xref: /linux/arch/x86/kvm/vmx/nested.c (revision c5951e7c8ee5cb04b8b41c32bf567b90117a2124)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/frame.h>
4 #include <linux/percpu.h>
5 
6 #include <asm/debugreg.h>
7 #include <asm/mmu_context.h>
8 
9 #include "cpuid.h"
10 #include "hyperv.h"
11 #include "mmu.h"
12 #include "nested.h"
13 #include "pmu.h"
14 #include "trace.h"
15 #include "x86.h"
16 
17 static bool __read_mostly enable_shadow_vmcs = 1;
18 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
19 
20 static bool __read_mostly nested_early_check = 0;
21 module_param(nested_early_check, bool, S_IRUGO);
22 
23 #define CC(consistency_check)						\
24 ({									\
25 	bool failed = (consistency_check);				\
26 	if (failed)							\
27 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
28 	failed;								\
29 })
30 
31 /*
32  * Hyper-V requires all of these, so mark them as supported even though
33  * they are just treated the same as all-context.
34  */
35 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
36 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
37 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
38 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
39 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
40 
41 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
42 
43 enum {
44 	VMX_VMREAD_BITMAP,
45 	VMX_VMWRITE_BITMAP,
46 	VMX_BITMAP_NR
47 };
48 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
49 
50 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
51 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
52 
53 struct shadow_vmcs_field {
54 	u16	encoding;
55 	u16	offset;
56 };
57 static struct shadow_vmcs_field shadow_read_only_fields[] = {
58 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
59 #include "vmcs_shadow_fields.h"
60 };
61 static int max_shadow_read_only_fields =
62 	ARRAY_SIZE(shadow_read_only_fields);
63 
64 static struct shadow_vmcs_field shadow_read_write_fields[] = {
65 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
66 #include "vmcs_shadow_fields.h"
67 };
68 static int max_shadow_read_write_fields =
69 	ARRAY_SIZE(shadow_read_write_fields);
70 
71 static void init_vmcs_shadow_fields(void)
72 {
73 	int i, j;
74 
75 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
76 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
77 
78 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
79 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
80 		u16 field = entry.encoding;
81 
82 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
83 		    (i + 1 == max_shadow_read_only_fields ||
84 		     shadow_read_only_fields[i + 1].encoding != field + 1))
85 			pr_err("Missing field from shadow_read_only_field %x\n",
86 			       field + 1);
87 
88 		clear_bit(field, vmx_vmread_bitmap);
89 		if (field & 1)
90 #ifdef CONFIG_X86_64
91 			continue;
92 #else
93 			entry.offset += sizeof(u32);
94 #endif
95 		shadow_read_only_fields[j++] = entry;
96 	}
97 	max_shadow_read_only_fields = j;
98 
99 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
100 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
101 		u16 field = entry.encoding;
102 
103 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
104 		    (i + 1 == max_shadow_read_write_fields ||
105 		     shadow_read_write_fields[i + 1].encoding != field + 1))
106 			pr_err("Missing field from shadow_read_write_field %x\n",
107 			       field + 1);
108 
109 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
110 			  field <= GUEST_TR_AR_BYTES,
111 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
112 
113 		/*
114 		 * PML and the preemption timer can be emulated, but the
115 		 * processor cannot vmwrite to fields that don't exist
116 		 * on bare metal.
117 		 */
118 		switch (field) {
119 		case GUEST_PML_INDEX:
120 			if (!cpu_has_vmx_pml())
121 				continue;
122 			break;
123 		case VMX_PREEMPTION_TIMER_VALUE:
124 			if (!cpu_has_vmx_preemption_timer())
125 				continue;
126 			break;
127 		case GUEST_INTR_STATUS:
128 			if (!cpu_has_vmx_apicv())
129 				continue;
130 			break;
131 		default:
132 			break;
133 		}
134 
135 		clear_bit(field, vmx_vmwrite_bitmap);
136 		clear_bit(field, vmx_vmread_bitmap);
137 		if (field & 1)
138 #ifdef CONFIG_X86_64
139 			continue;
140 #else
141 			entry.offset += sizeof(u32);
142 #endif
143 		shadow_read_write_fields[j++] = entry;
144 	}
145 	max_shadow_read_write_fields = j;
146 }
147 
148 /*
149  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
150  * set the success or error code of an emulated VMX instruction (as specified
151  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
152  * instruction.
153  */
154 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
155 {
156 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
157 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
158 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
159 	return kvm_skip_emulated_instruction(vcpu);
160 }
161 
162 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
163 {
164 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
165 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
166 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
167 			| X86_EFLAGS_CF);
168 	return kvm_skip_emulated_instruction(vcpu);
169 }
170 
171 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
172 				u32 vm_instruction_error)
173 {
174 	struct vcpu_vmx *vmx = to_vmx(vcpu);
175 
176 	/*
177 	 * failValid writes the error number to the current VMCS, which
178 	 * can't be done if there isn't a current VMCS.
179 	 */
180 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
181 		return nested_vmx_failInvalid(vcpu);
182 
183 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
184 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
185 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
186 			| X86_EFLAGS_ZF);
187 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
188 	/*
189 	 * We don't need to force a shadow sync because
190 	 * VM_INSTRUCTION_ERROR is not shadowed
191 	 */
192 	return kvm_skip_emulated_instruction(vcpu);
193 }
194 
195 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
196 {
197 	/* TODO: not to reset guest simply here. */
198 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
199 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
200 }
201 
202 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
203 {
204 	return fixed_bits_valid(control, low, high);
205 }
206 
207 static inline u64 vmx_control_msr(u32 low, u32 high)
208 {
209 	return low | ((u64)high << 32);
210 }
211 
212 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
213 {
214 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
215 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
216 	vmx->nested.need_vmcs12_to_shadow_sync = false;
217 }
218 
219 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
220 {
221 	struct vcpu_vmx *vmx = to_vmx(vcpu);
222 
223 	if (!vmx->nested.hv_evmcs)
224 		return;
225 
226 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
227 	vmx->nested.hv_evmcs_vmptr = -1ull;
228 	vmx->nested.hv_evmcs = NULL;
229 }
230 
231 /*
232  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
233  * just stops using VMX.
234  */
235 static void free_nested(struct kvm_vcpu *vcpu)
236 {
237 	struct vcpu_vmx *vmx = to_vmx(vcpu);
238 
239 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
240 		return;
241 
242 	kvm_clear_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
243 
244 	vmx->nested.vmxon = false;
245 	vmx->nested.smm.vmxon = false;
246 	free_vpid(vmx->nested.vpid02);
247 	vmx->nested.posted_intr_nv = -1;
248 	vmx->nested.current_vmptr = -1ull;
249 	if (enable_shadow_vmcs) {
250 		vmx_disable_shadow_vmcs(vmx);
251 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
252 		free_vmcs(vmx->vmcs01.shadow_vmcs);
253 		vmx->vmcs01.shadow_vmcs = NULL;
254 	}
255 	kfree(vmx->nested.cached_vmcs12);
256 	vmx->nested.cached_vmcs12 = NULL;
257 	kfree(vmx->nested.cached_shadow_vmcs12);
258 	vmx->nested.cached_shadow_vmcs12 = NULL;
259 	/* Unpin physical memory we referred to in the vmcs02 */
260 	if (vmx->nested.apic_access_page) {
261 		kvm_release_page_clean(vmx->nested.apic_access_page);
262 		vmx->nested.apic_access_page = NULL;
263 	}
264 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
265 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
266 	vmx->nested.pi_desc = NULL;
267 
268 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
269 
270 	nested_release_evmcs(vcpu);
271 
272 	free_loaded_vmcs(&vmx->nested.vmcs02);
273 }
274 
275 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
276 				     struct loaded_vmcs *prev)
277 {
278 	struct vmcs_host_state *dest, *src;
279 
280 	if (unlikely(!vmx->guest_state_loaded))
281 		return;
282 
283 	src = &prev->host_state;
284 	dest = &vmx->loaded_vmcs->host_state;
285 
286 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
287 	dest->ldt_sel = src->ldt_sel;
288 #ifdef CONFIG_X86_64
289 	dest->ds_sel = src->ds_sel;
290 	dest->es_sel = src->es_sel;
291 #endif
292 }
293 
294 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
295 {
296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
297 	struct loaded_vmcs *prev;
298 	int cpu;
299 
300 	if (vmx->loaded_vmcs == vmcs)
301 		return;
302 
303 	cpu = get_cpu();
304 	prev = vmx->loaded_vmcs;
305 	vmx->loaded_vmcs = vmcs;
306 	vmx_vcpu_load_vmcs(vcpu, cpu);
307 	vmx_sync_vmcs_host_state(vmx, prev);
308 	put_cpu();
309 
310 	vmx_segment_cache_clear(vmx);
311 }
312 
313 /*
314  * Ensure that the current vmcs of the logical processor is the
315  * vmcs01 of the vcpu before calling free_nested().
316  */
317 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
318 {
319 	vcpu_load(vcpu);
320 	vmx_leave_nested(vcpu);
321 	vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
322 	free_nested(vcpu);
323 	vcpu_put(vcpu);
324 }
325 
326 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
327 		struct x86_exception *fault)
328 {
329 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
331 	u32 exit_reason;
332 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
333 
334 	if (vmx->nested.pml_full) {
335 		exit_reason = EXIT_REASON_PML_FULL;
336 		vmx->nested.pml_full = false;
337 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
338 	} else if (fault->error_code & PFERR_RSVD_MASK)
339 		exit_reason = EXIT_REASON_EPT_MISCONFIG;
340 	else
341 		exit_reason = EXIT_REASON_EPT_VIOLATION;
342 
343 	nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
344 	vmcs12->guest_physical_address = fault->address;
345 }
346 
347 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
348 {
349 	WARN_ON(mmu_is_nested(vcpu));
350 
351 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
352 	kvm_init_shadow_ept_mmu(vcpu,
353 			to_vmx(vcpu)->nested.msrs.ept_caps &
354 			VMX_EPT_EXECUTE_ONLY_BIT,
355 			nested_ept_ad_enabled(vcpu),
356 			nested_ept_get_cr3(vcpu));
357 	vcpu->arch.mmu->set_cr3           = vmx_set_cr3;
358 	vcpu->arch.mmu->get_cr3           = nested_ept_get_cr3;
359 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
360 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
361 
362 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
363 }
364 
365 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
366 {
367 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
368 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
369 }
370 
371 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
372 					    u16 error_code)
373 {
374 	bool inequality, bit;
375 
376 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
377 	inequality =
378 		(error_code & vmcs12->page_fault_error_code_mask) !=
379 		 vmcs12->page_fault_error_code_match;
380 	return inequality ^ bit;
381 }
382 
383 
384 /*
385  * KVM wants to inject page-faults which it got to the guest. This function
386  * checks whether in a nested guest, we need to inject them to L1 or L2.
387  */
388 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
389 {
390 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
391 	unsigned int nr = vcpu->arch.exception.nr;
392 	bool has_payload = vcpu->arch.exception.has_payload;
393 	unsigned long payload = vcpu->arch.exception.payload;
394 
395 	if (nr == PF_VECTOR) {
396 		if (vcpu->arch.exception.nested_apf) {
397 			*exit_qual = vcpu->arch.apf.nested_apf_token;
398 			return 1;
399 		}
400 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
401 						    vcpu->arch.exception.error_code)) {
402 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
403 			return 1;
404 		}
405 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
406 		if (nr == DB_VECTOR) {
407 			if (!has_payload) {
408 				payload = vcpu->arch.dr6;
409 				payload &= ~(DR6_FIXED_1 | DR6_BT);
410 				payload ^= DR6_RTM;
411 			}
412 			*exit_qual = payload;
413 		} else
414 			*exit_qual = 0;
415 		return 1;
416 	}
417 
418 	return 0;
419 }
420 
421 
422 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
423 		struct x86_exception *fault)
424 {
425 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
426 
427 	WARN_ON(!is_guest_mode(vcpu));
428 
429 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
430 		!to_vmx(vcpu)->nested.nested_run_pending) {
431 		vmcs12->vm_exit_intr_error_code = fault->error_code;
432 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
433 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
434 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
435 				  fault->address);
436 	} else {
437 		kvm_inject_page_fault(vcpu, fault);
438 	}
439 }
440 
441 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
442 {
443 	return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
444 }
445 
446 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
447 					       struct vmcs12 *vmcs12)
448 {
449 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
450 		return 0;
451 
452 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
453 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
454 		return -EINVAL;
455 
456 	return 0;
457 }
458 
459 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
460 						struct vmcs12 *vmcs12)
461 {
462 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
463 		return 0;
464 
465 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
466 		return -EINVAL;
467 
468 	return 0;
469 }
470 
471 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
472 						struct vmcs12 *vmcs12)
473 {
474 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
475 		return 0;
476 
477 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
478 		return -EINVAL;
479 
480 	return 0;
481 }
482 
483 /*
484  * Check if MSR is intercepted for L01 MSR bitmap.
485  */
486 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
487 {
488 	unsigned long *msr_bitmap;
489 	int f = sizeof(unsigned long);
490 
491 	if (!cpu_has_vmx_msr_bitmap())
492 		return true;
493 
494 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
495 
496 	if (msr <= 0x1fff) {
497 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
498 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
499 		msr &= 0x1fff;
500 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
501 	}
502 
503 	return true;
504 }
505 
506 /*
507  * If a msr is allowed by L0, we should check whether it is allowed by L1.
508  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
509  */
510 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
511 					       unsigned long *msr_bitmap_nested,
512 					       u32 msr, int type)
513 {
514 	int f = sizeof(unsigned long);
515 
516 	/*
517 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
518 	 * have the write-low and read-high bitmap offsets the wrong way round.
519 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
520 	 */
521 	if (msr <= 0x1fff) {
522 		if (type & MSR_TYPE_R &&
523 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
524 			/* read-low */
525 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
526 
527 		if (type & MSR_TYPE_W &&
528 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
529 			/* write-low */
530 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
531 
532 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
533 		msr &= 0x1fff;
534 		if (type & MSR_TYPE_R &&
535 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
536 			/* read-high */
537 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
538 
539 		if (type & MSR_TYPE_W &&
540 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
541 			/* write-high */
542 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
543 
544 	}
545 }
546 
547 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap) {
548 	int msr;
549 
550 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
551 		unsigned word = msr / BITS_PER_LONG;
552 
553 		msr_bitmap[word] = ~0;
554 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
555 	}
556 }
557 
558 /*
559  * Merge L0's and L1's MSR bitmap, return false to indicate that
560  * we do not use the hardware.
561  */
562 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
563 						 struct vmcs12 *vmcs12)
564 {
565 	int msr;
566 	unsigned long *msr_bitmap_l1;
567 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
568 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
569 
570 	/* Nothing to do if the MSR bitmap is not in use.  */
571 	if (!cpu_has_vmx_msr_bitmap() ||
572 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
573 		return false;
574 
575 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
576 		return false;
577 
578 	msr_bitmap_l1 = (unsigned long *)map->hva;
579 
580 	/*
581 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
582 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
583 	 * the x2APIC MSR range and selectively disable them below.
584 	 */
585 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
586 
587 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
588 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
589 			/*
590 			 * L0 need not intercept reads for MSRs between 0x800
591 			 * and 0x8ff, it just lets the processor take the value
592 			 * from the virtual-APIC page; take those 256 bits
593 			 * directly from the L1 bitmap.
594 			 */
595 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
596 				unsigned word = msr / BITS_PER_LONG;
597 
598 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
599 			}
600 		}
601 
602 		nested_vmx_disable_intercept_for_msr(
603 			msr_bitmap_l1, msr_bitmap_l0,
604 			X2APIC_MSR(APIC_TASKPRI),
605 			MSR_TYPE_R | MSR_TYPE_W);
606 
607 		if (nested_cpu_has_vid(vmcs12)) {
608 			nested_vmx_disable_intercept_for_msr(
609 				msr_bitmap_l1, msr_bitmap_l0,
610 				X2APIC_MSR(APIC_EOI),
611 				MSR_TYPE_W);
612 			nested_vmx_disable_intercept_for_msr(
613 				msr_bitmap_l1, msr_bitmap_l0,
614 				X2APIC_MSR(APIC_SELF_IPI),
615 				MSR_TYPE_W);
616 		}
617 	}
618 
619 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
620 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
621 					     MSR_FS_BASE, MSR_TYPE_RW);
622 
623 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
624 					     MSR_GS_BASE, MSR_TYPE_RW);
625 
626 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
627 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
628 
629 	/*
630 	 * Checking the L0->L1 bitmap is trying to verify two things:
631 	 *
632 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
633 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
634 	 *    from the L12 MSR bitmap that is too permissive.
635 	 * 2. That L1 or L2s have actually used the MSR. This avoids
636 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
637 	 *    works properly because we only update the L01 MSR bitmap lazily.
638 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
639 	 *    updated to reflect this when L1 (or its L2s) actually write to
640 	 *    the MSR.
641 	 */
642 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
643 		nested_vmx_disable_intercept_for_msr(
644 					msr_bitmap_l1, msr_bitmap_l0,
645 					MSR_IA32_SPEC_CTRL,
646 					MSR_TYPE_R | MSR_TYPE_W);
647 
648 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
649 		nested_vmx_disable_intercept_for_msr(
650 					msr_bitmap_l1, msr_bitmap_l0,
651 					MSR_IA32_PRED_CMD,
652 					MSR_TYPE_W);
653 
654 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
655 
656 	return true;
657 }
658 
659 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
660 				       struct vmcs12 *vmcs12)
661 {
662 	struct kvm_host_map map;
663 	struct vmcs12 *shadow;
664 
665 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
666 	    vmcs12->vmcs_link_pointer == -1ull)
667 		return;
668 
669 	shadow = get_shadow_vmcs12(vcpu);
670 
671 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
672 		return;
673 
674 	memcpy(shadow, map.hva, VMCS12_SIZE);
675 	kvm_vcpu_unmap(vcpu, &map, false);
676 }
677 
678 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
679 					      struct vmcs12 *vmcs12)
680 {
681 	struct vcpu_vmx *vmx = to_vmx(vcpu);
682 
683 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
684 	    vmcs12->vmcs_link_pointer == -1ull)
685 		return;
686 
687 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
688 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
689 }
690 
691 /*
692  * In nested virtualization, check if L1 has set
693  * VM_EXIT_ACK_INTR_ON_EXIT
694  */
695 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
696 {
697 	return get_vmcs12(vcpu)->vm_exit_controls &
698 		VM_EXIT_ACK_INTR_ON_EXIT;
699 }
700 
701 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
702 {
703 	return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
704 }
705 
706 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
707 					  struct vmcs12 *vmcs12)
708 {
709 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
710 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
711 		return -EINVAL;
712 	else
713 		return 0;
714 }
715 
716 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
717 					   struct vmcs12 *vmcs12)
718 {
719 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
720 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
721 	    !nested_cpu_has_vid(vmcs12) &&
722 	    !nested_cpu_has_posted_intr(vmcs12))
723 		return 0;
724 
725 	/*
726 	 * If virtualize x2apic mode is enabled,
727 	 * virtualize apic access must be disabled.
728 	 */
729 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
730 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
731 		return -EINVAL;
732 
733 	/*
734 	 * If virtual interrupt delivery is enabled,
735 	 * we must exit on external interrupts.
736 	 */
737 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
738 		return -EINVAL;
739 
740 	/*
741 	 * bits 15:8 should be zero in posted_intr_nv,
742 	 * the descriptor address has been already checked
743 	 * in nested_get_vmcs12_pages.
744 	 *
745 	 * bits 5:0 of posted_intr_desc_addr should be zero.
