xref: /linux/arch/x86/kvm/vmx/nested.c (revision 0a94608f0f7de9b1135ffea3546afe68eafef57f)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/objtool.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 "evmcs.h"
11 #include "hyperv.h"
12 #include "mmu.h"
13 #include "nested.h"
14 #include "pmu.h"
15 #include "sgx.h"
16 #include "trace.h"
17 #include "vmx.h"
18 #include "x86.h"
19 
20 static bool __read_mostly enable_shadow_vmcs = 1;
21 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
22 
23 static bool __read_mostly nested_early_check = 0;
24 module_param(nested_early_check, bool, S_IRUGO);
25 
26 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
27 
28 /*
29  * Hyper-V requires all of these, so mark them as supported even though
30  * they are just treated the same as all-context.
31  */
32 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
33 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
34 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
35 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
36 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
37 
38 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
39 
40 enum {
41 	VMX_VMREAD_BITMAP,
42 	VMX_VMWRITE_BITMAP,
43 	VMX_BITMAP_NR
44 };
45 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
46 
47 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
48 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
49 
50 struct shadow_vmcs_field {
51 	u16	encoding;
52 	u16	offset;
53 };
54 static struct shadow_vmcs_field shadow_read_only_fields[] = {
55 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
56 #include "vmcs_shadow_fields.h"
57 };
58 static int max_shadow_read_only_fields =
59 	ARRAY_SIZE(shadow_read_only_fields);
60 
61 static struct shadow_vmcs_field shadow_read_write_fields[] = {
62 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
63 #include "vmcs_shadow_fields.h"
64 };
65 static int max_shadow_read_write_fields =
66 	ARRAY_SIZE(shadow_read_write_fields);
67 
68 static void init_vmcs_shadow_fields(void)
69 {
70 	int i, j;
71 
72 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
73 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
74 
75 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
76 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
77 		u16 field = entry.encoding;
78 
79 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
80 		    (i + 1 == max_shadow_read_only_fields ||
81 		     shadow_read_only_fields[i + 1].encoding != field + 1))
82 			pr_err("Missing field from shadow_read_only_field %x\n",
83 			       field + 1);
84 
85 		clear_bit(field, vmx_vmread_bitmap);
86 		if (field & 1)
87 #ifdef CONFIG_X86_64
88 			continue;
89 #else
90 			entry.offset += sizeof(u32);
91 #endif
92 		shadow_read_only_fields[j++] = entry;
93 	}
94 	max_shadow_read_only_fields = j;
95 
96 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
97 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
98 		u16 field = entry.encoding;
99 
100 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
101 		    (i + 1 == max_shadow_read_write_fields ||
102 		     shadow_read_write_fields[i + 1].encoding != field + 1))
103 			pr_err("Missing field from shadow_read_write_field %x\n",
104 			       field + 1);
105 
106 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
107 			  field <= GUEST_TR_AR_BYTES,
108 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
109 
110 		/*
111 		 * PML and the preemption timer can be emulated, but the
112 		 * processor cannot vmwrite to fields that don't exist
113 		 * on bare metal.
114 		 */
115 		switch (field) {
116 		case GUEST_PML_INDEX:
117 			if (!cpu_has_vmx_pml())
118 				continue;
119 			break;
120 		case VMX_PREEMPTION_TIMER_VALUE:
121 			if (!cpu_has_vmx_preemption_timer())
122 				continue;
123 			break;
124 		case GUEST_INTR_STATUS:
125 			if (!cpu_has_vmx_apicv())
126 				continue;
127 			break;
128 		default:
129 			break;
130 		}
131 
132 		clear_bit(field, vmx_vmwrite_bitmap);
133 		clear_bit(field, vmx_vmread_bitmap);
134 		if (field & 1)
135 #ifdef CONFIG_X86_64
136 			continue;
137 #else
138 			entry.offset += sizeof(u32);
139 #endif
140 		shadow_read_write_fields[j++] = entry;
141 	}
142 	max_shadow_read_write_fields = j;
143 }
144 
145 /*
146  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
147  * set the success or error code of an emulated VMX instruction (as specified
148  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
149  * instruction.
150  */
151 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
152 {
153 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
154 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
155 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
156 	return kvm_skip_emulated_instruction(vcpu);
157 }
158 
159 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
160 {
161 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
162 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
163 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
164 			| X86_EFLAGS_CF);
165 	return kvm_skip_emulated_instruction(vcpu);
166 }
167 
168 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
169 				u32 vm_instruction_error)
170 {
171 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
172 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
173 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
174 			| X86_EFLAGS_ZF);
175 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
176 	/*
177 	 * We don't need to force sync to shadow VMCS because
178 	 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
179 	 * fields and thus must be synced.
180 	 */
181 	if (to_vmx(vcpu)->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
182 		to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
183 
184 	return kvm_skip_emulated_instruction(vcpu);
185 }
186 
187 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
188 {
189 	struct vcpu_vmx *vmx = to_vmx(vcpu);
190 
191 	/*
192 	 * failValid writes the error number to the current VMCS, which
193 	 * can't be done if there isn't a current VMCS.
194 	 */
195 	if (vmx->nested.current_vmptr == INVALID_GPA &&
196 	    !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
197 		return nested_vmx_failInvalid(vcpu);
198 
199 	return nested_vmx_failValid(vcpu, vm_instruction_error);
200 }
201 
202 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
203 {
204 	/* TODO: not to reset guest simply here. */
205 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
206 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
207 }
208 
209 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
210 {
211 	return fixed_bits_valid(control, low, high);
212 }
213 
214 static inline u64 vmx_control_msr(u32 low, u32 high)
215 {
216 	return low | ((u64)high << 32);
217 }
218 
219 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
220 {
221 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
222 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
223 	vmx->nested.need_vmcs12_to_shadow_sync = false;
224 }
225 
226 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
227 {
228 	struct vcpu_vmx *vmx = to_vmx(vcpu);
229 
230 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
231 		kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
232 		vmx->nested.hv_evmcs = NULL;
233 	}
234 
235 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
236 }
237 
238 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
239 				     struct loaded_vmcs *prev)
240 {
241 	struct vmcs_host_state *dest, *src;
242 
243 	if (unlikely(!vmx->guest_state_loaded))
244 		return;
245 
246 	src = &prev->host_state;
247 	dest = &vmx->loaded_vmcs->host_state;
248 
249 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
250 	dest->ldt_sel = src->ldt_sel;
251 #ifdef CONFIG_X86_64
252 	dest->ds_sel = src->ds_sel;
253 	dest->es_sel = src->es_sel;
254 #endif
255 }
256 
257 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
258 {
259 	struct vcpu_vmx *vmx = to_vmx(vcpu);
260 	struct loaded_vmcs *prev;
261 	int cpu;
262 
263 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
264 		return;
265 
266 	cpu = get_cpu();
267 	prev = vmx->loaded_vmcs;
268 	vmx->loaded_vmcs = vmcs;
269 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
270 	vmx_sync_vmcs_host_state(vmx, prev);
271 	put_cpu();
272 
273 	vcpu->arch.regs_avail = ~VMX_REGS_LAZY_LOAD_SET;
274 
275 	/*
276 	 * All lazily updated registers will be reloaded from VMCS12 on both
277 	 * vmentry and vmexit.
278 	 */
279 	vcpu->arch.regs_dirty = 0;
280 }
281 
282 /*
283  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
284  * just stops using VMX.
285  */
286 static void free_nested(struct kvm_vcpu *vcpu)
287 {
288 	struct vcpu_vmx *vmx = to_vmx(vcpu);
289 
290 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
291 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
292 
293 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
294 		return;
295 
296 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
297 
298 	vmx->nested.vmxon = false;
299 	vmx->nested.smm.vmxon = false;
300 	vmx->nested.vmxon_ptr = INVALID_GPA;
301 	free_vpid(vmx->nested.vpid02);
302 	vmx->nested.posted_intr_nv = -1;
303 	vmx->nested.current_vmptr = INVALID_GPA;
304 	if (enable_shadow_vmcs) {
305 		vmx_disable_shadow_vmcs(vmx);
306 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
307 		free_vmcs(vmx->vmcs01.shadow_vmcs);
308 		vmx->vmcs01.shadow_vmcs = NULL;
309 	}
310 	kfree(vmx->nested.cached_vmcs12);
311 	vmx->nested.cached_vmcs12 = NULL;
312 	kfree(vmx->nested.cached_shadow_vmcs12);
313 	vmx->nested.cached_shadow_vmcs12 = NULL;
314 	/* Unpin physical memory we referred to in the vmcs02 */
315 	if (vmx->nested.apic_access_page) {
316 		kvm_release_page_clean(vmx->nested.apic_access_page);
317 		vmx->nested.apic_access_page = NULL;
318 	}
319 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
320 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
321 	vmx->nested.pi_desc = NULL;
322 
323 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
324 
325 	nested_release_evmcs(vcpu);
326 
327 	free_loaded_vmcs(&vmx->nested.vmcs02);
328 }
329 
330 /*
331  * Ensure that the current vmcs of the logical processor is the
332  * vmcs01 of the vcpu before calling free_nested().
333  */
334 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
335 {
336 	vcpu_load(vcpu);
337 	vmx_leave_nested(vcpu);
338 	vcpu_put(vcpu);
339 }
340 
341 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
342 
343 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
344 {
345 	return VALID_PAGE(root_hpa) &&
346 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
347 }
348 
349 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
350 				       gpa_t addr)
351 {
352 	uint i;
353 	struct kvm_mmu_root_info *cached_root;
354 
355 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
356 
357 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
358 		cached_root = &vcpu->arch.mmu->prev_roots[i];
359 
360 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
361 					    eptp))
362 			vcpu->arch.mmu->invlpg(vcpu, addr, cached_root->hpa);
363 	}
364 }
365 
366 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
367 		struct x86_exception *fault)
368 {
369 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
370 	struct vcpu_vmx *vmx = to_vmx(vcpu);
371 	u32 vm_exit_reason;
372 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
373 
374 	if (vmx->nested.pml_full) {
375 		vm_exit_reason = EXIT_REASON_PML_FULL;
376 		vmx->nested.pml_full = false;
377 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
378 	} else {
379 		if (fault->error_code & PFERR_RSVD_MASK)
380 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
381 		else
382 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
383 
384 		/*
385 		 * Although the caller (kvm_inject_emulated_page_fault) would
386 		 * have already synced the faulting address in the shadow EPT
387 		 * tables for the current EPTP12, we also need to sync it for
388 		 * any other cached EPTP02s based on the same EP4TA, since the
389 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
390 		 */
391 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
392 					   fault->address);
393 	}
394 
395 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
396 	vmcs12->guest_physical_address = fault->address;
397 }
398 
399 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
400 {
401 	struct vcpu_vmx *vmx = to_vmx(vcpu);
402 	bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
403 	int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
404 
405 	kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
406 				nested_ept_ad_enabled(vcpu),
407 				nested_ept_get_eptp(vcpu));
408 }
409 
410 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
411 {
412 	WARN_ON(mmu_is_nested(vcpu));
413 
414 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
415 	nested_ept_new_eptp(vcpu);
416 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
417 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
418 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
419 
420 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
421 }
422 
423 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
424 {
425 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
426 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
427 }
428 
429 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
430 					    u16 error_code)
431 {
432 	bool inequality, bit;
433 
434 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
435 	inequality =
436 		(error_code & vmcs12->page_fault_error_code_mask) !=
437 		 vmcs12->page_fault_error_code_match;
438 	return inequality ^ bit;
439 }
440 
441 
442 /*
443  * KVM wants to inject page-faults which it got to the guest. This function
444  * checks whether in a nested guest, we need to inject them to L1 or L2.
445  */
446 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
447 {
448 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
449 	unsigned int nr = vcpu->arch.exception.nr;
450 	bool has_payload = vcpu->arch.exception.has_payload;
451 	unsigned long payload = vcpu->arch.exception.payload;
452 
453 	if (nr == PF_VECTOR) {
454 		if (vcpu->arch.exception.nested_apf) {
455 			*exit_qual = vcpu->arch.apf.nested_apf_token;
456 			return 1;
457 		}
458 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
459 						    vcpu->arch.exception.error_code)) {
460 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
461 			return 1;
462 		}
463 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
464 		if (nr == DB_VECTOR) {
465 			if (!has_payload) {
466 				payload = vcpu->arch.dr6;
467 				payload &= ~DR6_BT;
468 				payload ^= DR6_ACTIVE_LOW;
469 			}
470 			*exit_qual = payload;
471 		} else
472 			*exit_qual = 0;
473 		return 1;
474 	}
475 
476 	return 0;
477 }
478 
479 
480 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
481 		struct x86_exception *fault)
482 {
483 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
484 
485 	WARN_ON(!is_guest_mode(vcpu));
486 
487 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
488 		!to_vmx(vcpu)->nested.nested_run_pending) {
489 		vmcs12->vm_exit_intr_error_code = fault->error_code;
490 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
491 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
492 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
493 				  fault->address);
494 	} else {
495 		kvm_inject_page_fault(vcpu, fault);
496 	}
497 }
498 
499 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
500 					       struct vmcs12 *vmcs12)
501 {
502 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
503 		return 0;
504 
505 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
506 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
507 		return -EINVAL;
508 
509 	return 0;
510 }
511 
512 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
513 						struct vmcs12 *vmcs12)
514 {
515 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
516 		return 0;
517 
518 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
519 		return -EINVAL;
520 
521 	return 0;
522 }
523 
524 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
525 						struct vmcs12 *vmcs12)
526 {
527 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
528 		return 0;
529 
530 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
531 		return -EINVAL;
532 
533 	return 0;
534 }
535 
536 /*
537  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
538  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
539  * only the "disable intercept" case needs to be handled.
540  */
541 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
542 							unsigned long *msr_bitmap_l0,
543 							u32 msr, int type)
544 {
545 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
546 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
547 
548 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
549 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
550 }
551 
552 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
553 {
554 	int msr;
555 
556 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
557 		unsigned word = msr / BITS_PER_LONG;
558 
559 		msr_bitmap[word] = ~0;
560 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
561 	}
562 }
563 
564 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
565 static inline									\
566 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
567 					 unsigned long *msr_bitmap_l1,		\
568 					 unsigned long *msr_bitmap_l0, u32 msr)	\
569 {										\
570 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
571 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
572 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
573 	else									\
574 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
575 }
576 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
577 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
578 
579 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
580 						    unsigned long *msr_bitmap_l1,
581 						    unsigned long *msr_bitmap_l0,
582 						    u32 msr, int types)
583 {
584 	if (types & MSR_TYPE_R)
585 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
586 						  msr_bitmap_l0, msr);
587 	if (types & MSR_TYPE_W)
588 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
589 						   msr_bitmap_l0, msr);
590 }
591 
592 /*
593  * Merge L0's and L1's MSR bitmap, return false to indicate that
594  * we do not use the hardware.
595  */
596 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
597 						 struct vmcs12 *vmcs12)
598 {
599 	struct vcpu_vmx *vmx = to_vmx(vcpu);
600 	int msr;
601 	unsigned long *msr_bitmap_l1;
602 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
603 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
604 	struct kvm_host_map *map = &vmx->nested.msr_bitmap_map;
605 
606 	/* Nothing to do if the MSR bitmap is not in use.  */
607 	if (!cpu_has_vmx_msr_bitmap() ||
608 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
609 		return false;
610 
611 	/*
612 	 * MSR bitmap update can be skipped when:
613 	 * - MSR bitmap for L1 hasn't changed.
614 	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
615 	 *   before.
616 	 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
617 	 *   and tells KVM (L0) there were no changes in MSR bitmap for L2.
618 	 */
619 	if (!vmx->nested.force_msr_bitmap_recalc && evmcs &&
620 	    evmcs->hv_enlightenments_control.msr_bitmap &&
621 	    evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
622 		return true;
623 
624 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
625 		return false;
626 
627 	msr_bitmap_l1 = (unsigned long *)map->hva;
628 
629 	/*
630 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
631 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
632 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
633 	 */
634 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
635 
636 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
637 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
638 			/*
639 			 * L0 need not intercept reads for MSRs between 0x800
640 			 * and 0x8ff, it just lets the processor take the value
641 			 * from the virtual-APIC page; take those 256 bits
642 			 * directly from the L1 bitmap.
643 			 */
644 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
645 				unsigned word = msr / BITS_PER_LONG;
646 
647 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
648 			}
649 		}
650 
651 		nested_vmx_disable_intercept_for_x2apic_msr(
652 			msr_bitmap_l1, msr_bitmap_l0,
653 			X2APIC_MSR(APIC_TASKPRI),
654 			MSR_TYPE_R | MSR_TYPE_W);
655 
656 		if (nested_cpu_has_vid(vmcs12)) {
657 			nested_vmx_disable_intercept_for_x2apic_msr(
658 				msr_bitmap_l1, msr_bitmap_l0,
659 				X2APIC_MSR(APIC_EOI),
660 				MSR_TYPE_W);
661 			nested_vmx_disable_intercept_for_x2apic_msr(
662 				msr_bitmap_l1, msr_bitmap_l0,
663 				X2APIC_MSR(APIC_SELF_IPI),
664 				MSR_TYPE_W);
665 		}
666 	}
667 
668 	/*
669 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
670 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
671 	 */
672 #ifdef CONFIG_X86_64
673 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
674 					 MSR_FS_BASE, MSR_TYPE_RW);
675 
676 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
677 					 MSR_GS_BASE, MSR_TYPE_RW);
678 
679 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
680 					 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
681 #endif
682 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
683 					 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
684 
685 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
686 					 MSR_IA32_PRED_CMD, MSR_TYPE_W);
687 
688 	kvm_vcpu_unmap(vcpu, &vmx->nested.msr_bitmap_map, false);
689 
690 	vmx->nested.force_msr_bitmap_recalc = false;
691 
692 	return true;
693 }
694 
695 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
696 				       struct vmcs12 *vmcs12)
697 {
698 	struct vcpu_vmx *vmx = to_vmx(vcpu);
699 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
700 
701 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
702 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
703 		return;
704 
705 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
706 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
707 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
708 		return;
709 
710 	kvm_read_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
711 			      VMCS12_SIZE);
712 }
713 
714 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
715 					      struct vmcs12 *vmcs12)
716 {
717 	struct vcpu_vmx *vmx = to_vmx(vcpu);
718 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
719 
720 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
721 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
722 		return;
723 
724 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
725 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
726 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
727 		return;
728 
729 	kvm_write_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
730 			       VMCS12_SIZE);
731 }
732 
733 /*
734  * In nested virtualization, check if L1 has set
735  * VM_EXIT_ACK_INTR_ON_EXIT
736  */
737 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
738 {
739 	return get_vmcs12(vcpu)->vm_exit_controls &
740 		VM_EXIT_ACK_INTR_ON_EXIT;
741 }
742 
743 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
744 					  struct vmcs12 *vmcs12)
745 {
746 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
747 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
748 		return -EINVAL;
749 	else
750 		return 0;
751 }
752 
753 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
754 					   struct vmcs12 *vmcs12)
755 {
756 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
757 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
758 	    !nested_cpu_has_vid(vmcs12) &&
759 	    !nested_cpu_has_posted_intr(vmcs12))
760 		return 0;
761 
762 	/*
763 	 * If virtualize x2apic mode is enabled,
764 	 * virtualize apic access must be disabled.
765 	 */
766 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
767 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
768 		return -EINVAL;
769 
770 	/*
771 	 * If virtual interrupt delivery is enabled,
772 	 * we must exit on external interrupts.
773 	 */
774 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
775 		return -EINVAL;
776 
777 	/*
778 	 * bits 15:8 should be zero in posted_intr_nv,
779 	 * the descriptor address has been already checked
780 	 * in nested_get_vmcs12_pages.
781 	 *
782 	 * bits 5:0 of posted_intr_desc_addr should be zero.
783 	 */
784 	if (nested_cpu_has_posted_intr(vmcs12) &&
785 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
786 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
787 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
788 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
789 		return -EINVAL;
790 
791 	/* tpr shadow is needed by all apicv features. */
792 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
793 		return -EINVAL;
794 
795 	return 0;
796 }
797 
798 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
799 				       u32 count, u64 addr)
800 {
801 	if (count == 0)
802 		return 0;
803 
804 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
805 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
806 		return -EINVAL;
807 
808 	return 0;
809 }
810 
811 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
812 						     struct vmcs12 *vmcs12)
813 {
814 	if (CC(nested_vmx_check_msr_switch(vcpu,
815 					   vmcs12->vm_exit_msr_load_count,
816 					   vmcs12->vm_exit_msr_load_addr)) ||
817 	    CC(nested_vmx_check_msr_switch(vcpu,
818 					   vmcs12->vm_exit_msr_store_count,
819 					   vmcs12->vm_exit_msr_store_addr)))
820 		return -EINVAL;
821 
822 	return 0;
823 }
824 
825 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
826                                                       struct vmcs12 *vmcs12)
827 {
828 	if (CC(nested_vmx_check_msr_switch(vcpu,
829 					   vmcs12->vm_entry_msr_load_count,
830 					   vmcs12->vm_entry_msr_load_addr)))
831                 return -EINVAL;
832 
833 	return 0;
834 }
835 
836 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
837 					 struct vmcs12 *vmcs12)
838 {
839 	if (!nested_cpu_has_pml(vmcs12))
840 		return 0;
841 
842 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
843 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
844 		return -EINVAL;
845 
846 	return 0;
847 }
848 
849 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
850 							struct vmcs12 *vmcs12)
851 {
852 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
853 	       !nested_cpu_has_ept(vmcs12)))
854 		return -EINVAL;
855 	return 0;
856 }
857 
858 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
859 							 struct vmcs12 *vmcs12)
860 {
861 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
862 	       !nested_cpu_has_ept(vmcs12)))
863 		return -EINVAL;
864 	return 0;
865 }
866 
867 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
868 						 struct vmcs12 *vmcs12)
869 {
870 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
871 		return 0;
872 
873 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
874 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
875 		return -EINVAL;
876 
877 	return 0;
878 }
879 
880 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
881 				       struct vmx_msr_entry *e)
882 {
883 	/* x2APIC MSR accesses are not allowed */
884 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
885 		return -EINVAL;
886 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
887 	    CC(e->index == MSR_IA32_UCODE_REV))
888 		return -EINVAL;
889 	if (CC(e->reserved != 0))
890 		return -EINVAL;
891 	return 0;
892 }
893 
894 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
895 				     struct vmx_msr_entry *e)
896 {
897 	if (CC(e->index == MSR_FS_BASE) ||
898 	    CC(e->index == MSR_GS_BASE) ||
899 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
900 	    nested_vmx_msr_check_common(vcpu, e))
901 		return -EINVAL;
902 	return 0;
903 }
904 
905 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
906 				      struct vmx_msr_entry *e)
907 {
908 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
909 	    nested_vmx_msr_check_common(vcpu, e))
910 		return -EINVAL;
911 	return 0;
912 }
913 
914 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
915 {
916 	struct vcpu_vmx *vmx = to_vmx(vcpu);
917 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
918 				       vmx->nested.msrs.misc_high);
919 
920 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
921 }
922 
923 /*
924  * Load guest's/host's msr at nested entry/exit.
