			goto out;
5096 		}
5097 		if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5098 			vcpu->fpu_active = 0;
5099 			kvm_x86_ops->fpu_deactivate(vcpu);
5100 		}
5101 		if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5102 			/* Page is swapped out. Do synthetic halt */
5103 			vcpu->arch.apf.halted = true;
5104 			r = 1;
5105 			goto out;
5106 		}
5107 		if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5108 			record_steal_time(vcpu);
5109 		if (kvm_check_request(KVM_REQ_NMI, vcpu))
5110 			process_nmi(vcpu);
5111 		req_immediate_exit =
5112 			kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5113 		if (kvm_check_request(KVM_REQ_PMU, vcpu))
5114 			kvm_handle_pmu_event(vcpu);
5115 		if (kvm_check_request(KVM_REQ_PMI, vcpu))
5116 			kvm_deliver_pmi(vcpu);
5117 	}
5118 
5119 	r = kvm_mmu_reload(vcpu);
5120 	if (unlikely(r))
5121 		goto out;
5122 
5123 	if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5124 		inject_pending_event(vcpu);
5125 
5126 		/* enable NMI/IRQ window open exits if needed */
5127 		if (vcpu->arch.nmi_pending)
5128 			kvm_x86_ops->enable_nmi_window(vcpu);
5129 		else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5130 			kvm_x86_ops->enable_irq_window(vcpu);
5131 
5132 		if (kvm_lapic_enabled(vcpu)) {
5133 			update_cr8_intercept(vcpu);
5134 			kvm_lapic_sync_to_vapic(vcpu);
5135 		}
5136 	}
5137 
5138 	preempt_disable();
5139 
5140 	kvm_x86_ops->prepare_guest_switch(vcpu);
5141 	if (vcpu->fpu_active)
5142 		kvm_load_guest_fpu(vcpu);
5143 	kvm_load_guest_xcr0(vcpu);
5144 
5145 	vcpu->mode = IN_GUEST_MODE;
5146 
5147 	/* We should set ->mode before check ->requests,
5148 	 * see the comment in make_all_cpus_request.
5149 	 */
5150 	smp_mb();
5151 
5152 	local_irq_disable();
5153 
5154 	if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5155 	    || need_resched() || signal_pending(current)) {
5156 		vcpu->mode = OUTSIDE_GUEST_MODE;
5157 		smp_wmb();
5158 		local_irq_enable();
5159 		preempt_enable();
5160 		kvm_x86_ops->cancel_injection(vcpu);
5161 		r = 1;
5162 		goto out;
5163 	}
5164 
5165 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5166 
5167 	if (req_immediate_exit)
5168 		smp_send_reschedule(vcpu->cpu);
5169 
5170 	kvm_guest_enter();
5171 
5172 	if (unlikely(vcpu->arch.switch_db_regs)) {
5173 		set_debugreg(0, 7);
5174 		set_debugreg(vcpu->arch.eff_db[0], 0);
5175 		set_debugreg(vcpu->arch.eff_db[1], 1);
5176 		set_debugreg(vcpu->arch.eff_db[2], 2);
5177 		set_debugreg(vcpu->arch.eff_db[3], 3);
5178 	}
5179 
5180 	trace_kvm_entry(vcpu->vcpu_id);
5181 	kvm_x86_ops->run(vcpu);
5182 
5183 	/*
5184 	 * If the guest has used debug registers, at least dr7
5185 	 * will be disabled while returning to the host.
5186 	 * If we don't have active breakpoints in the host, we don't
5187 	 * care about the messed up debug address registers. But if
5188 	 * we have some of them active, restore the old state.
5189 	 */
5190 	if (hw_breakpoint_active())
5191 		hw_breakpoint_restore();
5192 
5193 	vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
5194 
5195 	vcpu->mode = OUTSIDE_GUEST_MODE;
5196 	smp_wmb();
5197 	local_irq_enable();
5198 
5199 	++vcpu->stat.exits;
5200 
5201 	/*
5202 	 * We must have an instruction between local_irq_enable() and
5203 	 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5204 	 * the interrupt shadow.  The stat.exits increment will do nicely.