746 	 */
747 	if (nested_cpu_has_posted_intr(vmcs12) &&
748 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
749 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
750 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
751 	    CC((vmcs12->posted_intr_desc_addr & 0x3f)) ||
752 	    CC((vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu)))))
753 		return -EINVAL;
754 
755 	/* tpr shadow is needed by all apicv features. */
756 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
757 		return -EINVAL;
758 
759 	return 0;
760 }
761 
762 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
763 				       u32 count, u64 addr)
764 {
765 	int maxphyaddr;
766 
767 	if (count == 0)
768 		return 0;
769 	maxphyaddr = cpuid_maxphyaddr(vcpu);
770 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
771 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
772 		return -EINVAL;
773 
774 	return 0;
775 }
776 
777 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
778 						     struct vmcs12 *vmcs12)
779 {
780 	if (CC(nested_vmx_check_msr_switch(vcpu,
781 					   vmcs12->vm_exit_msr_load_count,
782 					   vmcs12->vm_exit_msr_load_addr)) ||
783 	    CC(nested_vmx_check_msr_switch(vcpu,
784 					   vmcs12->vm_exit_msr_store_count,
785 					   vmcs12->vm_exit_msr_store_addr)))
786 		return -EINVAL;
787 
788 	return 0;
789 }
790 
791 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
792                                                       struct vmcs12 *vmcs12)
793 {
794 	if (CC(nested_vmx_check_msr_switch(vcpu,
795 					   vmcs12->vm_entry_msr_load_count,
796 					   vmcs12->vm_entry_msr_load_addr)))
797                 return -EINVAL;
798 
799 	return 0;
800 }
801 
802 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
803 					 struct vmcs12 *vmcs12)
804 {
805 	if (!nested_cpu_has_pml(vmcs12))
806 		return 0;
807 
808 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
809 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
810 		return -EINVAL;
811 
812 	return 0;
813 }
814 
815 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
816 							struct vmcs12 *vmcs12)
817 {
818 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
819 	       !nested_cpu_has_ept(vmcs12)))
820 		return -EINVAL;
821 	return 0;
822 }
823 
824 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
825 							 struct vmcs12 *vmcs12)
826 {
827 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
828 	       !nested_cpu_has_ept(vmcs12)))
829 		return -EINVAL;
830 	return 0;
831 }
832 
833 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
834 						 struct vmcs12 *vmcs12)
835 {
836 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
837 		return 0;
838 
839 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
840 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
841 		return -EINVAL;
842 
843 	return 0;
844 }
845 
846 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
847 				       struct vmx_msr_entry *e)
848 {
849 	/* x2APIC MSR accesses are not allowed */
850 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
851 		return -EINVAL;
852 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
853 	    CC(e->index == MSR_IA32_UCODE_REV))
854 		return -EINVAL;
855 	if (CC(e->reserved != 0))
856 		return -EINVAL;
857 	return 0;
858 }
859 
860 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
861 				     struct vmx_msr_entry *e)
862 {
863 	if (CC(e->index == MSR_FS_BASE) ||
864 	    CC(e->index == MSR_GS_BASE) ||
865 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
866 	    nested_vmx_msr_check_common(vcpu, e))
867 		return -EINVAL;
868 	return 0;
869 }
870 
871 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
872 				      struct vmx_msr_entry *e)
873 {
874 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
875 	    nested_vmx_msr_check_common(vcpu, e))
876 		return -EINVAL;
877 	return 0;
878 }
879 
880 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
881 {
882 	struct vcpu_vmx *vmx = to_vmx(vcpu);
883 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
884 				       vmx->nested.msrs.misc_high);
885 
886 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
887 }
888 
889 /*
890  * Load guest's/host's msr at nested entry/exit.
891  * return 0 for success, entry index for failure.
892  *
893  * One of the failure modes for MSR load/store is when a list exceeds the
894  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
895  * as possible, process all valid entries before failing rather than precheck
896  * for a capacity violation.
897  */
898 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
899 {
900 	u32 i;
901 	struct vmx_msr_entry e;
902 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
903 
904 	for (i = 0; i < count; i++) {
905 		if (unlikely(i >= max_msr_list_size))
906 			goto fail;
907 
908 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
909 					&e, sizeof(e))) {
910 			pr_debug_ratelimited(
911 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
912 				__func__, i, gpa + i * sizeof(e));
913 			goto fail;
914 		}
915 		if (nested_vmx_load_msr_check(vcpu, &e)) {
916 			pr_debug_ratelimited(
917 				"%s check failed (%u, 0x%x, 0x%x)\n",
918 				__func__, i, e.index, e.reserved);
919 			goto fail;
920 		}
921 		if (kvm_set_msr(vcpu, e.index, e.value)) {
922 			pr_debug_ratelimited(
923 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
924 				__func__, i, e.index, e.value);
925 			goto fail;
926 		}
927 	}
928 	return 0;
929 fail:
930 	return i + 1;
931 }
932 
933 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
934 					    u32 msr_index,
935 					    u64 *data)
936 {
937 	struct vcpu_vmx *vmx = to_vmx(vcpu);
938 
939 	/*
940 	 * If the L0 hypervisor stored a more accurate value for the TSC that
941 	 * does not include the time taken for emulation of the L2->L1
942 	 * VM-exit in L0, use the more accurate value.
943 	 */
944 	if (msr_index == MSR_IA32_TSC) {
945 		int index = vmx_find_msr_index(&vmx->msr_autostore.guest,
946 					       MSR_IA32_TSC);
947 
948 		if (index >= 0) {
949 			u64 val = vmx->msr_autostore.guest.val[index].value;
950 
951 			*data = kvm_read_l1_tsc(vcpu, val);
952 			return true;
953 		}
954 	}
955 
956 	if (kvm_get_msr(vcpu, msr_index, data)) {
957 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
958 			msr_index);
959 		return false;
960 	}
961 	return true;
962 }
963 
964 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
965 				     struct vmx_msr_entry *e)
966 {
967 	if (kvm_vcpu_read_guest(vcpu,
968 				gpa + i * sizeof(*e),
969 				e, 2 * sizeof(u32))) {
970 		pr_debug_ratelimited(
971 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
972 			__func__, i, gpa + i * sizeof(*e));
973 		return false;
974 	}
975 	if (nested_vmx_store_msr_check(vcpu, e)) {
976 		pr_debug_ratelimited(
977 			"%s check failed (%u, 0x%x, 0x%x)\n",
978 			__func__, i, e->index, e->reserved);
979 		return false;
980 	}
981 	return true;
982 }
983 
984 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
985 {
986 	u64 data;
987 	u32 i;
988 	struct vmx_msr_entry e;
989 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
990 
991 	for (i = 0; i < count; i++) {
992 		if (unlikely(i >= max_msr_list_size))
993 			return -EINVAL;
994 
995 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
996 			return -EINVAL;
997 
998 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
999 			return -EINVAL;
1000 
1001 		if (kvm_vcpu_write_guest(vcpu,
1002 					 gpa + i * sizeof(e) +
1003 					     offsetof(struct vmx_msr_entry, value),
1004 					 &data, sizeof(data))) {
1005 			pr_debug_ratelimited(
1006 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1007 				__func__, i, e.index, data);
1008 			return -EINVAL;
1009 		}
1010 	}
1011 	return 0;
1012 }
1013 
1014 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1015 {
1016 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1017 	u32 count = vmcs12->vm_exit_msr_store_count;
1018 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1019 	struct vmx_msr_entry e;
1020 	u32 i;
1021 
1022 	for (i = 0; i < count; i++) {
1023 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1024 			return false;
1025 
1026 		if (e.index == msr_index)
1027 			return true;
1028 	}
1029 	return false;
1030 }
1031 
1032 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1033 					   u32 msr_index)
1034 {
1035 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1036 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1037 	bool in_vmcs12_store_list;
1038 	int msr_autostore_index;
1039 	bool in_autostore_list;
1040 	int last;
1041 
1042 	msr_autostore_index = vmx_find_msr_index(autostore, msr_index);
1043 	in_autostore_list = msr_autostore_index >= 0;
1044 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1045 
1046 	if (in_vmcs12_store_list && !in_autostore_list) {
1047 		if (autostore->nr == NR_LOADSTORE_MSRS) {
1048 			/*
1049 			 * Emulated VMEntry does not fail here.  Instead a less
1050 			 * accurate value will be returned by
1051 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1052 			 * instead of reading the value from the vmcs02 VMExit
1053 			 * MSR-store area.
1054 			 */
1055 			pr_warn_ratelimited(
1056 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1057 				msr_index);
1058 			return;
1059 		}
1060 		last = autostore->nr++;
1061 		autostore->val[last].index = msr_index;
1062 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1063 		last = --autostore->nr;
1064 		autostore->val[msr_autostore_index] = autostore->val[last];
1065 	}
1066 }
1067 
1068 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
1069 {
1070 	unsigned long invalid_mask;
1071 
1072 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1073 	return (val & invalid_mask) == 0;
1074 }
1075 
1076 /*
1077  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
1078  * emulating VM entry into a guest with EPT enabled.
1079  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
1080  * is assigned to entry_failure_code on failure.
1081  */
1082 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
1083 			       u32 *entry_failure_code)
1084 {
1085 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
1086 		if (CC(!nested_cr3_valid(vcpu, cr3))) {
1087 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
1088 			return -EINVAL;
1089 		}
1090 
1091 		/*
1092 		 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1093 		 * must not be dereferenced.
1094 		 */
1095 		if (is_pae_paging(vcpu) && !nested_ept) {
1096 			if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1097 				*entry_failure_code = ENTRY_FAIL_PDPTE;
1098 				return -EINVAL;
1099 			}
1100 		}
1101 	}
1102 
1103 	if (!nested_ept)
1104 		kvm_mmu_new_cr3(vcpu, cr3, false);
1105 
1106 	vcpu->arch.cr3 = cr3;
1107 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1108 
1109 	kvm_init_mmu(vcpu, false);
1110 
1111 	return 0;
1112 }
1113 
1114 /*
1115  * Returns if KVM is able to config CPU to tag TLB entries
1116  * populated by L2 differently than TLB entries populated
1117  * by L1.
1118  *
1119  * If L0 uses EPT, L1 and L2 run with different EPTP because
1120  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1121  * are tagged with different EPTP.
1122  *
1123  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1124  * with different VPID (L1 entries are tagged with vmx->vpid
1125  * while L2 entries are tagged with vmx->nested.vpid02).
1126  */
1127 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1128 {
1129 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1130 
1131 	return enable_ept ||
1132 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1133 }
1134 
1135 static u16 nested_get_vpid02(struct kvm_vcpu *vcpu)
1136 {
1137 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1138 
1139 	return vmx->nested.vpid02 ? vmx->nested.vpid02 : vmx->vpid;
1140 }
1141 
1142 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1143 {
1144 	superset &= mask;
1145 	subset &= mask;
1146 
1147 	return (superset | subset) == superset;
1148 }
1149 
1150 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1151 {
1152 	const u64 feature_and_reserved =
1153 		/* feature (except bit 48; see below) */
1154 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1155 		/* reserved */
1156 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1157 	u64 vmx_basic = vmx->nested.msrs.basic;
1158 
1159 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1160 		return -EINVAL;
1161 
1162 	/*
1163 	 * KVM does not emulate a version of VMX that constrains physical
1164 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1165 	 */
1166 	if (data & BIT_ULL(48))
1167 		return -EINVAL;
1168 
1169 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1170 	    vmx_basic_vmcs_revision_id(data))
1171 		return -EINVAL;
1172 
1173 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1174 		return -EINVAL;
1175 
1176 	vmx->nested.msrs.basic = data;
1177 	return 0;
1178 }
1179 
1180 static int
1181 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1182 {
1183 	u64 supported;
1184 	u32 *lowp, *highp;
1185 
1186 	switch (msr_index) {
1187 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1188 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1189 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1190 		break;
1191 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1192 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1193 		highp = &vmx->nested.msrs.procbased_ctls_high;
1194 		break;
1195 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1196 		lowp = &vmx->nested.msrs.exit_ctls_low;
1197 		highp = &vmx->nested.msrs.exit_ctls_high;
1198 		break;
1199 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1200 		lowp = &vmx->nested.msrs.entry_ctls_low;
1201 		highp = &vmx->nested.msrs.entry_ctls_high;
1202 		break;
1203 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1204 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1205 		highp = &vmx->nested.msrs.secondary_ctls_high;
1206 		break;
1207 	default:
1208 		BUG();
1209 	}
1210 
1211 	supported = vmx_control_msr(*lowp, *highp);
1212 
1213 	/* Check must-be-1 bits are still 1. */
1214 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1215 		return -EINVAL;
1216 
1217 	/* Check must-be-0 bits are still 0. */
1218 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1219 		return -EINVAL;
1220 
1221 	*lowp = data;
1222 	*highp = data >> 32;
1223 	return 0;
1224 }
1225 
1226 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1227 {
1228 	const u64 feature_and_reserved_bits =
1229 		/* feature */
1230 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1231 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1232 		/* reserved */
1233 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1234 	u64 vmx_misc;
1235 
1236 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1237 				   vmx->nested.msrs.misc_high);
1238 
1239 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1240 		return -EINVAL;
1241 
1242 	if ((vmx->nested.msrs.pinbased_ctls_high &
1243 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1244 	    vmx_misc_preemption_timer_rate(data) !=
1245 	    vmx_misc_preemption_timer_rate(vmx_misc))
1246 		return -EINVAL;
1247 
1248 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1249 		return -EINVAL;
1250 
1251 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1252 		return -EINVAL;
1253 
1254 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1255 		return -EINVAL;
1256 
1257 	vmx->nested.msrs.misc_low = data;
1258 	vmx->nested.msrs.misc_high = data >> 32;
1259 
1260 	return 0;
1261 }
1262 
1263 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1264 {
1265 	u64 vmx_ept_vpid_cap;
1266 
1267 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1268 					   vmx->nested.msrs.vpid_caps);
1269 
1270 	/* Every bit is either reserved or a feature bit. */
1271 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1272 		return -EINVAL;
1273 
1274 	vmx->nested.msrs.ept_caps = data;
1275 	vmx->nested.msrs.vpid_caps = data >> 32;
1276 	return 0;
1277 }
1278 
1279 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1280 {
1281 	u64 *msr;
1282 
1283 	switch (msr_index) {
1284 	case MSR_IA32_VMX_CR0_FIXED0:
1285 		msr = &vmx->nested.msrs.cr0_fixed0;
1286 		break;
1287 	case MSR_IA32_VMX_CR4_FIXED0:
1288 		msr = &vmx->nested.msrs.cr4_fixed0;
1289 		break;
1290 	default:
1291 		BUG();
1292 	}
1293 
1294 	/*
1295 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1296 	 * must be 1 in the restored value.
1297 	 */
1298 	if (!is_bitwise_subset(data, *msr, -1ULL))
1299 		return -EINVAL;
1300 
1301 	*msr = data;
1302 	return 0;
1303 }
1304 
1305 /*
1306  * Called when userspace is restoring VMX MSRs.
1307  *
1308  * Returns 0 on success, non-0 otherwise.
1309  */
1310 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1311 {
1312 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1313 
1314 	/*
1315 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1316 	 * is in VMX operation.
1317 	 */
1318 	if (vmx->nested.vmxon)
1319 		return -EBUSY;
1320 
1321 	switch (msr_index) {
1322 	case MSR_IA32_VMX_BASIC:
1323 		return vmx_restore_vmx_basic(vmx, data);
1324 	case MSR_IA32_VMX_PINBASED_CTLS:
1325 	case MSR_IA32_VMX_PROCBASED_CTLS:
1326 	case MSR_IA32_VMX_EXIT_CTLS:
1327 	case MSR_IA32_VMX_ENTRY_CTLS:
1328 		/*
1329 		 * The "non-true" VMX capability MSRs are generated from the
1330 		 * "true" MSRs, so we do not support restoring them directly.
1331 		 *
1332 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1333 		 * should restore the "true" MSRs with the must-be-1 bits
1334 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1335 		 * DEFAULT SETTINGS".
1336 		 */
1337 		return -EINVAL;
1338 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1339 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1340 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1341 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1342 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1343 		return vmx_restore_control_msr(vmx, msr_index, data);
1344 	case MSR_IA32_VMX_MISC:
1345 		return vmx_restore_vmx_misc(vmx, data);
1346 	case MSR_IA32_VMX_CR0_FIXED0:
1347 	case MSR_IA32_VMX_CR4_FIXED0:
1348 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1349 	case MSR_IA32_VMX_CR0_FIXED1:
1350 	case MSR_IA32_VMX_CR4_FIXED1:
1351 		/*
1352 		 * These MSRs are generated based on the vCPU's CPUID, so we
1353 		 * do not support restoring them directly.
1354 		 */
1355 		return -EINVAL;
1356 	case MSR_IA32_VMX_EPT_VPID_CAP:
1357 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1358 	case MSR_IA32_VMX_VMCS_ENUM:
1359 		vmx->nested.msrs.vmcs_enum = data;
1360 		return 0;
1361 	case MSR_IA32_VMX_VMFUNC:
1362 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1363 			return -EINVAL;
1364 		vmx->nested.msrs.vmfunc_controls = data;
1365 		return 0;
1366 	default:
1367 		/*
1368 		 * The rest of the VMX capability MSRs do not support restore.
1369 		 */
1370 		return -EINVAL;
1371 	}
1372 }
1373 
1374 /* Returns 0 on success, non-0 otherwise. */
1375 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1376 {
1377 	switch (msr_index) {
1378 	case MSR_IA32_VMX_BASIC:
1379 		*pdata = msrs->basic;
1380 		break;
1381 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1382 	case MSR_IA32_VMX_PINBASED_CTLS:
1383 		*pdata = vmx_control_msr(
1384 			msrs->pinbased_ctls_low,
1385 			msrs->pinbased_ctls_high);
1386 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1387 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1388 		break;
1389 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1390 	case MSR_IA32_VMX_PROCBASED_CTLS:
1391 		*pdata = vmx_control_msr(
1392 			msrs->procbased_ctls_low,
1393 			msrs->procbased_ctls_high);
1394 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1395 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1396 		break;
1397 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1398 	case MSR_IA32_VMX_EXIT_CTLS:
1399 		*pdata = vmx_control_msr(
1400 			msrs->exit_ctls_low,
1401 			msrs->exit_ctls_high);
1402 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1403 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1404 		break;
1405 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1406 	case MSR_IA32_VMX_ENTRY_CTLS:
1407 		*pdata = vmx_control_msr(
1408 			msrs->entry_ctls_low,
1409 			msrs->entry_ctls_high);
1410 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1411 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1412 		break;
1413 	case MSR_IA32_VMX_MISC:
1414 		*pdata = vmx_control_msr(
1415 			msrs->misc_low,
1416 			msrs->misc_high);
1417 		break;
1418 	case MSR_IA32_VMX_CR0_FIXED0:
1419 		*pdata = msrs->cr0_fixed0;
1420 		break;
1421 	case MSR_IA32_VMX_CR0_FIXED1:
1422 		*pdata = msrs->cr0_fixed1;
1423 		break;
1424 	case MSR_IA32_VMX_CR4_FIXED0:
1425 		*pdata = msrs->cr4_fixed0;
1426 		break;
1427 	case MSR_IA32_VMX_CR4_FIXED1:
1428 		*pdata = msrs->cr4_fixed1;
1429 		break;
1430 	case MSR_IA32_VMX_VMCS_ENUM:
1431 		*pdata = msrs->vmcs_enum;
1432 		break;
1433 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1434 		*pdata = vmx_control_msr(
1435 			msrs->secondary_ctls_low,
1436 			msrs->secondary_ctls_high);
1437 		break;
1438 	case MSR_IA32_VMX_EPT_VPID_CAP:
1439 		*pdata = msrs->ept_caps |
1440 			((u64)msrs->vpid_caps << 32);
1441 		break;
1442 	case MSR_IA32_VMX_VMFUNC:
1443 		*pdata = msrs->vmfunc_controls;
1444 		break;
1445 	default:
1446 		return 1;
1447 	}
1448 
1449 	return 0;
1450 }
1451 
1452 /*
1453  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1454  * been modified by the L1 guest.  Note, "writable" in this context means
1455  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1456  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1457  * VM-exit information fields (which are actually writable if the vCPU is
1458  * configured to support "VMWRITE to any supported field in the VMCS").