925  * return 0 for success, entry index for failure.
926  *
927  * One of the failure modes for MSR load/store is when a list exceeds the
928  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
929  * as possible, process all valid entries before failing rather than precheck
930  * for a capacity violation.
931  */
932 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
933 {
934 	u32 i;
935 	struct vmx_msr_entry e;
936 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
937 
938 	for (i = 0; i < count; i++) {
939 		if (unlikely(i >= max_msr_list_size))
940 			goto fail;
941 
942 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
943 					&e, sizeof(e))) {
944 			pr_debug_ratelimited(
945 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
946 				__func__, i, gpa + i * sizeof(e));
947 			goto fail;
948 		}
949 		if (nested_vmx_load_msr_check(vcpu, &e)) {
950 			pr_debug_ratelimited(
951 				"%s check failed (%u, 0x%x, 0x%x)\n",
952 				__func__, i, e.index, e.reserved);
953 			goto fail;
954 		}
955 		if (kvm_set_msr(vcpu, e.index, e.value)) {
956 			pr_debug_ratelimited(
957 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
958 				__func__, i, e.index, e.value);
959 			goto fail;
960 		}
961 	}
962 	return 0;
963 fail:
964 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
965 	return i + 1;
966 }
967 
968 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
969 					    u32 msr_index,
970 					    u64 *data)
971 {
972 	struct vcpu_vmx *vmx = to_vmx(vcpu);
973 
974 	/*
975 	 * If the L0 hypervisor stored a more accurate value for the TSC that
976 	 * does not include the time taken for emulation of the L2->L1
977 	 * VM-exit in L0, use the more accurate value.
978 	 */
979 	if (msr_index == MSR_IA32_TSC) {
980 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
981 						    MSR_IA32_TSC);
982 
983 		if (i >= 0) {
984 			u64 val = vmx->msr_autostore.guest.val[i].value;
985 
986 			*data = kvm_read_l1_tsc(vcpu, val);
987 			return true;
988 		}
989 	}
990 
991 	if (kvm_get_msr(vcpu, msr_index, data)) {
992 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
993 			msr_index);
994 		return false;
995 	}
996 	return true;
997 }
998 
999 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
1000 				     struct vmx_msr_entry *e)
1001 {
1002 	if (kvm_vcpu_read_guest(vcpu,
1003 				gpa + i * sizeof(*e),
1004 				e, 2 * sizeof(u32))) {
1005 		pr_debug_ratelimited(
1006 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
1007 			__func__, i, gpa + i * sizeof(*e));
1008 		return false;
1009 	}
1010 	if (nested_vmx_store_msr_check(vcpu, e)) {
1011 		pr_debug_ratelimited(
1012 			"%s check failed (%u, 0x%x, 0x%x)\n",
1013 			__func__, i, e->index, e->reserved);
1014 		return false;
1015 	}
1016 	return true;
1017 }
1018 
1019 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1020 {
1021 	u64 data;
1022 	u32 i;
1023 	struct vmx_msr_entry e;
1024 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1025 
1026 	for (i = 0; i < count; i++) {
1027 		if (unlikely(i >= max_msr_list_size))
1028 			return -EINVAL;
1029 
1030 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1031 			return -EINVAL;
1032 
1033 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1034 			return -EINVAL;
1035 
1036 		if (kvm_vcpu_write_guest(vcpu,
1037 					 gpa + i * sizeof(e) +
1038 					     offsetof(struct vmx_msr_entry, value),
1039 					 &data, sizeof(data))) {
1040 			pr_debug_ratelimited(
1041 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1042 				__func__, i, e.index, data);
1043 			return -EINVAL;
1044 		}
1045 	}
1046 	return 0;
1047 }
1048 
1049 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1050 {
1051 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1052 	u32 count = vmcs12->vm_exit_msr_store_count;
1053 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1054 	struct vmx_msr_entry e;
1055 	u32 i;
1056 
1057 	for (i = 0; i < count; i++) {
1058 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1059 			return false;
1060 
1061 		if (e.index == msr_index)
1062 			return true;
1063 	}
1064 	return false;
1065 }
1066 
1067 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1068 					   u32 msr_index)
1069 {
1070 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1071 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1072 	bool in_vmcs12_store_list;
1073 	int msr_autostore_slot;
1074 	bool in_autostore_list;
1075 	int last;
1076 
1077 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1078 	in_autostore_list = msr_autostore_slot >= 0;
1079 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1080 
1081 	if (in_vmcs12_store_list && !in_autostore_list) {
1082 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1083 			/*
1084 			 * Emulated VMEntry does not fail here.  Instead a less
1085 			 * accurate value will be returned by
1086 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1087 			 * instead of reading the value from the vmcs02 VMExit
1088 			 * MSR-store area.
1089 			 */
1090 			pr_warn_ratelimited(
1091 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1092 				msr_index);
1093 			return;
1094 		}
1095 		last = autostore->nr++;
1096 		autostore->val[last].index = msr_index;
1097 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1098 		last = --autostore->nr;
1099 		autostore->val[msr_autostore_slot] = autostore->val[last];
1100 	}
1101 }
1102 
1103 /*
1104  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1105  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1106  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1107  * @entry_failure_code.
1108  */
1109 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1110 			       bool nested_ept, bool reload_pdptrs,
1111 			       enum vm_entry_failure_code *entry_failure_code)
1112 {
1113 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1114 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1115 		return -EINVAL;
1116 	}
1117 
1118 	/*
1119 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1120 	 * must not be dereferenced.
1121 	 */
1122 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1123 	    CC(!load_pdptrs(vcpu, cr3))) {
1124 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1125 		return -EINVAL;
1126 	}
1127 
1128 	vcpu->arch.cr3 = cr3;
1129 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1130 
1131 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1132 	kvm_init_mmu(vcpu);
1133 
1134 	if (!nested_ept)
1135 		kvm_mmu_new_pgd(vcpu, cr3);
1136 
1137 	return 0;
1138 }
1139 
1140 /*
1141  * Returns if KVM is able to config CPU to tag TLB entries
1142  * populated by L2 differently than TLB entries populated
1143  * by L1.
1144  *
1145  * If L0 uses EPT, L1 and L2 run with different EPTP because
1146  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1147  * are tagged with different EPTP.
1148  *
1149  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1150  * with different VPID (L1 entries are tagged with vmx->vpid
1151  * while L2 entries are tagged with vmx->nested.vpid02).
1152  */
1153 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1154 {
1155 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1156 
1157 	return enable_ept ||
1158 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1159 }
1160 
1161 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1162 					    struct vmcs12 *vmcs12,
1163 					    bool is_vmenter)
1164 {
1165 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1166 
1167 	/*
1168 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1169 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1170 	 * full TLB flush from the guest's perspective.  This is required even
1171 	 * if VPID is disabled in the host as KVM may need to synchronize the
1172 	 * MMU in response to the guest TLB flush.
1173 	 *
1174 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1175 	 * EPT is a special snowflake, as guest-physical mappings aren't
1176 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1177 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1178 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1179 	 * those mappings.
1180 	 */
1181 	if (!nested_cpu_has_vpid(vmcs12)) {
1182 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1183 		return;
1184 	}
1185 
1186 	/* L2 should never have a VPID if VPID is disabled. */
1187 	WARN_ON(!enable_vpid);
1188 
1189 	/*
1190 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1191 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1192 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1193 	 * that the new vpid12 has never been used and thus represents a new
1194 	 * guest ASID that cannot have entries in the TLB.
1195 	 */
1196 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1197 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1198 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1199 		return;
1200 	}
1201 
1202 	/*
1203 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1204 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1205 	 * KVM was unable to allocate a VPID for L2, flush the current context
1206 	 * as the effective ASID is common to both L1 and L2.
1207 	 */
1208 	if (!nested_has_guest_tlb_tag(vcpu))
1209 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1210 }
1211 
1212 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1213 {
1214 	superset &= mask;
1215 	subset &= mask;
1216 
1217 	return (superset | subset) == superset;
1218 }
1219 
1220 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1221 {
1222 	const u64 feature_and_reserved =
1223 		/* feature (except bit 48; see below) */
1224 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1225 		/* reserved */
1226 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1227 	u64 vmx_basic = vmx->nested.msrs.basic;
1228 
1229 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1230 		return -EINVAL;
1231 
1232 	/*
1233 	 * KVM does not emulate a version of VMX that constrains physical
1234 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1235 	 */
1236 	if (data & BIT_ULL(48))
1237 		return -EINVAL;
1238 
1239 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1240 	    vmx_basic_vmcs_revision_id(data))
1241 		return -EINVAL;
1242 
1243 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1244 		return -EINVAL;
1245 
1246 	vmx->nested.msrs.basic = data;
1247 	return 0;
1248 }
1249 
1250 static int
1251 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1252 {
1253 	u64 supported;
1254 	u32 *lowp, *highp;
1255 
1256 	switch (msr_index) {
1257 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1258 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1259 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1260 		break;
1261 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1262 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1263 		highp = &vmx->nested.msrs.procbased_ctls_high;
1264 		break;
1265 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1266 		lowp = &vmx->nested.msrs.exit_ctls_low;
1267 		highp = &vmx->nested.msrs.exit_ctls_high;
1268 		break;
1269 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1270 		lowp = &vmx->nested.msrs.entry_ctls_low;
1271 		highp = &vmx->nested.msrs.entry_ctls_high;
1272 		break;
1273 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1274 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1275 		highp = &vmx->nested.msrs.secondary_ctls_high;
1276 		break;
1277 	default:
1278 		BUG();
1279 	}
1280 
1281 	supported = vmx_control_msr(*lowp, *highp);
1282 
1283 	/* Check must-be-1 bits are still 1. */
1284 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1285 		return -EINVAL;
1286 
1287 	/* Check must-be-0 bits are still 0. */
1288 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1289 		return -EINVAL;
1290 
1291 	*lowp = data;
1292 	*highp = data >> 32;
1293 	return 0;
1294 }
1295 
1296 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1297 {
1298 	const u64 feature_and_reserved_bits =
1299 		/* feature */
1300 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1301 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1302 		/* reserved */
1303 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1304 	u64 vmx_misc;
1305 
1306 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1307 				   vmx->nested.msrs.misc_high);
1308 
1309 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1310 		return -EINVAL;
1311 
1312 	if ((vmx->nested.msrs.pinbased_ctls_high &
1313 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1314 	    vmx_misc_preemption_timer_rate(data) !=
1315 	    vmx_misc_preemption_timer_rate(vmx_misc))
1316 		return -EINVAL;
1317 
1318 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1319 		return -EINVAL;
1320 
1321 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1322 		return -EINVAL;
1323 
1324 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1325 		return -EINVAL;
1326 
1327 	vmx->nested.msrs.misc_low = data;
1328 	vmx->nested.msrs.misc_high = data >> 32;
1329 
1330 	return 0;
1331 }
1332 
1333 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1334 {
1335 	u64 vmx_ept_vpid_cap;
1336 
1337 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1338 					   vmx->nested.msrs.vpid_caps);
1339 
1340 	/* Every bit is either reserved or a feature bit. */
1341 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1342 		return -EINVAL;
1343 
1344 	vmx->nested.msrs.ept_caps = data;
1345 	vmx->nested.msrs.vpid_caps = data >> 32;
1346 	return 0;
1347 }
1348 
1349 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1350 {
1351 	u64 *msr;
1352 
1353 	switch (msr_index) {
1354 	case MSR_IA32_VMX_CR0_FIXED0:
1355 		msr = &vmx->nested.msrs.cr0_fixed0;
1356 		break;
1357 	case MSR_IA32_VMX_CR4_FIXED0:
1358 		msr = &vmx->nested.msrs.cr4_fixed0;
1359 		break;
1360 	default:
1361 		BUG();
1362 	}
1363 
1364 	/*
1365 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1366 	 * must be 1 in the restored value.
1367 	 */
1368 	if (!is_bitwise_subset(data, *msr, -1ULL))
1369 		return -EINVAL;
1370 
1371 	*msr = data;
1372 	return 0;
1373 }
1374 
1375 /*
1376  * Called when userspace is restoring VMX MSRs.
1377  *
1378  * Returns 0 on success, non-0 otherwise.
1379  */
1380 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1381 {
1382 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1383 
1384 	/*
1385 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1386 	 * is in VMX operation.
1387 	 */
1388 	if (vmx->nested.vmxon)
1389 		return -EBUSY;
1390 
1391 	switch (msr_index) {
1392 	case MSR_IA32_VMX_BASIC:
1393 		return vmx_restore_vmx_basic(vmx, data);
1394 	case MSR_IA32_VMX_PINBASED_CTLS:
1395 	case MSR_IA32_VMX_PROCBASED_CTLS:
1396 	case MSR_IA32_VMX_EXIT_CTLS:
1397 	case MSR_IA32_VMX_ENTRY_CTLS:
1398 		/*
1399 		 * The "non-true" VMX capability MSRs are generated from the
1400 		 * "true" MSRs, so we do not support restoring them directly.
1401 		 *
1402 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1403 		 * should restore the "true" MSRs with the must-be-1 bits
1404 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1405 		 * DEFAULT SETTINGS".
1406 		 */
1407 		return -EINVAL;
1408 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1409 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1410 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1411 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1412 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1413 		return vmx_restore_control_msr(vmx, msr_index, data);
1414 	case MSR_IA32_VMX_MISC:
1415 		return vmx_restore_vmx_misc(vmx, data);
1416 	case MSR_IA32_VMX_CR0_FIXED0:
1417 	case MSR_IA32_VMX_CR4_FIXED0:
1418 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1419 	case MSR_IA32_VMX_CR0_FIXED1:
1420 	case MSR_IA32_VMX_CR4_FIXED1:
1421 		/*
1422 		 * These MSRs are generated based on the vCPU's CPUID, so we
1423 		 * do not support restoring them directly.
1424 		 */
1425 		return -EINVAL;
1426 	case MSR_IA32_VMX_EPT_VPID_CAP:
1427 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1428 	case MSR_IA32_VMX_VMCS_ENUM:
1429 		vmx->nested.msrs.vmcs_enum = data;
1430 		return 0;
1431 	case MSR_IA32_VMX_VMFUNC:
1432 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1433 			return -EINVAL;
1434 		vmx->nested.msrs.vmfunc_controls = data;
1435 		return 0;
1436 	default:
1437 		/*
1438 		 * The rest of the VMX capability MSRs do not support restore.
1439 		 */
1440 		return -EINVAL;
1441 	}
1442 }
1443 
1444 /* Returns 0 on success, non-0 otherwise. */
1445 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1446 {
1447 	switch (msr_index) {
1448 	case MSR_IA32_VMX_BASIC:
1449 		*pdata = msrs->basic;
1450 		break;
1451 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1452 	case MSR_IA32_VMX_PINBASED_CTLS:
1453 		*pdata = vmx_control_msr(
1454 			msrs->pinbased_ctls_low,
1455 			msrs->pinbased_ctls_high);
1456 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1457 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1458 		break;
1459 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1460 	case MSR_IA32_VMX_PROCBASED_CTLS:
1461 		*pdata = vmx_control_msr(
1462 			msrs->procbased_ctls_low,
1463 			msrs->procbased_ctls_high);
1464 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1465 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1466 		break;
1467 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1468 	case MSR_IA32_VMX_EXIT_CTLS:
1469 		*pdata = vmx_control_msr(
1470 			msrs->exit_ctls_low,
1471 			msrs->exit_ctls_high);
1472 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1473 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1474 		break;
1475 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1476 	case MSR_IA32_VMX_ENTRY_CTLS:
1477 		*pdata = vmx_control_msr(
1478 			msrs->entry_ctls_low,
1479 			msrs->entry_ctls_high);
1480 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1481 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1482 		break;
1483 	case MSR_IA32_VMX_MISC:
1484 		*pdata = vmx_control_msr(
1485 			msrs->misc_low,
1486 			msrs->misc_high);
1487 		break;
1488 	case MSR_IA32_VMX_CR0_FIXED0:
1489 		*pdata = msrs->cr0_fixed0;
1490 		break;
1491 	case MSR_IA32_VMX_CR0_FIXED1:
1492 		*pdata = msrs->cr0_fixed1;
1493 		break;
1494 	case MSR_IA32_VMX_CR4_FIXED0:
1495 		*pdata = msrs->cr4_fixed0;
1496 		break;
1497 	case MSR_IA32_VMX_CR4_FIXED1:
1498 		*pdata = msrs->cr4_fixed1;
1499 		break;
1500 	case MSR_IA32_VMX_VMCS_ENUM:
1501 		*pdata = msrs->vmcs_enum;
1502 		break;
1503 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1504 		*pdata = vmx_control_msr(
1505 			msrs->secondary_ctls_low,
1506 			msrs->secondary_ctls_high);
1507 		break;
1508 	case MSR_IA32_VMX_EPT_VPID_CAP:
1509 		*pdata = msrs->ept_caps |
1510 			((u64)msrs->vpid_caps << 32);
1511 		break;
1512 	case MSR_IA32_VMX_VMFUNC:
1513 		*pdata = msrs->vmfunc_controls;
1514 		break;
1515 	default:
1516 		return 1;
1517 	}
1518 
1519 	return 0;
1520 }
1521 
1522 /*
1523  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1524  * been modified by the L1 guest.  Note, "writable" in this context means
1525  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1526  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1527  * VM-exit information fields (which are actually writable if the vCPU is
1528  * configured to support "VMWRITE to any supported field in the VMCS").
1529  */
1530 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1531 {
1532 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1533 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1534 	struct shadow_vmcs_field field;
1535 	unsigned long val;
1536 	int i;
1537 
1538 	if (WARN_ON(!shadow_vmcs))
1539 		return;
1540 
1541 	preempt_disable();
1542 
1543 	vmcs_load(shadow_vmcs);
1544 
1545 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1546 		field = shadow_read_write_fields[i];
1547 		val = __vmcs_readl(field.encoding);
1548 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1549 	}
1550 
1551 	vmcs_clear(shadow_vmcs);
1552 	vmcs_load(vmx->loaded_vmcs->vmcs);
1553 
1554 	preempt_enable();
1555 }
1556 
1557 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1558 {
1559 	const struct shadow_vmcs_field *fields[] = {
1560 		shadow_read_write_fields,
1561 		shadow_read_only_fields
1562 	};
1563 	const int max_fields[] = {
1564 		max_shadow_read_write_fields,
1565 		max_shadow_read_only_fields
1566 	};
1567 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1568 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1569 	struct shadow_vmcs_field field;
1570 	unsigned long val;
1571 	int i, q;
1572 
1573 	if (WARN_ON(!shadow_vmcs))
1574 		return;
1575 
1576 	vmcs_load(shadow_vmcs);
1577 
1578 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1579 		for (i = 0; i < max_fields[q]; i++) {
1580 			field = fields[q][i];
1581 			val = vmcs12_read_any(vmcs12, field.encoding,
1582 					      field.offset);
1583 			__vmcs_writel(field.encoding, val);
1584 		}
1585 	}
1586 
1587 	vmcs_clear(shadow_vmcs);
1588 	vmcs_load(vmx->loaded_vmcs->vmcs);
1589 }
1590 
1591 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1592 {
1593 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1594 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1595 
1596 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1597 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1598 	vmcs12->guest_rip = evmcs->guest_rip;
1599 
1600 	if (unlikely(!(hv_clean_fields &
1601 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1602 		vmcs12->guest_rsp = evmcs->guest_rsp;
1603 		vmcs12->guest_rflags = evmcs->guest_rflags;
1604 		vmcs12->guest_interruptibility_info =
1605 			evmcs->guest_interruptibility_info;
1606 	}
1607 
1608 	if (unlikely(!(hv_clean_fields &
1609 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1610 		vmcs12->cpu_based_vm_exec_control =
1611 			evmcs->cpu_based_vm_exec_control;
1612 	}
1613 
1614 	if (unlikely(!(hv_clean_fields &
1615 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1616 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1617 	}
1618 
1619 	if (unlikely(!(hv_clean_fields &
1620 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1621 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1622 	}
1623 
1624 	if (unlikely(!(hv_clean_fields &
1625 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1626 		vmcs12->vm_entry_intr_info_field =
1627 			evmcs->vm_entry_intr_info_field;
1628 		vmcs12->vm_entry_exception_error_code =
1629 			evmcs->vm_entry_exception_error_code;
1630 		vmcs12->vm_entry_instruction_len =
1631 			evmcs->vm_entry_instruction_len;
1632 	}
1633 
1634 	if (unlikely(!(hv_clean_fields &
1635 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1636 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1637 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1638 		vmcs12->host_cr0 = evmcs->host_cr0;
1639 		vmcs12->host_cr3 = evmcs->host_cr3;
1640 		vmcs12->host_cr4 = evmcs->host_cr4;
1641 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1642 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1643 		vmcs12->host_rip = evmcs->host_rip;
1644 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1645 		vmcs12->host_es_selector = evmcs->host_es_selector;
1646 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1647 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1648 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1649 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1650 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1651 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1652 	}
1653 
1654 	if (unlikely(!(hv_clean_fields &
1655 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1656 		vmcs12->pin_based_vm_exec_control =
1657 			evmcs->pin_based_vm_exec_control;
1658 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1659 		vmcs12->secondary_vm_exec_control =
1660 			evmcs->secondary_vm_exec_control;
1661 	}
1662 
1663 	if (unlikely(!(hv_clean_fields &
1664 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1665 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1666 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1667 	}
1668 
1669 	if (unlikely(!(hv_clean_fields &
1670 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1671 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1672 	}
1673 
1674 	if (unlikely(!(hv_clean_fields &
1675 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1676 		vmcs12->guest_es_base = evmcs->guest_es_base;
1677 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1678 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1679 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1680 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1681 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1682 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1683 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1684 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1685 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1686 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1687 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1688 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1689 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1690 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1691 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1692 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1693 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1694 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1695 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1696 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1697 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1698 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1699 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1700 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1701 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1702 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1703 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1704 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1705 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1706 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1707 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1708 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1709 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1710 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1711 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1712 	}
1713 
1714 	if (unlikely(!(hv_clean_fields &
1715 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1716 		vmcs12->tsc_offset = evmcs->tsc_offset;
1717 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1718 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1719 	}
1720 
1721 	if (unlikely(!(hv_clean_fields &
1722 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1723 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1724 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1725 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1726 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1727 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1728 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1729 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1730 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1731 	}
1732 
1733 	if (unlikely(!(hv_clean_fields &
1734 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1735 		vmcs12->host_fs_base = evmcs->host_fs_base;
1736 		vmcs12->host_gs_base = evmcs->host_gs_base;
1737 		vmcs12->host_tr_base = evmcs->host_tr_base;
1738 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1739 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1740 		vmcs12->host_rsp = evmcs->host_rsp;
1741 	}
1742 
1743 	if (unlikely(!(hv_clean_fields &
1744 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1745 		vmcs12->ept_pointer = evmcs->ept_pointer;
1746 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1747 	}
1748 
1749 	if (unlikely(!(hv_clean_fields &
1750 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1751 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1752 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1753 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1754 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1755 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1756 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1757 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1758 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1759 		vmcs12->guest_pending_dbg_exceptions =
1760 			evmcs->guest_pending_dbg_exceptions;
1761 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1762 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1763 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1764 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1765 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1766 	}
1767 
1768 	/*
1769 	 * Not used?