5205 	 * But we need to prevent reordering, hence this barrier():
5206 	 */
5207 	barrier();
5208 
5209 	kvm_guest_exit();
5210 
5211 	preempt_enable();
5212 
5213 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5214 
5215 	/*
5216 	 * Profile KVM exit RIPs:
5217 	 */
5218 	if (unlikely(prof_on == KVM_PROFILING)) {
5219 		unsigned long rip = kvm_rip_read(vcpu);
5220 		profile_hit(KVM_PROFILING, (void *)rip);
5221 	}
5222 
5223 
5224 	kvm_lapic_sync_from_vapic(vcpu);
5225 
5226 	r = kvm_x86_ops->handle_exit(vcpu);
5227 out:
5228 	return r;
5229 }
5230 
5231 
5232 static int __vcpu_run(struct kvm_vcpu *vcpu)
5233 {
5234 	int r;
5235 	struct kvm *kvm = vcpu->kvm;
5236 
5237 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5238 		pr_debug("vcpu %d received sipi with vector # %x\n",
5239 			 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5240 		kvm_lapic_reset(vcpu);
5241 		r = kvm_arch_vcpu_reset(vcpu);
5242 		if (r)
5243 			return r;
5244 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5245 	}
5246 
5247 	vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5248 	vapic_enter(vcpu);
5249 
5250 	r = 1;
5251 	while (r > 0) {
5252 		if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5253 		    !vcpu->arch.apf.halted)
5254 			r = vcpu_enter_guest(vcpu);
5255 		else {
5256 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5257 			kvm_vcpu_block(vcpu);
5258 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5259 			if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5260 			{
5261 				switch(vcpu->arch.mp_state) {
5262 				case KVM_MP_STATE_HALTED:
5263 					vcpu->arch.mp_state =
5264 						KVM_MP_STATE_RUNNABLE;
5265 				case KVM_MP_STATE_RUNNABLE:
5266 					vcpu->arch.apf.halted = false;
5267 					break;
5268 				case KVM_MP_STATE_SIPI_RECEIVED:
5269 				default:
5270 					r = -EINTR;
5271 					break;
5272 				}
5273 			}
5274 		}
5275 
5276 		if (r <= 0)
5277 			break;
5278 
5279 		clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5280 		if (kvm_cpu_has_pending_timer(vcpu))
5281 			kvm_inject_pending_timer_irqs(vcpu);
5282 
5283 		if (dm_request_for_irq_injection(vcpu)) {
5284 			r = -EINTR;
5285 			vcpu->run->exit_reason = KVM_EXIT_INTR;
5286 			++vcpu->stat.request_irq_exits;
5287 		}
5288 
5289 		kvm_check_async_pf_completion(vcpu);
5290 
5291 		if (signal_pending(current)) {
5292 			r = -EINTR;
5293 			vcpu->run->exit_reason = KVM_EXIT_INTR;
5294 			++vcpu->stat.signal_exits;
5295 		}
5296 		if (need_resched()) {
5297 			srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5298 			kvm_resched(vcpu);
5299 			vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5300 		}
5301 	}
5302 
5303 	srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5304 
5305 	vapic_exit(vcpu);
5306 
5307 	return r;
5308 }
5309 
5310 static int complete_mmio(struct kvm_vcpu *vcpu)
5311 {
5312 	struct kvm_run *run = vcpu->run;
5313 	int r;
5314 
5315 	if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5316 		return 1;
5317 
5318 	if (vcpu->mmio_needed) {
5319 		vcpu->mmio_needed = 0;
5320 		if (!vcpu->mmio_is_write)
5321 			memcpy(vcpu->mmio_data + vcpu->mmio_index,
5322 			       run->mmio.data, 8);
5323 		vcpu->mmio_index += 8;
5324 		if (vcpu->mmio_index < vcpu->mmio_size) {
5325 			run->exit_reason = KVM_EXIT_MMIO;
5326 			run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5327 			memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5328 			run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5329 			run->mmio.is_write = vcpu->mmio_is_write;
5330 			vcpu->mmio_needed = 1;
5331 			return 0;
5332 		}
5333 		if (vcpu->mmio_is_write)
5334 			return 1;
5335 		vcpu->mmio_read_completed = 1;
5336 	}
5337 	vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5338 	r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5339 	srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5340 	if (r != EMULATE_DONE)
5341 		return 0;
5342 	return 1;
5343 }
5344 
5345 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5346 {
5347 	int r;
5348 	sigset_t sigsaved;
5349 
5350 	if (!tsk_used_math(current) && init_fpu(current))
5351 		return -ENOMEM;
5352 
5353 	if (vcpu->sigset_active)
5354 		sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5355 
5356 	if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5357 		kvm_vcpu_block(vcpu);
5358 		clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5359 		r = -EAGAIN;
5360 		goto out;
5361 	}
5362 
5363 	/* re-sync apic's tpr */
5364 	if (!irqchip_in_kernel(vcpu->kvm)) {
5365 		if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5366 			r = -EINVAL;
5367 			goto out;
5368 		}
5369 	}
5370 
5371 	r = complete_mmio(vcpu);
5372 	if (r <= 0)
5373 		goto out;
5374 
5375 	r = __vcpu_run(vcpu);
5376 
5377 out:
5378 	post_kvm_run_save(vcpu);
5379 	if (vcpu->sigset_active)
5380 		sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5381 
5382 	return r;
5383 }
5384 
5385 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5386 {
5387 	if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5388 		/*
5389 		 * We are here if userspace calls get_regs() in the middle of