1459  */
1460 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1461 {
1462 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1463 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1464 	struct shadow_vmcs_field field;
1465 	unsigned long val;
1466 	int i;
1467 
1468 	if (WARN_ON(!shadow_vmcs))
1469 		return;
1470 
1471 	preempt_disable();
1472 
1473 	vmcs_load(shadow_vmcs);
1474 
1475 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1476 		field = shadow_read_write_fields[i];
1477 		val = __vmcs_readl(field.encoding);
1478 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1479 	}
1480 
1481 	vmcs_clear(shadow_vmcs);
1482 	vmcs_load(vmx->loaded_vmcs->vmcs);
1483 
1484 	preempt_enable();
1485 }
1486 
1487 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1488 {
1489 	const struct shadow_vmcs_field *fields[] = {
1490 		shadow_read_write_fields,
1491 		shadow_read_only_fields
1492 	};
1493 	const int max_fields[] = {
1494 		max_shadow_read_write_fields,
1495 		max_shadow_read_only_fields
1496 	};
1497 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1498 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1499 	struct shadow_vmcs_field field;
1500 	unsigned long val;
1501 	int i, q;
1502 
1503 	if (WARN_ON(!shadow_vmcs))
1504 		return;
1505 
1506 	vmcs_load(shadow_vmcs);
1507 
1508 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1509 		for (i = 0; i < max_fields[q]; i++) {
1510 			field = fields[q][i];
1511 			val = vmcs12_read_any(vmcs12, field.encoding,
1512 					      field.offset);
1513 			__vmcs_writel(field.encoding, val);
1514 		}
1515 	}
1516 
1517 	vmcs_clear(shadow_vmcs);
1518 	vmcs_load(vmx->loaded_vmcs->vmcs);
1519 }
1520 
1521 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1522 {
1523 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1524 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1525 
1526 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1527 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1528 	vmcs12->guest_rip = evmcs->guest_rip;
1529 
1530 	if (unlikely(!(evmcs->hv_clean_fields &
1531 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1532 		vmcs12->guest_rsp = evmcs->guest_rsp;
1533 		vmcs12->guest_rflags = evmcs->guest_rflags;
1534 		vmcs12->guest_interruptibility_info =
1535 			evmcs->guest_interruptibility_info;
1536 	}
1537 
1538 	if (unlikely(!(evmcs->hv_clean_fields &
1539 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1540 		vmcs12->cpu_based_vm_exec_control =
1541 			evmcs->cpu_based_vm_exec_control;
1542 	}
1543 
1544 	if (unlikely(!(evmcs->hv_clean_fields &
1545 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1546 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1547 	}
1548 
1549 	if (unlikely(!(evmcs->hv_clean_fields &
1550 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1551 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1552 	}
1553 
1554 	if (unlikely(!(evmcs->hv_clean_fields &
1555 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1556 		vmcs12->vm_entry_intr_info_field =
1557 			evmcs->vm_entry_intr_info_field;
1558 		vmcs12->vm_entry_exception_error_code =
1559 			evmcs->vm_entry_exception_error_code;
1560 		vmcs12->vm_entry_instruction_len =
1561 			evmcs->vm_entry_instruction_len;
1562 	}
1563 
1564 	if (unlikely(!(evmcs->hv_clean_fields &
1565 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1566 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1567 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1568 		vmcs12->host_cr0 = evmcs->host_cr0;
1569 		vmcs12->host_cr3 = evmcs->host_cr3;
1570 		vmcs12->host_cr4 = evmcs->host_cr4;
1571 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1572 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1573 		vmcs12->host_rip = evmcs->host_rip;
1574 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1575 		vmcs12->host_es_selector = evmcs->host_es_selector;
1576 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1577 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1578 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1579 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1580 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1581 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1582 	}
1583 
1584 	if (unlikely(!(evmcs->hv_clean_fields &
1585 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1586 		vmcs12->pin_based_vm_exec_control =
1587 			evmcs->pin_based_vm_exec_control;
1588 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1589 		vmcs12->secondary_vm_exec_control =
1590 			evmcs->secondary_vm_exec_control;
1591 	}
1592 
1593 	if (unlikely(!(evmcs->hv_clean_fields &
1594 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1595 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1596 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1597 	}
1598 
1599 	if (unlikely(!(evmcs->hv_clean_fields &
1600 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1601 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1602 	}
1603 
1604 	if (unlikely(!(evmcs->hv_clean_fields &
1605 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1606 		vmcs12->guest_es_base = evmcs->guest_es_base;
1607 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1608 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1609 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1610 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1611 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1612 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1613 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1614 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1615 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1616 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1617 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1618 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1619 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1620 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1621 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1622 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1623 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1624 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1625 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1626 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1627 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1628 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1629 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1630 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1631 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1632 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1633 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1634 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1635 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1636 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1637 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1638 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1639 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1640 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1641 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1642 	}
1643 
1644 	if (unlikely(!(evmcs->hv_clean_fields &
1645 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1646 		vmcs12->tsc_offset = evmcs->tsc_offset;
1647 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1648 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1649 	}
1650 
1651 	if (unlikely(!(evmcs->hv_clean_fields &
1652 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1653 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1654 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1655 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1656 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1657 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1658 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1659 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1660 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1661 	}
1662 
1663 	if (unlikely(!(evmcs->hv_clean_fields &
1664 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1665 		vmcs12->host_fs_base = evmcs->host_fs_base;
1666 		vmcs12->host_gs_base = evmcs->host_gs_base;
1667 		vmcs12->host_tr_base = evmcs->host_tr_base;
1668 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1669 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1670 		vmcs12->host_rsp = evmcs->host_rsp;
1671 	}
1672 
1673 	if (unlikely(!(evmcs->hv_clean_fields &
1674 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1675 		vmcs12->ept_pointer = evmcs->ept_pointer;
1676 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1677 	}
1678 
1679 	if (unlikely(!(evmcs->hv_clean_fields &
1680 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1681 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1682 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1683 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1684 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1685 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1686 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1687 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1688 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1689 		vmcs12->guest_pending_dbg_exceptions =
1690 			evmcs->guest_pending_dbg_exceptions;
1691 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1692 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1693 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1694 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1695 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1696 	}
1697 
1698 	/*
1699 	 * Not used?
1700 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1701 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1702 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1703 	 * vmcs12->cr3_target_value0 = evmcs->cr3_target_value0;
1704 	 * vmcs12->cr3_target_value1 = evmcs->cr3_target_value1;
1705 	 * vmcs12->cr3_target_value2 = evmcs->cr3_target_value2;
1706 	 * vmcs12->cr3_target_value3 = evmcs->cr3_target_value3;
1707 	 * vmcs12->page_fault_error_code_mask =
1708 	 *		evmcs->page_fault_error_code_mask;
1709 	 * vmcs12->page_fault_error_code_match =
1710 	 *		evmcs->page_fault_error_code_match;
1711 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1712 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1713 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1714 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1715 	 */
1716 
1717 	/*
1718 	 * Read only fields:
1719 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1720 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1721 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1722 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1723 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1724 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1725 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1726 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1727 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1728 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1729 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1730 	 *
1731 	 * Not present in struct vmcs12:
1732 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1733 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1734 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1735 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1736 	 */
1737 
1738 	return 0;
1739 }
1740 
1741 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1742 {
1743 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1744 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1745 
1746 	/*
1747 	 * Should not be changed by KVM:
1748 	 *
1749 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1750 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1751 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1752 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1753 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1754 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1755 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1756 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1757 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1758 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1759 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1760 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1761 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1762 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1763 	 * evmcs->host_rip = vmcs12->host_rip;
1764 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1765 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1766 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1767 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1768 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1769 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1770 	 * evmcs->host_rsp = vmcs12->host_rsp;
1771 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1772 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1773 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1774 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1775 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1776 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1777 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1778 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1779 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1780 	 * evmcs->cr3_target_value0 = vmcs12->cr3_target_value0;
1781 	 * evmcs->cr3_target_value1 = vmcs12->cr3_target_value1;
1782 	 * evmcs->cr3_target_value2 = vmcs12->cr3_target_value2;
1783 	 * evmcs->cr3_target_value3 = vmcs12->cr3_target_value3;
1784 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1785 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1786 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1787 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1788 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1789 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1790 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1791 	 * evmcs->page_fault_error_code_mask =
1792 	 *		vmcs12->page_fault_error_code_mask;
1793 	 * evmcs->page_fault_error_code_match =
1794 	 *		vmcs12->page_fault_error_code_match;
1795 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1796 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1797 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1798 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1799 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1800 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1801 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1802 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1803 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1804 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1805 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1806 	 *
1807 	 * Not present in struct vmcs12:
1808 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1809 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1810 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1811 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1812 	 */
1813 
1814 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1815 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1816 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1817 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1818 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1819 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1820 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1821 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1822 
1823 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1824 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1825 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1826 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1827 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1828 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1829 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1830 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1831 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1832 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1833 
1834 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1835 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1836 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1837 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1838 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1839 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1840 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1841 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1842 
1843 	evmcs->guest_es_base = vmcs12->guest_es_base;
1844 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1845 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1846 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1847 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1848 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1849 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1850 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1851 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1852 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1853 
1854 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1855 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1856 
1857 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1858 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1859 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1860 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1861 
1862 	evmcs->guest_pending_dbg_exceptions =
1863 		vmcs12->guest_pending_dbg_exceptions;
1864 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1865 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1866 
1867 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1868 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1869 
1870 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1871 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1872 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1873 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1874 
1875 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1876 
1877 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1878 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1879 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1880 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1881 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1882 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1883 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1884 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1885 
1886 	evmcs->exit_qualification = vmcs12->exit_qualification;
1887 
1888 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1889 	evmcs->guest_rsp = vmcs12->guest_rsp;
1890 	evmcs->guest_rflags = vmcs12->guest_rflags;
1891 
1892 	evmcs->guest_interruptibility_info =
1893 		vmcs12->guest_interruptibility_info;
1894 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1895 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1896 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1897 	evmcs->vm_entry_exception_error_code =
1898 		vmcs12->vm_entry_exception_error_code;
1899 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1900 
1901 	evmcs->guest_rip = vmcs12->guest_rip;
1902 
1903 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1904 
1905 	return 0;
1906 }
1907 
1908 /*
1909  * This is an equivalent of the nested hypervisor executing the vmptrld
1910  * instruction.
1911  */
1912 static int nested_vmx_handle_enlightened_vmptrld(struct kvm_vcpu *vcpu,
1913 						 bool from_launch)
1914 {
1915 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1916 	bool evmcs_gpa_changed = false;
1917 	u64 evmcs_gpa;
1918 
1919 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1920 		return 1;
1921 
1922 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa))
1923 		return 1;
1924 
1925 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1926 		if (!vmx->nested.hv_evmcs)
1927 			vmx->nested.current_vmptr = -1ull;
1928 
1929 		nested_release_evmcs(vcpu);
1930 
1931 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1932 				 &vmx->nested.hv_evmcs_map))
1933 			return 0;
1934 
1935 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1936 
1937 		/*
1938 		 * Currently, KVM only supports eVMCS version 1
1939 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1940 		 * value to first u32 field of eVMCS which should specify eVMCS
1941 		 * VersionNumber.
1942 		 *
1943 		 * Guest should be aware of supported eVMCS versions by host by
1944 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1945 		 * expected to set this CPUID leaf according to the value
1946 		 * returned in vmcs_version from nested_enable_evmcs().
1947 		 *
1948 		 * However, it turns out that Microsoft Hyper-V fails to comply
1949 		 * to their own invented interface: When Hyper-V use eVMCS, it
1950 		 * just sets first u32 field of eVMCS to revision_id specified
1951 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1952 		 * which is one of the supported versions specified in
1953 		 * CPUID.0x4000000A.EAX[0:15].
1954 		 *
1955 		 * To overcome Hyper-V bug, we accept here either a supported
1956 		 * eVMCS version or VMCS12 revision_id as valid values for first
1957 		 * u32 field of eVMCS.
1958 		 */
1959 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1960 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1961 			nested_release_evmcs(vcpu);
1962 			return 0;
1963 		}
1964 
1965 		vmx->nested.dirty_vmcs12 = true;
1966 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
1967 
1968 		evmcs_gpa_changed = true;
1969 		/*
1970 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
1971 		 * reloaded from guest's memory (read only fields, fields not
1972 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
1973 		 * are no leftovers.
1974 		 */
1975 		if (from_launch) {
1976 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1977 			memset(vmcs12, 0, sizeof(*vmcs12));
1978 			vmcs12->hdr.revision_id = VMCS12_REVISION;
1979 		}
1980 
1981 	}
1982 
1983 	/*
1984 	 * Clean fields data can't de used on VMLAUNCH and when we switch
1985 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
1986 	 */
1987 	if (from_launch || evmcs_gpa_changed)
1988 		vmx->nested.hv_evmcs->hv_clean_fields &=
1989 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1990 
1991 	return 1;
1992 }
1993 
1994 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
1995 {
1996 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1997 
1998 	/*
1999 	 * hv_evmcs may end up being not mapped after migration (when
2000 	 * L2 was running), map it here to make sure vmcs12 changes are
2001 	 * properly reflected.
2002 	 */
2003 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs)
2004 		nested_vmx_handle_enlightened_vmptrld(vcpu, false);
2005 
2006 	if (vmx->nested.hv_evmcs) {
2007 		copy_vmcs12_to_enlightened(vmx);
2008 		/* All fields are clean */
2009 		vmx->nested.hv_evmcs->hv_clean_fields |=
2010 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2011 	} else {
2012 		copy_vmcs12_to_shadow(vmx);
2013 	}
2014 
2015 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2016 }
2017 
2018 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2019 {
2020 	struct vcpu_vmx *vmx =
2021 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2022 
2023 	vmx->nested.preemption_timer_expired = true;
2024 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2025 	kvm_vcpu_kick(&vmx->vcpu);
2026 
2027 	return HRTIMER_NORESTART;
2028 }
2029 
2030 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
2031 {
2032 	u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
2033 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2034 
2035 	/*
2036 	 * A timer value of zero is architecturally guaranteed to cause
2037 	 * a VMExit prior to executing any instructions in the guest.
2038 	 */
2039 	if (preemption_timeout == 0) {
2040 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2041 		return;
2042 	}
2043 
2044 	if (vcpu->arch.virtual_tsc_khz == 0)
2045 		return;
2046 
2047 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2048 	preemption_timeout *= 1000000;
2049 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2050 	hrtimer_start(&vmx->nested.preemption_timer,
2051 		      ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
2052 }
2053 
2054 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2055 {
2056 	if (vmx->nested.nested_run_pending &&
2057 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2058 		return vmcs12->guest_ia32_efer;
2059 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2060 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2061 	else
2062 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2063 }
2064 
2065 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2066 {
2067 	/*
2068 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2069 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2070 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2071 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2072 	 */
2073 	if (vmx->nested.vmcs02_initialized)
2074 		return;
2075 	vmx->nested.vmcs02_initialized = true;
2076 
2077 	/*
2078 	 * We don't care what the EPTP value is we just need to guarantee
2079 	 * it's valid so we don't get a false positive when doing early
2080 	 * consistency checks.
2081 	 */
2082 	if (enable_ept && nested_early_check)
2083 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
2084 
2085 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2086 	if (cpu_has_vmx_vmfunc())
2087 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2088 
2089 	if (cpu_has_vmx_posted_intr())
2090 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2091 
2092 	if (cpu_has_vmx_msr_bitmap())
2093 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2094 
2095 	/*
2096 	 * The PML address never changes, so it is constant in vmcs02.
2097 	 * Conceptually we want to copy the PML index from vmcs01 here,
2098 	 * and then back to vmcs01 on nested vmexit.  But since we flush
2099 	 * the log and reset GUEST_PML_INDEX on each vmexit, the PML
2100 	 * index is also effectively constant in vmcs02.
2101 	 */
2102 	if (enable_pml) {
2103 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
2104 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2105 	}
2106 
2107 	if (cpu_has_vmx_encls_vmexit())
2108 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2109 
2110 	/*
2111 	 * Set the MSR load/store lists to match L0's settings.  Only the
2112 	 * addresses are constant (for vmcs02), the counts can change based
2113 	 * on L2's behavior, e.g. switching to/from long mode.
2114 	 */
2115 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2116 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2117 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2118 
2119 	vmx_set_constant_host_state(vmx);
2120 }
2121 
2122 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2123 				      struct vmcs12 *vmcs12)
2124 {
2125 	prepare_vmcs02_constant_state(vmx);
2126 
2127 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2128 
2129 	if (enable_vpid) {
2130 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2131 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2132 		else
2133 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2134 	}
2135 }
2136 
2137 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2138 {
2139 	u32 exec_control, vmcs12_exec_ctrl;
2140 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2141 
2142 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
2143 		prepare_vmcs02_early_rare(vmx, vmcs12);
2144 
2145 	/*
2146 	 * PIN CONTROLS
2147 	 */
2148 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2149 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2150 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2151 
2152 	/* Posted interrupts setting is only taken from vmcs12.  */
2153 	if (nested_cpu_has_posted_intr(vmcs12)) {
2154 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2155 		vmx->nested.pi_pending = false;
2156 	} else {
2157 		exec_control &= ~PIN_BASED_POSTED_INTR;
2158 	}
2159 	pin_controls_set(vmx, exec_control);
2160 
2161 	/*
2162 	 * EXEC CONTROLS
2163 	 */
2164 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2165 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2166 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2167 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2168 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2169 
2170 	vmx->nested.l1_tpr_threshold = -1;
2171 	if (exec_control & CPU_BASED_TPR_SHADOW)
2172 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2173 #ifdef CONFIG_X86_64
2174 	else
2175 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2176 				CPU_BASED_CR8_STORE_EXITING;
2177 #endif
2178 
2179 	/*
2180 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2181 	 * for I/O port accesses.
2182 	 */
2183 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2184 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2185 
2186 	/*
2187 	 * This bit will be computed in nested_get_vmcs12_pages, because
2188 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2189 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2190 	 * only set/clear this bit.
2191 	 */
2192 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2193 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2194 
2195 	exec_controls_set(vmx, exec_control);
2196 
2197 	/*
2198 	 * SECONDARY EXEC CONTROLS
2199 	 */
2200 	if (cpu_has_secondary_exec_ctrls()) {
2201 		exec_control = vmx->secondary_exec_control;
2202 
2203 		/* Take the following fields only from vmcs12 */
2204 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2205 				  SECONDARY_EXEC_ENABLE_INVPCID |
2206 				  SECONDARY_EXEC_RDTSCP |
2207 				  SECONDARY_EXEC_XSAVES |
2208 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2209 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2210 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2211 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2212 		if (nested_cpu_has(vmcs12,
2213 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2214 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2215 				~SECONDARY_EXEC_ENABLE_PML;
2216 			exec_control |= vmcs12_exec_ctrl;
2217 		}
2218 
2219 		/* VMCS shadowing for L2 is emulated for now */
2220 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2221 
2222 		/*
2223 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2224 		 * will not have to rewrite the controls just for this bit.
2225 		 */
2226 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2227 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2228 			exec_control |= SECONDARY_EXEC_DESC;
2229 
2230 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2231 			vmcs_write16(GUEST_INTR_STATUS,
2232 				vmcs12->guest_intr_status);
2233 
2234 		secondary_exec_controls_set(vmx, exec_control);
2235 	}
2236 
2237 	/*
2238 	 * ENTRY CONTROLS
2239 	 *
2240 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2241 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2242 	 * on the related bits (if supported by the CPU) in the hope that
2243 	 * we can avoid VMWrites during vmx_set_efer().
2244 	 */
2245 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2246 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2247 	if (cpu_has_load_ia32_efer()) {
2248 		if (guest_efer & EFER_LMA)
2249 			exec_control |= VM_ENTRY_IA32E_MODE;
2250 		if (guest_efer != host_efer)
2251 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2252 	}
2253 	vm_entry_controls_set(vmx, exec_control);
2254 
2255 	/*
2256 	 * EXIT CONTROLS
2257 	 *
2258 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2259 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2260 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2261 	 */
2262 	exec_control = vmx_vmexit_ctrl();
2263 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2264 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2265 	vm_exit_controls_set(vmx, exec_control);
2266 
2267 	/*
2268 	 * Interrupt/Exception Fields
2269 	 */
2270 	if (vmx->nested.nested_run_pending) {
2271 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2272 			     vmcs12->vm_entry_intr_info_field);
2273 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2274 			     vmcs12->vm_entry_exception_error_code);
2275 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2276 			     vmcs12->vm_entry_instruction_len);
2277 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2278 			     vmcs12->guest_interruptibility_info);
2279 		vmx->loaded_vmcs->nmi_known_unmasked =
2280 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2281 	} else {
2282 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2283 	}
2284 }
2285 
2286 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2287 {
2288 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2289 
2290 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2291 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2292 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2293 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2294 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2295 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2296 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2297 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2298 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2299 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2300 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2301 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2302 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2303 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2304 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2305 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2306 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2307 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2308 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2309 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2310 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2311 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2312 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2313 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2314 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2315 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2316 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2317 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2318 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2319 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2320 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2321 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2322 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2323 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2324 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2325 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2326 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2327 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2328 	}
2329 
2330 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2331 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2332 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2333 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2334 			    vmcs12->guest_pending_dbg_exceptions);
2335 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2336 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2337 
2338 		/*
2339 		 * L1 may access the L2's PDPTR, so save them to construct
2340 		 * vmcs12
2341 		 */
2342 		if (enable_ept) {
2343 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2344 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2345 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2346 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2347 		}
2348 
2349 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2350 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2351 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2352 	}
2353 
2354 	if (nested_cpu_has_xsaves(vmcs12))
2355 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2356 
2357 	/*
2358 	 * Whether page-faults are trapped is determined by a combination of
2359 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2360 	 * If enable_ept, L0 doesn't care about page faults and we should
2361 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2362 	 * care about (at least some) page faults, and because it is not easy
2363 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2364 	 * to exit on each and every L2 page fault. This is done by setting
2365 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2366 	 * Note that below we don't need special code to set EB.PF beyond the
2367 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2368 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2369 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2370 	 */
2371 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2372 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2373 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2374 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2375 
2376 	if (cpu_has_vmx_apicv()) {
2377 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2378 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2379 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2380 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2381 	}
2382 
2383 	/*
2384 	 * Make sure the msr_autostore list is up to date before we set the
2385 	 * count in the vmcs02.
2386 	 */
2387 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2388 
2389 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2390 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2391 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2392 
2393 	set_cr4_guest_host_mask(vmx);
2394 }
2395 
2396 /*
2397  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2398  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2399  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2400  * guest in a way that will both be appropriate to L1's requests, and our
2401  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2402  * function also has additional necessary side-effects, like setting various
2403  * vcpu->arch fields.
2404  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2405  * is assigned to entry_failure_code on failure.
2406  */
2407 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2408 			  u32 *entry_failure_code)
2409 {
2410 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2411 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2412 	bool load_guest_pdptrs_vmcs12 = false;
2413 
2414 	if (vmx->nested.dirty_vmcs12 || hv_evmcs) {
2415 		prepare_vmcs02_rare(vmx, vmcs12);
2416 		vmx->nested.dirty_vmcs12 = false;
2417 
2418 		load_guest_pdptrs_vmcs12 = !hv_evmcs ||
2419 			!(hv_evmcs->hv_clean_fields &
2420 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2421 	}
2422 
2423 	if (vmx->nested.nested_run_pending &&
2424 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2425 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2426 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2427 	} else {
2428 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2429 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2430 	}
2431 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2432 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2433 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2434 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2435 
2436 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2437 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2438 	 * trap. Note that CR0.TS also needs updating - we do this later.