1770 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1771 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1772 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1773 	 * vmcs12->page_fault_error_code_mask =
1774 	 *		evmcs->page_fault_error_code_mask;
1775 	 * vmcs12->page_fault_error_code_match =
1776 	 *		evmcs->page_fault_error_code_match;
1777 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1778 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1779 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1780 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1781 	 */
1782 
1783 	/*
1784 	 * Read only fields:
1785 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1786 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1787 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1788 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1789 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1790 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1791 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1792 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1793 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1794 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1795 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1796 	 *
1797 	 * Not present in struct vmcs12:
1798 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1799 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1800 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1801 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1802 	 */
1803 
1804 	return;
1805 }
1806 
1807 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1808 {
1809 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1810 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1811 
1812 	/*
1813 	 * Should not be changed by KVM:
1814 	 *
1815 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1816 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1817 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1818 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1819 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1820 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1821 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1822 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1823 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1824 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1825 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1826 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1827 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1828 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1829 	 * evmcs->host_rip = vmcs12->host_rip;
1830 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1831 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1832 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1833 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1834 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1835 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1836 	 * evmcs->host_rsp = vmcs12->host_rsp;
1837 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1838 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1839 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1840 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1841 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1842 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1843 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1844 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1845 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1846 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1847 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1848 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1849 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1850 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1851 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1852 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1853 	 * evmcs->page_fault_error_code_mask =
1854 	 *		vmcs12->page_fault_error_code_mask;
1855 	 * evmcs->page_fault_error_code_match =
1856 	 *		vmcs12->page_fault_error_code_match;
1857 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1858 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1859 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1860 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1861 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1862 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1863 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1864 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1865 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1866 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1867 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1868 	 *
1869 	 * Not present in struct vmcs12:
1870 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1871 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1872 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1873 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1874 	 */
1875 
1876 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1877 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1878 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1879 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1880 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1881 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1882 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1883 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1884 
1885 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1886 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1887 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1888 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1889 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1890 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1891 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1892 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1893 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1894 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1895 
1896 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1897 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1898 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1899 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1900 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1901 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1902 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1903 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1904 
1905 	evmcs->guest_es_base = vmcs12->guest_es_base;
1906 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1907 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1908 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1909 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1910 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1911 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1912 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1913 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1914 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1915 
1916 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1917 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1918 
1919 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1920 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1921 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1922 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1923 
1924 	evmcs->guest_pending_dbg_exceptions =
1925 		vmcs12->guest_pending_dbg_exceptions;
1926 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1927 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1928 
1929 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1930 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1931 
1932 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1933 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1934 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1935 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1936 
1937 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1938 
1939 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1940 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1941 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1942 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1943 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1944 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1945 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1946 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1947 
1948 	evmcs->exit_qualification = vmcs12->exit_qualification;
1949 
1950 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1951 	evmcs->guest_rsp = vmcs12->guest_rsp;
1952 	evmcs->guest_rflags = vmcs12->guest_rflags;
1953 
1954 	evmcs->guest_interruptibility_info =
1955 		vmcs12->guest_interruptibility_info;
1956 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1957 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1958 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1959 	evmcs->vm_entry_exception_error_code =
1960 		vmcs12->vm_entry_exception_error_code;
1961 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1962 
1963 	evmcs->guest_rip = vmcs12->guest_rip;
1964 
1965 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1966 
1967 	return;
1968 }
1969 
1970 /*
1971  * This is an equivalent of the nested hypervisor executing the vmptrld
1972  * instruction.
1973  */
1974 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1975 	struct kvm_vcpu *vcpu, bool from_launch)
1976 {
1977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1978 	bool evmcs_gpa_changed = false;
1979 	u64 evmcs_gpa;
1980 
1981 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1982 		return EVMPTRLD_DISABLED;
1983 
1984 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa)) {
1985 		nested_release_evmcs(vcpu);
1986 		return EVMPTRLD_DISABLED;
1987 	}
1988 
1989 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1990 		vmx->nested.current_vmptr = INVALID_GPA;
1991 
1992 		nested_release_evmcs(vcpu);
1993 
1994 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1995 				 &vmx->nested.hv_evmcs_map))
1996 			return EVMPTRLD_ERROR;
1997 
1998 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1999 
2000 		/*
2001 		 * Currently, KVM only supports eVMCS version 1
2002 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2003 		 * value to first u32 field of eVMCS which should specify eVMCS
2004 		 * VersionNumber.
2005 		 *
2006 		 * Guest should be aware of supported eVMCS versions by host by
2007 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2008 		 * expected to set this CPUID leaf according to the value
2009 		 * returned in vmcs_version from nested_enable_evmcs().
2010 		 *
2011 		 * However, it turns out that Microsoft Hyper-V fails to comply
2012 		 * to their own invented interface: When Hyper-V use eVMCS, it
2013 		 * just sets first u32 field of eVMCS to revision_id specified
2014 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2015 		 * which is one of the supported versions specified in
2016 		 * CPUID.0x4000000A.EAX[0:15].
2017 		 *
2018 		 * To overcome Hyper-V bug, we accept here either a supported
2019 		 * eVMCS version or VMCS12 revision_id as valid values for first
2020 		 * u32 field of eVMCS.
2021 		 */
2022 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2023 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2024 			nested_release_evmcs(vcpu);
2025 			return EVMPTRLD_VMFAIL;
2026 		}
2027 
2028 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2029 
2030 		evmcs_gpa_changed = true;
2031 		/*
2032 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2033 		 * reloaded from guest's memory (read only fields, fields not
2034 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2035 		 * are no leftovers.
2036 		 */
2037 		if (from_launch) {
2038 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2039 			memset(vmcs12, 0, sizeof(*vmcs12));
2040 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2041 		}
2042 
2043 	}
2044 
2045 	/*
2046 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2047 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2048 	 */
2049 	if (from_launch || evmcs_gpa_changed) {
2050 		vmx->nested.hv_evmcs->hv_clean_fields &=
2051 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2052 
2053 		vmx->nested.force_msr_bitmap_recalc = true;
2054 	}
2055 
2056 	return EVMPTRLD_SUCCEEDED;
2057 }
2058 
2059 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2060 {
2061 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2062 
2063 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2064 		copy_vmcs12_to_enlightened(vmx);
2065 	else
2066 		copy_vmcs12_to_shadow(vmx);
2067 
2068 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2069 }
2070 
2071 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2072 {
2073 	struct vcpu_vmx *vmx =
2074 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2075 
2076 	vmx->nested.preemption_timer_expired = true;
2077 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2078 	kvm_vcpu_kick(&vmx->vcpu);
2079 
2080 	return HRTIMER_NORESTART;
2081 }
2082 
2083 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2084 {
2085 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2086 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2087 
2088 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2089 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2090 
2091 	if (!vmx->nested.has_preemption_timer_deadline) {
2092 		vmx->nested.preemption_timer_deadline =
2093 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2094 		vmx->nested.has_preemption_timer_deadline = true;
2095 	}
2096 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2097 }
2098 
2099 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2100 					u64 preemption_timeout)
2101 {
2102 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2103 
2104 	/*
2105 	 * A timer value of zero is architecturally guaranteed to cause
2106 	 * a VMExit prior to executing any instructions in the guest.
2107 	 */
2108 	if (preemption_timeout == 0) {
2109 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2110 		return;
2111 	}
2112 
2113 	if (vcpu->arch.virtual_tsc_khz == 0)
2114 		return;
2115 
2116 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2117 	preemption_timeout *= 1000000;
2118 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2119 	hrtimer_start(&vmx->nested.preemption_timer,
2120 		      ktime_add_ns(ktime_get(), preemption_timeout),
2121 		      HRTIMER_MODE_ABS_PINNED);
2122 }
2123 
2124 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2125 {
2126 	if (vmx->nested.nested_run_pending &&
2127 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2128 		return vmcs12->guest_ia32_efer;
2129 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2130 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2131 	else
2132 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2133 }
2134 
2135 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2136 {
2137 	/*
2138 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2139 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2140 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2141 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2142 	 */
2143 	if (vmx->nested.vmcs02_initialized)
2144 		return;
2145 	vmx->nested.vmcs02_initialized = true;
2146 
2147 	/*
2148 	 * We don't care what the EPTP value is we just need to guarantee
2149 	 * it's valid so we don't get a false positive when doing early
2150 	 * consistency checks.
2151 	 */
2152 	if (enable_ept && nested_early_check)
2153 		vmcs_write64(EPT_POINTER,
2154 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2155 
2156 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2157 	if (cpu_has_vmx_vmfunc())
2158 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2159 
2160 	if (cpu_has_vmx_posted_intr())
2161 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2162 
2163 	if (cpu_has_vmx_msr_bitmap())
2164 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2165 
2166 	/*
2167 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2168 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2169 	 * deal with filtering out L2 GPAs from the buffer.
2170 	 */
2171 	if (enable_pml) {
2172 		vmcs_write64(PML_ADDRESS, 0);
2173 		vmcs_write16(GUEST_PML_INDEX, -1);
2174 	}
2175 
2176 	if (cpu_has_vmx_encls_vmexit())
2177 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2178 
2179 	/*
2180 	 * Set the MSR load/store lists to match L0's settings.  Only the
2181 	 * addresses are constant (for vmcs02), the counts can change based
2182 	 * on L2's behavior, e.g. switching to/from long mode.
2183 	 */
2184 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2185 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2186 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2187 
2188 	vmx_set_constant_host_state(vmx);
2189 }
2190 
2191 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2192 				      struct vmcs12 *vmcs12)
2193 {
2194 	prepare_vmcs02_constant_state(vmx);
2195 
2196 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2197 
2198 	if (enable_vpid) {
2199 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2200 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2201 		else
2202 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2203 	}
2204 }
2205 
2206 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2207 				 struct vmcs12 *vmcs12)
2208 {
2209 	u32 exec_control;
2210 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2211 
2212 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2213 		prepare_vmcs02_early_rare(vmx, vmcs12);
2214 
2215 	/*
2216 	 * PIN CONTROLS
2217 	 */
2218 	exec_control = __pin_controls_get(vmcs01);
2219 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2220 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2221 
2222 	/* Posted interrupts setting is only taken from vmcs12.  */
2223 	vmx->nested.pi_pending = false;
2224 	if (nested_cpu_has_posted_intr(vmcs12))
2225 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2226 	else
2227 		exec_control &= ~PIN_BASED_POSTED_INTR;
2228 	pin_controls_set(vmx, exec_control);
2229 
2230 	/*
2231 	 * EXEC CONTROLS
2232 	 */
2233 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2234 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2235 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2236 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2237 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2238 
2239 	vmx->nested.l1_tpr_threshold = -1;
2240 	if (exec_control & CPU_BASED_TPR_SHADOW)
2241 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2242 #ifdef CONFIG_X86_64
2243 	else
2244 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2245 				CPU_BASED_CR8_STORE_EXITING;
2246 #endif
2247 
2248 	/*
2249 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2250 	 * for I/O port accesses.
2251 	 */
2252 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2253 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2254 
2255 	/*
2256 	 * This bit will be computed in nested_get_vmcs12_pages, because
2257 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2258 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2259 	 * only set/clear this bit.
2260 	 */
2261 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2262 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2263 
2264 	exec_controls_set(vmx, exec_control);
2265 
2266 	/*
2267 	 * SECONDARY EXEC CONTROLS
2268 	 */
2269 	if (cpu_has_secondary_exec_ctrls()) {
2270 		exec_control = __secondary_exec_controls_get(vmcs01);
2271 
2272 		/* Take the following fields only from vmcs12 */
2273 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2274 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2275 				  SECONDARY_EXEC_ENABLE_INVPCID |
2276 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2277 				  SECONDARY_EXEC_XSAVES |
2278 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2279 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2280 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2281 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2282 				  SECONDARY_EXEC_TSC_SCALING |
2283 				  SECONDARY_EXEC_DESC);
2284 
2285 		if (nested_cpu_has(vmcs12,
2286 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2287 			exec_control |= vmcs12->secondary_vm_exec_control;
2288 
2289 		/* PML is emulated and never enabled in hardware for L2. */
2290 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2291 
2292 		/* VMCS shadowing for L2 is emulated for now */
2293 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2294 
2295 		/*
2296 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2297 		 * will not have to rewrite the controls just for this bit.
2298 		 */
2299 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2300 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2301 			exec_control |= SECONDARY_EXEC_DESC;
2302 
2303 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2304 			vmcs_write16(GUEST_INTR_STATUS,
2305 				vmcs12->guest_intr_status);
2306 
2307 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2308 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2309 
2310 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2311 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2312 
2313 		secondary_exec_controls_set(vmx, exec_control);
2314 	}
2315 
2316 	/*
2317 	 * ENTRY CONTROLS
2318 	 *
2319 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2320 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2321 	 * on the related bits (if supported by the CPU) in the hope that
2322 	 * we can avoid VMWrites during vmx_set_efer().
2323 	 */
2324 	exec_control = __vm_entry_controls_get(vmcs01);
2325 	exec_control |= vmcs12->vm_entry_controls;
2326 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2327 	if (cpu_has_load_ia32_efer()) {
2328 		if (guest_efer & EFER_LMA)
2329 			exec_control |= VM_ENTRY_IA32E_MODE;
2330 		if (guest_efer != host_efer)
2331 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2332 	}
2333 	vm_entry_controls_set(vmx, exec_control);
2334 
2335 	/*
2336 	 * EXIT CONTROLS
2337 	 *
2338 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2339 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2340 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2341 	 */
2342 	exec_control = __vm_exit_controls_get(vmcs01);
2343 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2344 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2345 	else
2346 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2347 	vm_exit_controls_set(vmx, exec_control);
2348 
2349 	/*
2350 	 * Interrupt/Exception Fields
2351 	 */
2352 	if (vmx->nested.nested_run_pending) {
2353 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2354 			     vmcs12->vm_entry_intr_info_field);
2355 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2356 			     vmcs12->vm_entry_exception_error_code);
2357 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2358 			     vmcs12->vm_entry_instruction_len);
2359 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2360 			     vmcs12->guest_interruptibility_info);
2361 		vmx->loaded_vmcs->nmi_known_unmasked =
2362 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2363 	} else {
2364 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2365 	}
2366 }
2367 
2368 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2369 {
2370 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2371 
2372 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2373 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2374 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2375 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2376 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2377 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2378 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2379 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2380 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2381 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2382 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2383 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2384 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2385 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2386 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2387 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2388 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2389 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2390 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2391 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2392 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2393 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2394 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2395 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2396 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2397 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2398 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2399 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2400 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2401 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2402 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2403 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2404 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2405 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2406 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2407 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2408 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2409 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2410 
2411 		vmx->segment_cache.bitmask = 0;
2412 	}
2413 
2414 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2415 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2416 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2417 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2418 			    vmcs12->guest_pending_dbg_exceptions);
2419 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2420 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2421 
2422 		/*
2423 		 * L1 may access the L2's PDPTR, so save them to construct
2424 		 * vmcs12
2425 		 */
2426 		if (enable_ept) {
2427 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2428 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2429 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2430 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2431 		}
2432 
2433 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2434 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2435 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2436 	}
2437 
2438 	if (nested_cpu_has_xsaves(vmcs12))
2439 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2440 
2441 	/*
2442 	 * Whether page-faults are trapped is determined by a combination of
2443 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2444 	 * doesn't care about page faults then we should set all of these to
2445 	 * L1's desires. However, if L0 does care about (some) page faults, it
2446 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2447 	 * simply ask to exit on each and every L2 page fault. This is done by
2448 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2449 	 * Note that below we don't need special code to set EB.PF beyond the
2450 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2451 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2452 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2453 	 */
2454 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2455 		/*
2456 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2457 		 * go ahead and use it?
2458 		 */
2459 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2460 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2461 	} else {
2462 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2463 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2464 	}
2465 
2466 	if (cpu_has_vmx_apicv()) {
2467 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2468 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2469 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2470 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2471 	}
2472 
2473 	/*
2474 	 * Make sure the msr_autostore list is up to date before we set the
2475 	 * count in the vmcs02.
2476 	 */
2477 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2478 
2479 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2480 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2481 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2482 
2483 	set_cr4_guest_host_mask(vmx);
2484 }
2485 
2486 /*
2487  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2488  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2489  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2490  * guest in a way that will both be appropriate to L1's requests, and our
2491  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2492  * function also has additional necessary side-effects, like setting various
2493  * vcpu->arch fields.
2494  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2495  * is assigned to entry_failure_code on failure.
2496  */
2497 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2498 			  bool from_vmentry,
2499 			  enum vm_entry_failure_code *entry_failure_code)
2500 {
2501 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2502 	bool load_guest_pdptrs_vmcs12 = false;
2503 
2504 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2505 		prepare_vmcs02_rare(vmx, vmcs12);
2506 		vmx->nested.dirty_vmcs12 = false;
2507 
2508 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2509 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2510 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2511 	}
2512 
2513 	if (vmx->nested.nested_run_pending &&
2514 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2515 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2516 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2517 	} else {
2518 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2519 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2520 	}
2521 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2522 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2523 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2524 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2525 
2526 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2527 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2528 	 * trap. Note that CR0.TS also needs updating - we do this later.
2529 	 */
2530 	vmx_update_exception_bitmap(vcpu);
2531 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2532 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2533 
2534 	if (vmx->nested.nested_run_pending &&
2535 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2536 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2537 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2538 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2539 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2540 	}
2541 
2542 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2543 			vcpu->arch.l1_tsc_offset,
2544 			vmx_get_l2_tsc_offset(vcpu),
2545 			vmx_get_l2_tsc_multiplier(vcpu));
2546 
2547 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2548 			vcpu->arch.l1_tsc_scaling_ratio,
2549 			vmx_get_l2_tsc_multiplier(vcpu));
2550 
2551 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2552 	if (kvm_has_tsc_control)
2553 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2554 
2555 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2556 
2557 	if (nested_cpu_has_ept(vmcs12))
2558 		nested_ept_init_mmu_context(vcpu);
2559 
2560 	/*
2561 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2562 	 * bits which we consider mandatory enabled.
2563 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2564 	 * the specifications by L1; It's not enough to take
2565 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2566 	 * have more bits than L1 expected.
2567 	 */
2568 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2569 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2570 
2571 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2572 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2573 
2574 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2575 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2576 	vmx_set_efer(vcpu, vcpu->arch.efer);
2577 
2578 	/*
2579 	 * Guest state is invalid and unrestricted guest is disabled,
2580 	 * which means L1 attempted VMEntry to L2 with invalid state.
2581 	 * Fail the VMEntry.
2582 	 *
2583 	 * However when force loading the guest state (SMM exit or
2584 	 * loading nested state after migration, it is possible to
2585 	 * have invalid guest state now, which will be later fixed by
2586 	 * restoring L2 register state
2587 	 */
2588 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2589 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2590 		return -EINVAL;
2591 	}
2592 
2593 	/* Shadow page tables on either EPT or shadow page tables. */
2594 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2595 				from_vmentry, entry_failure_code))
2596 		return -EINVAL;
2597 
2598 	/*
2599 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2600 	 * on nested VM-Exit, which can occur without actually running L2 and
2601 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2602 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2603 	 * transition to HLT instead of running L2.
2604 	 */
2605 	if (enable_ept)
2606 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2607 
2608 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2609 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2610 	    is_pae_paging(vcpu)) {
2611 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2612 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2613 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2614 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2615 	}
2616 
2617 	if (!enable_ept)
2618 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2619 
2620 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2621 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2622 				     vmcs12->guest_ia32_perf_global_ctrl))) {
2623 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2624 		return -EINVAL;
2625 	}
2626 
2627 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2628 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2629 
2630 	/*
2631 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2632 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2633 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2634 	 * here.