5390 		 * instruction emulation. Registers state needs to be copied
5391 		 * back from emulation context to vcpu. Usrapace shouldn't do
5392 		 * that usually, but some bad designed PV devices (vmware
5393 		 * backdoor interface) need this to work
5394 		 */
5395 		struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5396 		memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5397 		vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5398 	}
5399 	regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5400 	regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5401 	regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5402 	regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5403 	regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5404 	regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5405 	regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5406 	regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5407 #ifdef CONFIG_X86_64
5408 	regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5409 	regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5410 	regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5411 	regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5412 	regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5413 	regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5414 	regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5415 	regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5416 #endif
5417 
5418 	regs->rip = kvm_rip_read(vcpu);
5419 	regs->rflags = kvm_get_rflags(vcpu);
5420 
5421 	return 0;
5422 }
5423 
5424 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5425 {
5426 	vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5427 	vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5428 
5429 	kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5430 	kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5431 	kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5432 	kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5433 	kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5434 	kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5435 	kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5436 	kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5437 #ifdef CONFIG_X86_64
5438 	kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5439 	kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5440 	kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5441 	kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5442 	kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5443 	kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5444 	kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5445 	kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5446 #endif
5447 
5448 	kvm_rip_write(vcpu, regs->rip);
5449 	kvm_set_rflags(vcpu, regs->rflags);
5450 
5451 	vcpu->arch.exception.pending = false;
5452 
5453 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5454 
5455 	return 0;
5456 }
5457 
5458 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5459 {
5460 	struct kvm_segment cs;
5461 
5462 	kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5463 	*db = cs.db;
5464 	*l = cs.l;
5465 }
5466 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5467 
5468 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5469 				  struct kvm_sregs *sregs)
5470 {
5471 	struct desc_ptr dt;
5472 
5473 	kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5474 	kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5475 	kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5476 	kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5477 	kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5478 	kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5479 
5480 	kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5481 	kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5482 
5483 	kvm_x86_ops->get_idt(vcpu, &dt);
5484 	sregs->idt.limit = dt.size;
5485 	sregs->idt.base = dt.address;
5486 	kvm_x86_ops->get_gdt(vcpu, &dt);
5487 	sregs->gdt.limit = dt.size;
5488 	sregs->gdt.base = dt.address;
5489 
5490 	sregs->cr0 = kvm_read_cr0(vcpu);
5491 	sregs->cr2 = vcpu->arch.cr2;
5492 	sregs->cr3 = kvm_read_cr3(vcpu);
5493 	sregs->cr4 = kvm_read_cr4(vcpu);
5494 	sregs->cr8 = kvm_get_cr8(vcpu);
5495 	sregs->efer = vcpu->arch.efer;
5496 	sregs->apic_base = kvm_get_apic_base(vcpu);
5497 
5498 	memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5499 
5500 	if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5501 		set_bit(vcpu->arch.interrupt.nr,
5502 			(unsigned long *)sregs->interrupt_bitmap);
5503 
5504 	return 0;
5505 }
5506 
5507 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5508 				    struct kvm_mp_state *mp_state)
5509 {
5510 	mp_state->mp_state = vcpu->arch.mp_state;
5511 	return 0;
5512 }
5513 
5514 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5515 				    struct kvm_mp_state *mp_state)
5516 {
5517 	vcpu->arch.mp_state = mp_state->mp_state;
5518 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5519 	return 0;
5520 }
5521 