2439 	 */
2440 	update_exception_bitmap(vcpu);
2441 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2442 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2443 
2444 	if (vmx->nested.nested_run_pending &&
2445 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2446 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2447 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2448 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2449 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2450 	}
2451 
2452 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2453 
2454 	if (kvm_has_tsc_control)
2455 		decache_tsc_multiplier(vmx);
2456 
2457 	if (enable_vpid) {
2458 		/*
2459 		 * There is no direct mapping between vpid02 and vpid12, the
2460 		 * vpid02 is per-vCPU for L0 and reused while the value of
2461 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2462 		 * The vpid12 is allocated by L1 for L2, so it will not
2463 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2464 		 * even if spawn a lot of nested vCPUs.
2465 		 */
2466 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2467 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2468 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2469 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2470 			}
2471 		} else {
2472 			/*
2473 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2474 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2475 			 * flush until vmcs02->eptp is fully updated by
2476 			 * KVM_REQ_LOAD_CR3. Note that this assumes
2477 			 * KVM_REQ_TLB_FLUSH is evaluated after
2478 			 * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
2479 			 */
2480 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2481 		}
2482 	}
2483 
2484 	if (nested_cpu_has_ept(vmcs12))
2485 		nested_ept_init_mmu_context(vcpu);
2486 
2487 	/*
2488 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2489 	 * bits which we consider mandatory enabled.
2490 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2491 	 * the specifications by L1; It's not enough to take
2492 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2493 	 * have more bits than L1 expected.
2494 	 */
2495 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2496 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2497 
2498 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2499 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2500 
2501 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2502 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2503 	vmx_set_efer(vcpu, vcpu->arch.efer);
2504 
2505 	/*
2506 	 * Guest state is invalid and unrestricted guest is disabled,
2507 	 * which means L1 attempted VMEntry to L2 with invalid state.
2508 	 * Fail the VMEntry.
2509 	 */
2510 	if (vmx->emulation_required) {
2511 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2512 		return -EINVAL;
2513 	}
2514 
2515 	/* Shadow page tables on either EPT or shadow page tables. */
2516 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2517 				entry_failure_code))
2518 		return -EINVAL;
2519 
2520 	/*
2521 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2522 	 * on nested VM-Exit, which can occur without actually running L2 and
2523 	 * thus without hitting vmx_set_cr3(), e.g. if L1 is entering L2 with
2524 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2525 	 * transition to HLT instead of running L2.
2526 	 */
2527 	if (enable_ept)
2528 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2529 
2530 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2531 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2532 	    is_pae_paging(vcpu)) {
2533 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2534 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2535 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2536 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2537 	}
2538 
2539 	if (!enable_ept)
2540 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2541 
2542 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2543 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2544 				     vmcs12->guest_ia32_perf_global_ctrl)))
2545 		return -EINVAL;
2546 
2547 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2548 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2549 	return 0;
2550 }
2551 
2552 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2553 {
2554 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2555 	       nested_cpu_has_virtual_nmis(vmcs12)))
2556 		return -EINVAL;
2557 
2558 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2559 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2560 		return -EINVAL;
2561 
2562 	return 0;
2563 }
2564 
2565 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
2566 {
2567 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2568 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2569 
2570 	/* Check for memory type validity */
2571 	switch (address & VMX_EPTP_MT_MASK) {
2572 	case VMX_EPTP_MT_UC:
2573 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2574 			return false;
2575 		break;
2576 	case VMX_EPTP_MT_WB:
2577 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2578 			return false;
2579 		break;
2580 	default:
2581 		return false;
2582 	}
2583 
2584 	/* only 4 levels page-walk length are valid */
2585 	if (CC((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4))
2586 		return false;
2587 
2588 	/* Reserved bits should not be set */
2589 	if (CC(address >> maxphyaddr || ((address >> 7) & 0x1f)))
2590 		return false;
2591 
2592 	/* AD, if set, should be supported */
2593 	if (address & VMX_EPTP_AD_ENABLE_BIT) {
2594 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2595 			return false;
2596 	}
2597 
2598 	return true;
2599 }
2600 
2601 /*
2602  * Checks related to VM-Execution Control Fields
2603  */
2604 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2605                                               struct vmcs12 *vmcs12)
2606 {
2607 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2608 
2609 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2610 				   vmx->nested.msrs.pinbased_ctls_low,
2611 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2612 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2613 				   vmx->nested.msrs.procbased_ctls_low,
2614 				   vmx->nested.msrs.procbased_ctls_high)))
2615 		return -EINVAL;
2616 
2617 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2618 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2619 				   vmx->nested.msrs.secondary_ctls_low,
2620 				   vmx->nested.msrs.secondary_ctls_high)))
2621 		return -EINVAL;
2622 
2623 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2624 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2625 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2626 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2627 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2628 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2629 	    nested_vmx_check_nmi_controls(vmcs12) ||
2630 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2631 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2632 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2633 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2634 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2635 		return -EINVAL;
2636 
2637 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2638 	    nested_cpu_has_save_preemption_timer(vmcs12))
2639 		return -EINVAL;
2640 
2641 	if (nested_cpu_has_ept(vmcs12) &&
2642 	    CC(!valid_ept_address(vcpu, vmcs12->ept_pointer)))
2643 		return -EINVAL;
2644 
2645 	if (nested_cpu_has_vmfunc(vmcs12)) {
2646 		if (CC(vmcs12->vm_function_control &
2647 		       ~vmx->nested.msrs.vmfunc_controls))
2648 			return -EINVAL;
2649 
2650 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2651 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2652 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2653 				return -EINVAL;
2654 		}
2655 	}
2656 
2657 	return 0;
2658 }
2659 
2660 /*
2661  * Checks related to VM-Exit Control Fields
2662  */
2663 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2664                                          struct vmcs12 *vmcs12)
2665 {
2666 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2667 
2668 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2669 				    vmx->nested.msrs.exit_ctls_low,
2670 				    vmx->nested.msrs.exit_ctls_high)) ||
2671 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2672 		return -EINVAL;
2673 
2674 	return 0;
2675 }
2676 
2677 /*
2678  * Checks related to VM-Entry Control Fields
2679  */
2680 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2681 					  struct vmcs12 *vmcs12)
2682 {
2683 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2684 
2685 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2686 				    vmx->nested.msrs.entry_ctls_low,
2687 				    vmx->nested.msrs.entry_ctls_high)))
2688 		return -EINVAL;
2689 
2690 	/*
2691 	 * From the Intel SDM, volume 3:
2692 	 * Fields relevant to VM-entry event injection must be set properly.
2693 	 * These fields are the VM-entry interruption-information field, the
2694 	 * VM-entry exception error code, and the VM-entry instruction length.
2695 	 */
2696 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2697 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2698 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2699 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2700 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2701 		bool should_have_error_code;
2702 		bool urg = nested_cpu_has2(vmcs12,
2703 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2704 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2705 
2706 		/* VM-entry interruption-info field: interruption type */
2707 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2708 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2709 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2710 			return -EINVAL;
2711 
2712 		/* VM-entry interruption-info field: vector */
2713 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2714 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2715 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2716 			return -EINVAL;
2717 
2718 		/* VM-entry interruption-info field: deliver error code */
2719 		should_have_error_code =
2720 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2721 			x86_exception_has_error_code(vector);
2722 		if (CC(has_error_code != should_have_error_code))
2723 			return -EINVAL;
2724 
2725 		/* VM-entry exception error code */
2726 		if (CC(has_error_code &&
2727 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2728 			return -EINVAL;
2729 
2730 		/* VM-entry interruption-info field: reserved bits */
2731 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2732 			return -EINVAL;
2733 
2734 		/* VM-entry instruction length */
2735 		switch (intr_type) {
2736 		case INTR_TYPE_SOFT_EXCEPTION:
2737 		case INTR_TYPE_SOFT_INTR:
2738 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2739 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2740 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2741 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2742 				return -EINVAL;
2743 		}
2744 	}
2745 
2746 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2747 		return -EINVAL;
2748 
2749 	return 0;
2750 }
2751 
2752 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2753 				     struct vmcs12 *vmcs12)
2754 {
2755 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2756 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2757 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2758 		return -EINVAL;
2759 
2760 	return 0;
2761 }
2762 
2763 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2764 				       struct vmcs12 *vmcs12)
2765 {
2766 	bool ia32e;
2767 
2768 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2769 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2770 	    CC(!nested_cr3_valid(vcpu, vmcs12->host_cr3)))
2771 		return -EINVAL;
2772 
2773 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2774 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2775 		return -EINVAL;
2776 
2777 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2778 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2779 		return -EINVAL;
2780 
2781 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2782 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2783 					   vmcs12->host_ia32_perf_global_ctrl)))
2784 		return -EINVAL;
2785 
2786 #ifdef CONFIG_X86_64
2787 	ia32e = !!(vcpu->arch.efer & EFER_LMA);
2788 #else
2789 	ia32e = false;
2790 #endif
2791 
2792 	if (ia32e) {
2793 		if (CC(!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)) ||
2794 		    CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2795 			return -EINVAL;
2796 	} else {
2797 		if (CC(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) ||
2798 		    CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2799 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2800 		    CC((vmcs12->host_rip) >> 32))
2801 			return -EINVAL;
2802 	}
2803 
2804 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2805 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2806 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2807 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2808 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2809 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2810 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2811 	    CC(vmcs12->host_cs_selector == 0) ||
2812 	    CC(vmcs12->host_tr_selector == 0) ||
2813 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2814 		return -EINVAL;
2815 
2816 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2817 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2818 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2819 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2820 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2821 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2822 		return -EINVAL;
2823 
2824 	/*
2825 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2826 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2827 	 * the values of the LMA and LME bits in the field must each be that of
2828 	 * the host address-space size VM-exit control.
2829 	 */
2830 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2831 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2832 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2833 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2834 			return -EINVAL;
2835 	}
2836 
2837 	return 0;
2838 }
2839 
2840 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2841 					  struct vmcs12 *vmcs12)
2842 {
2843 	int r = 0;
2844 	struct vmcs12 *shadow;
2845 	struct kvm_host_map map;
2846 
2847 	if (vmcs12->vmcs_link_pointer == -1ull)
2848 		return 0;
2849 
2850 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2851 		return -EINVAL;
2852 
2853 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2854 		return -EINVAL;
2855 
2856 	shadow = map.hva;
2857 
2858 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2859 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2860 		r = -EINVAL;
2861 
2862 	kvm_vcpu_unmap(vcpu, &map, false);
2863 	return r;
2864 }
2865 
2866 /*
2867  * Checks related to Guest Non-register State
2868  */
2869 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2870 {
2871 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2872 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT))
2873 		return -EINVAL;
2874 
2875 	return 0;
2876 }
2877 
2878 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2879 					struct vmcs12 *vmcs12,
2880 					u32 *exit_qual)
2881 {
2882 	bool ia32e;
2883 
2884 	*exit_qual = ENTRY_FAIL_DEFAULT;
2885 
2886 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2887 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2888 		return -EINVAL;
2889 
2890 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2891 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2892 		return -EINVAL;
2893 
2894 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2895 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2896 		return -EINVAL;
2897 
2898 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2899 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2900 		return -EINVAL;
2901 	}
2902 
2903 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2904 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2905 					   vmcs12->guest_ia32_perf_global_ctrl)))
2906 		return -EINVAL;
2907 
2908 	/*
2909 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2910 	 * are performed on the field for the IA32_EFER MSR:
2911 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2912 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2913 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2914 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2915 	 *   CR0.PG) is 1.
2916 	 */
2917 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2918 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2919 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2920 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
2921 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
2922 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
2923 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
2924 			return -EINVAL;
2925 	}
2926 
2927 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2928 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
2929 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
2930 		return -EINVAL;
2931 
2932 	if (nested_check_guest_non_reg_state(vmcs12))
2933 		return -EINVAL;
2934 
2935 	return 0;
2936 }
2937 
2938 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2939 {
2940 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2941 	unsigned long cr3, cr4;
2942 	bool vm_fail;
2943 
2944 	if (!nested_early_check)
2945 		return 0;
2946 
2947 	if (vmx->msr_autoload.host.nr)
2948 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2949 	if (vmx->msr_autoload.guest.nr)
2950 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2951 
2952 	preempt_disable();
2953 
2954 	vmx_prepare_switch_to_guest(vcpu);
2955 
2956 	/*
2957 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2958 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2959 	 * be written (by preparve_vmcs02()) before the "real" VMEnter, i.e.
2960 	 * there is no need to preserve other bits or save/restore the field.
2961 	 */
2962 	vmcs_writel(GUEST_RFLAGS, 0);
2963 
2964 	cr3 = __get_current_cr3_fast();
2965 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2966 		vmcs_writel(HOST_CR3, cr3);
2967 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2968 	}
2969 
2970 	cr4 = cr4_read_shadow();
2971 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2972 		vmcs_writel(HOST_CR4, cr4);
2973 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2974 	}
2975 
2976 	asm(
2977 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2978 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2979 		"je 1f \n\t"
2980 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
2981 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2982 		"1: \n\t"
2983 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2984 
2985 		/* Check if vmlaunch or vmresume is needed */
2986 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
2987 
2988 		/*
2989 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
2990 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
2991 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
2992 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
2993 		 */
2994 		"call vmx_vmenter\n\t"
2995 
2996 		CC_SET(be)
2997 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
2998 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
2999 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
3000 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
3001 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
3002 		[wordsize]"i"(sizeof(ulong))
3003 	      : "memory"
3004 	);
3005 
3006 	if (vmx->msr_autoload.host.nr)
3007 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3008 	if (vmx->msr_autoload.guest.nr)
3009 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3010 
3011 	if (vm_fail) {
3012 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3013 
3014 		preempt_enable();
3015 
3016 		trace_kvm_nested_vmenter_failed(
3017 			"early hardware check VM-instruction error: ", error);
3018 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3019 		return 1;
3020 	}
3021 
3022 	/*
3023 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3024 	 */
3025 	local_irq_enable();
3026 	if (hw_breakpoint_active())
3027 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3028 	preempt_enable();
3029 
3030 	/*
3031 	 * A non-failing VMEntry means we somehow entered guest mode with
3032 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3033 	 * is no telling what memory has been modified or what state has
3034 	 * been exposed to unknown code.  Hitting this all but guarantees
3035 	 * a (very critical) hardware issue.
3036 	 */
3037 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3038 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3039 
3040 	return 0;
3041 }
3042 
3043 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3044 {
3045 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3046 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3047 	struct kvm_host_map *map;
3048 	struct page *page;
3049 	u64 hpa;
3050 
3051 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3052 		/*
3053 		 * Translate L1 physical address to host physical
3054 		 * address for vmcs02. Keep the page pinned, so this
3055 		 * physical address remains valid. We keep a reference
3056 		 * to it so we can release it later.
3057 		 */
3058 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3059 			kvm_release_page_clean(vmx->nested.apic_access_page);
3060 			vmx->nested.apic_access_page = NULL;
3061 		}
3062 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3063 		if (!is_error_page(page)) {
3064 			vmx->nested.apic_access_page = page;
3065 			hpa = page_to_phys(vmx->nested.apic_access_page);
3066 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3067 		} else {
3068 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3069 					     __func__);
3070 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3071 			vcpu->run->internal.suberror =
3072 				KVM_INTERNAL_ERROR_EMULATION;
3073 			vcpu->run->internal.ndata = 0;
3074 			return false;
3075 		}
3076 	}
3077 
3078 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3079 		map = &vmx->nested.virtual_apic_map;
3080 
3081 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3082 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3083 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3084 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3085 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3086 			/*
3087 			 * The processor will never use the TPR shadow, simply
3088 			 * clear the bit from the execution control.  Such a
3089 			 * configuration is useless, but it happens in tests.
3090 			 * For any other configuration, failing the vm entry is
3091 			 * _not_ what the processor does but it's basically the
3092 			 * only possibility we have.
3093 			 */
3094 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3095 		} else {
3096 			/*
3097 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3098 			 * force VM-Entry to fail.
3099 			 */
3100 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
3101 		}
3102 	}
3103 
3104 	if (nested_cpu_has_posted_intr(vmcs12)) {
3105 		map = &vmx->nested.pi_desc_map;
3106 
3107 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3108 			vmx->nested.pi_desc =
3109 				(struct pi_desc *)(((void *)map->hva) +
3110 				offset_in_page(vmcs12->posted_intr_desc_addr));
3111 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3112 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3113 		}
3114 	}
3115 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3116 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3117 	else
3118 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3119 	return true;
3120 }
3121 
3122 /*
3123  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3124  * for running VMX instructions (except VMXON, whose prerequisites are
3125  * slightly different). It also specifies what exception to inject otherwise.
3126  * Note that many of these exceptions have priority over VM exits, so they
3127  * don't have to be checked again here.
3128  */
3129 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3130 {
3131 	if (!to_vmx(vcpu)->nested.vmxon) {
3132 		kvm_queue_exception(vcpu, UD_VECTOR);
3133 		return 0;
3134 	}
3135 
3136 	if (vmx_get_cpl(vcpu)) {
3137 		kvm_inject_gp(vcpu, 0);
3138 		return 0;
3139 	}
3140 
3141 	return 1;
3142 }
3143 
3144 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3145 {
3146 	u8 rvi = vmx_get_rvi();
3147 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3148 
3149 	return ((rvi & 0xf0) > (vppr & 0xf0));
3150 }
3151 
3152 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3153 				   struct vmcs12 *vmcs12);
3154 
3155 /*
3156  * If from_vmentry is false, this is being called from state restore (either RSM
3157  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3158  *
3159  * Returns:
3160  *	NVMX_ENTRY_SUCCESS: Entered VMX non-root mode
3161  *	NVMX_ENTRY_VMFAIL:  Consistency check VMFail
3162  *	NVMX_ENTRY_VMEXIT:  Consistency check VMExit
3163  *	NVMX_ENTRY_KVM_INTERNAL_ERROR: KVM internal error
3164  */
3165 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3166 							bool from_vmentry)
3167 {
3168 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3169 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3170 	bool evaluate_pending_interrupts;
3171 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
3172 	u32 exit_qual;
3173 
3174 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3175 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3176 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3177 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3178 
3179 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3180 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3181 	if (kvm_mpx_supported() &&
3182 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3183 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3184 
3185 	/*
3186 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3187 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3188 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3189 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3190 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3191 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3192 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3193 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3194 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3195 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3196 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3197 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3198 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3199 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3200 	 */
3201 	if (!enable_ept && !nested_early_check)
3202 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3203 
3204 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3205 
3206 	prepare_vmcs02_early(vmx, vmcs12);
3207 
3208 	if (from_vmentry) {
3209 		if (unlikely(!nested_get_vmcs12_pages(vcpu)))
3210 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3211 
3212 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3213 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3214 			return NVMX_VMENTRY_VMFAIL;
3215 		}
3216 
3217 		if (nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
3218 			goto vmentry_fail_vmexit;
3219 	}
3220 
3221 	enter_guest_mode(vcpu);
3222 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3223 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3224 
3225 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
3226 		goto vmentry_fail_vmexit_guest_mode;
3227 
3228 	if (from_vmentry) {
3229 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3230 		exit_qual = nested_vmx_load_msr(vcpu,
3231 						vmcs12->vm_entry_msr_load_addr,
3232 						vmcs12->vm_entry_msr_load_count);
3233 		if (exit_qual)
3234 			goto vmentry_fail_vmexit_guest_mode;
3235 	} else {
3236 		/*
3237 		 * The MMU is not initialized to point at the right entities yet and
3238 		 * "get pages" would need to read data from the guest (i.e. we will
3239 		 * need to perform gpa to hpa translation). Request a call
3240 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3241 		 * have already been set at vmentry time and should not be reset.
3242 		 */
3243 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3244 	}
3245 
3246 	/*
3247 	 * If L1 had a pending IRQ/NMI until it executed
3248 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3249 	 * disallowed (e.g. interrupts disabled), L0 needs to
3250 	 * evaluate if this pending event should cause an exit from L2
3251 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3252 	 * intercept EXTERNAL_INTERRUPT).
3253 	 *
3254 	 * Usually this would be handled by the processor noticing an
3255 	 * IRQ/NMI window request, or checking RVI during evaluation of
3256 	 * pending virtual interrupts.  However, this setting was done
3257 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3258 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3259 	 */
3260 	if (unlikely(evaluate_pending_interrupts))
3261 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3262 
3263 	/*
3264 	 * Do not start the preemption timer hrtimer until after we know
3265 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3266 	 * the timer.
3267 	 */
3268 	vmx->nested.preemption_timer_expired = false;
3269 	if (nested_cpu_has_preemption_timer(vmcs12))
3270 		vmx_start_preemption_timer(vcpu);
3271 
3272 	/*
3273 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3274 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3275 	 * returned as far as L1 is concerned. It will only return (and set
3276 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3277 	 */
3278 	return NVMX_VMENTRY_SUCCESS;
3279 
3280 	/*
3281 	 * A failed consistency check that leads to a VMExit during L1's
3282 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3283 	 * 26.7 "VM-entry failures during or after loading guest state".