2635 	 */
2636 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2637 		vmx->nested.hv_evmcs->hv_clean_fields |=
2638 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2639 
2640 	return 0;
2641 }
2642 
2643 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2644 {
2645 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2646 	       nested_cpu_has_virtual_nmis(vmcs12)))
2647 		return -EINVAL;
2648 
2649 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2650 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2651 		return -EINVAL;
2652 
2653 	return 0;
2654 }
2655 
2656 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2657 {
2658 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2659 
2660 	/* Check for memory type validity */
2661 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2662 	case VMX_EPTP_MT_UC:
2663 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2664 			return false;
2665 		break;
2666 	case VMX_EPTP_MT_WB:
2667 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2668 			return false;
2669 		break;
2670 	default:
2671 		return false;
2672 	}
2673 
2674 	/* Page-walk levels validity. */
2675 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2676 	case VMX_EPTP_PWL_5:
2677 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2678 			return false;
2679 		break;
2680 	case VMX_EPTP_PWL_4:
2681 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2682 			return false;
2683 		break;
2684 	default:
2685 		return false;
2686 	}
2687 
2688 	/* Reserved bits should not be set */
2689 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2690 		return false;
2691 
2692 	/* AD, if set, should be supported */
2693 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2694 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2695 			return false;
2696 	}
2697 
2698 	return true;
2699 }
2700 
2701 /*
2702  * Checks related to VM-Execution Control Fields
2703  */
2704 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2705                                               struct vmcs12 *vmcs12)
2706 {
2707 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2708 
2709 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2710 				   vmx->nested.msrs.pinbased_ctls_low,
2711 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2712 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2713 				   vmx->nested.msrs.procbased_ctls_low,
2714 				   vmx->nested.msrs.procbased_ctls_high)))
2715 		return -EINVAL;
2716 
2717 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2718 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2719 				   vmx->nested.msrs.secondary_ctls_low,
2720 				   vmx->nested.msrs.secondary_ctls_high)))
2721 		return -EINVAL;
2722 
2723 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2724 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2725 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2726 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2727 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2728 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2729 	    nested_vmx_check_nmi_controls(vmcs12) ||
2730 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2731 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2732 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2733 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2734 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2735 		return -EINVAL;
2736 
2737 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2738 	    nested_cpu_has_save_preemption_timer(vmcs12))
2739 		return -EINVAL;
2740 
2741 	if (nested_cpu_has_ept(vmcs12) &&
2742 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2743 		return -EINVAL;
2744 
2745 	if (nested_cpu_has_vmfunc(vmcs12)) {
2746 		if (CC(vmcs12->vm_function_control &
2747 		       ~vmx->nested.msrs.vmfunc_controls))
2748 			return -EINVAL;
2749 
2750 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2751 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2752 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2753 				return -EINVAL;
2754 		}
2755 	}
2756 
2757 	return 0;
2758 }
2759 
2760 /*
2761  * Checks related to VM-Exit Control Fields
2762  */
2763 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2764                                          struct vmcs12 *vmcs12)
2765 {
2766 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2767 
2768 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2769 				    vmx->nested.msrs.exit_ctls_low,
2770 				    vmx->nested.msrs.exit_ctls_high)) ||
2771 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2772 		return -EINVAL;
2773 
2774 	return 0;
2775 }
2776 
2777 /*
2778  * Checks related to VM-Entry Control Fields
2779  */
2780 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2781 					  struct vmcs12 *vmcs12)
2782 {
2783 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2784 
2785 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2786 				    vmx->nested.msrs.entry_ctls_low,
2787 				    vmx->nested.msrs.entry_ctls_high)))
2788 		return -EINVAL;
2789 
2790 	/*
2791 	 * From the Intel SDM, volume 3:
2792 	 * Fields relevant to VM-entry event injection must be set properly.
2793 	 * These fields are the VM-entry interruption-information field, the
2794 	 * VM-entry exception error code, and the VM-entry instruction length.
2795 	 */
2796 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2797 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2798 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2799 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2800 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2801 		bool should_have_error_code;
2802 		bool urg = nested_cpu_has2(vmcs12,
2803 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2804 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2805 
2806 		/* VM-entry interruption-info field: interruption type */
2807 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2808 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2809 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2810 			return -EINVAL;
2811 
2812 		/* VM-entry interruption-info field: vector */
2813 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2814 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2815 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2816 			return -EINVAL;
2817 
2818 		/* VM-entry interruption-info field: deliver error code */
2819 		should_have_error_code =
2820 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2821 			x86_exception_has_error_code(vector);
2822 		if (CC(has_error_code != should_have_error_code))
2823 			return -EINVAL;
2824 
2825 		/* VM-entry exception error code */
2826 		if (CC(has_error_code &&
2827 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2828 			return -EINVAL;
2829 
2830 		/* VM-entry interruption-info field: reserved bits */
2831 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2832 			return -EINVAL;
2833 
2834 		/* VM-entry instruction length */
2835 		switch (intr_type) {
2836 		case INTR_TYPE_SOFT_EXCEPTION:
2837 		case INTR_TYPE_SOFT_INTR:
2838 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2839 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2840 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2841 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2842 				return -EINVAL;
2843 		}
2844 	}
2845 
2846 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2847 		return -EINVAL;
2848 
2849 	return 0;
2850 }
2851 
2852 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2853 				     struct vmcs12 *vmcs12)
2854 {
2855 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2856 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2857 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2858 		return -EINVAL;
2859 
2860 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2861 		return nested_evmcs_check_controls(vmcs12);
2862 
2863 	return 0;
2864 }
2865 
2866 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
2867 				       struct vmcs12 *vmcs12)
2868 {
2869 #ifdef CONFIG_X86_64
2870 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
2871 		!!(vcpu->arch.efer & EFER_LMA)))
2872 		return -EINVAL;
2873 #endif
2874 	return 0;
2875 }
2876 
2877 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2878 				       struct vmcs12 *vmcs12)
2879 {
2880 	bool ia32e;
2881 
2882 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2883 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2884 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2885 		return -EINVAL;
2886 
2887 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2888 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2889 		return -EINVAL;
2890 
2891 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2892 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2893 		return -EINVAL;
2894 
2895 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2896 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2897 					   vmcs12->host_ia32_perf_global_ctrl)))
2898 		return -EINVAL;
2899 
2900 #ifdef CONFIG_X86_64
2901 	ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
2902 #else
2903 	ia32e = false;
2904 #endif
2905 
2906 	if (ia32e) {
2907 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2908 			return -EINVAL;
2909 	} else {
2910 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2911 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2912 		    CC((vmcs12->host_rip) >> 32))
2913 			return -EINVAL;
2914 	}
2915 
2916 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2917 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2918 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2919 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2920 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2921 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2922 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2923 	    CC(vmcs12->host_cs_selector == 0) ||
2924 	    CC(vmcs12->host_tr_selector == 0) ||
2925 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2926 		return -EINVAL;
2927 
2928 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2929 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2930 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2931 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2932 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2933 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2934 		return -EINVAL;
2935 
2936 	/*
2937 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2938 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2939 	 * the values of the LMA and LME bits in the field must each be that of
2940 	 * the host address-space size VM-exit control.
2941 	 */
2942 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2943 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2944 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2945 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2946 			return -EINVAL;
2947 	}
2948 
2949 	return 0;
2950 }
2951 
2952 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2953 					  struct vmcs12 *vmcs12)
2954 {
2955 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2956 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
2957 	struct vmcs_hdr hdr;
2958 
2959 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
2960 		return 0;
2961 
2962 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2963 		return -EINVAL;
2964 
2965 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
2966 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
2967 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
2968                 return -EINVAL;
2969 
2970 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
2971 					    offsetof(struct vmcs12, hdr),
2972 					    sizeof(hdr))))
2973 		return -EINVAL;
2974 
2975 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
2976 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2977 		return -EINVAL;
2978 
2979 	return 0;
2980 }
2981 
2982 /*
2983  * Checks related to Guest Non-register State
2984  */
2985 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2986 {
2987 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2988 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2989 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2990 		return -EINVAL;
2991 
2992 	return 0;
2993 }
2994 
2995 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2996 					struct vmcs12 *vmcs12,
2997 					enum vm_entry_failure_code *entry_failure_code)
2998 {
2999 	bool ia32e;
3000 
3001 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
3002 
3003 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3004 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3005 		return -EINVAL;
3006 
3007 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
3008 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
3009 		return -EINVAL;
3010 
3011 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3012 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3013 		return -EINVAL;
3014 
3015 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3016 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3017 		return -EINVAL;
3018 	}
3019 
3020 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3021 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3022 					   vmcs12->guest_ia32_perf_global_ctrl)))
3023 		return -EINVAL;
3024 
3025 	/*
3026 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3027 	 * are performed on the field for the IA32_EFER MSR:
3028 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3029 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3030 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3031 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3032 	 *   CR0.PG) is 1.
3033 	 */
3034 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3035 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3036 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
3037 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3038 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3039 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3040 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3041 			return -EINVAL;
3042 	}
3043 
3044 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3045 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3046 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3047 		return -EINVAL;
3048 
3049 	if (nested_check_guest_non_reg_state(vmcs12))
3050 		return -EINVAL;
3051 
3052 	return 0;
3053 }
3054 
3055 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3056 {
3057 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3058 	unsigned long cr3, cr4;
3059 	bool vm_fail;
3060 
3061 	if (!nested_early_check)
3062 		return 0;
3063 
3064 	if (vmx->msr_autoload.host.nr)
3065 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3066 	if (vmx->msr_autoload.guest.nr)
3067 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3068 
3069 	preempt_disable();
3070 
3071 	vmx_prepare_switch_to_guest(vcpu);
3072 
3073 	/*
3074 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3075 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3076 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3077 	 * there is no need to preserve other bits or save/restore the field.
3078 	 */
3079 	vmcs_writel(GUEST_RFLAGS, 0);
3080 
3081 	cr3 = __get_current_cr3_fast();
3082 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3083 		vmcs_writel(HOST_CR3, cr3);
3084 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3085 	}
3086 
3087 	cr4 = cr4_read_shadow();
3088 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3089 		vmcs_writel(HOST_CR4, cr4);
3090 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3091 	}
3092 
3093 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3094 				 vmx->loaded_vmcs->launched);
3095 
3096 	if (vmx->msr_autoload.host.nr)
3097 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3098 	if (vmx->msr_autoload.guest.nr)
3099 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3100 
3101 	if (vm_fail) {
3102 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3103 
3104 		preempt_enable();
3105 
3106 		trace_kvm_nested_vmenter_failed(
3107 			"early hardware check VM-instruction error: ", error);
3108 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3109 		return 1;
3110 	}
3111 
3112 	/*
3113 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3114 	 */
3115 	if (hw_breakpoint_active())
3116 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3117 	local_irq_enable();
3118 	preempt_enable();
3119 
3120 	/*
3121 	 * A non-failing VMEntry means we somehow entered guest mode with
3122 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3123 	 * is no telling what memory has been modified or what state has
3124 	 * been exposed to unknown code.  Hitting this all but guarantees
3125 	 * a (very critical) hardware issue.
3126 	 */
3127 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3128 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3129 
3130 	return 0;
3131 }
3132 
3133 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3134 {
3135 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3136 
3137 	/*
3138 	 * hv_evmcs may end up being not mapped after migration (when
3139 	 * L2 was running), map it here to make sure vmcs12 changes are
3140 	 * properly reflected.
3141 	 */
3142 	if (vmx->nested.enlightened_vmcs_enabled &&
3143 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3144 		enum nested_evmptrld_status evmptrld_status =
3145 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3146 
3147 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3148 		    evmptrld_status == EVMPTRLD_ERROR)
3149 			return false;
3150 
3151 		/*
3152 		 * Post migration VMCS12 always provides the most actual
3153 		 * information, copy it to eVMCS upon entry.
3154 		 */
3155 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3156 	}
3157 
3158 	return true;
3159 }
3160 
3161 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3162 {
3163 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3164 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3165 	struct kvm_host_map *map;
3166 	struct page *page;
3167 	u64 hpa;
3168 
3169 	if (!vcpu->arch.pdptrs_from_userspace &&
3170 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3171 		/*
3172 		 * Reload the guest's PDPTRs since after a migration
3173 		 * the guest CR3 might be restored prior to setting the nested
3174 		 * state which can lead to a load of wrong PDPTRs.
3175 		 */
3176 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3177 			return false;
3178 	}
3179 
3180 
3181 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3182 		/*
3183 		 * Translate L1 physical address to host physical
3184 		 * address for vmcs02. Keep the page pinned, so this
3185 		 * physical address remains valid. We keep a reference
3186 		 * to it so we can release it later.
3187 		 */
3188 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3189 			kvm_release_page_clean(vmx->nested.apic_access_page);
3190 			vmx->nested.apic_access_page = NULL;
3191 		}
3192 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3193 		if (!is_error_page(page)) {
3194 			vmx->nested.apic_access_page = page;
3195 			hpa = page_to_phys(vmx->nested.apic_access_page);
3196 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3197 		} else {
3198 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3199 					     __func__);
3200 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3201 			vcpu->run->internal.suberror =
3202 				KVM_INTERNAL_ERROR_EMULATION;
3203 			vcpu->run->internal.ndata = 0;
3204 			return false;
3205 		}
3206 	}
3207 
3208 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3209 		map = &vmx->nested.virtual_apic_map;
3210 
3211 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3212 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3213 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3214 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3215 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3216 			/*
3217 			 * The processor will never use the TPR shadow, simply
3218 			 * clear the bit from the execution control.  Such a
3219 			 * configuration is useless, but it happens in tests.
3220 			 * For any other configuration, failing the vm entry is
3221 			 * _not_ what the processor does but it's basically the
3222 			 * only possibility we have.
3223 			 */
3224 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3225 		} else {
3226 			/*
3227 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3228 			 * force VM-Entry to fail.
3229 			 */
3230 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3231 		}
3232 	}
3233 
3234 	if (nested_cpu_has_posted_intr(vmcs12)) {
3235 		map = &vmx->nested.pi_desc_map;
3236 
3237 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3238 			vmx->nested.pi_desc =
3239 				(struct pi_desc *)(((void *)map->hva) +
3240 				offset_in_page(vmcs12->posted_intr_desc_addr));
3241 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3242 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3243 		} else {
3244 			/*
3245 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3246 			 * access the contents of the VMCS12 posted interrupt
3247 			 * descriptor. (Note that KVM may do this when it
3248 			 * should not, per the architectural specification.)
3249 			 */
3250 			vmx->nested.pi_desc = NULL;
3251 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3252 		}
3253 	}
3254 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3255 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3256 	else
3257 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3258 
3259 	return true;
3260 }
3261 
3262 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3263 {
3264 	if (!nested_get_evmcs_page(vcpu)) {
3265 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3266 				     __func__);
3267 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3268 		vcpu->run->internal.suberror =
3269 			KVM_INTERNAL_ERROR_EMULATION;
3270 		vcpu->run->internal.ndata = 0;
3271 
3272 		return false;
3273 	}
3274 
3275 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3276 		return false;
3277 
3278 	return true;
3279 }
3280 
3281 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3282 {
3283 	struct vmcs12 *vmcs12;
3284 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3285 	gpa_t dst;
3286 
3287 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3288 		return 0;
3289 
3290 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3291 		return 1;
3292 
3293 	/*
3294 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3295 	 * set is already checked as part of A/D emulation.
3296 	 */
3297 	vmcs12 = get_vmcs12(vcpu);
3298 	if (!nested_cpu_has_pml(vmcs12))
3299 		return 0;
3300 
3301 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3302 		vmx->nested.pml_full = true;
3303 		return 1;
3304 	}
3305 
3306 	gpa &= ~0xFFFull;
3307 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3308 
3309 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3310 				 offset_in_page(dst), sizeof(gpa)))
3311 		return 0;
3312 
3313 	vmcs12->guest_pml_index--;
3314 
3315 	return 0;
3316 }
3317 
3318 /*
3319  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3320  * for running VMX instructions (except VMXON, whose prerequisites are
3321  * slightly different). It also specifies what exception to inject otherwise.
3322  * Note that many of these exceptions have priority over VM exits, so they
3323  * don't have to be checked again here.
3324  */
3325 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3326 {
3327 	if (!to_vmx(vcpu)->nested.vmxon) {
3328 		kvm_queue_exception(vcpu, UD_VECTOR);
3329 		return 0;
3330 	}
3331 
3332 	if (vmx_get_cpl(vcpu)) {
3333 		kvm_inject_gp(vcpu, 0);
3334 		return 0;
3335 	}
3336 
3337 	return 1;
3338 }
3339 
3340 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3341 {
3342 	u8 rvi = vmx_get_rvi();
3343 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3344 
3345 	return ((rvi & 0xf0) > (vppr & 0xf0));
3346 }
3347 
3348 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3349 				   struct vmcs12 *vmcs12);
3350 
3351 /*
3352  * If from_vmentry is false, this is being called from state restore (either RSM
3353  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3354  *
3355  * Returns:
3356  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3357  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3358  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3359  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3360  */
3361 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3362 							bool from_vmentry)
3363 {
3364 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3365 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3366 	enum vm_entry_failure_code entry_failure_code;
3367 	bool evaluate_pending_interrupts;
3368 	union vmx_exit_reason exit_reason = {
3369 		.basic = EXIT_REASON_INVALID_STATE,
3370 		.failed_vmentry = 1,
3371 	};
3372 	u32 failed_index;
3373 
3374 	kvm_service_local_tlb_flush_requests(vcpu);
3375 
3376 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3377 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3378 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3379 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3380 
3381 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3382 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3383 	if (kvm_mpx_supported() &&
3384 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3385 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3386 
3387 	/*
3388 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3389 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3390 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3391 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3392 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3393 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3394 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3395 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3396 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3397 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3398 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3399 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3400 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3401 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3402 	 */
3403 	if (!enable_ept && !nested_early_check)
3404 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3405 
3406 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3407 
3408 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3409 
3410 	if (from_vmentry) {
3411 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3412 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3413 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3414 		}
3415 
3416 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3417 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3418 			return NVMX_VMENTRY_VMFAIL;
3419 		}
3420 
3421 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3422 						 &entry_failure_code)) {
3423 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3424 			vmcs12->exit_qualification = entry_failure_code;
3425 			goto vmentry_fail_vmexit;
3426 		}
3427 	}
3428 
3429 	enter_guest_mode(vcpu);
3430 
3431 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3432 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3433 		vmcs12->exit_qualification = entry_failure_code;
3434 		goto vmentry_fail_vmexit_guest_mode;
3435 	}
3436 
3437 	if (from_vmentry) {
3438 		failed_index = nested_vmx_load_msr(vcpu,
3439 						   vmcs12->vm_entry_msr_load_addr,
3440 						   vmcs12->vm_entry_msr_load_count);
3441 		if (failed_index) {
3442 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3443 			vmcs12->exit_qualification = failed_index;
3444 			goto vmentry_fail_vmexit_guest_mode;
3445 		}
3446 	} else {
3447 		/*
3448 		 * The MMU is not initialized to point at the right entities yet and
3449 		 * "get pages" would need to read data from the guest (i.e. we will
3450 		 * need to perform gpa to hpa translation). Request a call
3451 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3452 		 * have already been set at vmentry time and should not be reset.
3453 		 */
3454 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3455 	}
3456 
3457 	/*
3458 	 * If L1 had a pending IRQ/NMI until it executed
3459 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3460 	 * disallowed (e.g. interrupts disabled), L0 needs to
3461 	 * evaluate if this pending event should cause an exit from L2
3462 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3463 	 * intercept EXTERNAL_INTERRUPT).
3464 	 *
3465 	 * Usually this would be handled by the processor noticing an
3466 	 * IRQ/NMI window request, or checking RVI during evaluation of
3467 	 * pending virtual interrupts.  However, this setting was done
3468 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3469 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3470 	 */
3471 	if (unlikely(evaluate_pending_interrupts))
3472 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3473 
3474 	/*
3475 	 * Do not start the preemption timer hrtimer until after we know
3476 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3477 	 * the timer.
3478 	 */
3479 	vmx->nested.preemption_timer_expired = false;
3480 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3481 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3482 		vmx_start_preemption_timer(vcpu, timer_value);
3483 	}
3484 
3485 	/*
3486 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3487 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3488 	 * returned as far as L1 is concerned. It will only return (and set
3489 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3490 	 */
3491 	return NVMX_VMENTRY_SUCCESS;
3492 
3493 	/*
3494 	 * A failed consistency check that leads to a VMExit during L1's
3495 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3496 	 * 26.7 "VM-entry failures during or after loading guest state".
3497 	 */
3498 vmentry_fail_vmexit_guest_mode:
3499 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3500 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3501 	leave_guest_mode(vcpu);
3502 
3503 vmentry_fail_vmexit:
3504 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3505 
3506 	if (!from_vmentry)
3507 		return NVMX_VMENTRY_VMEXIT;
3508 
3509 	load_vmcs12_host_state(vcpu, vmcs12);
3510 	vmcs12->vm_exit_reason = exit_reason.full;
3511 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3512 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3513 	return NVMX_VMENTRY_VMEXIT;
3514 }
3515 
3516 /*
3517  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3518  * for running an L2 nested guest.
3519  */
3520 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3521 {
3522 	struct vmcs12 *vmcs12;
3523 	enum nvmx_vmentry_status status;
3524 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3525 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3526 	enum nested_evmptrld_status evmptrld_status;
3527 
3528 	if (!nested_vmx_check_permission(vcpu))
3529 		return 1;
3530 
3531 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3532 	if (evmptrld_status == EVMPTRLD_ERROR) {
3533 		kvm_queue_exception(vcpu, UD_VECTOR);
3534 		return 1;
3535 	}
3536 
3537 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
3538 
3539 	if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3540 		return nested_vmx_failInvalid(vcpu);
3541 
3542 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3543 	       vmx->nested.current_vmptr == INVALID_GPA))
3544 		return nested_vmx_failInvalid(vcpu);
3545 
3546 	vmcs12 = get_vmcs12(vcpu);
3547 
3548 	/*
3549 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3550 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3551 	 * rather than RFLAGS.ZF, and no error number is stored to the
3552 	 * VM-instruction error field.
3553 	 */
3554 	if (CC(vmcs12->hdr.shadow_vmcs))
3555 		return nested_vmx_failInvalid(vcpu);
3556 
3557 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3558 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3559 		/* Enlightened VMCS doesn't have launch state */
3560 		vmcs12->launch_state = !launch;
3561 	} else if (enable_shadow_vmcs) {
3562 		copy_shadow_to_vmcs12(vmx);
3563 	}
3564 
3565 	/*
3566 	 * The nested entry process starts with enforcing various prerequisites
3567 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3568 	 * they fail: As the SDM explains, some conditions should cause the
3569 	 * instruction to fail, while others will cause the instruction to seem
3570 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3571 	 * To speed up the normal (success) code path, we should avoid checking
3572 	 * for misconfigurations which will anyway be caught by the processor
3573 	 * when using the merged vmcs02.
3574 	 */
3575 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3576 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3577 
3578 	if (CC(vmcs12->launch_state == launch))
3579 		return nested_vmx_fail(vcpu,
3580 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3581 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3582 
3583 	if (nested_vmx_check_controls(vcpu, vmcs12))
3584 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3585 
3586 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3587 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3588 
3589 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3590 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3591 
3592 	/*
3593 	 * We're finally done with prerequisite checking, and can start with
3594 	 * the nested entry.
3595 	 */
3596 	vmx->nested.nested_run_pending = 1;
3597 	vmx->nested.has_preemption_timer_deadline = false;
3598 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3599 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3600 		goto vmentry_failed;
3601 
3602 	/* Emulate processing of posted interrupts on VM-Enter. */
3603 	if (nested_cpu_has_posted_intr(vmcs12) &&
3604 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3605 		vmx->nested.pi_pending = true;
3606 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3607 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3608 	}
3609 
3610 	/* Hide L1D cache contents from the nested guest.  */
3611 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3612 
3613 	/*
3614 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3615 	 * also be used as part of restoring nVMX state for
3616 	 * snapshot restore (migration).