5522 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5523 		    bool has_error_code, u32 error_code)
5524 {
5525 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5526 	int ret;
5527 
5528 	init_emulate_ctxt(vcpu);
5529 
5530 	ret = emulator_task_switch(ctxt, tss_selector, reason,
5531 				   has_error_code, error_code);
5532 
5533 	if (ret)
5534 		return EMULATE_FAIL;
5535 
5536 	memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5537 	kvm_rip_write(vcpu, ctxt->eip);
5538 	kvm_set_rflags(vcpu, ctxt->eflags);
5539 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5540 	return EMULATE_DONE;
5541 }
5542 EXPORT_SYMBOL_GPL(kvm_task_switch);
5543 
5544 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5545 				  struct kvm_sregs *sregs)
5546 {
5547 	int mmu_reset_needed = 0;
5548 	int pending_vec, max_bits, idx;
5549 	struct desc_ptr dt;
5550 
5551 	dt.size = sregs->idt.limit;
5552 	dt.address = sregs->idt.base;
5553 	kvm_x86_ops->set_idt(vcpu, &dt);
5554 	dt.size = sregs->gdt.limit;
5555 	dt.address = sregs->gdt.base;
5556 	kvm_x86_ops->set_gdt(vcpu, &dt);
5557 
5558 	vcpu->arch.cr2 = sregs->cr2;
5559 	mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5560 	vcpu->arch.cr3 = sregs->cr3;
5561 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5562 
5563 	kvm_set_cr8(vcpu, sregs->cr8);
5564 
5565 	mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5566 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
5567 	kvm_set_apic_base(vcpu, sregs->apic_base);
5568 
5569 	mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5570 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5571 	vcpu->arch.cr0 = sregs->cr0;
5572 
5573 	mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5574 	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5575 	if (sregs->cr4 & X86_CR4_OSXSAVE)
5576 		kvm_update_cpuid(vcpu);
5577 
5578 	idx = srcu_read_lock(&vcpu->kvm->srcu);
5579 	if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5580 		load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5581 		mmu_reset_needed = 1;
5582 	}
5583 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
5584 
5585 	if (mmu_reset_needed)
5586 		kvm_mmu_reset_context(vcpu);
5587 
5588 	max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5589 	pending_vec = find_first_bit(
5590 		(const unsigned long *)sregs->interrupt_bitmap, max_bits);
5591 	if (pending_vec < max_bits) {
5592 		kvm_queue_interrupt(vcpu, pending_vec, false);
5593 		pr_debug("Set back pending irq %d\n", pending_vec);
5594 	}
5595 
5596 	kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5597 	kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5598 	kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5599 	kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5600 	kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5601 	kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5602 
5603 	kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5604 	kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5605 
5606 	update_cr8_intercept(vcpu);
5607 
5608 	/* Older userspace won't unhalt the vcpu on reset. */
5609 	if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5610 	    sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5611 	    !is_protmode(vcpu))
5612 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5613 
5614 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5615 
5616 	return 0;
5617 }
5618 
5619 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5620 					struct kvm_guest_debug *dbg)
5621 {
5622 	unsigned long rflags;
5623 	int i, r;
5624 
5625 	if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5626 		r = -EBUSY;
5627 		if (vcpu->arch.exception.pending)
5628 			goto out;
5629 		if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5630 			kvm_queue_exception(vcpu, DB_VECTOR);
5631 		else
5632 			kvm_queue_exception(vcpu, BP_VECTOR);
5633 	}
5634 
5635 	/*
5636 	 * Read rflags as long as potentially injected trace flags are still
5637 	 * filtered out.
5638 	 */
5639 	rflags = kvm_get_rflags(vcpu);
5640 
5641 	vcpu->guest_debug = dbg->control;
5642 	if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5643 		vcpu->guest_debug = 0;
5644 
5645 	if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5646 		for (i = 0; i < KVM_NR_DB_REGS; ++i)
5647 			vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5648 		vcpu->arch.switch_db_regs =
5649 			(dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5650 	} else {
5651 		for (i = 0; i < KVM_NR_DB_REGS; i++)
5652 			vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5653 		vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5654 	}
5655 
5656 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5657 		vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5658 			get_segment_base(vcpu, VCPU_SREG_CS);
5659 
5660 	/*
5661 	 * Trigger an rflags update that will inject or remove the trace
5662 	 * flags.
5663 	 */
5664 	kvm_set_rflags(vcpu, rflags);
5665 
5666 	kvm_x86_ops->set_guest_debug(vcpu, dbg);
5667 
5668 	r = 0;
5669 
5670 out:
5671 
5672 	return r;
5673 }
5674 
5675 /*
5676  * Translate a guest virtual address to a guest physical address.