3284 	 */
3285 vmentry_fail_vmexit_guest_mode:
3286 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3287 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3288 	leave_guest_mode(vcpu);
3289 
3290 vmentry_fail_vmexit:
3291 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3292 
3293 	if (!from_vmentry)
3294 		return NVMX_VMENTRY_VMEXIT;
3295 
3296 	load_vmcs12_host_state(vcpu, vmcs12);
3297 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3298 	vmcs12->exit_qualification = exit_qual;
3299 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3300 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3301 	return NVMX_VMENTRY_VMEXIT;
3302 }
3303 
3304 /*
3305  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3306  * for running an L2 nested guest.
3307  */
3308 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3309 {
3310 	struct vmcs12 *vmcs12;
3311 	enum nvmx_vmentry_status status;
3312 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3313 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3314 
3315 	if (!nested_vmx_check_permission(vcpu))
3316 		return 1;
3317 
3318 	if (!nested_vmx_handle_enlightened_vmptrld(vcpu, launch))
3319 		return 1;
3320 
3321 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3322 		return nested_vmx_failInvalid(vcpu);
3323 
3324 	vmcs12 = get_vmcs12(vcpu);
3325 
3326 	/*
3327 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3328 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3329 	 * rather than RFLAGS.ZF, and no error number is stored to the
3330 	 * VM-instruction error field.
3331 	 */
3332 	if (vmcs12->hdr.shadow_vmcs)
3333 		return nested_vmx_failInvalid(vcpu);
3334 
3335 	if (vmx->nested.hv_evmcs) {
3336 		copy_enlightened_to_vmcs12(vmx);
3337 		/* Enlightened VMCS doesn't have launch state */
3338 		vmcs12->launch_state = !launch;
3339 	} else if (enable_shadow_vmcs) {
3340 		copy_shadow_to_vmcs12(vmx);
3341 	}
3342 
3343 	/*
3344 	 * The nested entry process starts with enforcing various prerequisites
3345 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3346 	 * they fail: As the SDM explains, some conditions should cause the
3347 	 * instruction to fail, while others will cause the instruction to seem
3348 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3349 	 * To speed up the normal (success) code path, we should avoid checking
3350 	 * for misconfigurations which will anyway be caught by the processor
3351 	 * when using the merged vmcs02.
3352 	 */
3353 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3354 		return nested_vmx_failValid(vcpu,
3355 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3356 
3357 	if (vmcs12->launch_state == launch)
3358 		return nested_vmx_failValid(vcpu,
3359 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3360 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3361 
3362 	if (nested_vmx_check_controls(vcpu, vmcs12))
3363 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3364 
3365 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3366 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3367 
3368 	/*
3369 	 * We're finally done with prerequisite checking, and can start with
3370 	 * the nested entry.
3371 	 */
3372 	vmx->nested.nested_run_pending = 1;
3373 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3374 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3375 		goto vmentry_failed;
3376 
3377 	/* Hide L1D cache contents from the nested guest.  */
3378 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3379 
3380 	/*
3381 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3382 	 * also be used as part of restoring nVMX state for
3383 	 * snapshot restore (migration).
3384 	 *
3385 	 * In this flow, it is assumed that vmcs12 cache was
3386 	 * trasferred as part of captured nVMX state and should
3387 	 * therefore not be read from guest memory (which may not
3388 	 * exist on destination host yet).
3389 	 */
3390 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3391 
3392 	/*
3393 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3394 	 * awakened by event injection or by an NMI-window VM-exit or
3395 	 * by an interrupt-window VM-exit, halt the vcpu.
3396 	 */
3397 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3398 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3399 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_NMI_WINDOW_EXITING) &&
3400 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_INTR_WINDOW_EXITING) &&
3401 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3402 		vmx->nested.nested_run_pending = 0;
3403 		return kvm_vcpu_halt(vcpu);
3404 	}
3405 	return 1;
3406 
3407 vmentry_failed:
3408 	vmx->nested.nested_run_pending = 0;
3409 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3410 		return 0;
3411 	if (status == NVMX_VMENTRY_VMEXIT)
3412 		return 1;
3413 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3414 	return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3415 }
3416 
3417 /*
3418  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3419  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3420  * This function returns the new value we should put in vmcs12.guest_cr0.
3421  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3422  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3423  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3424  *     didn't trap the bit, because if L1 did, so would L0).
3425  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3426  *     been modified by L2, and L1 knows it. So just leave the old value of
3427  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3428  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3429  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3430  *     changed these bits, and therefore they need to be updated, but L0
3431  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3432  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3433  */
3434 static inline unsigned long
3435 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3436 {
3437 	return
3438 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3439 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3440 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3441 			vcpu->arch.cr0_guest_owned_bits));
3442 }
3443 
3444 static inline unsigned long
3445 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3446 {
3447 	return
3448 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3449 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3450 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3451 			vcpu->arch.cr4_guest_owned_bits));
3452 }
3453 
3454 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3455 				      struct vmcs12 *vmcs12)
3456 {
3457 	u32 idt_vectoring;
3458 	unsigned int nr;
3459 
3460 	if (vcpu->arch.exception.injected) {
3461 		nr = vcpu->arch.exception.nr;
3462 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3463 
3464 		if (kvm_exception_is_soft(nr)) {
3465 			vmcs12->vm_exit_instruction_len =
3466 				vcpu->arch.event_exit_inst_len;
3467 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3468 		} else
3469 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3470 
3471 		if (vcpu->arch.exception.has_error_code) {
3472 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3473 			vmcs12->idt_vectoring_error_code =
3474 				vcpu->arch.exception.error_code;
3475 		}
3476 
3477 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3478 	} else if (vcpu->arch.nmi_injected) {
3479 		vmcs12->idt_vectoring_info_field =
3480 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3481 	} else if (vcpu->arch.interrupt.injected) {
3482 		nr = vcpu->arch.interrupt.nr;
3483 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3484 
3485 		if (vcpu->arch.interrupt.soft) {
3486 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3487 			vmcs12->vm_entry_instruction_len =
3488 				vcpu->arch.event_exit_inst_len;
3489 		} else
3490 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3491 
3492 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3493 	}
3494 }
3495 
3496 
3497 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3498 {
3499 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3500 	gfn_t gfn;
3501 
3502 	/*
3503 	 * Don't need to mark the APIC access page dirty; it is never
3504 	 * written to by the CPU during APIC virtualization.
3505 	 */
3506 
3507 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3508 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3509 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3510 	}
3511 
3512 	if (nested_cpu_has_posted_intr(vmcs12)) {
3513 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3514 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3515 	}
3516 }
3517 
3518 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3519 {
3520 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3521 	int max_irr;
3522 	void *vapic_page;
3523 	u16 status;
3524 
3525 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3526 		return;
3527 
3528 	vmx->nested.pi_pending = false;
3529 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3530 		return;
3531 
3532 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3533 	if (max_irr != 256) {
3534 		vapic_page = vmx->nested.virtual_apic_map.hva;
3535 		if (!vapic_page)
3536 			return;
3537 
3538 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3539 			vapic_page, &max_irr);
3540 		status = vmcs_read16(GUEST_INTR_STATUS);
3541 		if ((u8)max_irr > ((u8)status & 0xff)) {
3542 			status &= ~0xff;
3543 			status |= (u8)max_irr;
3544 			vmcs_write16(GUEST_INTR_STATUS, status);
3545 		}
3546 	}
3547 
3548 	nested_mark_vmcs12_pages_dirty(vcpu);
3549 }
3550 
3551 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3552 					       unsigned long exit_qual)
3553 {
3554 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3555 	unsigned int nr = vcpu->arch.exception.nr;
3556 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3557 
3558 	if (vcpu->arch.exception.has_error_code) {
3559 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3560 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3561 	}
3562 
3563 	if (kvm_exception_is_soft(nr))
3564 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3565 	else
3566 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3567 
3568 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3569 	    vmx_get_nmi_mask(vcpu))
3570 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3571 
3572 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3573 }
3574 
3575 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
3576 {
3577 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3578 	unsigned long exit_qual;
3579 	bool block_nested_events =
3580 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3581 	struct kvm_lapic *apic = vcpu->arch.apic;
3582 
3583 	if (lapic_in_kernel(vcpu) &&
3584 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3585 		if (block_nested_events)
3586 			return -EBUSY;
3587 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3588 		nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3589 		return 0;
3590 	}
3591 
3592 	if (vcpu->arch.exception.pending &&
3593 		nested_vmx_check_exception(vcpu, &exit_qual)) {
3594 		if (block_nested_events)
3595 			return -EBUSY;
3596 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3597 		return 0;
3598 	}
3599 
3600 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3601 	    vmx->nested.preemption_timer_expired) {
3602 		if (block_nested_events)
3603 			return -EBUSY;
3604 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3605 		return 0;
3606 	}
3607 
3608 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3609 		if (block_nested_events)
3610 			return -EBUSY;
3611 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3612 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3613 				  INTR_INFO_VALID_MASK, 0);
3614 		/*
3615 		 * The NMI-triggered VM exit counts as injection:
3616 		 * clear this one and block further NMIs.
3617 		 */
3618 		vcpu->arch.nmi_pending = 0;
3619 		vmx_set_nmi_mask(vcpu, true);
3620 		return 0;
3621 	}
3622 
3623 	if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
3624 	    nested_exit_on_intr(vcpu)) {
3625 		if (block_nested_events)
3626 			return -EBUSY;
3627 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3628 		return 0;
3629 	}
3630 
3631 	vmx_complete_nested_posted_interrupt(vcpu);
3632 	return 0;
3633 }
3634 
3635 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3636 {
3637 	ktime_t remaining =
3638 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3639 	u64 value;
3640 
3641 	if (ktime_to_ns(remaining) <= 0)
3642 		return 0;
3643 
3644 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3645 	do_div(value, 1000000);
3646 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3647 }
3648 
3649 static bool is_vmcs12_ext_field(unsigned long field)
3650 {
3651 	switch (field) {
3652 	case GUEST_ES_SELECTOR:
3653 	case GUEST_CS_SELECTOR:
3654 	case GUEST_SS_SELECTOR:
3655 	case GUEST_DS_SELECTOR:
3656 	case GUEST_FS_SELECTOR:
3657 	case GUEST_GS_SELECTOR:
3658 	case GUEST_LDTR_SELECTOR:
3659 	case GUEST_TR_SELECTOR:
3660 	case GUEST_ES_LIMIT:
3661 	case GUEST_CS_LIMIT:
3662 	case GUEST_SS_LIMIT:
3663 	case GUEST_DS_LIMIT:
3664 	case GUEST_FS_LIMIT:
3665 	case GUEST_GS_LIMIT:
3666 	case GUEST_LDTR_LIMIT:
3667 	case GUEST_TR_LIMIT:
3668 	case GUEST_GDTR_LIMIT:
3669 	case GUEST_IDTR_LIMIT:
3670 	case GUEST_ES_AR_BYTES:
3671 	case GUEST_DS_AR_BYTES:
3672 	case GUEST_FS_AR_BYTES:
3673 	case GUEST_GS_AR_BYTES:
3674 	case GUEST_LDTR_AR_BYTES:
3675 	case GUEST_TR_AR_BYTES:
3676 	case GUEST_ES_BASE:
3677 	case GUEST_CS_BASE:
3678 	case GUEST_SS_BASE:
3679 	case GUEST_DS_BASE:
3680 	case GUEST_FS_BASE:
3681 	case GUEST_GS_BASE:
3682 	case GUEST_LDTR_BASE:
3683 	case GUEST_TR_BASE:
3684 	case GUEST_GDTR_BASE:
3685 	case GUEST_IDTR_BASE:
3686 	case GUEST_PENDING_DBG_EXCEPTIONS:
3687 	case GUEST_BNDCFGS:
3688 		return true;
3689 	default:
3690 		break;
3691 	}
3692 
3693 	return false;
3694 }
3695 
3696 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3697 				       struct vmcs12 *vmcs12)
3698 {
3699 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3700 
3701 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3702 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3703 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3704 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3705 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3706 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3707 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3708 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3709 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3710 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3711 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3712 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3713 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3714 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3715 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3716 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3717 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3718 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3719 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3720 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3721 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3722 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3723 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3724 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3725 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3726 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3727 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3728 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3729 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3730 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3731 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3732 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3733 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3734 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3735 	vmcs12->guest_pending_dbg_exceptions =
3736 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3737 	if (kvm_mpx_supported())
3738 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3739 
3740 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
3741 }
3742 
3743 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3744 				       struct vmcs12 *vmcs12)
3745 {
3746 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3747 	int cpu;
3748 
3749 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
3750 		return;
3751 
3752 
3753 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
3754 
3755 	cpu = get_cpu();
3756 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
3757 	vmx_vcpu_load(&vmx->vcpu, cpu);
3758 
3759 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3760 
3761 	vmx->loaded_vmcs = &vmx->vmcs01;
3762 	vmx_vcpu_load(&vmx->vcpu, cpu);
3763 	put_cpu();
3764 }
3765 
3766 /*
3767  * Update the guest state fields of vmcs12 to reflect changes that
3768  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3769  * VM-entry controls is also updated, since this is really a guest
3770  * state bit.)
3771  */
3772 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3773 {
3774 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3775 
3776 	if (vmx->nested.hv_evmcs)
3777 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3778 
3779 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = !vmx->nested.hv_evmcs;
3780 
3781 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3782 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3783 
3784 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
3785 	vmcs12->guest_rip = kvm_rip_read(vcpu);
3786 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3787 
3788 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3789 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3790 
3791 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3792 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3793 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3794 
3795 	vmcs12->guest_interruptibility_info =
3796 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3797 
3798 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3799 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3800 	else
3801 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3802 
3803 	if (nested_cpu_has_preemption_timer(vmcs12) &&
3804 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3805 			vmcs12->vmx_preemption_timer_value =
3806 				vmx_get_preemption_timer_value(vcpu);
3807 
3808 	/*
3809 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3810 	 * own CR3 without exiting. If it has changed it, we must keep it.
3811 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3812 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3813 	 *
3814 	 * Additionally, restore L2's PDPTR to vmcs12.
3815 	 */
3816 	if (enable_ept) {
3817 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3818 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3819 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3820 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3821 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3822 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3823 		}
3824 	}
3825 
3826 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3827 
3828 	if (nested_cpu_has_vid(vmcs12))
3829 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3830 
3831 	vmcs12->vm_entry_controls =
3832 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3833 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3834 
3835 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
3836 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3837 
3838 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3839 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3840 }
3841 
3842 /*
3843  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3844  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3845  * and this function updates it to reflect the changes to the guest state while
3846  * L2 was running (and perhaps made some exits which were handled directly by L0
3847  * without going back to L1), and to reflect the exit reason.
3848  * Note that we do not have to copy here all VMCS fields, just those that
3849  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3850  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3851  * which already writes to vmcs12 directly.
3852  */
3853 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3854 			   u32 exit_reason, u32 exit_intr_info,
3855 			   unsigned long exit_qualification)
3856 {
3857 	/* update exit information fields: */
3858 	vmcs12->vm_exit_reason = exit_reason;
3859 	vmcs12->exit_qualification = exit_qualification;
3860 	vmcs12->vm_exit_intr_info = exit_intr_info;
3861 
3862 	vmcs12->idt_vectoring_info_field = 0;
3863 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3864 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3865 
3866 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3867 		vmcs12->launch_state = 1;
3868 
3869 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3870 		 * instead of reading the real value. */
3871 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3872 
3873 		/*
3874 		 * Transfer the event that L0 or L1 may wanted to inject into
3875 		 * L2 to IDT_VECTORING_INFO_FIELD.
3876 		 */
3877 		vmcs12_save_pending_event(vcpu, vmcs12);
3878 
3879 		/*
3880 		 * According to spec, there's no need to store the guest's
3881 		 * MSRs if the exit is due to a VM-entry failure that occurs
3882 		 * during or after loading the guest state. Since this exit
3883 		 * does not fall in that category, we need to save the MSRs.
3884 		 */
3885 		if (nested_vmx_store_msr(vcpu,
3886 					 vmcs12->vm_exit_msr_store_addr,
3887 					 vmcs12->vm_exit_msr_store_count))
3888 			nested_vmx_abort(vcpu,
3889 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3890 	}
3891 
3892 	/*
3893 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3894 	 * preserved above and would only end up incorrectly in L1.
3895 	 */
3896 	vcpu->arch.nmi_injected = false;
3897 	kvm_clear_exception_queue(vcpu);
3898 	kvm_clear_interrupt_queue(vcpu);
3899 }
3900 
3901 /*
3902  * A part of what we need to when the nested L2 guest exits and we want to
3903  * run its L1 parent, is to reset L1's guest state to the host state specified
3904  * in vmcs12.
3905  * This function is to be called not only on normal nested exit, but also on
3906  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3907  * Failures During or After Loading Guest State").
3908  * This function should be called when the active VMCS is L1's (vmcs01).
3909  */
3910 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3911 				   struct vmcs12 *vmcs12)
3912 {
3913 	struct kvm_segment seg;
3914 	u32 entry_failure_code;
3915 
3916 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
3917 		vcpu->arch.efer = vmcs12->host_ia32_efer;
3918 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3919 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
3920 	else
3921 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
3922 	vmx_set_efer(vcpu, vcpu->arch.efer);
3923 
3924 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
3925 	kvm_rip_write(vcpu, vmcs12->host_rip);
3926 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
3927 	vmx_set_interrupt_shadow(vcpu, 0);
3928 
3929 	/*
3930 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
3931 	 * actually changed, because vmx_set_cr0 refers to efer set above.
3932 	 *
3933 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
3934 	 * (KVM doesn't change it);
3935 	 */
3936 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3937 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
3938 
3939 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
3940 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3941 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
3942 
3943 	nested_ept_uninit_mmu_context(vcpu);
3944 
3945 	/*
3946 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
3947 	 * couldn't have changed.
3948 	 */
3949 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
3950 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
3951 
3952 	if (!enable_ept)
3953 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
3954 
3955 	/*
3956 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
3957 	 * VMEntry/VMExit. Thus, no need to flush TLB.
3958 	 *
3959 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
3960 	 * flushed on every VMEntry/VMExit.
3961 	 *
3962 	 * Otherwise, we can preserve TLB entries as long as we are
3963 	 * able to tag L1 TLB entries differently than L2 TLB entries.
3964 	 *
3965 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
3966 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
3967 	 * has been set by KVM_REQ_LOAD_CR3.
3968 	 */
3969 	if (enable_vpid &&
3970 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
3971 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3972 	}
3973 
3974 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
3975 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
3976 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
3977 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
3978 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
3979 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
3980 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
3981 
3982 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
3983 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
3984 		vmcs_write64(GUEST_BNDCFGS, 0);
3985 
3986 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
3987 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
3988 		vcpu->arch.pat = vmcs12->host_ia32_pat;
3989 	}
3990 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
3991 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
3992 					 vmcs12->host_ia32_perf_global_ctrl));
3993 
3994 	/* Set L1 segment info according to Intel SDM
3995 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
3996 	seg = (struct kvm_segment) {
3997 		.base = 0,
3998 		.limit = 0xFFFFFFFF,
3999 		.selector = vmcs12->host_cs_selector,
4000 		.type = 11,
4001 		.present = 1,
4002 		.s = 1,
4003 		.g = 1
4004 	};
4005 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4006 		seg.l = 1;
4007 	else
4008 		seg.db = 1;
4009 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4010 	seg = (struct kvm_segment) {
4011 		.base = 0,
4012 		.limit = 0xFFFFFFFF,
4013 		.type = 3,
4014 		.present = 1,
4015 		.s = 1,
4016 		.db = 1,
4017 		.g = 1
4018 	};
4019 	seg.selector = vmcs12->host_ds_selector;
4020 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4021 	seg.selector = vmcs12->host_es_selector;
4022 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4023 	seg.selector = vmcs12->host_ss_selector;
4024 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4025 	seg.selector = vmcs12->host_fs_selector;
4026 	seg.base = vmcs12->host_fs_base;
4027 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4028 	seg.selector = vmcs12->host_gs_selector;
4029 	seg.base = vmcs12->host_gs_base;
4030 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4031 	seg = (struct kvm_segment) {
4032 		.base = vmcs12->host_tr_base,
4033 		.limit = 0x67,
4034 		.selector = vmcs12->host_tr_selector,
4035 		.type = 11,
4036 		.present = 1
4037 	};
4038 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4039 
4040 	kvm_set_dr(vcpu, 7, 0x400);
4041 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4042 
4043 	if (cpu_has_vmx_msr_bitmap())
4044 		vmx_update_msr_bitmap(vcpu);
4045 
4046 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4047 				vmcs12->vm_exit_msr_load_count))
4048 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4049 }
4050 
4051 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4052 {
4053 	struct shared_msr_entry *efer_msr;
4054 	unsigned int i;
4055 
4056 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4057 		return vmcs_read64(GUEST_IA32_EFER);
4058 
4059 	if (cpu_has_load_ia32_efer())
4060 		return host_efer;
4061 
4062 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4063 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4064 			return vmx->msr_autoload.guest.val[i].value;
4065 	}
4066 
4067 	efer_msr = find_msr_entry(vmx, MSR_EFER);
4068 	if (efer_msr)
4069 		return efer_msr->data;
4070 
4071 	return host_efer;
4072 }
4073 
4074 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4075 {
4076 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4077 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4078 	struct vmx_msr_entry g, h;
4079 	gpa_t gpa;
4080 	u32 i, j;
4081 
4082 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4083 
4084 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4085 		/*
4086 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4087 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4088 		 * and vcpu->arch.dr7 is not squirreled away before the
4089 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4090 		 */
4091 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4092 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4093 		else
4094 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4095 	}
4096 
4097 	/*
4098 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4099 	 * handle a variety of side effects to KVM's software model.