3617 	 *
3618 	 * In this flow, it is assumed that vmcs12 cache was
3619 	 * transferred as part of captured nVMX state and should
3620 	 * therefore not be read from guest memory (which may not
3621 	 * exist on destination host yet).
3622 	 */
3623 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3624 
3625 	switch (vmcs12->guest_activity_state) {
3626 	case GUEST_ACTIVITY_HLT:
3627 		/*
3628 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3629 		 * awakened by event injection or by an NMI-window VM-exit or
3630 		 * by an interrupt-window VM-exit, halt the vcpu.
3631 		 */
3632 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3633 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3634 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3635 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3636 			vmx->nested.nested_run_pending = 0;
3637 			return kvm_emulate_halt_noskip(vcpu);
3638 		}
3639 		break;
3640 	case GUEST_ACTIVITY_WAIT_SIPI:
3641 		vmx->nested.nested_run_pending = 0;
3642 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3643 		break;
3644 	default:
3645 		break;
3646 	}
3647 
3648 	return 1;
3649 
3650 vmentry_failed:
3651 	vmx->nested.nested_run_pending = 0;
3652 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3653 		return 0;
3654 	if (status == NVMX_VMENTRY_VMEXIT)
3655 		return 1;
3656 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3657 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3658 }
3659 
3660 /*
3661  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3662  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3663  * This function returns the new value we should put in vmcs12.guest_cr0.
3664  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3665  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3666  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3667  *     didn't trap the bit, because if L1 did, so would L0).
3668  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3669  *     been modified by L2, and L1 knows it. So just leave the old value of
3670  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3671  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3672  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3673  *     changed these bits, and therefore they need to be updated, but L0
3674  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3675  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3676  */
3677 static inline unsigned long
3678 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3679 {
3680 	return
3681 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3682 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3683 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3684 			vcpu->arch.cr0_guest_owned_bits));
3685 }
3686 
3687 static inline unsigned long
3688 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3689 {
3690 	return
3691 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3692 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3693 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3694 			vcpu->arch.cr4_guest_owned_bits));
3695 }
3696 
3697 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3698 				      struct vmcs12 *vmcs12)
3699 {
3700 	u32 idt_vectoring;
3701 	unsigned int nr;
3702 
3703 	if (vcpu->arch.exception.injected) {
3704 		nr = vcpu->arch.exception.nr;
3705 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3706 
3707 		if (kvm_exception_is_soft(nr)) {
3708 			vmcs12->vm_exit_instruction_len =
3709 				vcpu->arch.event_exit_inst_len;
3710 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3711 		} else
3712 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3713 
3714 		if (vcpu->arch.exception.has_error_code) {
3715 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3716 			vmcs12->idt_vectoring_error_code =
3717 				vcpu->arch.exception.error_code;
3718 		}
3719 
3720 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3721 	} else if (vcpu->arch.nmi_injected) {
3722 		vmcs12->idt_vectoring_info_field =
3723 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3724 	} else if (vcpu->arch.interrupt.injected) {
3725 		nr = vcpu->arch.interrupt.nr;
3726 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3727 
3728 		if (vcpu->arch.interrupt.soft) {
3729 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3730 			vmcs12->vm_entry_instruction_len =
3731 				vcpu->arch.event_exit_inst_len;
3732 		} else
3733 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3734 
3735 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3736 	}
3737 }
3738 
3739 
3740 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3741 {
3742 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3743 	gfn_t gfn;
3744 
3745 	/*
3746 	 * Don't need to mark the APIC access page dirty; it is never
3747 	 * written to by the CPU during APIC virtualization.
3748 	 */
3749 
3750 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3751 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3752 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3753 	}
3754 
3755 	if (nested_cpu_has_posted_intr(vmcs12)) {
3756 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3757 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3758 	}
3759 }
3760 
3761 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3762 {
3763 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3764 	int max_irr;
3765 	void *vapic_page;
3766 	u16 status;
3767 
3768 	if (!vmx->nested.pi_pending)
3769 		return 0;
3770 
3771 	if (!vmx->nested.pi_desc)
3772 		goto mmio_needed;
3773 
3774 	vmx->nested.pi_pending = false;
3775 
3776 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3777 		return 0;
3778 
3779 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3780 	if (max_irr != 256) {
3781 		vapic_page = vmx->nested.virtual_apic_map.hva;
3782 		if (!vapic_page)
3783 			goto mmio_needed;
3784 
3785 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3786 			vapic_page, &max_irr);
3787 		status = vmcs_read16(GUEST_INTR_STATUS);
3788 		if ((u8)max_irr > ((u8)status & 0xff)) {
3789 			status &= ~0xff;
3790 			status |= (u8)max_irr;
3791 			vmcs_write16(GUEST_INTR_STATUS, status);
3792 		}
3793 	}
3794 
3795 	nested_mark_vmcs12_pages_dirty(vcpu);
3796 	return 0;
3797 
3798 mmio_needed:
3799 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3800 	return -ENXIO;
3801 }
3802 
3803 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3804 					       unsigned long exit_qual)
3805 {
3806 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3807 	unsigned int nr = vcpu->arch.exception.nr;
3808 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3809 
3810 	if (vcpu->arch.exception.has_error_code) {
3811 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3812 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3813 	}
3814 
3815 	if (kvm_exception_is_soft(nr))
3816 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3817 	else
3818 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3819 
3820 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3821 	    vmx_get_nmi_mask(vcpu))
3822 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3823 
3824 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3825 }
3826 
3827 /*
3828  * Returns true if a debug trap is pending delivery.
3829  *
3830  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3831  * exception may be inferred from the presence of an exception payload.
3832  */
3833 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3834 {
3835 	return vcpu->arch.exception.pending &&
3836 			vcpu->arch.exception.nr == DB_VECTOR &&
3837 			vcpu->arch.exception.payload;
3838 }
3839 
3840 /*
3841  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3842  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3843  * represents these debug traps with a payload that is said to be compatible
3844  * with the 'pending debug exceptions' field, write the payload to the VMCS
3845  * field if a VM-exit is delivered before the debug trap.
3846  */
3847 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3848 {
3849 	if (vmx_pending_dbg_trap(vcpu))
3850 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3851 			    vcpu->arch.exception.payload);
3852 }
3853 
3854 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3855 {
3856 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3857 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3858 }
3859 
3860 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3861 {
3862 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3863 	unsigned long exit_qual;
3864 	bool block_nested_events =
3865 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3866 	bool mtf_pending = vmx->nested.mtf_pending;
3867 	struct kvm_lapic *apic = vcpu->arch.apic;
3868 
3869 	/*
3870 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3871 	 * this state is discarded.
3872 	 */
3873 	if (!block_nested_events)
3874 		vmx->nested.mtf_pending = false;
3875 
3876 	if (lapic_in_kernel(vcpu) &&
3877 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3878 		if (block_nested_events)
3879 			return -EBUSY;
3880 		nested_vmx_update_pending_dbg(vcpu);
3881 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3882 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3883 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3884 		return 0;
3885 	}
3886 
3887 	if (lapic_in_kernel(vcpu) &&
3888 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3889 		if (block_nested_events)
3890 			return -EBUSY;
3891 
3892 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3893 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3894 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3895 						apic->sipi_vector & 0xFFUL);
3896 		return 0;
3897 	}
3898 
3899 	/*
3900 	 * Process any exceptions that are not debug traps before MTF.
3901 	 *
3902 	 * Note that only a pending nested run can block a pending exception.
3903 	 * Otherwise an injected NMI/interrupt should either be
3904 	 * lost or delivered to the nested hypervisor in the IDT_VECTORING_INFO,
3905 	 * while delivering the pending exception.
3906 	 */
3907 
3908 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3909 		if (vmx->nested.nested_run_pending)
3910 			return -EBUSY;
3911 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3912 			goto no_vmexit;
3913 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3914 		return 0;
3915 	}
3916 
3917 	if (mtf_pending) {
3918 		if (block_nested_events)
3919 			return -EBUSY;
3920 		nested_vmx_update_pending_dbg(vcpu);
3921 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3922 		return 0;
3923 	}
3924 
3925 	if (vcpu->arch.exception.pending) {
3926 		if (vmx->nested.nested_run_pending)
3927 			return -EBUSY;
3928 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3929 			goto no_vmexit;
3930 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3931 		return 0;
3932 	}
3933 
3934 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3935 		if (block_nested_events)
3936 			return -EBUSY;
3937 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3938 		return 0;
3939 	}
3940 
3941 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3942 		if (block_nested_events)
3943 			return -EBUSY;
3944 		goto no_vmexit;
3945 	}
3946 
3947 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3948 		if (block_nested_events)
3949 			return -EBUSY;
3950 		if (!nested_exit_on_nmi(vcpu))
3951 			goto no_vmexit;
3952 
3953 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3954 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3955 				  INTR_INFO_VALID_MASK, 0);
3956 		/*
3957 		 * The NMI-triggered VM exit counts as injection:
3958 		 * clear this one and block further NMIs.
3959 		 */
3960 		vcpu->arch.nmi_pending = 0;
3961 		vmx_set_nmi_mask(vcpu, true);
3962 		return 0;
3963 	}
3964 
3965 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3966 		if (block_nested_events)
3967 			return -EBUSY;
3968 		if (!nested_exit_on_intr(vcpu))
3969 			goto no_vmexit;
3970 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3971 		return 0;
3972 	}
3973 
3974 no_vmexit:
3975 	return vmx_complete_nested_posted_interrupt(vcpu);
3976 }
3977 
3978 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3979 {
3980 	ktime_t remaining =
3981 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3982 	u64 value;
3983 
3984 	if (ktime_to_ns(remaining) <= 0)
3985 		return 0;
3986 
3987 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3988 	do_div(value, 1000000);
3989 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3990 }
3991 
3992 static bool is_vmcs12_ext_field(unsigned long field)
3993 {
3994 	switch (field) {
3995 	case GUEST_ES_SELECTOR:
3996 	case GUEST_CS_SELECTOR:
3997 	case GUEST_SS_SELECTOR:
3998 	case GUEST_DS_SELECTOR:
3999 	case GUEST_FS_SELECTOR:
4000 	case GUEST_GS_SELECTOR:
4001 	case GUEST_LDTR_SELECTOR:
4002 	case GUEST_TR_SELECTOR:
4003 	case GUEST_ES_LIMIT:
4004 	case GUEST_CS_LIMIT:
4005 	case GUEST_SS_LIMIT:
4006 	case GUEST_DS_LIMIT:
4007 	case GUEST_FS_LIMIT:
4008 	case GUEST_GS_LIMIT:
4009 	case GUEST_LDTR_LIMIT:
4010 	case GUEST_TR_LIMIT:
4011 	case GUEST_GDTR_LIMIT:
4012 	case GUEST_IDTR_LIMIT:
4013 	case GUEST_ES_AR_BYTES:
4014 	case GUEST_DS_AR_BYTES:
4015 	case GUEST_FS_AR_BYTES:
4016 	case GUEST_GS_AR_BYTES:
4017 	case GUEST_LDTR_AR_BYTES:
4018 	case GUEST_TR_AR_BYTES:
4019 	case GUEST_ES_BASE:
4020 	case GUEST_CS_BASE:
4021 	case GUEST_SS_BASE:
4022 	case GUEST_DS_BASE:
4023 	case GUEST_FS_BASE:
4024 	case GUEST_GS_BASE:
4025 	case GUEST_LDTR_BASE:
4026 	case GUEST_TR_BASE:
4027 	case GUEST_GDTR_BASE:
4028 	case GUEST_IDTR_BASE:
4029 	case GUEST_PENDING_DBG_EXCEPTIONS:
4030 	case GUEST_BNDCFGS:
4031 		return true;
4032 	default:
4033 		break;
4034 	}
4035 
4036 	return false;
4037 }
4038 
4039 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4040 				       struct vmcs12 *vmcs12)
4041 {
4042 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4043 
4044 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4045 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4046 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4047 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4048 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4049 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4050 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4051 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4052 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4053 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4054 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4055 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4056 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4057 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4058 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4059 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4060 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4061 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4062 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4063 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4064 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4065 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4066 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4067 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4068 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4069 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4070 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4071 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4072 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4073 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4074 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4075 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4076 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4077 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4078 	vmcs12->guest_pending_dbg_exceptions =
4079 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4080 	if (kvm_mpx_supported())
4081 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
4082 
4083 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4084 }
4085 
4086 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4087 				       struct vmcs12 *vmcs12)
4088 {
4089 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4090 	int cpu;
4091 
4092 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4093 		return;
4094 
4095 
4096 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4097 
4098 	cpu = get_cpu();
4099 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4100 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4101 
4102 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4103 
4104 	vmx->loaded_vmcs = &vmx->vmcs01;
4105 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4106 	put_cpu();
4107 }
4108 
4109 /*
4110  * Update the guest state fields of vmcs12 to reflect changes that
4111  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4112  * VM-entry controls is also updated, since this is really a guest
4113  * state bit.)
4114  */
4115 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4116 {
4117 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4118 
4119 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4120 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4121 
4122 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4123 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4124 
4125 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4126 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4127 
4128 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4129 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4130 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4131 
4132 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4133 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4134 
4135 	vmcs12->guest_interruptibility_info =
4136 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4137 
4138 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4139 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4140 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4141 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4142 	else
4143 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4144 
4145 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4146 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4147 	    !vmx->nested.nested_run_pending)
4148 		vmcs12->vmx_preemption_timer_value =
4149 			vmx_get_preemption_timer_value(vcpu);
4150 
4151 	/*
4152 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4153 	 * own CR3 without exiting. If it has changed it, we must keep it.
4154 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4155 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4156 	 *
4157 	 * Additionally, restore L2's PDPTR to vmcs12.
4158 	 */
4159 	if (enable_ept) {
4160 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4161 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4162 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4163 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4164 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4165 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4166 		}
4167 	}
4168 
4169 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4170 
4171 	if (nested_cpu_has_vid(vmcs12))
4172 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4173 
4174 	vmcs12->vm_entry_controls =
4175 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4176 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4177 
4178 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4179 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4180 
4181 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4182 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4183 }
4184 
4185 /*
4186  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4187  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4188  * and this function updates it to reflect the changes to the guest state while
4189  * L2 was running (and perhaps made some exits which were handled directly by L0
4190  * without going back to L1), and to reflect the exit reason.
4191  * Note that we do not have to copy here all VMCS fields, just those that
4192  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4193  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4194  * which already writes to vmcs12 directly.
4195  */
4196 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4197 			   u32 vm_exit_reason, u32 exit_intr_info,
4198 			   unsigned long exit_qualification)
4199 {
4200 	/* update exit information fields: */
4201 	vmcs12->vm_exit_reason = vm_exit_reason;
4202 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4203 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4204 	vmcs12->exit_qualification = exit_qualification;
4205 	vmcs12->vm_exit_intr_info = exit_intr_info;
4206 
4207 	vmcs12->idt_vectoring_info_field = 0;
4208 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4209 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4210 
4211 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4212 		vmcs12->launch_state = 1;
4213 
4214 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4215 		 * instead of reading the real value. */
4216 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4217 
4218 		/*
4219 		 * Transfer the event that L0 or L1 may wanted to inject into
4220 		 * L2 to IDT_VECTORING_INFO_FIELD.
4221 		 */
4222 		vmcs12_save_pending_event(vcpu, vmcs12);
4223 
4224 		/*
4225 		 * According to spec, there's no need to store the guest's
4226 		 * MSRs if the exit is due to a VM-entry failure that occurs
4227 		 * during or after loading the guest state. Since this exit
4228 		 * does not fall in that category, we need to save the MSRs.
4229 		 */
4230 		if (nested_vmx_store_msr(vcpu,
4231 					 vmcs12->vm_exit_msr_store_addr,
4232 					 vmcs12->vm_exit_msr_store_count))
4233 			nested_vmx_abort(vcpu,
4234 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4235 	}
4236 
4237 	/*
4238 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4239 	 * preserved above and would only end up incorrectly in L1.
4240 	 */
4241 	vcpu->arch.nmi_injected = false;
4242 	kvm_clear_exception_queue(vcpu);
4243 	kvm_clear_interrupt_queue(vcpu);
4244 }
4245 
4246 /*
4247  * A part of what we need to when the nested L2 guest exits and we want to
4248  * run its L1 parent, is to reset L1's guest state to the host state specified
4249  * in vmcs12.
4250  * This function is to be called not only on normal nested exit, but also on
4251  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4252  * Failures During or After Loading Guest State").
4253  * This function should be called when the active VMCS is L1's (vmcs01).
4254  */
4255 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4256 				   struct vmcs12 *vmcs12)
4257 {
4258 	enum vm_entry_failure_code ignored;
4259 	struct kvm_segment seg;
4260 
4261 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4262 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4263 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4264 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4265 	else
4266 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4267 	vmx_set_efer(vcpu, vcpu->arch.efer);
4268 
4269 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4270 	kvm_rip_write(vcpu, vmcs12->host_rip);
4271 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4272 	vmx_set_interrupt_shadow(vcpu, 0);
4273 
4274 	/*
4275 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4276 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4277 	 *
4278 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4279 	 * (KVM doesn't change it);
4280 	 */
4281 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4282 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4283 
4284 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4285 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4286 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4287 
4288 	nested_ept_uninit_mmu_context(vcpu);
4289 
4290 	/*
4291 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4292 	 * couldn't have changed.
4293 	 */
4294 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4295 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4296 
4297 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4298 
4299 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4300 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4301 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4302 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4303 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4304 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4305 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4306 
4307 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4308 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4309 		vmcs_write64(GUEST_BNDCFGS, 0);
4310 
4311 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4312 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4313 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4314 	}
4315 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4316 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4317 					 vmcs12->host_ia32_perf_global_ctrl));
4318 
4319 	/* Set L1 segment info according to Intel SDM
4320 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4321 	seg = (struct kvm_segment) {
4322 		.base = 0,
4323 		.limit = 0xFFFFFFFF,
4324 		.selector = vmcs12->host_cs_selector,
4325 		.type = 11,
4326 		.present = 1,
4327 		.s = 1,
4328 		.g = 1
4329 	};
4330 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4331 		seg.l = 1;
4332 	else
4333 		seg.db = 1;
4334 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4335 	seg = (struct kvm_segment) {
4336 		.base = 0,
4337 		.limit = 0xFFFFFFFF,
4338 		.type = 3,
4339 		.present = 1,
4340 		.s = 1,
4341 		.db = 1,
4342 		.g = 1
4343 	};
4344 	seg.selector = vmcs12->host_ds_selector;
4345 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4346 	seg.selector = vmcs12->host_es_selector;
4347 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4348 	seg.selector = vmcs12->host_ss_selector;
4349 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4350 	seg.selector = vmcs12->host_fs_selector;
4351 	seg.base = vmcs12->host_fs_base;
4352 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4353 	seg.selector = vmcs12->host_gs_selector;
4354 	seg.base = vmcs12->host_gs_base;
4355 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4356 	seg = (struct kvm_segment) {
4357 		.base = vmcs12->host_tr_base,
4358 		.limit = 0x67,
4359 		.selector = vmcs12->host_tr_selector,
4360 		.type = 11,
4361 		.present = 1
4362 	};
4363 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4364 
4365 	memset(&seg, 0, sizeof(seg));
4366 	seg.unusable = 1;
4367 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4368 
4369 	kvm_set_dr(vcpu, 7, 0x400);
4370 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4371 
4372 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4373 				vmcs12->vm_exit_msr_load_count))
4374 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4375 
4376 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4377 }
4378 
4379 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4380 {
4381 	struct vmx_uret_msr *efer_msr;
4382 	unsigned int i;
4383 
4384 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4385 		return vmcs_read64(GUEST_IA32_EFER);
4386 
4387 	if (cpu_has_load_ia32_efer())
4388 		return host_efer;
4389 
4390 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4391 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4392 			return vmx->msr_autoload.guest.val[i].value;
4393 	}
4394 
4395 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4396 	if (efer_msr)
4397 		return efer_msr->data;
4398 
4399 	return host_efer;
4400 }
4401 
4402 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4403 {
4404 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4405 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4406 	struct vmx_msr_entry g, h;
4407 	gpa_t gpa;
4408 	u32 i, j;
4409 
4410 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4411 
4412 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4413 		/*
4414 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4415 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4416 		 * and vcpu->arch.dr7 is not squirreled away before the
4417 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4418 		 */
4419 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4420 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4421 		else
4422 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4423 	}
4424 
4425 	/*
4426 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4427 	 * handle a variety of side effects to KVM's software model.
4428 	 */
4429 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4430 
4431 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4432 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4433 
4434 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4435 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4436 
4437 	nested_ept_uninit_mmu_context(vcpu);
4438 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4439 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4440 
4441 	/*
4442 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4443 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4444 	 * VMFail, like everything else we just need to ensure our
4445 	 * software model is up-to-date.
4446 	 */
4447 	if (enable_ept && is_pae_paging(vcpu))
4448 		ept_save_pdptrs(vcpu);
4449 
4450 	kvm_mmu_reset_context(vcpu);
4451 
4452 	/*
4453 	 * This nasty bit of open coding is a compromise between blindly
4454 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4455 	 * of VMFail), leaving the nested VM's MSRs in the software model
4456 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4457 	 * expensive since the lists are unbound by hardware).  For each
4458 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4459 	 * list, reload it from the exit load list if it exists and differs
4460 	 * from the guest value.  The intent is to stuff host state as
4461 	 * silently as possible, not to fully process the exit load list.
4462 	 */
4463 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4464 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4465 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4466 			pr_debug_ratelimited(
4467 				"%s read MSR index failed (%u, 0x%08llx)\n",
4468 				__func__, i, gpa);
4469 			goto vmabort;
4470 		}
4471 
4472 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4473 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4474 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4475 				pr_debug_ratelimited(
4476 					"%s read MSR failed (%u, 0x%08llx)\n",
4477 					__func__, j, gpa);
4478 				goto vmabort;
4479 			}
4480 			if (h.index != g.index)
4481 				continue;
4482 			if (h.value == g.value)
4483 				break;
4484 
4485 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4486 				pr_debug_ratelimited(
4487 					"%s check failed (%u, 0x%x, 0x%x)\n",
4488 					__func__, j, h.index, h.reserved);
4489 				goto vmabort;
4490 			}
4491 
4492 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4493 				pr_debug_ratelimited(
4494 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4495 					__func__, j, h.index, h.value);
4496 				goto vmabort;
4497 			}
4498 		}
4499 	}
4500 
4501 	return;
4502 
4503 vmabort:
4504 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4505 }
4506 
4507 /*
4508  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4509  * and modify vmcs12 to make it see what it would expect to see there if
4510  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4511  */
4512 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4513 		       u32 exit_intr_info, unsigned long exit_qualification)
4514 {
4515 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4516 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4517 
4518 	/* trying to cancel vmlaunch/vmresume is a bug */
4519 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4520 
4521 	/* Similarly, triple faults in L2 should never escape. */
4522 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
4523 
4524 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4525 		/*
4526 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4527 		 * Enlightened VMCS after migration and we still need to
4528 		 * do that when something is forcing L2->L1 exit prior to
4529 		 * the first L2 run.