5677  */
5678 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5679 				    struct kvm_translation *tr)
5680 {
5681 	unsigned long vaddr = tr->linear_address;
5682 	gpa_t gpa;
5683 	int idx;
5684 
5685 	idx = srcu_read_lock(&vcpu->kvm->srcu);
5686 	gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5687 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
5688 	tr->physical_address = gpa;
5689 	tr->valid = gpa != UNMAPPED_GVA;
5690 	tr->writeable = 1;
5691 	tr->usermode = 0;
5692 
5693 	return 0;
5694 }
5695 
5696 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5697 {
5698 	struct i387_fxsave_struct *fxsave =
5699 			&vcpu->arch.guest_fpu.state->fxsave;
5700 
5701 	memcpy(fpu->fpr, fxsave->st_space, 128);
5702 	fpu->fcw = fxsave->cwd;
5703 	fpu->fsw = fxsave->swd;
5704 	fpu->ftwx = fxsave->twd;
5705 	fpu->last_opcode = fxsave->fop;
5706 	fpu->last_ip = fxsave->rip;
5707 	fpu->last_dp = fxsave->rdp;
5708 	memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5709 
5710 	return 0;
5711 }
5712 
5713 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5714 {
5715 	struct i387_fxsave_struct *fxsave =
5716 			&vcpu->arch.guest_fpu.state->fxsave;
5717 
5718 	memcpy(fxsave->st_space, fpu->fpr, 128);
5719 	fxsave->cwd = fpu->fcw;
5720 	fxsave->swd = fpu->fsw;
5721 	fxsave->twd = fpu->ftwx;
5722 	fxsave->fop = fpu->last_opcode;
5723 	fxsave->rip = fpu->last_ip;
5724 	fxsave->rdp = fpu->last_dp;
5725 	memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5726 
5727 	return 0;
5728 }
5729 
5730 int fx_init(struct kvm_vcpu *vcpu)
5731 {
5732 	int err;
5733 
5734 	err = fpu_alloc(&vcpu->arch.guest_fpu);
5735 	if (err)
5736 		return err;
5737 
5738 	fpu_finit(&vcpu->arch.guest_fpu);
5739 
5740 	/*
5741 	 * Ensure guest xcr0 is valid for loading
5742 	 */
5743 	vcpu->arch.xcr0 = XSTATE_FP;
5744 
5745 	vcpu->arch.cr0 |= X86_CR0_ET;
5746 
5747 	return 0;
5748 }
5749 EXPORT_SYMBOL_GPL(fx_init);
5750 
5751 static void fx_free(struct kvm_vcpu *vcpu)
5752 {
5753 	fpu_free(&vcpu->arch.guest_fpu);
5754 }
5755 
5756 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5757 {
5758 	if (vcpu->guest_fpu_loaded)
5759 		return;
5760 
5761 	/*
5762 	 * Restore all possible states in the guest,
5763 	 * and assume host would use all available bits.
5764 	 * Guest xcr0 would be loaded later.
5765 	 */
5766 	kvm_put_guest_xcr0(vcpu);
5767 	vcpu->guest_fpu_loaded = 1;
5768 	unlazy_fpu(current);
5769 	fpu_restore_checking(&vcpu->arch.guest_fpu);
5770 	trace_kvm_fpu(1);
5771 }
5772 
5773 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5774 {
5775 	kvm_put_guest_xcr0(vcpu);
5776 
5777 	if (!vcpu->guest_fpu_loaded)
5778 		return;
5779 
5780 	vcpu->guest_fpu_loaded = 0;
5781 	fpu_save_init(&vcpu->arch.guest_fpu);
5782 	++vcpu->stat.fpu_reload;
5783 	kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5784 	trace_kvm_fpu(0);
5785 }
5786 
5787 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5788 {
5789 	kvmclock_reset(vcpu);
5790 
5791 	free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5792 	fx_free(vcpu);
5793 	kvm_x86_ops->vcpu_free(vcpu);
5794 }
5795 
5796 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5797 						unsigned int id)
5798 {
5799 	if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5800 		printk_once(KERN_WARNING
5801 		"kvm: SMP vm created on host with unstable TSC; "
5802 		"guest TSC will not be reliable\n");
5803 	return kvm_x86_ops->vcpu_create(kvm, id);
5804 }
5805 
5806 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5807 {
5808 	int r;
5809 
5810 	vcpu->arch.mtrr_state.have_fixed = 1;
5811 	vcpu_load(vcpu);
5812 	r = kvm_arch_vcpu_reset(vcpu);
5813 	if (r == 0)
5814 		r = kvm_mmu_setup(vcpu);
5815 	vcpu_put(vcpu);
5816 
5817 	return r;
5818 }
5819 
5820 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5821 {
5822 	vcpu->arch.apf.msr_val = 0;
5823 
5824 	vcpu_load(vcpu);
5825 	kvm_mmu_unload(vcpu);
5826 	vcpu_put(vcpu);
5827 
5828 	fx_free(vcpu);
5829 	kvm_x86_ops->vcpu_free(vcpu);
5830 }
5831 
5832 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5833 {
5834 	atomic_set(&vcpu->arch.nmi_queued, 0);
5835 	vcpu->arch.nmi_pending = 0;
5836 	vcpu->arch.nmi_injected = false;