4100 	 */
4101 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4102 
4103 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
4104 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4105 
4106 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4107 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4108 
4109 	nested_ept_uninit_mmu_context(vcpu);
4110 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4111 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4112 
4113 	/*
4114 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4115 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4116 	 * VMFail, like everything else we just need to ensure our
4117 	 * software model is up-to-date.
4118 	 */
4119 	if (enable_ept)
4120 		ept_save_pdptrs(vcpu);
4121 
4122 	kvm_mmu_reset_context(vcpu);
4123 
4124 	if (cpu_has_vmx_msr_bitmap())
4125 		vmx_update_msr_bitmap(vcpu);
4126 
4127 	/*
4128 	 * This nasty bit of open coding is a compromise between blindly
4129 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4130 	 * of VMFail), leaving the nested VM's MSRs in the software model
4131 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4132 	 * expensive since the lists are unbound by hardware).  For each
4133 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4134 	 * list, reload it from the exit load list if it exists and differs
4135 	 * from the guest value.  The intent is to stuff host state as
4136 	 * silently as possible, not to fully process the exit load list.
4137 	 */
4138 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4139 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4140 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4141 			pr_debug_ratelimited(
4142 				"%s read MSR index failed (%u, 0x%08llx)\n",
4143 				__func__, i, gpa);
4144 			goto vmabort;
4145 		}
4146 
4147 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4148 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4149 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4150 				pr_debug_ratelimited(
4151 					"%s read MSR failed (%u, 0x%08llx)\n",
4152 					__func__, j, gpa);
4153 				goto vmabort;
4154 			}
4155 			if (h.index != g.index)
4156 				continue;
4157 			if (h.value == g.value)
4158 				break;
4159 
4160 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4161 				pr_debug_ratelimited(
4162 					"%s check failed (%u, 0x%x, 0x%x)\n",
4163 					__func__, j, h.index, h.reserved);
4164 				goto vmabort;
4165 			}
4166 
4167 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4168 				pr_debug_ratelimited(
4169 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4170 					__func__, j, h.index, h.value);
4171 				goto vmabort;
4172 			}
4173 		}
4174 	}
4175 
4176 	return;
4177 
4178 vmabort:
4179 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4180 }
4181 
4182 /*
4183  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4184  * and modify vmcs12 to make it see what it would expect to see there if
4185  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4186  */
4187 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
4188 		       u32 exit_intr_info, unsigned long exit_qualification)
4189 {
4190 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4191 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4192 
4193 	/* trying to cancel vmlaunch/vmresume is a bug */
4194 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4195 
4196 	leave_guest_mode(vcpu);
4197 
4198 	if (nested_cpu_has_preemption_timer(vmcs12))
4199 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4200 
4201 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
4202 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
4203 
4204 	if (likely(!vmx->fail)) {
4205 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4206 
4207 		if (exit_reason != -1)
4208 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
4209 				       exit_qualification);
4210 
4211 		/*
4212 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4213 		 * also be used to capture vmcs12 cache as part of
4214 		 * capturing nVMX state for snapshot (migration).
4215 		 *
4216 		 * Otherwise, this flush will dirty guest memory at a
4217 		 * point it is already assumed by user-space to be
4218 		 * immutable.
4219 		 */
4220 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4221 	} else {
4222 		/*
4223 		 * The only expected VM-instruction error is "VM entry with
4224 		 * invalid control field(s)." Anything else indicates a
4225 		 * problem with L0.  And we should never get here with a
4226 		 * VMFail of any type if early consistency checks are enabled.
4227 		 */
4228 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4229 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4230 		WARN_ON_ONCE(nested_early_check);
4231 	}
4232 
4233 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4234 
4235 	/* Update any VMCS fields that might have changed while L2 ran */
4236 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4237 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4238 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4239 	if (vmx->nested.l1_tpr_threshold != -1)
4240 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4241 
4242 	if (kvm_has_tsc_control)
4243 		decache_tsc_multiplier(vmx);
4244 
4245 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4246 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4247 		vmx_set_virtual_apic_mode(vcpu);
4248 	}
4249 
4250 	/* Unpin physical memory we referred to in vmcs02 */
4251 	if (vmx->nested.apic_access_page) {
4252 		kvm_release_page_clean(vmx->nested.apic_access_page);
4253 		vmx->nested.apic_access_page = NULL;
4254 	}
4255 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4256 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4257 	vmx->nested.pi_desc = NULL;
4258 
4259 	/*
4260 	 * We are now running in L2, mmu_notifier will force to reload the
4261 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
4262 	 */
4263 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4264 
4265 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
4266 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4267 
4268 	/* in case we halted in L2 */
4269 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4270 
4271 	if (likely(!vmx->fail)) {
4272 		/*
4273 		 * TODO: SDM says that with acknowledge interrupt on
4274 		 * exit, bit 31 of the VM-exit interrupt information
4275 		 * (valid interrupt) is always set to 1 on
4276 		 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
4277 		 * need kvm_cpu_has_interrupt().  See the commit
4278 		 * message for details.
4279 		 */
4280 		if (nested_exit_intr_ack_set(vcpu) &&
4281 		    exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4282 		    kvm_cpu_has_interrupt(vcpu)) {
4283 			int irq = kvm_cpu_get_interrupt(vcpu);
4284 			WARN_ON(irq < 0);
4285 			vmcs12->vm_exit_intr_info = irq |
4286 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4287 		}
4288 
4289 		if (exit_reason != -1)
4290 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4291 						       vmcs12->exit_qualification,
4292 						       vmcs12->idt_vectoring_info_field,
4293 						       vmcs12->vm_exit_intr_info,
4294 						       vmcs12->vm_exit_intr_error_code,
4295 						       KVM_ISA_VMX);
4296 
4297 		load_vmcs12_host_state(vcpu, vmcs12);
4298 
4299 		return;
4300 	}
4301 
4302 	/*
4303 	 * After an early L2 VM-entry failure, we're now back
4304 	 * in L1 which thinks it just finished a VMLAUNCH or
4305 	 * VMRESUME instruction, so we need to set the failure
4306 	 * flag and the VM-instruction error field of the VMCS
4307 	 * accordingly, and skip the emulated instruction.
4308 	 */
4309 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4310 
4311 	/*
4312 	 * Restore L1's host state to KVM's software model.  We're here
4313 	 * because a consistency check was caught by hardware, which
4314 	 * means some amount of guest state has been propagated to KVM's
4315 	 * model and needs to be unwound to the host's state.
4316 	 */
4317 	nested_vmx_restore_host_state(vcpu);
4318 
4319 	vmx->fail = 0;
4320 }
4321 
4322 /*
4323  * Decode the memory-address operand of a vmx instruction, as recorded on an
4324  * exit caused by such an instruction (run by a guest hypervisor).
4325  * On success, returns 0. When the operand is invalid, returns 1 and throws
4326  * #UD or #GP.
4327  */
4328 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4329 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4330 {
4331 	gva_t off;
4332 	bool exn;
4333 	struct kvm_segment s;
4334 
4335 	/*
4336 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4337 	 * Execution", on an exit, vmx_instruction_info holds most of the
4338 	 * addressing components of the operand. Only the displacement part
4339 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4340 	 * For how an actual address is calculated from all these components,
4341 	 * refer to Vol. 1, "Operand Addressing".
4342 	 */
4343 	int  scaling = vmx_instruction_info & 3;
4344 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4345 	bool is_reg = vmx_instruction_info & (1u << 10);
4346 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4347 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4348 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4349 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4350 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4351 
4352 	if (is_reg) {
4353 		kvm_queue_exception(vcpu, UD_VECTOR);
4354 		return 1;
4355 	}
4356 
4357 	/* Addr = segment_base + offset */
4358 	/* offset = base + [index * scale] + displacement */
4359 	off = exit_qualification; /* holds the displacement */
4360 	if (addr_size == 1)
4361 		off = (gva_t)sign_extend64(off, 31);
4362 	else if (addr_size == 0)
4363 		off = (gva_t)sign_extend64(off, 15);
4364 	if (base_is_valid)
4365 		off += kvm_register_read(vcpu, base_reg);
4366 	if (index_is_valid)
4367 		off += kvm_register_read(vcpu, index_reg)<<scaling;
4368 	vmx_get_segment(vcpu, &s, seg_reg);
4369 
4370 	/*
4371 	 * The effective address, i.e. @off, of a memory operand is truncated
4372 	 * based on the address size of the instruction.  Note that this is
4373 	 * the *effective address*, i.e. the address prior to accounting for
4374 	 * the segment's base.
4375 	 */
4376 	if (addr_size == 1) /* 32 bit */
4377 		off &= 0xffffffff;
4378 	else if (addr_size == 0) /* 16 bit */
4379 		off &= 0xffff;
4380 
4381 	/* Checks for #GP/#SS exceptions. */
4382 	exn = false;
4383 	if (is_long_mode(vcpu)) {
4384 		/*
4385 		 * The virtual/linear address is never truncated in 64-bit
4386 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4387 		 * address when using FS/GS with a non-zero base.
4388 		 */
4389 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4390 			*ret = s.base + off;
4391 		else
4392 			*ret = off;
4393 
4394 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4395 		 * non-canonical form. This is the only check on the memory
4396 		 * destination for long mode!
4397 		 */
4398 		exn = is_noncanonical_address(*ret, vcpu);
4399 	} else {
4400 		/*
4401 		 * When not in long mode, the virtual/linear address is
4402 		 * unconditionally truncated to 32 bits regardless of the
4403 		 * address size.
4404 		 */
4405 		*ret = (s.base + off) & 0xffffffff;
4406 
4407 		/* Protected mode: apply checks for segment validity in the
4408 		 * following order:
4409 		 * - segment type check (#GP(0) may be thrown)
4410 		 * - usability check (#GP(0)/#SS(0))
4411 		 * - limit check (#GP(0)/#SS(0))
4412 		 */
4413 		if (wr)
4414 			/* #GP(0) if the destination operand is located in a
4415 			 * read-only data segment or any code segment.
4416 			 */
4417 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4418 		else
4419 			/* #GP(0) if the source operand is located in an
4420 			 * execute-only code segment
4421 			 */
4422 			exn = ((s.type & 0xa) == 8);
4423 		if (exn) {
4424 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4425 			return 1;
4426 		}
4427 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4428 		 */
4429 		exn = (s.unusable != 0);
4430 
4431 		/*
4432 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4433 		 * outside the segment limit.  All CPUs that support VMX ignore
4434 		 * limit checks for flat segments, i.e. segments with base==0,
4435 		 * limit==0xffffffff and of type expand-up data or code.
4436 		 */
4437 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4438 		     ((s.type & 8) || !(s.type & 4))))
4439 			exn = exn || ((u64)off + len - 1 > s.limit);
4440 	}
4441 	if (exn) {
4442 		kvm_queue_exception_e(vcpu,
4443 				      seg_reg == VCPU_SREG_SS ?
4444 						SS_VECTOR : GP_VECTOR,
4445 				      0);
4446 		return 1;
4447 	}
4448 
4449 	return 0;
4450 }
4451 
4452 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4453 {
4454 	struct vcpu_vmx *vmx;
4455 
4456 	if (!nested_vmx_allowed(vcpu))
4457 		return;
4458 
4459 	vmx = to_vmx(vcpu);
4460 	if (kvm_x86_ops->pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4461 		vmx->nested.msrs.entry_ctls_high |=
4462 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4463 		vmx->nested.msrs.exit_ctls_high |=
4464 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4465 	} else {
4466 		vmx->nested.msrs.entry_ctls_high &=
4467 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4468 		vmx->nested.msrs.exit_ctls_high &=
4469 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4470 	}
4471 }
4472 
4473 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4474 {
4475 	gva_t gva;
4476 	struct x86_exception e;
4477 
4478 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4479 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4480 				sizeof(*vmpointer), &gva))
4481 		return 1;
4482 
4483 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4484 		kvm_inject_page_fault(vcpu, &e);
4485 		return 1;
4486 	}
4487 
4488 	return 0;
4489 }
4490 
4491 /*
4492  * Allocate a shadow VMCS and associate it with the currently loaded
4493  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4494  * VMCS is also VMCLEARed, so that it is ready for use.
4495  */
4496 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4497 {
4498 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4499 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4500 
4501 	/*
4502 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4503 	 * executes VMXON and free it when L1 executes VMXOFF.
4504 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4505 	 * here when vmcs01 already have an allocated shadow vmcs.
4506 	 */
4507 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4508 
4509 	if (!loaded_vmcs->shadow_vmcs) {
4510 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4511 		if (loaded_vmcs->shadow_vmcs)
4512 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4513 	}
4514 	return loaded_vmcs->shadow_vmcs;
4515 }
4516 
4517 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4518 {
4519 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4520 	int r;
4521 
4522 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4523 	if (r < 0)
4524 		goto out_vmcs02;
4525 
4526 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4527 	if (!vmx->nested.cached_vmcs12)
4528 		goto out_cached_vmcs12;
4529 
4530 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4531 	if (!vmx->nested.cached_shadow_vmcs12)
4532 		goto out_cached_shadow_vmcs12;
4533 
4534 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4535 		goto out_shadow_vmcs;
4536 
4537 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4538 		     HRTIMER_MODE_REL_PINNED);
4539 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4540 
4541 	vmx->nested.vpid02 = allocate_vpid();
4542 
4543 	vmx->nested.vmcs02_initialized = false;
4544 	vmx->nested.vmxon = true;
4545 
4546 	if (pt_mode == PT_MODE_HOST_GUEST) {
4547 		vmx->pt_desc.guest.ctl = 0;
4548 		pt_update_intercept_for_msr(vmx);
4549 	}
4550 
4551 	return 0;
4552 
4553 out_shadow_vmcs:
4554 	kfree(vmx->nested.cached_shadow_vmcs12);
4555 
4556 out_cached_shadow_vmcs12:
4557 	kfree(vmx->nested.cached_vmcs12);
4558 
4559 out_cached_vmcs12:
4560 	free_loaded_vmcs(&vmx->nested.vmcs02);
4561 
4562 out_vmcs02:
4563 	return -ENOMEM;
4564 }
4565 
4566 /*
4567  * Emulate the VMXON instruction.
4568  * Currently, we just remember that VMX is active, and do not save or even
4569  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4570  * do not currently need to store anything in that guest-allocated memory
4571  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4572  * argument is different from the VMXON pointer (which the spec says they do).
4573  */
4574 static int handle_vmon(struct kvm_vcpu *vcpu)
4575 {
4576 	int ret;
4577 	gpa_t vmptr;
4578 	uint32_t revision;
4579 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4580 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4581 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4582 
4583 	/*
4584 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4585 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4586 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4587 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4588 	 * have already been checked by hardware, prior to the VM-exit for
4589 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4590 	 * that bit set to 1 in non-root mode.
4591 	 */
4592 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4593 		kvm_queue_exception(vcpu, UD_VECTOR);
4594 		return 1;
4595 	}
4596 
4597 	/* CPL=0 must be checked manually. */
4598 	if (vmx_get_cpl(vcpu)) {
4599 		kvm_inject_gp(vcpu, 0);
4600 		return 1;
4601 	}
4602 
4603 	if (vmx->nested.vmxon)
4604 		return nested_vmx_failValid(vcpu,
4605 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4606 
4607 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4608 			!= VMXON_NEEDED_FEATURES) {
4609 		kvm_inject_gp(vcpu, 0);
4610 		return 1;
4611 	}
4612 
4613 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4614 		return 1;
4615 
4616 	/*
4617 	 * SDM 3: 24.11.5
4618 	 * The first 4 bytes of VMXON region contain the supported
4619 	 * VMCS revision identifier
4620 	 *
4621 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4622 	 * which replaces physical address width with 32
4623 	 */
4624 	if (!page_address_valid(vcpu, vmptr))
4625 		return nested_vmx_failInvalid(vcpu);
4626 
4627 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4628 	    revision != VMCS12_REVISION)
4629 		return nested_vmx_failInvalid(vcpu);
4630 
4631 	vmx->nested.vmxon_ptr = vmptr;
4632 	ret = enter_vmx_operation(vcpu);
4633 	if (ret)
4634 		return ret;
4635 
4636 	return nested_vmx_succeed(vcpu);
4637 }
4638 
4639 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4640 {
4641 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4642 
4643 	if (vmx->nested.current_vmptr == -1ull)
4644 		return;
4645 
4646 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4647 
4648 	if (enable_shadow_vmcs) {
4649 		/* copy to memory all shadowed fields in case
4650 		   they were modified */
4651 		copy_shadow_to_vmcs12(vmx);
4652 		vmx_disable_shadow_vmcs(vmx);
4653 	}
4654 	vmx->nested.posted_intr_nv = -1;
4655 
4656 	/* Flush VMCS12 to guest memory */
4657 	kvm_vcpu_write_guest_page(vcpu,
4658 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4659 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4660 
4661 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4662 
4663 	vmx->nested.current_vmptr = -1ull;
4664 }
4665 
4666 /* Emulate the VMXOFF instruction */
4667 static int handle_vmoff(struct kvm_vcpu *vcpu)
4668 {
4669 	if (!nested_vmx_check_permission(vcpu))
4670 		return 1;
4671 
4672 	free_nested(vcpu);
4673 
4674 	/* Process a latched INIT during time CPU was in VMX operation */
4675 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4676 
4677 	return nested_vmx_succeed(vcpu);
4678 }
4679 
4680 /* Emulate the VMCLEAR instruction */
4681 static int handle_vmclear(struct kvm_vcpu *vcpu)
4682 {
4683 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4684 	u32 zero = 0;
4685 	gpa_t vmptr;
4686 	u64 evmcs_gpa;
4687 
4688 	if (!nested_vmx_check_permission(vcpu))
4689 		return 1;
4690 
4691 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4692 		return 1;
4693 
4694 	if (!page_address_valid(vcpu, vmptr))
4695 		return nested_vmx_failValid(vcpu,
4696 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4697 
4698 	if (vmptr == vmx->nested.vmxon_ptr)
4699 		return nested_vmx_failValid(vcpu,
4700 			VMXERR_VMCLEAR_VMXON_POINTER);
4701 
4702 	/*
4703 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4704 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4705 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4706 	 * writing to the non-existent 'launch_state' field. The area doesn't
4707 	 * have to be the currently active EVMCS on the calling CPU and there's
4708 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
4709 	 * state. It is possible that the area will stay mapped as
4710 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
4711 	 */
4712 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
4713 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
4714 		if (vmptr == vmx->nested.current_vmptr)
4715 			nested_release_vmcs12(vcpu);
4716 
4717 		kvm_vcpu_write_guest(vcpu,
4718 				     vmptr + offsetof(struct vmcs12,
4719 						      launch_state),
4720 				     &zero, sizeof(zero));
4721 	}
4722 
4723 	return nested_vmx_succeed(vcpu);
4724 }
4725 
4726 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4727 
4728 /* Emulate the VMLAUNCH instruction */
4729 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4730 {
4731 	return nested_vmx_run(vcpu, true);
4732 }
4733 
4734 /* Emulate the VMRESUME instruction */
4735 static int handle_vmresume(struct kvm_vcpu *vcpu)
4736 {
4737 
4738 	return nested_vmx_run(vcpu, false);
4739 }
4740 
4741 static int handle_vmread(struct kvm_vcpu *vcpu)
4742 {
4743 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4744 						    : get_vmcs12(vcpu);
4745 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4746 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4747 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4748 	struct x86_exception e;
4749 	unsigned long field;
4750 	u64 value;
4751 	gva_t gva = 0;
4752 	short offset;
4753 	int len;
4754 
4755 	if (!nested_vmx_check_permission(vcpu))
4756 		return 1;
4757 
4758 	/*
4759 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
4760 	 * any VMREAD sets the ALU flags for VMfailInvalid.
4761 	 */
4762 	if (vmx->nested.current_vmptr == -1ull ||
4763 	    (is_guest_mode(vcpu) &&
4764 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
4765 		return nested_vmx_failInvalid(vcpu);
4766 
4767 	/* Decode instruction info and find the field to read */
4768 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
4769 
4770 	offset = vmcs_field_to_offset(field);
4771 	if (offset < 0)
4772 		return nested_vmx_failValid(vcpu,
4773 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4774 
4775 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
4776 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4777 
4778 	/* Read the field, zero-extended to a u64 value */
4779 	value = vmcs12_read_any(vmcs12, field, offset);
4780 
4781 	/*
4782 	 * Now copy part of this value to register or memory, as requested.