4530 		 */
4531 		(void)nested_get_evmcs_page(vcpu);
4532 	}
4533 
4534 	/* Service pending TLB flush requests for L2 before switching to L1. */
4535 	kvm_service_local_tlb_flush_requests(vcpu);
4536 
4537 	/*
4538 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4539 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4540 	 * up-to-date before switching to L1.
4541 	 */
4542 	if (enable_ept && is_pae_paging(vcpu))
4543 		vmx_ept_load_pdptrs(vcpu);
4544 
4545 	leave_guest_mode(vcpu);
4546 
4547 	if (nested_cpu_has_preemption_timer(vmcs12))
4548 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4549 
4550 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4551 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4552 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4553 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4554 	}
4555 
4556 	if (likely(!vmx->fail)) {
4557 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4558 
4559 		if (vm_exit_reason != -1)
4560 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4561 				       exit_intr_info, exit_qualification);
4562 
4563 		/*
4564 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4565 		 * also be used to capture vmcs12 cache as part of
4566 		 * capturing nVMX state for snapshot (migration).
4567 		 *
4568 		 * Otherwise, this flush will dirty guest memory at a
4569 		 * point it is already assumed by user-space to be
4570 		 * immutable.
4571 		 */
4572 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4573 	} else {
4574 		/*
4575 		 * The only expected VM-instruction error is "VM entry with
4576 		 * invalid control field(s)." Anything else indicates a
4577 		 * problem with L0.  And we should never get here with a
4578 		 * VMFail of any type if early consistency checks are enabled.
4579 		 */
4580 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4581 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4582 		WARN_ON_ONCE(nested_early_check);
4583 	}
4584 
4585 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4586 
4587 	/* Update any VMCS fields that might have changed while L2 ran */
4588 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4589 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4590 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4591 	if (kvm_has_tsc_control)
4592 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4593 
4594 	if (vmx->nested.l1_tpr_threshold != -1)
4595 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4596 
4597 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4598 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4599 		vmx_set_virtual_apic_mode(vcpu);
4600 	}
4601 
4602 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4603 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4604 		vmx_update_cpu_dirty_logging(vcpu);
4605 	}
4606 
4607 	/* Unpin physical memory we referred to in vmcs02 */
4608 	if (vmx->nested.apic_access_page) {
4609 		kvm_release_page_clean(vmx->nested.apic_access_page);
4610 		vmx->nested.apic_access_page = NULL;
4611 	}
4612 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4613 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4614 	vmx->nested.pi_desc = NULL;
4615 
4616 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4617 		vmx->nested.reload_vmcs01_apic_access_page = false;
4618 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4619 	}
4620 
4621 	if (vmx->nested.update_vmcs01_apicv_status) {
4622 		vmx->nested.update_vmcs01_apicv_status = false;
4623 		kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
4624 	}
4625 
4626 	if ((vm_exit_reason != -1) &&
4627 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4628 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4629 
4630 	/* in case we halted in L2 */
4631 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4632 
4633 	if (likely(!vmx->fail)) {
4634 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4635 		    nested_exit_intr_ack_set(vcpu)) {
4636 			int irq = kvm_cpu_get_interrupt(vcpu);
4637 			WARN_ON(irq < 0);
4638 			vmcs12->vm_exit_intr_info = irq |
4639 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4640 		}
4641 
4642 		if (vm_exit_reason != -1)
4643 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4644 						       vmcs12->exit_qualification,
4645 						       vmcs12->idt_vectoring_info_field,
4646 						       vmcs12->vm_exit_intr_info,
4647 						       vmcs12->vm_exit_intr_error_code,
4648 						       KVM_ISA_VMX);
4649 
4650 		load_vmcs12_host_state(vcpu, vmcs12);
4651 
4652 		return;
4653 	}
4654 
4655 	/*
4656 	 * After an early L2 VM-entry failure, we're now back
4657 	 * in L1 which thinks it just finished a VMLAUNCH or
4658 	 * VMRESUME instruction, so we need to set the failure
4659 	 * flag and the VM-instruction error field of the VMCS
4660 	 * accordingly, and skip the emulated instruction.
4661 	 */
4662 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4663 
4664 	/*
4665 	 * Restore L1's host state to KVM's software model.  We're here
4666 	 * because a consistency check was caught by hardware, which
4667 	 * means some amount of guest state has been propagated to KVM's
4668 	 * model and needs to be unwound to the host's state.
4669 	 */
4670 	nested_vmx_restore_host_state(vcpu);
4671 
4672 	vmx->fail = 0;
4673 }
4674 
4675 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4676 {
4677 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4678 }
4679 
4680 /*
4681  * Decode the memory-address operand of a vmx instruction, as recorded on an
4682  * exit caused by such an instruction (run by a guest hypervisor).
4683  * On success, returns 0. When the operand is invalid, returns 1 and throws
4684  * #UD, #GP, or #SS.
4685  */
4686 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4687 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4688 {
4689 	gva_t off;
4690 	bool exn;
4691 	struct kvm_segment s;
4692 
4693 	/*
4694 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4695 	 * Execution", on an exit, vmx_instruction_info holds most of the
4696 	 * addressing components of the operand. Only the displacement part
4697 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4698 	 * For how an actual address is calculated from all these components,
4699 	 * refer to Vol. 1, "Operand Addressing".
4700 	 */
4701 	int  scaling = vmx_instruction_info & 3;
4702 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4703 	bool is_reg = vmx_instruction_info & (1u << 10);
4704 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4705 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4706 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4707 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4708 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4709 
4710 	if (is_reg) {
4711 		kvm_queue_exception(vcpu, UD_VECTOR);
4712 		return 1;
4713 	}
4714 
4715 	/* Addr = segment_base + offset */
4716 	/* offset = base + [index * scale] + displacement */
4717 	off = exit_qualification; /* holds the displacement */
4718 	if (addr_size == 1)
4719 		off = (gva_t)sign_extend64(off, 31);
4720 	else if (addr_size == 0)
4721 		off = (gva_t)sign_extend64(off, 15);
4722 	if (base_is_valid)
4723 		off += kvm_register_read(vcpu, base_reg);
4724 	if (index_is_valid)
4725 		off += kvm_register_read(vcpu, index_reg) << scaling;
4726 	vmx_get_segment(vcpu, &s, seg_reg);
4727 
4728 	/*
4729 	 * The effective address, i.e. @off, of a memory operand is truncated
4730 	 * based on the address size of the instruction.  Note that this is
4731 	 * the *effective address*, i.e. the address prior to accounting for
4732 	 * the segment's base.
4733 	 */
4734 	if (addr_size == 1) /* 32 bit */
4735 		off &= 0xffffffff;
4736 	else if (addr_size == 0) /* 16 bit */
4737 		off &= 0xffff;
4738 
4739 	/* Checks for #GP/#SS exceptions. */
4740 	exn = false;
4741 	if (is_long_mode(vcpu)) {
4742 		/*
4743 		 * The virtual/linear address is never truncated in 64-bit
4744 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4745 		 * address when using FS/GS with a non-zero base.
4746 		 */
4747 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4748 			*ret = s.base + off;
4749 		else
4750 			*ret = off;
4751 
4752 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4753 		 * non-canonical form. This is the only check on the memory
4754 		 * destination for long mode!
4755 		 */
4756 		exn = is_noncanonical_address(*ret, vcpu);
4757 	} else {
4758 		/*
4759 		 * When not in long mode, the virtual/linear address is
4760 		 * unconditionally truncated to 32 bits regardless of the
4761 		 * address size.
4762 		 */
4763 		*ret = (s.base + off) & 0xffffffff;
4764 
4765 		/* Protected mode: apply checks for segment validity in the
4766 		 * following order:
4767 		 * - segment type check (#GP(0) may be thrown)
4768 		 * - usability check (#GP(0)/#SS(0))
4769 		 * - limit check (#GP(0)/#SS(0))
4770 		 */
4771 		if (wr)
4772 			/* #GP(0) if the destination operand is located in a
4773 			 * read-only data segment or any code segment.
4774 			 */
4775 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4776 		else
4777 			/* #GP(0) if the source operand is located in an
4778 			 * execute-only code segment
4779 			 */
4780 			exn = ((s.type & 0xa) == 8);
4781 		if (exn) {
4782 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4783 			return 1;
4784 		}
4785 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4786 		 */
4787 		exn = (s.unusable != 0);
4788 
4789 		/*
4790 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4791 		 * outside the segment limit.  All CPUs that support VMX ignore
4792 		 * limit checks for flat segments, i.e. segments with base==0,
4793 		 * limit==0xffffffff and of type expand-up data or code.
4794 		 */
4795 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4796 		     ((s.type & 8) || !(s.type & 4))))
4797 			exn = exn || ((u64)off + len - 1 > s.limit);
4798 	}
4799 	if (exn) {
4800 		kvm_queue_exception_e(vcpu,
4801 				      seg_reg == VCPU_SREG_SS ?
4802 						SS_VECTOR : GP_VECTOR,
4803 				      0);
4804 		return 1;
4805 	}
4806 
4807 	return 0;
4808 }
4809 
4810 void nested_vmx_pmu_refresh(struct kvm_vcpu *vcpu,
4811 			    bool vcpu_has_perf_global_ctrl)
4812 {
4813 	struct vcpu_vmx *vmx;
4814 
4815 	if (!nested_vmx_allowed(vcpu))
4816 		return;
4817 
4818 	vmx = to_vmx(vcpu);
4819 	if (vcpu_has_perf_global_ctrl) {
4820 		vmx->nested.msrs.entry_ctls_high |=
4821 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4822 		vmx->nested.msrs.exit_ctls_high |=
4823 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4824 	} else {
4825 		vmx->nested.msrs.entry_ctls_high &=
4826 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4827 		vmx->nested.msrs.exit_ctls_high &=
4828 				~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4829 	}
4830 }
4831 
4832 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4833 				int *ret)
4834 {
4835 	gva_t gva;
4836 	struct x86_exception e;
4837 	int r;
4838 
4839 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4840 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4841 				sizeof(*vmpointer), &gva)) {
4842 		*ret = 1;
4843 		return -EINVAL;
4844 	}
4845 
4846 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4847 	if (r != X86EMUL_CONTINUE) {
4848 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4849 		return -EINVAL;
4850 	}
4851 
4852 	return 0;
4853 }
4854 
4855 /*
4856  * Allocate a shadow VMCS and associate it with the currently loaded
4857  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4858  * VMCS is also VMCLEARed, so that it is ready for use.
4859  */
4860 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4861 {
4862 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4863 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4864 
4865 	/*
4866 	 * KVM allocates a shadow VMCS only when L1 executes VMXON and frees it
4867 	 * when L1 executes VMXOFF or the vCPU is forced out of nested
4868 	 * operation.  VMXON faults if the CPU is already post-VMXON, so it
4869 	 * should be impossible to already have an allocated shadow VMCS.  KVM
4870 	 * doesn't support virtualization of VMCS shadowing, so vmcs01 should
4871 	 * always be the loaded VMCS.
4872 	 */
4873 	if (WARN_ON(loaded_vmcs != &vmx->vmcs01 || loaded_vmcs->shadow_vmcs))
4874 		return loaded_vmcs->shadow_vmcs;
4875 
4876 	loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4877 	if (loaded_vmcs->shadow_vmcs)
4878 		vmcs_clear(loaded_vmcs->shadow_vmcs);
4879 
4880 	return loaded_vmcs->shadow_vmcs;
4881 }
4882 
4883 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4884 {
4885 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4886 	int r;
4887 
4888 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4889 	if (r < 0)
4890 		goto out_vmcs02;
4891 
4892 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4893 	if (!vmx->nested.cached_vmcs12)
4894 		goto out_cached_vmcs12;
4895 
4896 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
4897 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4898 	if (!vmx->nested.cached_shadow_vmcs12)
4899 		goto out_cached_shadow_vmcs12;
4900 
4901 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4902 		goto out_shadow_vmcs;
4903 
4904 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4905 		     HRTIMER_MODE_ABS_PINNED);
4906 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4907 
4908 	vmx->nested.vpid02 = allocate_vpid();
4909 
4910 	vmx->nested.vmcs02_initialized = false;
4911 	vmx->nested.vmxon = true;
4912 
4913 	if (vmx_pt_mode_is_host_guest()) {
4914 		vmx->pt_desc.guest.ctl = 0;
4915 		pt_update_intercept_for_msr(vcpu);
4916 	}
4917 
4918 	return 0;
4919 
4920 out_shadow_vmcs:
4921 	kfree(vmx->nested.cached_shadow_vmcs12);
4922 
4923 out_cached_shadow_vmcs12:
4924 	kfree(vmx->nested.cached_vmcs12);
4925 
4926 out_cached_vmcs12:
4927 	free_loaded_vmcs(&vmx->nested.vmcs02);
4928 
4929 out_vmcs02:
4930 	return -ENOMEM;
4931 }
4932 
4933 /* Emulate the VMXON instruction. */
4934 static int handle_vmon(struct kvm_vcpu *vcpu)
4935 {
4936 	int ret;
4937 	gpa_t vmptr;
4938 	uint32_t revision;
4939 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4940 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4941 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4942 
4943 	/*
4944 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4945 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4946 	 * 1 (see vmx_is_valid_cr4() for when we allow the guest to set this).
4947 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4948 	 * have already been checked by hardware, prior to the VM-exit for
4949 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4950 	 * that bit set to 1 in non-root mode.
4951 	 */
4952 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4953 		kvm_queue_exception(vcpu, UD_VECTOR);
4954 		return 1;
4955 	}
4956 
4957 	/* CPL=0 must be checked manually. */
4958 	if (vmx_get_cpl(vcpu)) {
4959 		kvm_inject_gp(vcpu, 0);
4960 		return 1;
4961 	}
4962 
4963 	if (vmx->nested.vmxon)
4964 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4965 
4966 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4967 			!= VMXON_NEEDED_FEATURES) {
4968 		kvm_inject_gp(vcpu, 0);
4969 		return 1;
4970 	}
4971 
4972 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4973 		return ret;
4974 
4975 	/*
4976 	 * SDM 3: 24.11.5
4977 	 * The first 4 bytes of VMXON region contain the supported
4978 	 * VMCS revision identifier
4979 	 *
4980 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4981 	 * which replaces physical address width with 32
4982 	 */
4983 	if (!page_address_valid(vcpu, vmptr))
4984 		return nested_vmx_failInvalid(vcpu);
4985 
4986 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4987 	    revision != VMCS12_REVISION)
4988 		return nested_vmx_failInvalid(vcpu);
4989 
4990 	vmx->nested.vmxon_ptr = vmptr;
4991 	ret = enter_vmx_operation(vcpu);
4992 	if (ret)
4993 		return ret;
4994 
4995 	return nested_vmx_succeed(vcpu);
4996 }
4997 
4998 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4999 {
5000 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5001 
5002 	if (vmx->nested.current_vmptr == INVALID_GPA)
5003 		return;
5004 
5005 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
5006 
5007 	if (enable_shadow_vmcs) {
5008 		/* copy to memory all shadowed fields in case
5009 		   they were modified */
5010 		copy_shadow_to_vmcs12(vmx);
5011 		vmx_disable_shadow_vmcs(vmx);
5012 	}
5013 	vmx->nested.posted_intr_nv = -1;
5014 
5015 	/* Flush VMCS12 to guest memory */
5016 	kvm_vcpu_write_guest_page(vcpu,
5017 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
5018 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5019 
5020 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5021 
5022 	vmx->nested.current_vmptr = INVALID_GPA;
5023 }
5024 
5025 /* Emulate the VMXOFF instruction */
5026 static int handle_vmoff(struct kvm_vcpu *vcpu)
5027 {
5028 	if (!nested_vmx_check_permission(vcpu))
5029 		return 1;
5030 
5031 	free_nested(vcpu);
5032 
5033 	/* Process a latched INIT during time CPU was in VMX operation */
5034 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5035 
5036 	return nested_vmx_succeed(vcpu);
5037 }
5038 
5039 /* Emulate the VMCLEAR instruction */
5040 static int handle_vmclear(struct kvm_vcpu *vcpu)
5041 {
5042 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5043 	u32 zero = 0;
5044 	gpa_t vmptr;
5045 	u64 evmcs_gpa;
5046 	int r;
5047 
5048 	if (!nested_vmx_check_permission(vcpu))
5049 		return 1;
5050 
5051 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5052 		return r;
5053 
5054 	if (!page_address_valid(vcpu, vmptr))
5055 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5056 
5057 	if (vmptr == vmx->nested.vmxon_ptr)
5058 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5059 
5060 	/*
5061 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
5062 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
5063 	 * way to distinguish it from VMCS12) and we must not corrupt it by
5064 	 * writing to the non-existent 'launch_state' field. The area doesn't
5065 	 * have to be the currently active EVMCS on the calling CPU and there's
5066 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5067 	 * state. It is possible that the area will stay mapped as
5068 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5069 	 */
5070 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
5071 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
5072 		if (vmptr == vmx->nested.current_vmptr)
5073 			nested_release_vmcs12(vcpu);
5074 
5075 		kvm_vcpu_write_guest(vcpu,
5076 				     vmptr + offsetof(struct vmcs12,
5077 						      launch_state),
5078 				     &zero, sizeof(zero));
5079 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5080 		nested_release_evmcs(vcpu);
5081 	}
5082 
5083 	return nested_vmx_succeed(vcpu);
5084 }
5085 
5086 /* Emulate the VMLAUNCH instruction */
5087 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5088 {
5089 	return nested_vmx_run(vcpu, true);
5090 }
5091 
5092 /* Emulate the VMRESUME instruction */
5093 static int handle_vmresume(struct kvm_vcpu *vcpu)
5094 {
5095 
5096 	return nested_vmx_run(vcpu, false);
5097 }
5098 
5099 static int handle_vmread(struct kvm_vcpu *vcpu)
5100 {
5101 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5102 						    : get_vmcs12(vcpu);
5103 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5104 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5105 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5106 	struct x86_exception e;
5107 	unsigned long field;
5108 	u64 value;
5109 	gva_t gva = 0;
5110 	short offset;
5111 	int len, r;
5112 
5113 	if (!nested_vmx_check_permission(vcpu))
5114 		return 1;
5115 
5116 	/* Decode instruction info and find the field to read */
5117 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5118 
5119 	if (!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
5120 		/*
5121 		 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5122 		 * any VMREAD sets the ALU flags for VMfailInvalid.
5123 		 */
5124 		if (vmx->nested.current_vmptr == INVALID_GPA ||
5125 		    (is_guest_mode(vcpu) &&
5126 		     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5127 			return nested_vmx_failInvalid(vcpu);
5128 
5129 		offset = get_vmcs12_field_offset(field);
5130 		if (offset < 0)
5131 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5132 
5133 		if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5134 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5135 
5136 		/* Read the field, zero-extended to a u64 value */
5137 		value = vmcs12_read_any(vmcs12, field, offset);
5138 	} else {
5139 		/*
5140 		 * Hyper-V TLFS (as of 6.0b) explicitly states, that while an
5141 		 * enlightened VMCS is active VMREAD/VMWRITE instructions are
5142 		 * unsupported. Unfortunately, certain versions of Windows 11
5143 		 * don't comply with this requirement which is not enforced in
5144 		 * genuine Hyper-V. Allow VMREAD from an enlightened VMCS as a
5145 		 * workaround, as misbehaving guests will panic on VM-Fail.
5146 		 * Note, enlightened VMCS is incompatible with shadow VMCS so
5147 		 * all VMREADs from L2 should go to L1.
5148 		 */
5149 		if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5150 			return nested_vmx_failInvalid(vcpu);
5151 
5152 		offset = evmcs_field_offset(field, NULL);
5153 		if (offset < 0)
5154 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5155 
5156 		/* Read the field, zero-extended to a u64 value */
5157 		value = evmcs_read_any(vmx->nested.hv_evmcs, field, offset);
5158 	}
5159 
5160 	/*
5161 	 * Now copy part of this value to register or memory, as requested.
5162 	 * Note that the number of bits actually copied is 32 or 64 depending
5163 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5164 	 */
5165 	if (instr_info & BIT(10)) {
5166 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5167 	} else {
5168 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5169 		if (get_vmx_mem_address(vcpu, exit_qualification,
5170 					instr_info, true, len, &gva))
5171 			return 1;
5172 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5173 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5174 		if (r != X86EMUL_CONTINUE)
5175 			return kvm_handle_memory_failure(vcpu, r, &e);
5176 	}
5177 
5178 	return nested_vmx_succeed(vcpu);
5179 }
5180 
5181 static bool is_shadow_field_rw(unsigned long field)
5182 {
5183 	switch (field) {
5184 #define SHADOW_FIELD_RW(x, y) case x:
5185 #include "vmcs_shadow_fields.h"
5186 		return true;
5187 	default:
5188 		break;
5189 	}
5190 	return false;
5191 }
5192 
5193 static bool is_shadow_field_ro(unsigned long field)
5194 {
5195 	switch (field) {
5196 #define SHADOW_FIELD_RO(x, y) case x:
5197 #include "vmcs_shadow_fields.h"
5198 		return true;
5199 	default:
5200 		break;
5201 	}
5202 	return false;
5203 }
5204 
5205 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5206 {
5207 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5208 						    : get_vmcs12(vcpu);
5209 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5210 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5211 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5212 	struct x86_exception e;
5213 	unsigned long field;
5214 	short offset;
5215 	gva_t gva;
5216 	int len, r;
5217 
5218 	/*
5219 	 * The value to write might be 32 or 64 bits, depending on L1's long
5220 	 * mode, and eventually we need to write that into a field of several
5221 	 * possible lengths. The code below first zero-extends the value to 64
5222 	 * bit (value), and then copies only the appropriate number of
5223 	 * bits into the vmcs12 field.