5837 
5838 	vcpu->arch.switch_db_regs = 0;
5839 	memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5840 	vcpu->arch.dr6 = DR6_FIXED_1;
5841 	vcpu->arch.dr7 = DR7_FIXED_1;
5842 
5843 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5844 	vcpu->arch.apf.msr_val = 0;
5845 	vcpu->arch.st.msr_val = 0;
5846 
5847 	kvmclock_reset(vcpu);
5848 
5849 	kvm_clear_async_pf_completion_queue(vcpu);
5850 	kvm_async_pf_hash_reset(vcpu);
5851 	vcpu->arch.apf.halted = false;
5852 
5853 	kvm_pmu_reset(vcpu);
5854 
5855 	return kvm_x86_ops->vcpu_reset(vcpu);
5856 }
5857 
5858 int kvm_arch_hardware_enable(void *garbage)
5859 {
5860 	struct kvm *kvm;
5861 	struct kvm_vcpu *vcpu;
5862 	int i;
5863 
5864 	kvm_shared_msr_cpu_online();
5865 	list_for_each_entry(kvm, &vm_list, vm_list)
5866 		kvm_for_each_vcpu(i, vcpu, kvm)
5867 			if (vcpu->cpu == smp_processor_id())
5868 				kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5869 	return kvm_x86_ops->hardware_enable(garbage);
5870 }
5871 
5872 void kvm_arch_hardware_disable(void *garbage)
5873 {
5874 	kvm_x86_ops->hardware_disable(garbage);
5875 	drop_user_return_notifiers(garbage);
5876 }
5877 
5878 int kvm_arch_hardware_setup(void)
5879 {
5880 	return kvm_x86_ops->hardware_setup();
5881 }
5882 
5883 void kvm_arch_hardware_unsetup(void)
5884 {
5885 	kvm_x86_ops->hardware_unsetup();
5886 }
5887 
5888 void kvm_arch_check_processor_compat(void *rtn)
5889 {
5890 	kvm_x86_ops->check_processor_compatibility(rtn);
5891 }
5892 
5893 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5894 {
5895 	struct page *page;
5896 	struct kvm *kvm;
5897 	int r;
5898 
5899 	BUG_ON(vcpu->kvm == NULL);
5900 	kvm = vcpu->kvm;
5901 
5902 	vcpu->arch.emulate_ctxt.ops = &emulate_ops;
5903 	if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5904 		vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5905 	else
5906 		vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5907 
5908 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5909 	if (!page) {
5910 		r = -ENOMEM;
5911 		goto fail;
5912 	}
5913 	vcpu->arch.pio_data = page_address(page);
5914 
5915 	kvm_init_tsc_catchup(vcpu, max_tsc_khz);
5916 
5917 	r = kvm_mmu_create(vcpu);
5918 	if (r < 0)
5919 		goto fail_free_pio_data;
5920 
5921 	if (irqchip_in_kernel(kvm)) {
5922 		r = kvm_create_lapic(vcpu);
5923 		if (r < 0)
5924 			goto fail_mmu_destroy;
5925 	}
5926 
5927 	vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5928 				       GFP_KERNEL);
5929 	if (!vcpu->arch.mce_banks) {
5930 		r = -ENOMEM;
5931 		goto fail_free_lapic;
5932 	}
5933 	vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5934 
5935 	if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5936 		goto fail_free_mce_banks;
5937 
5938 	kvm_async_pf_hash_reset(vcpu);
5939 	kvm_pmu_init(vcpu);
5940 
5941 	return 0;
5942 fail_free_mce_banks:
5943 	kfree(vcpu->arch.mce_banks);
5944 fail_free_lapic:
5945 	kvm_free_lapic(vcpu);
5946 fail_mmu_destroy:
5947 	kvm_mmu_destroy(vcpu);
5948 fail_free_pio_data:
5949 	free_page((unsigned long)vcpu->arch.pio_data);
5950 fail:
5951 	return r;
5952 }
5953 
5954 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5955 {
5956 	int idx;
5957 
5958 	kvm_pmu_destroy(vcpu);
5959 	kfree(vcpu->arch.mce_banks);
5960 	kvm_free_lapic(vcpu);
5961 	idx = srcu_read_lock(&vcpu->kvm->srcu);
5962 	kvm_mmu_destroy(vcpu);
5963 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
5964 	free_page((unsigned long)vcpu->arch.pio_data);
5965 }
5966 
5967 int kvm_arch_init_vm(struct kvm *kvm)
5968 {
5969 	INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5970 	INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5971 
5972 	/* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5973 	set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5974 
5975 	raw_spin_lock_init(&kvm->arch.tsc_write_lock);
5976 
5977 	return 0;
5978 }
5979 
5980 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5981 {
5982 	vcpu_load(vcpu);
5983 	kvm_mmu_unload(vcpu);
5984 	vcpu_put(vcpu);
5985 }
5986 
5987 static void kvm_free_vcpus(struct kvm *kvm)
5988 {
5989 	unsigned int i;
5990 	struct kvm_vcpu *vcpu;
5991 
5992 	/*
5993 	 * Unpin any mmu pages first.