4783 	 * Note that the number of bits actually copied is 32 or 64 depending
4784 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4785 	 */
4786 	if (instr_info & BIT(10)) {
4787 		kvm_register_writel(vcpu, (((instr_info) >> 3) & 0xf), value);
4788 	} else {
4789 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4790 		if (get_vmx_mem_address(vcpu, exit_qualification,
4791 					instr_info, true, len, &gva))
4792 			return 1;
4793 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4794 		if (kvm_write_guest_virt_system(vcpu, gva, &value, len, &e)) {
4795 			kvm_inject_page_fault(vcpu, &e);
4796 			return 1;
4797 		}
4798 	}
4799 
4800 	return nested_vmx_succeed(vcpu);
4801 }
4802 
4803 static bool is_shadow_field_rw(unsigned long field)
4804 {
4805 	switch (field) {
4806 #define SHADOW_FIELD_RW(x, y) case x:
4807 #include "vmcs_shadow_fields.h"
4808 		return true;
4809 	default:
4810 		break;
4811 	}
4812 	return false;
4813 }
4814 
4815 static bool is_shadow_field_ro(unsigned long field)
4816 {
4817 	switch (field) {
4818 #define SHADOW_FIELD_RO(x, y) case x:
4819 #include "vmcs_shadow_fields.h"
4820 		return true;
4821 	default:
4822 		break;
4823 	}
4824 	return false;
4825 }
4826 
4827 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4828 {
4829 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4830 						    : get_vmcs12(vcpu);
4831 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4832 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4833 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4834 	struct x86_exception e;
4835 	unsigned long field;
4836 	short offset;
4837 	gva_t gva;
4838 	int len;
4839 
4840 	/*
4841 	 * The value to write might be 32 or 64 bits, depending on L1's long
4842 	 * mode, and eventually we need to write that into a field of several
4843 	 * possible lengths. The code below first zero-extends the value to 64
4844 	 * bit (value), and then copies only the appropriate number of
4845 	 * bits into the vmcs12 field.
4846 	 */
4847 	u64 value = 0;
4848 
4849 	if (!nested_vmx_check_permission(vcpu))
4850 		return 1;
4851 
4852 	/*
4853 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
4854 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
4855 	 */
4856 	if (vmx->nested.current_vmptr == -1ull ||
4857 	    (is_guest_mode(vcpu) &&
4858 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
4859 		return nested_vmx_failInvalid(vcpu);
4860 
4861 	if (instr_info & BIT(10))
4862 		value = kvm_register_readl(vcpu, (((instr_info) >> 3) & 0xf));
4863 	else {
4864 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4865 		if (get_vmx_mem_address(vcpu, exit_qualification,
4866 					instr_info, false, len, &gva))
4867 			return 1;
4868 		if (kvm_read_guest_virt(vcpu, gva, &value, len, &e)) {
4869 			kvm_inject_page_fault(vcpu, &e);
4870 			return 1;
4871 		}
4872 	}
4873 
4874 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
4875 
4876 	offset = vmcs_field_to_offset(field);
4877 	if (offset < 0)
4878 		return nested_vmx_failValid(vcpu,
4879 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4880 
4881 	/*
4882 	 * If the vCPU supports "VMWRITE to any supported field in the
4883 	 * VMCS," then the "read-only" fields are actually read/write.
4884 	 */
4885 	if (vmcs_field_readonly(field) &&
4886 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4887 		return nested_vmx_failValid(vcpu,
4888 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4889 
4890 	/*
4891 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
4892 	 * vmcs12, else we may crush a field or consume a stale value.
4893 	 */
4894 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
4895 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4896 
4897 	/*
4898 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
4899 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
4900 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
4901 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
4902 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
4903 	 * the stripped down value, L2 sees the full value as stored by KVM).
4904 	 */
4905 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
4906 		value &= 0x1f0ff;
4907 
4908 	vmcs12_write_any(vmcs12, field, offset, value);
4909 
4910 	/*
4911 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
4912 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
4913 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
4914 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
4915 	 */
4916 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
4917 		/*
4918 		 * L1 can read these fields without exiting, ensure the
4919 		 * shadow VMCS is up-to-date.
4920 		 */
4921 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
4922 			preempt_disable();
4923 			vmcs_load(vmx->vmcs01.shadow_vmcs);
4924 
4925 			__vmcs_writel(field, value);
4926 
4927 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
4928 			vmcs_load(vmx->loaded_vmcs->vmcs);
4929 			preempt_enable();
4930 		}
4931 		vmx->nested.dirty_vmcs12 = true;
4932 	}
4933 
4934 	return nested_vmx_succeed(vcpu);
4935 }
4936 
4937 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
4938 {
4939 	vmx->nested.current_vmptr = vmptr;
4940 	if (enable_shadow_vmcs) {
4941 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
4942 		vmcs_write64(VMCS_LINK_POINTER,
4943 			     __pa(vmx->vmcs01.shadow_vmcs));
4944 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4945 	}
4946 	vmx->nested.dirty_vmcs12 = true;
4947 }
4948 
4949 /* Emulate the VMPTRLD instruction */
4950 static int handle_vmptrld(struct kvm_vcpu *vcpu)
4951 {
4952 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4953 	gpa_t vmptr;
4954 
4955 	if (!nested_vmx_check_permission(vcpu))
4956 		return 1;
4957 
4958 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4959 		return 1;
4960 
4961 	if (!page_address_valid(vcpu, vmptr))
4962 		return nested_vmx_failValid(vcpu,
4963 			VMXERR_VMPTRLD_INVALID_ADDRESS);
4964 
4965 	if (vmptr == vmx->nested.vmxon_ptr)
4966 		return nested_vmx_failValid(vcpu,
4967 			VMXERR_VMPTRLD_VMXON_POINTER);
4968 
4969 	/* Forbid normal VMPTRLD if Enlightened version was used */
4970 	if (vmx->nested.hv_evmcs)
4971 		return 1;
4972 
4973 	if (vmx->nested.current_vmptr != vmptr) {
4974 		struct kvm_host_map map;
4975 		struct vmcs12 *new_vmcs12;
4976 
4977 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
4978 			/*
4979 			 * Reads from an unbacked page return all 1s,
4980 			 * which means that the 32 bits located at the
4981 			 * given physical address won't match the required
4982 			 * VMCS12_REVISION identifier.
4983 			 */
4984 			return nested_vmx_failValid(vcpu,
4985 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4986 		}
4987 
4988 		new_vmcs12 = map.hva;
4989 
4990 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
4991 		    (new_vmcs12->hdr.shadow_vmcs &&
4992 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
4993 			kvm_vcpu_unmap(vcpu, &map, false);
4994 			return nested_vmx_failValid(vcpu,
4995 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4996 		}
4997 
4998 		nested_release_vmcs12(vcpu);
4999 
5000 		/*
5001 		 * Load VMCS12 from guest memory since it is not already
5002 		 * cached.
5003 		 */
5004 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
5005 		kvm_vcpu_unmap(vcpu, &map, false);
5006 
5007 		set_current_vmptr(vmx, vmptr);
5008 	}
5009 
5010 	return nested_vmx_succeed(vcpu);
5011 }
5012 
5013 /* Emulate the VMPTRST instruction */
5014 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5015 {
5016 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
5017 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5018 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5019 	struct x86_exception e;
5020 	gva_t gva;
5021 
5022 	if (!nested_vmx_check_permission(vcpu))
5023 		return 1;
5024 
5025 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
5026 		return 1;
5027 
5028 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5029 				true, sizeof(gpa_t), &gva))
5030 		return 1;
5031 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5032 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5033 					sizeof(gpa_t), &e)) {
5034 		kvm_inject_page_fault(vcpu, &e);
5035 		return 1;
5036 	}
5037 	return nested_vmx_succeed(vcpu);
5038 }
5039 
5040 /* Emulate the INVEPT instruction */
5041 static int handle_invept(struct kvm_vcpu *vcpu)
5042 {
5043 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5044 	u32 vmx_instruction_info, types;
5045 	unsigned long type;
5046 	gva_t gva;
5047 	struct x86_exception e;
5048 	struct {
5049 		u64 eptp, gpa;
5050 	} operand;
5051 
5052 	if (!(vmx->nested.msrs.secondary_ctls_high &
5053 	      SECONDARY_EXEC_ENABLE_EPT) ||
5054 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5055 		kvm_queue_exception(vcpu, UD_VECTOR);
5056 		return 1;
5057 	}
5058 
5059 	if (!nested_vmx_check_permission(vcpu))
5060 		return 1;
5061 
5062 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5063 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5064 
5065 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5066 
5067 	if (type >= 32 || !(types & (1 << type)))
5068 		return nested_vmx_failValid(vcpu,
5069 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5070 
5071 	/* According to the Intel VMX instruction reference, the memory
5072 	 * operand is read even if it isn't needed (e.g., for type==global)
5073 	 */
5074 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5075 			vmx_instruction_info, false, sizeof(operand), &gva))
5076 		return 1;
5077 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5078 		kvm_inject_page_fault(vcpu, &e);
5079 		return 1;
5080 	}
5081 
5082 	switch (type) {
5083 	case VMX_EPT_EXTENT_GLOBAL:
5084 	case VMX_EPT_EXTENT_CONTEXT:
5085 	/*
5086 	 * TODO: Sync the necessary shadow EPT roots here, rather than
5087 	 * at the next emulated VM-entry.
5088 	 */
5089 		break;
5090 	default:
5091 		BUG_ON(1);
5092 		break;
5093 	}
5094 
5095 	return nested_vmx_succeed(vcpu);
5096 }
5097 
5098 static int handle_invvpid(struct kvm_vcpu *vcpu)
5099 {
5100 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5101 	u32 vmx_instruction_info;
5102 	unsigned long type, types;
5103 	gva_t gva;
5104 	struct x86_exception e;
5105 	struct {
5106 		u64 vpid;
5107 		u64 gla;
5108 	} operand;
5109 	u16 vpid02;
5110 
5111 	if (!(vmx->nested.msrs.secondary_ctls_high &
5112 	      SECONDARY_EXEC_ENABLE_VPID) ||
5113 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5114 		kvm_queue_exception(vcpu, UD_VECTOR);
5115 		return 1;
5116 	}
5117 
5118 	if (!nested_vmx_check_permission(vcpu))
5119 		return 1;
5120 
5121 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5122 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5123 
5124 	types = (vmx->nested.msrs.vpid_caps &
5125 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5126 
5127 	if (type >= 32 || !(types & (1 << type)))
5128 		return nested_vmx_failValid(vcpu,
5129 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5130 
5131 	/* according to the intel vmx instruction reference, the memory
5132 	 * operand is read even if it isn't needed (e.g., for type==global)
5133 	 */
5134 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5135 			vmx_instruction_info, false, sizeof(operand), &gva))
5136 		return 1;
5137 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5138 		kvm_inject_page_fault(vcpu, &e);
5139 		return 1;
5140 	}
5141 	if (operand.vpid >> 16)
5142 		return nested_vmx_failValid(vcpu,
5143 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5144 
5145 	vpid02 = nested_get_vpid02(vcpu);
5146 	switch (type) {
5147 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5148 		if (!operand.vpid ||
5149 		    is_noncanonical_address(operand.gla, vcpu))
5150 			return nested_vmx_failValid(vcpu,
5151 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5152 		if (cpu_has_vmx_invvpid_individual_addr()) {
5153 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
5154 				vpid02, operand.gla);
5155 		} else
5156 			__vmx_flush_tlb(vcpu, vpid02, false);
5157 		break;
5158 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5159 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5160 		if (!operand.vpid)
5161 			return nested_vmx_failValid(vcpu,
5162 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5163 		__vmx_flush_tlb(vcpu, vpid02, false);
5164 		break;
5165 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5166 		__vmx_flush_tlb(vcpu, vpid02, false);
5167 		break;
5168 	default:
5169 		WARN_ON_ONCE(1);
5170 		return kvm_skip_emulated_instruction(vcpu);
5171 	}
5172 
5173 	return nested_vmx_succeed(vcpu);
5174 }
5175 
5176 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5177 				     struct vmcs12 *vmcs12)
5178 {
5179 	u32 index = kvm_rcx_read(vcpu);
5180 	u64 address;
5181 	bool accessed_dirty;
5182 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
5183 
5184 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
5185 	    !nested_cpu_has_ept(vmcs12))
5186 		return 1;
5187 
5188 	if (index >= VMFUNC_EPTP_ENTRIES)
5189 		return 1;
5190 
5191 
5192 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5193 				     &address, index * 8, 8))
5194 		return 1;
5195 
5196 	accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
5197 
5198 	/*
5199 	 * If the (L2) guest does a vmfunc to the currently
5200 	 * active ept pointer, we don't have to do anything else
5201 	 */
5202 	if (vmcs12->ept_pointer != address) {
5203 		if (!valid_ept_address(vcpu, address))
5204 			return 1;
5205 
5206 		kvm_mmu_unload(vcpu);
5207 		mmu->ept_ad = accessed_dirty;
5208 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
5209 		vmcs12->ept_pointer = address;
5210 		/*
5211 		 * TODO: Check what's the correct approach in case
5212 		 * mmu reload fails. Currently, we just let the next
5213 		 * reload potentially fail
5214 		 */
5215 		kvm_mmu_reload(vcpu);
5216 	}
5217 
5218 	return 0;
5219 }
5220 
5221 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5222 {
5223 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5224 	struct vmcs12 *vmcs12;
5225 	u32 function = kvm_rax_read(vcpu);
5226 
5227 	/*
5228 	 * VMFUNC is only supported for nested guests, but we always enable the
5229 	 * secondary control for simplicity; for non-nested mode, fake that we
5230 	 * didn't by injecting #UD.
5231 	 */
5232 	if (!is_guest_mode(vcpu)) {
5233 		kvm_queue_exception(vcpu, UD_VECTOR);
5234 		return 1;
5235 	}
5236 
5237 	vmcs12 = get_vmcs12(vcpu);
5238 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
5239 		goto fail;
5240 
5241 	switch (function) {
5242 	case 0:
5243 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5244 			goto fail;
5245 		break;
5246 	default:
5247 		goto fail;
5248 	}
5249 	return kvm_skip_emulated_instruction(vcpu);
5250 
5251 fail:
5252 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
5253 			  vmcs_read32(VM_EXIT_INTR_INFO),
5254 			  vmcs_readl(EXIT_QUALIFICATION));
5255 	return 1;
5256 }
5257 
5258 
5259 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5260 				       struct vmcs12 *vmcs12)
5261 {
5262 	unsigned long exit_qualification;
5263 	gpa_t bitmap, last_bitmap;
5264 	unsigned int port;
5265 	int size;
5266 	u8 b;
5267 
5268 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5269 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5270 
5271 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5272 
5273 	port = exit_qualification >> 16;
5274 	size = (exit_qualification & 7) + 1;
5275 
5276 	last_bitmap = (gpa_t)-1;
5277 	b = -1;
5278 
5279 	while (size > 0) {
5280 		if (port < 0x8000)
5281 			bitmap = vmcs12->io_bitmap_a;
5282 		else if (port < 0x10000)
5283 			bitmap = vmcs12->io_bitmap_b;
5284 		else
5285 			return true;
5286 		bitmap += (port & 0x7fff) / 8;
5287 
5288 		if (last_bitmap != bitmap)
5289 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5290 				return true;
5291 		if (b & (1 << (port & 7)))
5292 			return true;
5293 
5294 		port++;
5295 		size--;
5296 		last_bitmap = bitmap;
5297 	}
5298 
5299 	return false;
5300 }
5301 
5302 /*
5303  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5304  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5305  * disinterest in the current event (read or write a specific MSR) by using an
5306  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5307  */
5308 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5309 	struct vmcs12 *vmcs12, u32 exit_reason)
5310 {
5311 	u32 msr_index = kvm_rcx_read(vcpu);
5312 	gpa_t bitmap;
5313 
5314 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5315 		return true;
5316 
5317 	/*
5318 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5319 	 * for the four combinations of read/write and low/high MSR numbers.
5320 	 * First we need to figure out which of the four to use:
5321 	 */
5322 	bitmap = vmcs12->msr_bitmap;
5323 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5324 		bitmap += 2048;
5325 	if (msr_index >= 0xc0000000) {
5326 		msr_index -= 0xc0000000;
5327 		bitmap += 1024;
5328 	}
5329 
5330 	/* Then read the msr_index'th bit from this bitmap: */
5331 	if (msr_index < 1024*8) {
5332 		unsigned char b;
5333 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5334 			return true;
5335 		return 1 & (b >> (msr_index & 7));
5336 	} else
5337 		return true; /* let L1 handle the wrong parameter */
5338 }
5339 
5340 /*
5341  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5342  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5343  * intercept (via guest_host_mask etc.) the current event.
5344  */
5345 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5346 	struct vmcs12 *vmcs12)
5347 {
5348 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5349 	int cr = exit_qualification & 15;
5350 	int reg;
5351 	unsigned long val;
5352 
5353 	switch ((exit_qualification >> 4) & 3) {
5354 	case 0: /* mov to cr */
5355 		reg = (exit_qualification >> 8) & 15;
5356 		val = kvm_register_readl(vcpu, reg);
5357 		switch (cr) {
5358 		case 0:
5359 			if (vmcs12->cr0_guest_host_mask &
5360 			    (val ^ vmcs12->cr0_read_shadow))
5361 				return true;
5362 			break;
5363 		case 3:
5364 			if ((vmcs12->cr3_target_count >= 1 &&
5365 					vmcs12->cr3_target_value0 == val) ||
5366 				(vmcs12->cr3_target_count >= 2 &&
5367 					vmcs12->cr3_target_value1 == val) ||
5368 				(vmcs12->cr3_target_count >= 3 &&
5369 					vmcs12->cr3_target_value2 == val) ||
5370 				(vmcs12->cr3_target_count >= 4 &&
5371 					vmcs12->cr3_target_value3 == val))
5372 				return false;
5373 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5374 				return true;
5375 			break;
5376 		case 4:
5377 			if (vmcs12->cr4_guest_host_mask &
5378 			    (vmcs12->cr4_read_shadow ^ val))
5379 				return true;
5380 			break;
5381 		case 8:
5382 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5383 				return true;
5384 			break;
5385 		}
5386 		break;
5387 	case 2: /* clts */
5388 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5389 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5390 			return true;
5391 		break;
5392 	case 1: /* mov from cr */
5393 		switch (cr) {
5394 		case 3:
5395 			if (vmcs12->cpu_based_vm_exec_control &
5396 			    CPU_BASED_CR3_STORE_EXITING)
5397 				return true;
5398 			break;
5399 		case 8:
5400 			if (vmcs12->cpu_based_vm_exec_control &
5401 			    CPU_BASED_CR8_STORE_EXITING)
5402 				return true;
5403 			break;
5404 		}
5405 		break;
5406 	case 3: /* lmsw */
5407 		/*
5408 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5409 		 * cr0. Other attempted changes are ignored, with no exit.
5410 		 */
5411 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5412 		if (vmcs12->cr0_guest_host_mask & 0xe &
5413 		    (val ^ vmcs12->cr0_read_shadow))
5414 			return true;
5415 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5416 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5417 		    (val & 0x1))
5418 			return true;
5419 		break;
5420 	}
5421 	return false;
5422 }
5423 
5424 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5425 	struct vmcs12 *vmcs12, gpa_t bitmap)
5426 {
5427 	u32 vmx_instruction_info;
5428 	unsigned long field;
5429 	u8 b;
5430 
5431 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5432 		return true;
5433 
5434 	/* Decode instruction info and find the field to access */
5435 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5436 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5437 
5438 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5439 	if (field >> 15)
5440 		return true;
5441 
5442 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5443 		return true;
5444 
5445 	return 1 & (b >> (field & 7));
5446 }
5447 
5448 /*
5449  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5450  * should handle it ourselves in L0 (and then continue L2). Only call this
5451  * when in is_guest_mode (L2).
5452  */
5453 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5454 {
5455 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5456 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5457 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5458 
5459 	if (vmx->nested.nested_run_pending)
5460 		return false;
5461 
5462 	if (unlikely(vmx->fail)) {
5463 		trace_kvm_nested_vmenter_failed(
5464 			"hardware VM-instruction error: ",
5465 			vmcs_read32(VM_INSTRUCTION_ERROR));
5466 		return true;
5467 	}
5468 
5469 	/*
5470 	 * The host physical addresses of some pages of guest memory
5471 	 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5472 	 * Page). The CPU may write to these pages via their host
5473 	 * physical address while L2 is running, bypassing any
5474 	 * address-translation-based dirty tracking (e.g. EPT write
5475 	 * protection).
5476 	 *
5477 	 * Mark them dirty on every exit from L2 to prevent them from
5478 	 * getting out of sync with dirty tracking.