5224 	 */
5225 	u64 value = 0;
5226 
5227 	if (!nested_vmx_check_permission(vcpu))
5228 		return 1;
5229 
5230 	/*
5231 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5232 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5233 	 */
5234 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5235 	    (is_guest_mode(vcpu) &&
5236 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5237 		return nested_vmx_failInvalid(vcpu);
5238 
5239 	if (instr_info & BIT(10))
5240 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5241 	else {
5242 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5243 		if (get_vmx_mem_address(vcpu, exit_qualification,
5244 					instr_info, false, len, &gva))
5245 			return 1;
5246 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5247 		if (r != X86EMUL_CONTINUE)
5248 			return kvm_handle_memory_failure(vcpu, r, &e);
5249 	}
5250 
5251 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5252 
5253 	offset = get_vmcs12_field_offset(field);
5254 	if (offset < 0)
5255 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5256 
5257 	/*
5258 	 * If the vCPU supports "VMWRITE to any supported field in the
5259 	 * VMCS," then the "read-only" fields are actually read/write.
5260 	 */
5261 	if (vmcs_field_readonly(field) &&
5262 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5263 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5264 
5265 	/*
5266 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5267 	 * vmcs12, else we may crush a field or consume a stale value.
5268 	 */
5269 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5270 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5271 
5272 	/*
5273 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5274 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5275 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5276 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5277 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5278 	 * the stripped down value, L2 sees the full value as stored by KVM).
5279 	 */
5280 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5281 		value &= 0x1f0ff;
5282 
5283 	vmcs12_write_any(vmcs12, field, offset, value);
5284 
5285 	/*
5286 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5287 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5288 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5289 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5290 	 */
5291 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5292 		/*
5293 		 * L1 can read these fields without exiting, ensure the
5294 		 * shadow VMCS is up-to-date.
5295 		 */
5296 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5297 			preempt_disable();
5298 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5299 
5300 			__vmcs_writel(field, value);
5301 
5302 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5303 			vmcs_load(vmx->loaded_vmcs->vmcs);
5304 			preempt_enable();
5305 		}
5306 		vmx->nested.dirty_vmcs12 = true;
5307 	}
5308 
5309 	return nested_vmx_succeed(vcpu);
5310 }
5311 
5312 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5313 {
5314 	vmx->nested.current_vmptr = vmptr;
5315 	if (enable_shadow_vmcs) {
5316 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5317 		vmcs_write64(VMCS_LINK_POINTER,
5318 			     __pa(vmx->vmcs01.shadow_vmcs));
5319 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5320 	}
5321 	vmx->nested.dirty_vmcs12 = true;
5322 	vmx->nested.force_msr_bitmap_recalc = true;
5323 }
5324 
5325 /* Emulate the VMPTRLD instruction */
5326 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5327 {
5328 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5329 	gpa_t vmptr;
5330 	int r;
5331 
5332 	if (!nested_vmx_check_permission(vcpu))
5333 		return 1;
5334 
5335 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5336 		return r;
5337 
5338 	if (!page_address_valid(vcpu, vmptr))
5339 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5340 
5341 	if (vmptr == vmx->nested.vmxon_ptr)
5342 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5343 
5344 	/* Forbid normal VMPTRLD if Enlightened version was used */
5345 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5346 		return 1;
5347 
5348 	if (vmx->nested.current_vmptr != vmptr) {
5349 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5350 		struct vmcs_hdr hdr;
5351 
5352 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5353 			/*
5354 			 * Reads from an unbacked page return all 1s,
5355 			 * which means that the 32 bits located at the
5356 			 * given physical address won't match the required
5357 			 * VMCS12_REVISION identifier.
5358 			 */
5359 			return nested_vmx_fail(vcpu,
5360 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5361 		}
5362 
5363 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5364 						 offsetof(struct vmcs12, hdr),
5365 						 sizeof(hdr))) {
5366 			return nested_vmx_fail(vcpu,
5367 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5368 		}
5369 
5370 		if (hdr.revision_id != VMCS12_REVISION ||
5371 		    (hdr.shadow_vmcs &&
5372 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5373 			return nested_vmx_fail(vcpu,
5374 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5375 		}
5376 
5377 		nested_release_vmcs12(vcpu);
5378 
5379 		/*
5380 		 * Load VMCS12 from guest memory since it is not already
5381 		 * cached.
5382 		 */
5383 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5384 					  VMCS12_SIZE)) {
5385 			return nested_vmx_fail(vcpu,
5386 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5387 		}
5388 
5389 		set_current_vmptr(vmx, vmptr);
5390 	}
5391 
5392 	return nested_vmx_succeed(vcpu);
5393 }
5394 
5395 /* Emulate the VMPTRST instruction */
5396 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5397 {
5398 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5399 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5400 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5401 	struct x86_exception e;
5402 	gva_t gva;
5403 	int r;
5404 
5405 	if (!nested_vmx_check_permission(vcpu))
5406 		return 1;
5407 
5408 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5409 		return 1;
5410 
5411 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5412 				true, sizeof(gpa_t), &gva))
5413 		return 1;
5414 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5415 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5416 					sizeof(gpa_t), &e);
5417 	if (r != X86EMUL_CONTINUE)
5418 		return kvm_handle_memory_failure(vcpu, r, &e);
5419 
5420 	return nested_vmx_succeed(vcpu);
5421 }
5422 
5423 /* Emulate the INVEPT instruction */
5424 static int handle_invept(struct kvm_vcpu *vcpu)
5425 {
5426 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5427 	u32 vmx_instruction_info, types;
5428 	unsigned long type, roots_to_free;
5429 	struct kvm_mmu *mmu;
5430 	gva_t gva;
5431 	struct x86_exception e;
5432 	struct {
5433 		u64 eptp, gpa;
5434 	} operand;
5435 	int i, r, gpr_index;
5436 
5437 	if (!(vmx->nested.msrs.secondary_ctls_high &
5438 	      SECONDARY_EXEC_ENABLE_EPT) ||
5439 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5440 		kvm_queue_exception(vcpu, UD_VECTOR);
5441 		return 1;
5442 	}
5443 
5444 	if (!nested_vmx_check_permission(vcpu))
5445 		return 1;
5446 
5447 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5448 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5449 	type = kvm_register_read(vcpu, gpr_index);
5450 
5451 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5452 
5453 	if (type >= 32 || !(types & (1 << type)))
5454 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5455 
5456 	/* According to the Intel VMX instruction reference, the memory
5457 	 * operand is read even if it isn't needed (e.g., for type==global)
5458 	 */
5459 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5460 			vmx_instruction_info, false, sizeof(operand), &gva))
5461 		return 1;
5462 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5463 	if (r != X86EMUL_CONTINUE)
5464 		return kvm_handle_memory_failure(vcpu, r, &e);
5465 
5466 	/*
5467 	 * Nested EPT roots are always held through guest_mmu,
5468 	 * not root_mmu.
5469 	 */
5470 	mmu = &vcpu->arch.guest_mmu;
5471 
5472 	switch (type) {
5473 	case VMX_EPT_EXTENT_CONTEXT:
5474 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5475 			return nested_vmx_fail(vcpu,
5476 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5477 
5478 		roots_to_free = 0;
5479 		if (nested_ept_root_matches(mmu->root.hpa, mmu->root.pgd,
5480 					    operand.eptp))
5481 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5482 
5483 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5484 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5485 						    mmu->prev_roots[i].pgd,
5486 						    operand.eptp))
5487 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5488 		}
5489 		break;
5490 	case VMX_EPT_EXTENT_GLOBAL:
5491 		roots_to_free = KVM_MMU_ROOTS_ALL;
5492 		break;
5493 	default:
5494 		BUG();
5495 		break;
5496 	}
5497 
5498 	if (roots_to_free)
5499 		kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
5500 
5501 	return nested_vmx_succeed(vcpu);
5502 }
5503 
5504 static int handle_invvpid(struct kvm_vcpu *vcpu)
5505 {
5506 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5507 	u32 vmx_instruction_info;
5508 	unsigned long type, types;
5509 	gva_t gva;
5510 	struct x86_exception e;
5511 	struct {
5512 		u64 vpid;
5513 		u64 gla;
5514 	} operand;
5515 	u16 vpid02;
5516 	int r, gpr_index;
5517 
5518 	if (!(vmx->nested.msrs.secondary_ctls_high &
5519 	      SECONDARY_EXEC_ENABLE_VPID) ||
5520 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5521 		kvm_queue_exception(vcpu, UD_VECTOR);
5522 		return 1;
5523 	}
5524 
5525 	if (!nested_vmx_check_permission(vcpu))
5526 		return 1;
5527 
5528 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5529 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5530 	type = kvm_register_read(vcpu, gpr_index);
5531 
5532 	types = (vmx->nested.msrs.vpid_caps &
5533 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5534 
5535 	if (type >= 32 || !(types & (1 << type)))
5536 		return nested_vmx_fail(vcpu,
5537 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5538 
5539 	/* according to the intel vmx instruction reference, the memory
5540 	 * operand is read even if it isn't needed (e.g., for type==global)
5541 	 */
5542 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5543 			vmx_instruction_info, false, sizeof(operand), &gva))
5544 		return 1;
5545 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5546 	if (r != X86EMUL_CONTINUE)
5547 		return kvm_handle_memory_failure(vcpu, r, &e);
5548 
5549 	if (operand.vpid >> 16)
5550 		return nested_vmx_fail(vcpu,
5551 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5552 
5553 	vpid02 = nested_get_vpid02(vcpu);
5554 	switch (type) {
5555 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5556 		if (!operand.vpid ||
5557 		    is_noncanonical_address(operand.gla, vcpu))
5558 			return nested_vmx_fail(vcpu,
5559 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5560 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5561 		break;
5562 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5563 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5564 		if (!operand.vpid)
5565 			return nested_vmx_fail(vcpu,
5566 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5567 		vpid_sync_context(vpid02);
5568 		break;
5569 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5570 		vpid_sync_context(vpid02);
5571 		break;
5572 	default:
5573 		WARN_ON_ONCE(1);
5574 		return kvm_skip_emulated_instruction(vcpu);
5575 	}
5576 
5577 	/*
5578 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5579 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5580 	 * roots as VPIDs are not tracked in the MMU role.
5581 	 *
5582 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5583 	 * an MMU when EPT is disabled.
5584 	 *
5585 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5586 	 */
5587 	if (!enable_ept)
5588 		kvm_mmu_free_guest_mode_roots(vcpu->kvm, &vcpu->arch.root_mmu);
5589 
5590 	return nested_vmx_succeed(vcpu);
5591 }
5592 
5593 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5594 				     struct vmcs12 *vmcs12)
5595 {
5596 	u32 index = kvm_rcx_read(vcpu);
5597 	u64 new_eptp;
5598 
5599 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5600 		return 1;
5601 	if (index >= VMFUNC_EPTP_ENTRIES)
5602 		return 1;
5603 
5604 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5605 				     &new_eptp, index * 8, 8))
5606 		return 1;
5607 
5608 	/*
5609 	 * If the (L2) guest does a vmfunc to the currently
5610 	 * active ept pointer, we don't have to do anything else
5611 	 */
5612 	if (vmcs12->ept_pointer != new_eptp) {
5613 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5614 			return 1;
5615 
5616 		vmcs12->ept_pointer = new_eptp;
5617 		nested_ept_new_eptp(vcpu);
5618 
5619 		if (!nested_cpu_has_vpid(vmcs12))
5620 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5621 	}
5622 
5623 	return 0;
5624 }
5625 
5626 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5627 {
5628 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5629 	struct vmcs12 *vmcs12;
5630 	u32 function = kvm_rax_read(vcpu);
5631 
5632 	/*
5633 	 * VMFUNC is only supported for nested guests, but we always enable the
5634 	 * secondary control for simplicity; for non-nested mode, fake that we
5635 	 * didn't by injecting #UD.
5636 	 */
5637 	if (!is_guest_mode(vcpu)) {
5638 		kvm_queue_exception(vcpu, UD_VECTOR);
5639 		return 1;
5640 	}
5641 
5642 	vmcs12 = get_vmcs12(vcpu);
5643 
5644 	/*
5645 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5646 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5647 	 */
5648 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5649 		kvm_queue_exception(vcpu, UD_VECTOR);
5650 		return 1;
5651 	}
5652 
5653 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5654 		goto fail;
5655 
5656 	switch (function) {
5657 	case 0:
5658 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5659 			goto fail;
5660 		break;
5661 	default:
5662 		goto fail;
5663 	}
5664 	return kvm_skip_emulated_instruction(vcpu);
5665 
5666 fail:
5667 	/*
5668 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5669 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5670 	 * EXIT_REASON_VMFUNC as the exit reason.
5671 	 */
5672 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5673 			  vmx_get_intr_info(vcpu),
5674 			  vmx_get_exit_qual(vcpu));
5675 	return 1;
5676 }
5677 
5678 /*
5679  * Return true if an IO instruction with the specified port and size should cause
5680  * a VM-exit into L1.
5681  */
5682 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5683 				 int size)
5684 {
5685 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5686 	gpa_t bitmap, last_bitmap;
5687 	u8 b;
5688 
5689 	last_bitmap = INVALID_GPA;
5690 	b = -1;
5691 
5692 	while (size > 0) {
5693 		if (port < 0x8000)
5694 			bitmap = vmcs12->io_bitmap_a;
5695 		else if (port < 0x10000)
5696 			bitmap = vmcs12->io_bitmap_b;
5697 		else
5698 			return true;
5699 		bitmap += (port & 0x7fff) / 8;
5700 
5701 		if (last_bitmap != bitmap)
5702 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5703 				return true;
5704 		if (b & (1 << (port & 7)))
5705 			return true;
5706 
5707 		port++;
5708 		size--;
5709 		last_bitmap = bitmap;
5710 	}
5711 
5712 	return false;
5713 }
5714 
5715 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5716 				       struct vmcs12 *vmcs12)
5717 {
5718 	unsigned long exit_qualification;
5719 	unsigned short port;
5720 	int size;
5721 
5722 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5723 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5724 
5725 	exit_qualification = vmx_get_exit_qual(vcpu);
5726 
5727 	port = exit_qualification >> 16;
5728 	size = (exit_qualification & 7) + 1;
5729 
5730 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5731 }
5732 
5733 /*
5734  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5735  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5736  * disinterest in the current event (read or write a specific MSR) by using an
5737  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5738  */
5739 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5740 					struct vmcs12 *vmcs12,
5741 					union vmx_exit_reason exit_reason)
5742 {
5743 	u32 msr_index = kvm_rcx_read(vcpu);
5744 	gpa_t bitmap;
5745 
5746 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5747 		return true;
5748 
5749 	/*
5750 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5751 	 * for the four combinations of read/write and low/high MSR numbers.
5752 	 * First we need to figure out which of the four to use:
5753 	 */
5754 	bitmap = vmcs12->msr_bitmap;
5755 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5756 		bitmap += 2048;
5757 	if (msr_index >= 0xc0000000) {
5758 		msr_index -= 0xc0000000;
5759 		bitmap += 1024;
5760 	}
5761 
5762 	/* Then read the msr_index'th bit from this bitmap: */
5763 	if (msr_index < 1024*8) {
5764 		unsigned char b;
5765 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5766 			return true;
5767 		return 1 & (b >> (msr_index & 7));
5768 	} else
5769 		return true; /* let L1 handle the wrong parameter */
5770 }
5771 
5772 /*
5773  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5774  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5775  * intercept (via guest_host_mask etc.) the current event.
5776  */
5777 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5778 	struct vmcs12 *vmcs12)
5779 {
5780 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5781 	int cr = exit_qualification & 15;
5782 	int reg;
5783 	unsigned long val;
5784 
5785 	switch ((exit_qualification >> 4) & 3) {
5786 	case 0: /* mov to cr */
5787 		reg = (exit_qualification >> 8) & 15;
5788 		val = kvm_register_read(vcpu, reg);
5789 		switch (cr) {
5790 		case 0:
5791 			if (vmcs12->cr0_guest_host_mask &
5792 			    (val ^ vmcs12->cr0_read_shadow))
5793 				return true;
5794 			break;
5795 		case 3:
5796 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5797 				return true;
5798 			break;
5799 		case 4:
5800 			if (vmcs12->cr4_guest_host_mask &
5801 			    (vmcs12->cr4_read_shadow ^ val))
5802 				return true;
5803 			break;
5804 		case 8:
5805 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5806 				return true;
5807 			break;
5808 		}
5809 		break;
5810 	case 2: /* clts */
5811 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5812 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5813 			return true;
5814 		break;
5815 	case 1: /* mov from cr */
5816 		switch (cr) {
5817 		case 3:
5818 			if (vmcs12->cpu_based_vm_exec_control &
5819 			    CPU_BASED_CR3_STORE_EXITING)
5820 				return true;
5821 			break;
5822 		case 8:
5823 			if (vmcs12->cpu_based_vm_exec_control &
5824 			    CPU_BASED_CR8_STORE_EXITING)
5825 				return true;
5826 			break;
5827 		}
5828 		break;
5829 	case 3: /* lmsw */
5830 		/*
5831 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5832 		 * cr0. Other attempted changes are ignored, with no exit.
5833 		 */
5834 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5835 		if (vmcs12->cr0_guest_host_mask & 0xe &
5836 		    (val ^ vmcs12->cr0_read_shadow))
5837 			return true;
5838 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5839 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5840 		    (val & 0x1))
5841 			return true;
5842 		break;
5843 	}
5844 	return false;
5845 }
5846 
5847 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
5848 					  struct vmcs12 *vmcs12)
5849 {
5850 	u32 encls_leaf;
5851 
5852 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
5853 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
5854 		return false;
5855 
5856 	encls_leaf = kvm_rax_read(vcpu);
5857 	if (encls_leaf > 62)
5858 		encls_leaf = 63;
5859 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
5860 }
5861 
5862 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5863 	struct vmcs12 *vmcs12, gpa_t bitmap)
5864 {
5865 	u32 vmx_instruction_info;
5866 	unsigned long field;
5867 	u8 b;
5868 
5869 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5870 		return true;
5871 
5872 	/* Decode instruction info and find the field to access */
5873 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5874 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5875 
5876 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5877 	if (field >> 15)
5878 		return true;
5879 
5880 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5881 		return true;
5882 
5883 	return 1 & (b >> (field & 7));
5884 }
5885 
5886 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5887 {
5888 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5889 
5890 	if (nested_cpu_has_mtf(vmcs12))
5891 		return true;
5892 
5893 	/*
5894 	 * An MTF VM-exit may be injected into the guest by setting the
5895 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5896 	 * is the case regardless of the 'monitor trap flag' VM-execution
5897 	 * control.
5898 	 */
5899 	return entry_intr_info == (INTR_INFO_VALID_MASK
5900 				   | INTR_TYPE_OTHER_EVENT);
5901 }
5902 
5903 /*
5904  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5905  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5906  */
5907 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
5908 				     union vmx_exit_reason exit_reason)
5909 {
5910 	u32 intr_info;
5911 
5912 	switch ((u16)exit_reason.basic) {
5913 	case EXIT_REASON_EXCEPTION_NMI:
5914 		intr_info = vmx_get_intr_info(vcpu);
5915 		if (is_nmi(intr_info))
5916 			return true;
5917 		else if (is_page_fault(intr_info))
5918 			return vcpu->arch.apf.host_apf_flags ||
5919 			       vmx_need_pf_intercept(vcpu);
5920 		else if (is_debug(intr_info) &&
5921 			 vcpu->guest_debug &
5922 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5923 			return true;
5924 		else if (is_breakpoint(intr_info) &&
5925 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5926 			return true;
5927 		else if (is_alignment_check(intr_info) &&
5928 			 !vmx_guest_inject_ac(vcpu))
5929 			return true;
5930 		return false;
5931 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5932 		return true;
5933 	case EXIT_REASON_MCE_DURING_VMENTRY:
5934 		return true;
5935 	case EXIT_REASON_EPT_VIOLATION:
5936 		/*
5937 		 * L0 always deals with the EPT violation. If nested EPT is
5938 		 * used, and the nested mmu code discovers that the address is
5939 		 * missing in the guest EPT table (EPT12), the EPT violation
5940 		 * will be injected with nested_ept_inject_page_fault()
5941 		 */
5942 		return true;
5943 	case EXIT_REASON_EPT_MISCONFIG:
5944 		/*
5945 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5946 		 * table (shadow on EPT) or a merged EPT table that L0 built
5947 		 * (EPT on EPT). So any problems with the structure of the
5948 		 * table is L0's fault.
5949 		 */
5950 		return true;
5951 	case EXIT_REASON_PREEMPTION_TIMER:
5952 		return true;
5953 	case EXIT_REASON_PML_FULL:
5954 		/*
5955 		 * PML is emulated for an L1 VMM and should never be enabled in
5956 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
5957 		 */
5958 		return true;
5959 	case EXIT_REASON_VMFUNC:
5960 		/* VM functions are emulated through L2->L0 vmexits. */
5961 		return true;
5962 	case EXIT_REASON_BUS_LOCK:
5963 		/*
5964 		 * At present, bus lock VM exit is never exposed to L1.
5965 		 * Handle L2's bus locks in L0 directly.
5966 		 */
5967 		return true;
5968 	default:
5969 		break;
5970 	}
5971 	return false;
5972 }
5973 
5974 /*
5975  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5976  * is_guest_mode (L2).
5977  */
5978 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
5979 				     union vmx_exit_reason exit_reason)
5980 {
5981 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5982 	u32 intr_info;
5983 
5984 	switch ((u16)exit_reason.basic) {
5985 	case EXIT_REASON_EXCEPTION_NMI:
5986 		intr_info = vmx_get_intr_info(vcpu);
5987 		if (is_nmi(intr_info))
5988 			return true;
5989 		else if (is_page_fault(intr_info))
5990 			return true;
5991 		return vmcs12->exception_bitmap &
5992 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5993 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5994 		return nested_exit_on_intr(vcpu);
5995 	case EXIT_REASON_TRIPLE_FAULT:
5996 		return true;
5997 	case EXIT_REASON_INTERRUPT_WINDOW:
5998 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5999 	case EXIT_REASON_NMI_WINDOW:
6000 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
6001 	case EXIT_REASON_TASK_SWITCH:
6002 		return true;
6003 	case EXIT_REASON_CPUID:
6004 		return true;
6005 	case EXIT_REASON_HLT:
6006 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6007 	case EXIT_REASON_INVD:
6008 		return true;
6009 	case EXIT_REASON_INVLPG:
6010 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6011 	case EXIT_REASON_RDPMC:
6012 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6013 	case EXIT_REASON_RDRAND:
6014 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
6015 	case EXIT_REASON_RDSEED:
6016 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
6017 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
6018 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6019 	case EXIT_REASON_VMREAD:
6020 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6021 			vmcs12->vmread_bitmap);
6022 	case EXIT_REASON_VMWRITE:
6023 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6024 			vmcs12->vmwrite_bitmap);
6025 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6026 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6027 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
6028 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6029 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
6030 		/*
6031 		 * VMX instructions trap unconditionally. This allows L1 to
6032 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6033 		 */
6034 		return true;
6035 	case EXIT_REASON_CR_ACCESS:
6036 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6037 	case EXIT_REASON_DR_ACCESS:
6038 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6039 	case EXIT_REASON_IO_INSTRUCTION:
6040 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
6041 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6042 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6043 	case EXIT_REASON_MSR_READ:
6044 	case EXIT_REASON_MSR_WRITE:
6045 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6046 	case EXIT_REASON_INVALID_STATE:
6047 		return true;
6048 	case EXIT_REASON_MWAIT_INSTRUCTION:
6049 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6050 	case EXIT_REASON_MONITOR_TRAP_FLAG:
6051 		return nested_vmx_exit_handled_mtf(vmcs12);
6052 	case EXIT_REASON_MONITOR_INSTRUCTION:
6053 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6054 	case EXIT_REASON_PAUSE_INSTRUCTION:
6055 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6056 			nested_cpu_has2(vmcs12,
6057 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6058 	case EXIT_REASON_MCE_DURING_VMENTRY:
6059 		return true;
6060 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6061 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6062 	case EXIT_REASON_APIC_ACCESS:
6063 	case EXIT_REASON_APIC_WRITE:
6064 	case EXIT_REASON_EOI_INDUCED:
6065 		/*
6066 		 * The controls for "virtualize APIC accesses," "APIC-
6067 		 * register virtualization," and "virtual-interrupt
6068 		 * delivery" only come from vmcs12.