5994 	 */
5995 	kvm_for_each_vcpu(i, vcpu, kvm) {
5996 		kvm_clear_async_pf_completion_queue(vcpu);
5997 		kvm_unload_vcpu_mmu(vcpu);
5998 	}
5999 	kvm_for_each_vcpu(i, vcpu, kvm)
6000 		kvm_arch_vcpu_free(vcpu);
6001 
6002 	mutex_lock(&kvm->lock);
6003 	for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6004 		kvm->vcpus[i] = NULL;
6005 
6006 	atomic_set(&kvm->online_vcpus, 0);
6007 	mutex_unlock(&kvm->lock);
6008 }
6009 
6010 void kvm_arch_sync_events(struct kvm *kvm)
6011 {
6012 	kvm_free_all_assigned_devices(kvm);
6013 	kvm_free_pit(kvm);
6014 }
6015 
6016 void kvm_arch_destroy_vm(struct kvm *kvm)
6017 {
6018 	kvm_iommu_unmap_guest(kvm);
6019 	kfree(kvm->arch.vpic);
6020 	kfree(kvm->arch.vioapic);
6021 	kvm_free_vcpus(kvm);
6022 	if (kvm->arch.apic_access_page)
6023 		put_page(kvm->arch.apic_access_page);
6024 	if (kvm->arch.ept_identity_pagetable)
6025 		put_page(kvm->arch.ept_identity_pagetable);
6026 }
6027 
6028 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6029 				struct kvm_memory_slot *memslot,
6030 				struct kvm_memory_slot old,
6031 				struct kvm_userspace_memory_region *mem,
6032 				int user_alloc)
6033 {
6034 	int npages = memslot->npages;
6035 	int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6036 
6037 	/* Prevent internal slot pages from being moved by fork()/COW. */
6038 	if (memslot->id >= KVM_MEMORY_SLOTS)
6039 		map_flags = MAP_SHARED | MAP_ANONYMOUS;
6040 
6041 	/*To keep backward compatibility with older userspace,
6042 	 *x86 needs to hanlde !user_alloc case.
6043 	 */
6044 	if (!user_alloc) {
6045 		if (npages && !old.rmap) {
6046 			unsigned long userspace_addr;
6047 
6048 			down_write(&current->mm->mmap_sem);
6049 			userspace_addr = do_mmap(NULL, 0,
6050 						 npages * PAGE_SIZE,
6051 						 PROT_READ | PROT_WRITE,
6052 						 map_flags,
6053 						 0);
6054 			up_write(&current->mm->mmap_sem);
6055 
6056 			if (IS_ERR((void *)userspace_addr))
6057 				return PTR_ERR((void *)userspace_addr);
6058 
6059 			memslot->userspace_addr = userspace_addr;
6060 		}
6061 	}
6062 
6063 
6064 	return 0;
6065 }
6066 
6067 void kvm_arch_commit_memory_region(struct kvm *kvm,
6068 				struct kvm_userspace_memory_region *mem,
6069 				struct kvm_memory_slot old,
6070 				int user_alloc)
6071 {
6072 
6073 	int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6074 
6075 	if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6076 		int ret;
6077 
6078 		down_write(&current->mm->mmap_sem);
6079 		ret = do_munmap(current->mm, old.userspace_addr,
6080 				old.npages * PAGE_SIZE);
6081 		up_write(&current->mm->mmap_sem);
6082 		if (ret < 0)
6083 			printk(KERN_WARNING
6084 			       "kvm_vm_ioctl_set_memory_region: "
6085 			       "failed to munmap memory\n");
6086 	}
6087 
6088 	if (!kvm->arch.n_requested_mmu_pages)
6089 		nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6090 
6091 	spin_lock(&kvm->mmu_lock);
6092 	if (nr_mmu_pages)
6093 		kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6094 	kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6095 	spin_unlock(&kvm->mmu_lock);
6096 }
6097 
6098 void kvm_arch_flush_shadow(struct kvm *kvm)
6099 {
6100 	kvm_mmu_zap_all(kvm);
6101 	kvm_reload_remote_mmus(kvm);
6102 }
6103 
6104 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6105 {
6106 	return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6107 		!vcpu->arch.apf.halted)
6108 		|| !list_empty_careful(&vcpu->async_pf.done)
6109 		|| vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6110 		|| atomic_read(&vcpu->arch.nmi_queued) ||
6111 		(kvm_arch_interrupt_allowed(vcpu) &&
6112 		 kvm_cpu_has_interrupt(vcpu));
6113 }
6114 
6115 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6116 {
6117 	int me;
6118 	int cpu = vcpu->cpu;
6119 
6120 	if (waitqueue_active(&vcpu->wq)) {
6121 		wake_up_interruptible(&vcpu->wq);
6122 		++vcpu->stat.halt_wakeup;
6123 	}
6124 
6125 	me = get_cpu();
6126 	if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6127 		if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6128 			smp_send_reschedule(cpu);
6129 	put_cpu();
6130 }
6131 
6132 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6133 {
6134 	return kvm_x86_ops->interrupt_allowed(vcpu);
6135 }
6136 
6137 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6138 {
6139 	unsigned long current_rip = kvm_rip_read(vcpu) +
6140 		get_segment_base(vcpu, VCPU_SREG_CS);
6141 
6142 	return current_rip == linear_rip;
6143 }
6144 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6145 
6146 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6147 {
6148 	unsigned long rflags;
6149 
6150 	rflags = kvm_x86_ops->get_rflags(vcpu);
6151 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6152 		rflags &= ~X86_EFLAGS_TF;
6153 	return rflags;
6154 }
6155 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6156 
6157 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6158 {
6159 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6160 	    kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6161 		rflags |= X86_EFLAGS_TF;