5479 	 */
5480 	nested_mark_vmcs12_pages_dirty(vcpu);
5481 
5482 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5483 				vmcs_readl(EXIT_QUALIFICATION),
5484 				vmx->idt_vectoring_info,
5485 				intr_info,
5486 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5487 				KVM_ISA_VMX);
5488 
5489 	switch (exit_reason) {
5490 	case EXIT_REASON_EXCEPTION_NMI:
5491 		if (is_nmi(intr_info))
5492 			return false;
5493 		else if (is_page_fault(intr_info))
5494 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5495 		else if (is_debug(intr_info) &&
5496 			 vcpu->guest_debug &
5497 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5498 			return false;
5499 		else if (is_breakpoint(intr_info) &&
5500 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5501 			return false;
5502 		return vmcs12->exception_bitmap &
5503 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5504 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5505 		return false;
5506 	case EXIT_REASON_TRIPLE_FAULT:
5507 		return true;
5508 	case EXIT_REASON_INTERRUPT_WINDOW:
5509 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5510 	case EXIT_REASON_NMI_WINDOW:
5511 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5512 	case EXIT_REASON_TASK_SWITCH:
5513 		return true;
5514 	case EXIT_REASON_CPUID:
5515 		return true;
5516 	case EXIT_REASON_HLT:
5517 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5518 	case EXIT_REASON_INVD:
5519 		return true;
5520 	case EXIT_REASON_INVLPG:
5521 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5522 	case EXIT_REASON_RDPMC:
5523 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5524 	case EXIT_REASON_RDRAND:
5525 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5526 	case EXIT_REASON_RDSEED:
5527 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5528 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5529 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5530 	case EXIT_REASON_VMREAD:
5531 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5532 			vmcs12->vmread_bitmap);
5533 	case EXIT_REASON_VMWRITE:
5534 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5535 			vmcs12->vmwrite_bitmap);
5536 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5537 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5538 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5539 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5540 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5541 		/*
5542 		 * VMX instructions trap unconditionally. This allows L1 to
5543 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5544 		 */
5545 		return true;
5546 	case EXIT_REASON_CR_ACCESS:
5547 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5548 	case EXIT_REASON_DR_ACCESS:
5549 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5550 	case EXIT_REASON_IO_INSTRUCTION:
5551 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5552 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5553 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5554 	case EXIT_REASON_MSR_READ:
5555 	case EXIT_REASON_MSR_WRITE:
5556 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5557 	case EXIT_REASON_INVALID_STATE:
5558 		return true;
5559 	case EXIT_REASON_MWAIT_INSTRUCTION:
5560 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5561 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5562 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
5563 	case EXIT_REASON_MONITOR_INSTRUCTION:
5564 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5565 	case EXIT_REASON_PAUSE_INSTRUCTION:
5566 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5567 			nested_cpu_has2(vmcs12,
5568 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5569 	case EXIT_REASON_MCE_DURING_VMENTRY:
5570 		return false;
5571 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5572 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5573 	case EXIT_REASON_APIC_ACCESS:
5574 	case EXIT_REASON_APIC_WRITE:
5575 	case EXIT_REASON_EOI_INDUCED:
5576 		/*
5577 		 * The controls for "virtualize APIC accesses," "APIC-
5578 		 * register virtualization," and "virtual-interrupt
5579 		 * delivery" only come from vmcs12.
5580 		 */
5581 		return true;
5582 	case EXIT_REASON_EPT_VIOLATION:
5583 		/*
5584 		 * L0 always deals with the EPT violation. If nested EPT is
5585 		 * used, and the nested mmu code discovers that the address is
5586 		 * missing in the guest EPT table (EPT12), the EPT violation
5587 		 * will be injected with nested_ept_inject_page_fault()
5588 		 */
5589 		return false;
5590 	case EXIT_REASON_EPT_MISCONFIG:
5591 		/*
5592 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5593 		 * table (shadow on EPT) or a merged EPT table that L0 built
5594 		 * (EPT on EPT). So any problems with the structure of the
5595 		 * table is L0's fault.
5596 		 */
5597 		return false;
5598 	case EXIT_REASON_INVPCID:
5599 		return
5600 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5601 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5602 	case EXIT_REASON_WBINVD:
5603 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5604 	case EXIT_REASON_XSETBV:
5605 		return true;
5606 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5607 		/*
5608 		 * This should never happen, since it is not possible to
5609 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5610 		 * If if it were, XSS would have to be checked against
5611 		 * the XSS exit bitmap in vmcs12.
5612 		 */
5613 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5614 	case EXIT_REASON_PREEMPTION_TIMER:
5615 		return false;
5616 	case EXIT_REASON_PML_FULL:
5617 		/* We emulate PML support to L1. */
5618 		return false;
5619 	case EXIT_REASON_VMFUNC:
5620 		/* VM functions are emulated through L2->L0 vmexits. */
5621 		return false;
5622 	case EXIT_REASON_ENCLS:
5623 		/* SGX is never exposed to L1 */
5624 		return false;
5625 	case EXIT_REASON_UMWAIT:
5626 	case EXIT_REASON_TPAUSE:
5627 		return nested_cpu_has2(vmcs12,
5628 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
5629 	default:
5630 		return true;
5631 	}
5632 }
5633 
5634 
5635 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5636 				struct kvm_nested_state __user *user_kvm_nested_state,
5637 				u32 user_data_size)
5638 {
5639 	struct vcpu_vmx *vmx;
5640 	struct vmcs12 *vmcs12;
5641 	struct kvm_nested_state kvm_state = {
5642 		.flags = 0,
5643 		.format = KVM_STATE_NESTED_FORMAT_VMX,
5644 		.size = sizeof(kvm_state),
5645 		.hdr.vmx.vmxon_pa = -1ull,
5646 		.hdr.vmx.vmcs12_pa = -1ull,
5647 	};
5648 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5649 		&user_kvm_nested_state->data.vmx[0];
5650 
5651 	if (!vcpu)
5652 		return kvm_state.size + sizeof(*user_vmx_nested_state);
5653 
5654 	vmx = to_vmx(vcpu);
5655 	vmcs12 = get_vmcs12(vcpu);
5656 
5657 	if (nested_vmx_allowed(vcpu) &&
5658 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5659 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5660 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
5661 
5662 		if (vmx_has_valid_vmcs12(vcpu)) {
5663 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
5664 
5665 			if (vmx->nested.hv_evmcs)
5666 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5667 
5668 			if (is_guest_mode(vcpu) &&
5669 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5670 			    vmcs12->vmcs_link_pointer != -1ull)
5671 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
5672 		}
5673 
5674 		if (vmx->nested.smm.vmxon)
5675 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5676 
5677 		if (vmx->nested.smm.guest_mode)
5678 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5679 
5680 		if (is_guest_mode(vcpu)) {
5681 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5682 
5683 			if (vmx->nested.nested_run_pending)
5684 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5685 		}
5686 	}
5687 
5688 	if (user_data_size < kvm_state.size)
5689 		goto out;
5690 
5691 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5692 		return -EFAULT;
5693 
5694 	if (!vmx_has_valid_vmcs12(vcpu))
5695 		goto out;
5696 
5697 	/*
5698 	 * When running L2, the authoritative vmcs12 state is in the
5699 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5700 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5701 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
5702 	 * vmcs12 state is in the vmcs12 already.
5703 	 */
5704 	if (is_guest_mode(vcpu)) {
5705 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
5706 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5707 	} else if (!vmx->nested.need_vmcs12_to_shadow_sync) {
5708 		if (vmx->nested.hv_evmcs)
5709 			copy_enlightened_to_vmcs12(vmx);
5710 		else if (enable_shadow_vmcs)
5711 			copy_shadow_to_vmcs12(vmx);
5712 	}
5713 
5714 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
5715 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
5716 
5717 	/*
5718 	 * Copy over the full allocated size of vmcs12 rather than just the size
5719 	 * of the struct.
5720 	 */
5721 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
5722 		return -EFAULT;
5723 
5724 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5725 	    vmcs12->vmcs_link_pointer != -1ull) {
5726 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
5727 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
5728 			return -EFAULT;
5729 	}
5730 
5731 out:
5732 	return kvm_state.size;
5733 }
5734 
5735 /*
5736  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5737  */
5738 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5739 {
5740 	if (is_guest_mode(vcpu)) {
5741 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5742 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5743 	}
5744 	free_nested(vcpu);
5745 }
5746 
5747 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5748 				struct kvm_nested_state __user *user_kvm_nested_state,
5749 				struct kvm_nested_state *kvm_state)
5750 {
5751 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5752 	struct vmcs12 *vmcs12;
5753 	u32 exit_qual;
5754 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5755 		&user_kvm_nested_state->data.vmx[0];
5756 	int ret;
5757 
5758 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
5759 		return -EINVAL;
5760 
5761 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
5762 		if (kvm_state->hdr.vmx.smm.flags)
5763 			return -EINVAL;
5764 
5765 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
5766 			return -EINVAL;
5767 
5768 		/*
5769 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
5770 		 * enable eVMCS capability on vCPU. However, since then
5771 		 * code was changed such that flag signals vmcs12 should
5772 		 * be copied into eVMCS in guest memory.
5773 		 *
5774 		 * To preserve backwards compatability, allow user
5775 		 * to set this flag even when there is no VMXON region.
5776 		 */
5777 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
5778 			return -EINVAL;
5779 	} else {
5780 		if (!nested_vmx_allowed(vcpu))
5781 			return -EINVAL;
5782 
5783 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
5784 			return -EINVAL;
5785 	}
5786 
5787 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5788 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5789 		return -EINVAL;
5790 
5791 	if (kvm_state->hdr.vmx.smm.flags &
5792 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5793 		return -EINVAL;
5794 
5795 	/*
5796 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5797 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5798 	 * must be zero.
5799 	 */
5800 	if (is_smm(vcpu) ?
5801 		(kvm_state->flags &
5802 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
5803 		: kvm_state->hdr.vmx.smm.flags)
5804 		return -EINVAL;
5805 
5806 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5807 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5808 		return -EINVAL;
5809 
5810 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
5811 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
5812 			return -EINVAL;
5813 
5814 	vmx_leave_nested(vcpu);
5815 
5816 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
5817 		return 0;
5818 
5819 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
5820 	ret = enter_vmx_operation(vcpu);
5821 	if (ret)
5822 		return ret;
5823 
5824 	/* Empty 'VMXON' state is permitted */
5825 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12))
5826 		return 0;
5827 
5828 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
5829 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
5830 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
5831 			return -EINVAL;
5832 
5833 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
5834 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5835 		/*
5836 		 * Sync eVMCS upon entry as we may not have
5837 		 * HV_X64_MSR_VP_ASSIST_PAGE set up yet.
5838 		 */
5839 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5840 	} else {
5841 		return -EINVAL;
5842 	}
5843 
5844 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5845 		vmx->nested.smm.vmxon = true;
5846 		vmx->nested.vmxon = false;
5847 
5848 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5849 			vmx->nested.smm.guest_mode = true;
5850 	}
5851 
5852 	vmcs12 = get_vmcs12(vcpu);
5853 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
5854 		return -EFAULT;
5855 
5856 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5857 		return -EINVAL;
5858 
5859 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5860 		return 0;
5861 
5862 	vmx->nested.nested_run_pending =
5863 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5864 
5865 	ret = -EINVAL;
5866 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5867 	    vmcs12->vmcs_link_pointer != -1ull) {
5868 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5869 
5870 		if (kvm_state->size <
5871 		    sizeof(*kvm_state) +
5872 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
5873 			goto error_guest_mode;
5874 
5875 		if (copy_from_user(shadow_vmcs12,
5876 				   user_vmx_nested_state->shadow_vmcs12,
5877 				   sizeof(*shadow_vmcs12))) {
5878 			ret = -EFAULT;
5879 			goto error_guest_mode;
5880 		}
5881 
5882 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5883 		    !shadow_vmcs12->hdr.shadow_vmcs)
5884 			goto error_guest_mode;
5885 	}
5886 
5887 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
5888 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
5889 	    nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
5890 		goto error_guest_mode;
5891 
5892 	vmx->nested.dirty_vmcs12 = true;
5893 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5894 	if (ret)
5895 		goto error_guest_mode;
5896 
5897 	return 0;
5898 
5899 error_guest_mode:
5900 	vmx->nested.nested_run_pending = 0;
5901 	return ret;
5902 }
5903 
5904 void nested_vmx_set_vmcs_shadowing_bitmap(void)
5905 {
5906 	if (enable_shadow_vmcs) {
5907 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5908 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5909 	}
5910 }
5911 
5912 /*
5913  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5914  * returned for the various VMX controls MSRs when nested VMX is enabled.
5915  * The same values should also be used to verify that vmcs12 control fields are
5916  * valid during nested entry from L1 to L2.
5917  * Each of these control msrs has a low and high 32-bit half: A low bit is on
5918  * if the corresponding bit in the (32-bit) control field *must* be on, and a
5919  * bit in the high half is on if the corresponding bit in the control field
5920  * may be on. See also vmx_control_verify().
5921  */
5922 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps,
5923 				bool apicv)
5924 {
5925 	/*
5926 	 * Note that as a general rule, the high half of the MSRs (bits in
5927 	 * the control fields which may be 1) should be initialized by the
5928 	 * intersection of the underlying hardware's MSR (i.e., features which
5929 	 * can be supported) and the list of features we want to expose -
5930 	 * because they are known to be properly supported in our code.
5931 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
5932 	 * be set to 0, meaning that L1 may turn off any of these bits. The
5933 	 * reason is that if one of these bits is necessary, it will appear
5934 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
5935 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
5936 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
5937 	 * These rules have exceptions below.
5938 	 */
5939 
5940 	/* pin-based controls */
5941 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
5942 		msrs->pinbased_ctls_low,
5943 		msrs->pinbased_ctls_high);
5944 	msrs->pinbased_ctls_low |=
5945 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5946 	msrs->pinbased_ctls_high &=
5947 		PIN_BASED_EXT_INTR_MASK |
5948 		PIN_BASED_NMI_EXITING |
5949 		PIN_BASED_VIRTUAL_NMIS |
5950 		(apicv ? PIN_BASED_POSTED_INTR : 0);
5951 	msrs->pinbased_ctls_high |=
5952 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5953 		PIN_BASED_VMX_PREEMPTION_TIMER;
5954 
5955 	/* exit controls */
5956 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
5957 		msrs->exit_ctls_low,
5958 		msrs->exit_ctls_high);
5959 	msrs->exit_ctls_low =
5960 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
5961 
5962 	msrs->exit_ctls_high &=
5963 #ifdef CONFIG_X86_64
5964 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
5965 #endif
5966 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
5967 	msrs->exit_ctls_high |=
5968 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
5969 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
5970 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
5971 
5972 	/* We support free control of debug control saving. */
5973 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
5974 
5975 	/* entry controls */
5976 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
5977 		msrs->entry_ctls_low,
5978 		msrs->entry_ctls_high);
5979 	msrs->entry_ctls_low =
5980 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
5981 	msrs->entry_ctls_high &=
5982 #ifdef CONFIG_X86_64
5983 		VM_ENTRY_IA32E_MODE |
5984 #endif
5985 		VM_ENTRY_LOAD_IA32_PAT;
5986 	msrs->entry_ctls_high |=
5987 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
5988 
5989 	/* We support free control of debug control loading. */
5990 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
5991 
5992 	/* cpu-based controls */
5993 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
5994 		msrs->procbased_ctls_low,
5995 		msrs->procbased_ctls_high);
5996 	msrs->procbased_ctls_low =
5997 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5998 	msrs->procbased_ctls_high &=
5999 		CPU_BASED_INTR_WINDOW_EXITING |
6000 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6001 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6002 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6003 		CPU_BASED_CR3_STORE_EXITING |
6004 #ifdef CONFIG_X86_64
6005 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6006 #endif
6007 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6008 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6009 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6010 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6011 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6012 	/*
6013 	 * We can allow some features even when not supported by the
6014 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6015 	 * can use it to avoid exits to L1 - even when L0 runs L2
6016 	 * without MSR bitmaps.
6017 	 */
6018 	msrs->procbased_ctls_high |=
6019 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6020 		CPU_BASED_USE_MSR_BITMAPS;
6021 
6022 	/* We support free control of CR3 access interception. */
6023 	msrs->procbased_ctls_low &=
6024 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6025 
6026 	/*
6027 	 * secondary cpu-based controls.  Do not include those that
6028 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
6029 	 */
6030 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6031 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6032 		      msrs->secondary_ctls_low,
6033 		      msrs->secondary_ctls_high);
6034 
6035 	msrs->secondary_ctls_low = 0;
6036 	msrs->secondary_ctls_high &=
6037 		SECONDARY_EXEC_DESC |
6038 		SECONDARY_EXEC_RDTSCP |
6039 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6040 		SECONDARY_EXEC_WBINVD_EXITING |
6041 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6042 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6043 		SECONDARY_EXEC_RDRAND_EXITING |
6044 		SECONDARY_EXEC_ENABLE_INVPCID |
6045 		SECONDARY_EXEC_RDSEED_EXITING |
6046 		SECONDARY_EXEC_XSAVES;
6047 
6048 	/*
6049 	 * We can emulate "VMCS shadowing," even if the hardware
6050 	 * doesn't support it.
6051 	 */
6052 	msrs->secondary_ctls_high |=
6053 		SECONDARY_EXEC_SHADOW_VMCS;
6054 
6055 	if (enable_ept) {
6056 		/* nested EPT: emulate EPT also to L1 */
6057 		msrs->secondary_ctls_high |=
6058 			SECONDARY_EXEC_ENABLE_EPT;
6059 		msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
6060 			 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
6061 		if (cpu_has_vmx_ept_execute_only())
6062 			msrs->ept_caps |=
6063 				VMX_EPT_EXECUTE_ONLY_BIT;
6064 		msrs->ept_caps &= ept_caps;
6065 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6066 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6067 			VMX_EPT_1GB_PAGE_BIT;
6068 		if (enable_ept_ad_bits) {
6069 			msrs->secondary_ctls_high |=
6070 				SECONDARY_EXEC_ENABLE_PML;
6071 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6072 		}
6073 	}
6074 
6075 	if (cpu_has_vmx_vmfunc()) {
6076 		msrs->secondary_ctls_high |=
6077 			SECONDARY_EXEC_ENABLE_VMFUNC;
6078 		/*
6079 		 * Advertise EPTP switching unconditionally
6080 		 * since we emulate it
6081 		 */
6082 		if (enable_ept)
6083 			msrs->vmfunc_controls =
6084 				VMX_VMFUNC_EPTP_SWITCHING;
6085 	}
6086 
6087 	/*
6088 	 * Old versions of KVM use the single-context version without
6089 	 * checking for support, so declare that it is supported even
6090 	 * though it is treated as global context.  The alternative is
6091 	 * not failing the single-context invvpid, and it is worse.
6092 	 */
6093 	if (enable_vpid) {
6094 		msrs->secondary_ctls_high |=
6095 			SECONDARY_EXEC_ENABLE_VPID;
6096 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6097 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6098 	}
6099 
6100 	if (enable_unrestricted_guest)
6101 		msrs->secondary_ctls_high |=
6102 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6103 
6104 	if (flexpriority_enabled)
6105 		msrs->secondary_ctls_high |=
6106 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6107 
6108 	/* miscellaneous data */
6109 	rdmsr(MSR_IA32_VMX_MISC,
6110 		msrs->misc_low,
6111 		msrs->misc_high);
6112 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6113 	msrs->misc_low |=
6114 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6115 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6116 		VMX_MISC_ACTIVITY_HLT;
6117 	msrs->misc_high = 0;
6118 
6119 	/*
6120 	 * This MSR reports some information about VMX support. We
6121 	 * should return information about the VMX we emulate for the
6122 	 * guest, and the VMCS structure we give it - not about the
6123 	 * VMX support of the underlying hardware.
6124 	 */
6125 	msrs->basic =
6126 		VMCS12_REVISION |
6127 		VMX_BASIC_TRUE_CTLS |
6128 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6129 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6130 
6131 	if (cpu_has_vmx_basic_inout())
6132 		msrs->basic |= VMX_BASIC_INOUT;
6133 
6134 	/*
6135 	 * These MSRs specify bits which the guest must keep fixed on
6136 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6137 	 * We picked the standard core2 setting.
6138 	 */
6139 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6140 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6141 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6142 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6143 
6144 	/* These MSRs specify bits which the guest must keep fixed off. */
6145 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6146 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6147 
6148 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
6149 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
6150 }
6151 
6152 void nested_vmx_hardware_unsetup(void)
6153 {
6154 	int i;
6155 
6156 	if (enable_shadow_vmcs) {
6157 		for (i = 0; i < VMX_BITMAP_NR; i++)
6158 			free_page((unsigned long)vmx_bitmap[i]);
6159 	}
6160 }
6161 
6162 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6163 {
6164 	int i;
6165 
6166 	if (!cpu_has_vmx_shadow_vmcs())
6167 		enable_shadow_vmcs = 0;
6168 	if (enable_shadow_vmcs) {
6169 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6170 			/*
6171 			 * The vmx_bitmap is not tied to a VM and so should
6172 			 * not be charged to a memcg.
6173 			 */
6174 			vmx_bitmap[i] = (unsigned long *)
6175 				__get_free_page(GFP_KERNEL);
6176 			if (!vmx_bitmap[i]) {
6177 				nested_vmx_hardware_unsetup();
6178 				return -ENOMEM;
6179 			}
6180 		}
6181 
6182 		init_vmcs_shadow_fields();
6183 	}
6184 
6185 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6186 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6187 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6188 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6189 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6190 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6191 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6192 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6193 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6194 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6195 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6196 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6197 
6198 	kvm_x86_ops->check_nested_events = vmx_check_nested_events;
6199 	kvm_x86_ops->get_nested_state = vmx_get_nested_state;
6200 	kvm_x86_ops->set_nested_state = vmx_set_nested_state;
6201 	kvm_x86_ops->get_vmcs12_pages = nested_get_vmcs12_pages;
6202 	kvm_x86_ops->nested_enable_evmcs = nested_enable_evmcs;
6203 	kvm_x86_ops->nested_get_evmcs_version = nested_get_evmcs_version;
6204 
6205 	return 0;
6206 }
6207