6069 		 */
6070 		return true;
6071 	case EXIT_REASON_INVPCID:
6072 		return
6073 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6074 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6075 	case EXIT_REASON_WBINVD:
6076 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6077 	case EXIT_REASON_XSETBV:
6078 		return true;
6079 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6080 		/*
6081 		 * This should never happen, since it is not possible to
6082 		 * set XSS to a non-zero value---neither in L1 nor in L2.
6083 		 * If if it were, XSS would have to be checked against
6084 		 * the XSS exit bitmap in vmcs12.
6085 		 */
6086 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
6087 	case EXIT_REASON_UMWAIT:
6088 	case EXIT_REASON_TPAUSE:
6089 		return nested_cpu_has2(vmcs12,
6090 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6091 	case EXIT_REASON_ENCLS:
6092 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6093 	default:
6094 		return true;
6095 	}
6096 }
6097 
6098 /*
6099  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6100  * reflected into L1.
6101  */
6102 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6103 {
6104 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6105 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6106 	unsigned long exit_qual;
6107 	u32 exit_intr_info;
6108 
6109 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6110 
6111 	/*
6112 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6113 	 * has already loaded L2's state.
6114 	 */
6115 	if (unlikely(vmx->fail)) {
6116 		trace_kvm_nested_vmenter_failed(
6117 			"hardware VM-instruction error: ",
6118 			vmcs_read32(VM_INSTRUCTION_ERROR));
6119 		exit_intr_info = 0;
6120 		exit_qual = 0;
6121 		goto reflect_vmexit;
6122 	}
6123 
6124 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6125 
6126 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6127 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6128 		return false;
6129 
6130 	/* If L1 doesn't want the exit, handle it in L0. */
6131 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6132 		return false;
6133 
6134 	/*
6135 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6136 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6137 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6138 	 * interrupts are never reflected to L1 so it's a non-issue.
6139 	 */
6140 	exit_intr_info = vmx_get_intr_info(vcpu);
6141 	if (is_exception_with_error_code(exit_intr_info)) {
6142 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6143 
6144 		vmcs12->vm_exit_intr_error_code =
6145 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6146 	}
6147 	exit_qual = vmx_get_exit_qual(vcpu);
6148 
6149 reflect_vmexit:
6150 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6151 	return true;
6152 }
6153 
6154 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6155 				struct kvm_nested_state __user *user_kvm_nested_state,
6156 				u32 user_data_size)
6157 {
6158 	struct vcpu_vmx *vmx;
6159 	struct vmcs12 *vmcs12;
6160 	struct kvm_nested_state kvm_state = {
6161 		.flags = 0,
6162 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6163 		.size = sizeof(kvm_state),
6164 		.hdr.vmx.flags = 0,
6165 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6166 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6167 		.hdr.vmx.preemption_timer_deadline = 0,
6168 	};
6169 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6170 		&user_kvm_nested_state->data.vmx[0];
6171 
6172 	if (!vcpu)
6173 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6174 
6175 	vmx = to_vmx(vcpu);
6176 	vmcs12 = get_vmcs12(vcpu);
6177 
6178 	if (nested_vmx_allowed(vcpu) &&
6179 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6180 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6181 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6182 
6183 		if (vmx_has_valid_vmcs12(vcpu)) {
6184 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6185 
6186 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6187 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6188 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6189 
6190 			if (is_guest_mode(vcpu) &&
6191 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6192 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6193 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6194 		}
6195 
6196 		if (vmx->nested.smm.vmxon)
6197 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6198 
6199 		if (vmx->nested.smm.guest_mode)
6200 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6201 
6202 		if (is_guest_mode(vcpu)) {
6203 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6204 
6205 			if (vmx->nested.nested_run_pending)
6206 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6207 
6208 			if (vmx->nested.mtf_pending)
6209 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6210 
6211 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6212 			    vmx->nested.has_preemption_timer_deadline) {
6213 				kvm_state.hdr.vmx.flags |=
6214 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6215 				kvm_state.hdr.vmx.preemption_timer_deadline =
6216 					vmx->nested.preemption_timer_deadline;
6217 			}
6218 		}
6219 	}
6220 
6221 	if (user_data_size < kvm_state.size)
6222 		goto out;
6223 
6224 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6225 		return -EFAULT;
6226 
6227 	if (!vmx_has_valid_vmcs12(vcpu))
6228 		goto out;
6229 
6230 	/*
6231 	 * When running L2, the authoritative vmcs12 state is in the
6232 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6233 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6234 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6235 	 * vmcs12 state is in the vmcs12 already.
6236 	 */
6237 	if (is_guest_mode(vcpu)) {
6238 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6239 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6240 	} else  {
6241 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6242 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6243 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6244 				/*
6245 				 * L1 hypervisor is not obliged to keep eVMCS
6246 				 * clean fields data always up-to-date while
6247 				 * not in guest mode, 'hv_clean_fields' is only
6248 				 * supposed to be actual upon vmentry so we need
6249 				 * to ignore it here and do full copy.
6250 				 */
6251 				copy_enlightened_to_vmcs12(vmx, 0);
6252 			else if (enable_shadow_vmcs)
6253 				copy_shadow_to_vmcs12(vmx);
6254 		}
6255 	}
6256 
6257 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6258 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6259 
6260 	/*
6261 	 * Copy over the full allocated size of vmcs12 rather than just the size
6262 	 * of the struct.
6263 	 */
6264 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6265 		return -EFAULT;
6266 
6267 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6268 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6269 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6270 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6271 			return -EFAULT;
6272 	}
6273 out:
6274 	return kvm_state.size;
6275 }
6276 
6277 /*
6278  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6279  */
6280 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6281 {
6282 	if (is_guest_mode(vcpu)) {
6283 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6284 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6285 	}
6286 	free_nested(vcpu);
6287 }
6288 
6289 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6290 				struct kvm_nested_state __user *user_kvm_nested_state,
6291 				struct kvm_nested_state *kvm_state)
6292 {
6293 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6294 	struct vmcs12 *vmcs12;
6295 	enum vm_entry_failure_code ignored;
6296 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6297 		&user_kvm_nested_state->data.vmx[0];
6298 	int ret;
6299 
6300 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6301 		return -EINVAL;
6302 
6303 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6304 		if (kvm_state->hdr.vmx.smm.flags)
6305 			return -EINVAL;
6306 
6307 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6308 			return -EINVAL;
6309 
6310 		/*
6311 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6312 		 * enable eVMCS capability on vCPU. However, since then
6313 		 * code was changed such that flag signals vmcs12 should
6314 		 * be copied into eVMCS in guest memory.
6315 		 *
6316 		 * To preserve backwards compatability, allow user
6317 		 * to set this flag even when there is no VMXON region.
6318 		 */
6319 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6320 			return -EINVAL;
6321 	} else {
6322 		if (!nested_vmx_allowed(vcpu))
6323 			return -EINVAL;
6324 
6325 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6326 			return -EINVAL;
6327 	}
6328 
6329 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6330 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6331 		return -EINVAL;
6332 
6333 	if (kvm_state->hdr.vmx.smm.flags &
6334 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6335 		return -EINVAL;
6336 
6337 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6338 		return -EINVAL;
6339 
6340 	/*
6341 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6342 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6343 	 * must be zero.
6344 	 */
6345 	if (is_smm(vcpu) ?
6346 		(kvm_state->flags &
6347 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6348 		: kvm_state->hdr.vmx.smm.flags)
6349 		return -EINVAL;
6350 
6351 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6352 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6353 		return -EINVAL;
6354 
6355 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6356 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6357 			return -EINVAL;
6358 
6359 	vmx_leave_nested(vcpu);
6360 
6361 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6362 		return 0;
6363 
6364 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6365 	ret = enter_vmx_operation(vcpu);
6366 	if (ret)
6367 		return ret;
6368 
6369 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6370 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6371 		/* See vmx_has_valid_vmcs12.  */
6372 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6373 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6374 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6375 			return -EINVAL;
6376 		else
6377 			return 0;
6378 	}
6379 
6380 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6381 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6382 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6383 			return -EINVAL;
6384 
6385 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6386 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6387 		/*
6388 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6389 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6390 		 * restored yet. EVMCS will be mapped from
6391 		 * nested_get_vmcs12_pages().
6392 		 */
6393 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6394 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6395 	} else {
6396 		return -EINVAL;
6397 	}
6398 
6399 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6400 		vmx->nested.smm.vmxon = true;
6401 		vmx->nested.vmxon = false;
6402 
6403 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6404 			vmx->nested.smm.guest_mode = true;
6405 	}
6406 
6407 	vmcs12 = get_vmcs12(vcpu);
6408 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6409 		return -EFAULT;
6410 
6411 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6412 		return -EINVAL;
6413 
6414 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6415 		return 0;
6416 
6417 	vmx->nested.nested_run_pending =
6418 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6419 
6420 	vmx->nested.mtf_pending =
6421 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6422 
6423 	ret = -EINVAL;
6424 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6425 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6426 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6427 
6428 		if (kvm_state->size <
6429 		    sizeof(*kvm_state) +
6430 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6431 			goto error_guest_mode;
6432 
6433 		if (copy_from_user(shadow_vmcs12,
6434 				   user_vmx_nested_state->shadow_vmcs12,
6435 				   sizeof(*shadow_vmcs12))) {
6436 			ret = -EFAULT;
6437 			goto error_guest_mode;
6438 		}
6439 
6440 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6441 		    !shadow_vmcs12->hdr.shadow_vmcs)
6442 			goto error_guest_mode;
6443 	}
6444 
6445 	vmx->nested.has_preemption_timer_deadline = false;
6446 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6447 		vmx->nested.has_preemption_timer_deadline = true;
6448 		vmx->nested.preemption_timer_deadline =
6449 			kvm_state->hdr.vmx.preemption_timer_deadline;
6450 	}
6451 
6452 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6453 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6454 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6455 		goto error_guest_mode;
6456 
6457 	vmx->nested.dirty_vmcs12 = true;
6458 	vmx->nested.force_msr_bitmap_recalc = true;
6459 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6460 	if (ret)
6461 		goto error_guest_mode;
6462 
6463 	return 0;
6464 
6465 error_guest_mode:
6466 	vmx->nested.nested_run_pending = 0;
6467 	return ret;
6468 }
6469 
6470 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6471 {
6472 	if (enable_shadow_vmcs) {
6473 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6474 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6475 	}
6476 }
6477 
6478 /*
6479  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6480  * that madness to get the encoding for comparison.
6481  */
6482 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6483 
6484 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6485 {
6486 	/*
6487 	 * Note these are the so called "index" of the VMCS field encoding, not
6488 	 * the index into vmcs12.
6489 	 */
6490 	unsigned int max_idx, idx;
6491 	int i;
6492 
6493 	/*
6494 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6495 	 * vmcs12, regardless of whether or not the associated feature is
6496 	 * exposed to L1.  Simply find the field with the highest index.
6497 	 */
6498 	max_idx = 0;
6499 	for (i = 0; i < nr_vmcs12_fields; i++) {
6500 		/* The vmcs12 table is very, very sparsely populated. */
6501 		if (!vmcs12_field_offsets[i])
6502 			continue;
6503 
6504 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6505 		if (idx > max_idx)
6506 			max_idx = idx;
6507 	}
6508 
6509 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6510 }
6511 
6512 /*
6513  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6514  * returned for the various VMX controls MSRs when nested VMX is enabled.
6515  * The same values should also be used to verify that vmcs12 control fields are
6516  * valid during nested entry from L1 to L2.
6517  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6518  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6519  * bit in the high half is on if the corresponding bit in the control field
6520  * may be on. See also vmx_control_verify().
6521  */
6522 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6523 {
6524 	/*
6525 	 * Note that as a general rule, the high half of the MSRs (bits in
6526 	 * the control fields which may be 1) should be initialized by the
6527 	 * intersection of the underlying hardware's MSR (i.e., features which
6528 	 * can be supported) and the list of features we want to expose -
6529 	 * because they are known to be properly supported in our code.
6530 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6531 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6532 	 * reason is that if one of these bits is necessary, it will appear
6533 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6534 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6535 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6536 	 * These rules have exceptions below.
6537 	 */
6538 
6539 	/* pin-based controls */
6540 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6541 		msrs->pinbased_ctls_low,
6542 		msrs->pinbased_ctls_high);
6543 	msrs->pinbased_ctls_low |=
6544 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6545 	msrs->pinbased_ctls_high &=
6546 		PIN_BASED_EXT_INTR_MASK |
6547 		PIN_BASED_NMI_EXITING |
6548 		PIN_BASED_VIRTUAL_NMIS |
6549 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6550 	msrs->pinbased_ctls_high |=
6551 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6552 		PIN_BASED_VMX_PREEMPTION_TIMER;
6553 
6554 	/* exit controls */
6555 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6556 		msrs->exit_ctls_low,
6557 		msrs->exit_ctls_high);
6558 	msrs->exit_ctls_low =
6559 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6560 
6561 	msrs->exit_ctls_high &=
6562 #ifdef CONFIG_X86_64
6563 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6564 #endif
6565 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6566 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6567 	msrs->exit_ctls_high |=
6568 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6569 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6570 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6571 
6572 	/* We support free control of debug control saving. */
6573 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6574 
6575 	/* entry controls */
6576 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6577 		msrs->entry_ctls_low,
6578 		msrs->entry_ctls_high);
6579 	msrs->entry_ctls_low =
6580 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6581 	msrs->entry_ctls_high &=
6582 #ifdef CONFIG_X86_64
6583 		VM_ENTRY_IA32E_MODE |
6584 #endif
6585 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6586 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6587 	msrs->entry_ctls_high |=
6588 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6589 
6590 	/* We support free control of debug control loading. */
6591 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6592 
6593 	/* cpu-based controls */
6594 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6595 		msrs->procbased_ctls_low,
6596 		msrs->procbased_ctls_high);
6597 	msrs->procbased_ctls_low =
6598 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6599 	msrs->procbased_ctls_high &=
6600 		CPU_BASED_INTR_WINDOW_EXITING |
6601 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6602 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6603 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6604 		CPU_BASED_CR3_STORE_EXITING |
6605 #ifdef CONFIG_X86_64
6606 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6607 #endif
6608 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6609 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6610 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6611 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6612 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6613 	/*
6614 	 * We can allow some features even when not supported by the
6615 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6616 	 * can use it to avoid exits to L1 - even when L0 runs L2
6617 	 * without MSR bitmaps.
6618 	 */
6619 	msrs->procbased_ctls_high |=
6620 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6621 		CPU_BASED_USE_MSR_BITMAPS;
6622 
6623 	/* We support free control of CR3 access interception. */
6624 	msrs->procbased_ctls_low &=
6625 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6626 
6627 	/*
6628 	 * secondary cpu-based controls.  Do not include those that
6629 	 * depend on CPUID bits, they are added later by
6630 	 * vmx_vcpu_after_set_cpuid.
6631 	 */
6632 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6633 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6634 		      msrs->secondary_ctls_low,
6635 		      msrs->secondary_ctls_high);
6636 
6637 	msrs->secondary_ctls_low = 0;
6638 	msrs->secondary_ctls_high &=
6639 		SECONDARY_EXEC_DESC |
6640 		SECONDARY_EXEC_ENABLE_RDTSCP |
6641 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6642 		SECONDARY_EXEC_WBINVD_EXITING |
6643 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6644 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6645 		SECONDARY_EXEC_RDRAND_EXITING |
6646 		SECONDARY_EXEC_ENABLE_INVPCID |
6647 		SECONDARY_EXEC_RDSEED_EXITING |
6648 		SECONDARY_EXEC_XSAVES |
6649 		SECONDARY_EXEC_TSC_SCALING;
6650 
6651 	/*
6652 	 * We can emulate "VMCS shadowing," even if the hardware
6653 	 * doesn't support it.
6654 	 */
6655 	msrs->secondary_ctls_high |=
6656 		SECONDARY_EXEC_SHADOW_VMCS;
6657 
6658 	if (enable_ept) {
6659 		/* nested EPT: emulate EPT also to L1 */
6660 		msrs->secondary_ctls_high |=
6661 			SECONDARY_EXEC_ENABLE_EPT;
6662 		msrs->ept_caps =
6663 			VMX_EPT_PAGE_WALK_4_BIT |
6664 			VMX_EPT_PAGE_WALK_5_BIT |
6665 			VMX_EPTP_WB_BIT |
6666 			VMX_EPT_INVEPT_BIT |
6667 			VMX_EPT_EXECUTE_ONLY_BIT;
6668 
6669 		msrs->ept_caps &= ept_caps;
6670 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6671 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6672 			VMX_EPT_1GB_PAGE_BIT;
6673 		if (enable_ept_ad_bits) {
6674 			msrs->secondary_ctls_high |=
6675 				SECONDARY_EXEC_ENABLE_PML;
6676 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6677 		}
6678 	}
6679 
6680 	if (cpu_has_vmx_vmfunc()) {
6681 		msrs->secondary_ctls_high |=
6682 			SECONDARY_EXEC_ENABLE_VMFUNC;
6683 		/*
6684 		 * Advertise EPTP switching unconditionally
6685 		 * since we emulate it
6686 		 */
6687 		if (enable_ept)
6688 			msrs->vmfunc_controls =
6689 				VMX_VMFUNC_EPTP_SWITCHING;
6690 	}
6691 
6692 	/*
6693 	 * Old versions of KVM use the single-context version without
6694 	 * checking for support, so declare that it is supported even
6695 	 * though it is treated as global context.  The alternative is
6696 	 * not failing the single-context invvpid, and it is worse.
6697 	 */
6698 	if (enable_vpid) {
6699 		msrs->secondary_ctls_high |=
6700 			SECONDARY_EXEC_ENABLE_VPID;
6701 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6702 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6703 	}
6704 
6705 	if (enable_unrestricted_guest)
6706 		msrs->secondary_ctls_high |=
6707 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6708 
6709 	if (flexpriority_enabled)
6710 		msrs->secondary_ctls_high |=
6711 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6712 
6713 	if (enable_sgx)
6714 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6715 
6716 	/* miscellaneous data */
6717 	rdmsr(MSR_IA32_VMX_MISC,
6718 		msrs->misc_low,
6719 		msrs->misc_high);
6720 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6721 	msrs->misc_low |=
6722 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6723 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6724 		VMX_MISC_ACTIVITY_HLT |
6725 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6726 	msrs->misc_high = 0;
6727 
6728 	/*
6729 	 * This MSR reports some information about VMX support. We
6730 	 * should return information about the VMX we emulate for the
6731 	 * guest, and the VMCS structure we give it - not about the
6732 	 * VMX support of the underlying hardware.
6733 	 */
6734 	msrs->basic =
6735 		VMCS12_REVISION |
6736 		VMX_BASIC_TRUE_CTLS |
6737 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6738 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6739 
6740 	if (cpu_has_vmx_basic_inout())
6741 		msrs->basic |= VMX_BASIC_INOUT;
6742 
6743 	/*
6744 	 * These MSRs specify bits which the guest must keep fixed on
6745 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6746 	 * We picked the standard core2 setting.
6747 	 */
6748 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6749 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6750 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6751 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6752 
6753 	/* These MSRs specify bits which the guest must keep fixed off. */
6754 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6755 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6756 
6757 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
6758 }
6759 
6760 void nested_vmx_hardware_unsetup(void)
6761 {
6762 	int i;
6763 
6764 	if (enable_shadow_vmcs) {
6765 		for (i = 0; i < VMX_BITMAP_NR; i++)
6766 			free_page((unsigned long)vmx_bitmap[i]);
6767 	}
6768 }
6769 
6770 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6771 {
6772 	int i;
6773 
6774 	if (!cpu_has_vmx_shadow_vmcs())
6775 		enable_shadow_vmcs = 0;
6776 	if (enable_shadow_vmcs) {
6777 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6778 			/*
6779 			 * The vmx_bitmap is not tied to a VM and so should
6780 			 * not be charged to a memcg.
6781 			 */
6782 			vmx_bitmap[i] = (unsigned long *)
6783 				__get_free_page(GFP_KERNEL);
6784 			if (!vmx_bitmap[i]) {
6785 				nested_vmx_hardware_unsetup();
6786 				return -ENOMEM;
6787 			}
6788 		}
6789 
6790 		init_vmcs_shadow_fields();
6791 	}
6792 
6793 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6794 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6795 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6796 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6797 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6798 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6799 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6800 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6801 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6802 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6803 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6804 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6805 
6806 	return 0;
6807 }
6808 
6809 struct kvm_x86_nested_ops vmx_nested_ops = {
6810 	.leave_nested = vmx_leave_nested,
6811 	.check_events = vmx_check_nested_events,
6812 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6813 	.triple_fault = nested_vmx_triple_fault,
6814 	.get_state = vmx_get_nested_state,
6815 	.set_state = vmx_set_nested_state,
6816 	.get_nested_state_pages = vmx_get_nested_state_pages,
6817 	.write_log_dirty = nested_vmx_write_pml_buffer,
6818 	.enable_evmcs = nested_enable_evmcs,
6819 	.get_evmcs_version = nested_get_evmcs_version,
6820 };
6821