6162 	kvm_x86_ops->set_rflags(vcpu, rflags);
6163 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6164 }
6165 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6166 
6167 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6168 {
6169 	int r;
6170 
6171 	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6172 	      is_error_page(work->page))
6173 		return;
6174 
6175 	r = kvm_mmu_reload(vcpu);
6176 	if (unlikely(r))
6177 		return;
6178 
6179 	if (!vcpu->arch.mmu.direct_map &&
6180 	      work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6181 		return;
6182 
6183 	vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6184 }
6185 
6186 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6187 {
6188 	return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6189 }
6190 
6191 static inline u32 kvm_async_pf_next_probe(u32 key)
6192 {
6193 	return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6194 }
6195 
6196 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6197 {
6198 	u32 key = kvm_async_pf_hash_fn(gfn);
6199 
6200 	while (vcpu->arch.apf.gfns[key] != ~0)
6201 		key = kvm_async_pf_next_probe(key);
6202 
6203 	vcpu->arch.apf.gfns[key] = gfn;
6204 }
6205 
6206 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6207 {
6208 	int i;
6209 	u32 key = kvm_async_pf_hash_fn(gfn);
6210 
6211 	for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6212 		     (vcpu->arch.apf.gfns[key] != gfn &&
6213 		      vcpu->arch.apf.gfns[key] != ~0); i++)
6214 		key = kvm_async_pf_next_probe(key);
6215 
6216 	return key;
6217 }
6218 
6219 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6220 {
6221 	return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6222 }
6223 
6224 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6225 {
6226 	u32 i, j, k;
6227 
6228 	i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6229 	while (true) {
6230 		vcpu->arch.apf.gfns[i] = ~0;
6231 		do {
6232 			j = kvm_async_pf_next_probe(j);
6233 			if (vcpu->arch.apf.gfns[j] == ~0)
6234 				return;
6235 			k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6236 			/*
6237 			 * k lies cyclically in ]i,j]
6238 			 * |    i.k.j |
6239 			 * |....j i.k.| or  |.k..j i...|
6240 			 */
6241 		} while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6242 		vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6243 		i = j;
6244 	}
6245 }
6246 
6247 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6248 {
6249 
6250 	return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6251 				      sizeof(val));
6252 }
6253 
6254 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6255 				     struct kvm_async_pf *work)
6256 {
6257 	struct x86_exception fault;
6258 
6259 	trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6260 	kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6261 
6262 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6263 	    (vcpu->arch.apf.send_user_only &&
6264 	     kvm_x86_ops->get_cpl(vcpu) == 0))
6265 		kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6266 	else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6267 		fault.vector = PF_VECTOR;
6268 		fault.error_code_valid = true;
6269 		fault.error_code = 0;
6270 		fault.nested_page_fault = false;
6271 		fault.address = work->arch.token;
6272 		kvm_inject_page_fault(vcpu, &fault);
6273 	}
6274 }
6275 
6276 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6277 				 struct kvm_async_pf *work)
6278 {
6279 	struct x86_exception fault;
6280 
6281 	trace_kvm_async_pf_ready(work->arch.token, work->gva);
6282 	if (is_error_page(work->page))
6283 		work->arch.token = ~0; /* broadcast wakeup */
6284 	else
6285 		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6286 
6287 	if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6288 	    !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6289 		fault.vector = PF_VECTOR;
6290 		fault.error_code_valid = true;
6291 		fault.error_code = 0;
6292 		fault.nested_page_fault = false;
6293 		fault.address = work->arch.token;
6294 		kvm_inject_page_fault(vcpu, &fault);
6295 	}
6296 	vcpu->arch.apf.halted = false;
6297 }
6298 
6299 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6300 {
6301 	if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6302 		return true;
6303 	else
6304 		return !kvm_event_needs_reinjection(vcpu) &&
6305 			kvm_x86_ops->interrupt_allowed(vcpu);
6306 }
6307 
6308 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6309 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6310 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6311 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6312 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6313 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6314 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6315 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6316 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6317 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6318 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6319 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
6320