xref: /linux/arch/x86/kvm/vmx/nested.c (revision 46ff24efe04ac96a129dd01138640c3447a525e1)
1 // SPDX-License-Identifier: GPL-2.0
2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3 
4 #include <linux/objtool.h>
5 #include <linux/percpu.h>
6 
7 #include <asm/debugreg.h>
8 #include <asm/mmu_context.h>
9 
10 #include "cpuid.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 #include "smm.h"
20 
21 static bool __read_mostly enable_shadow_vmcs = 1;
22 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
23 
24 static bool __read_mostly nested_early_check = 0;
25 module_param(nested_early_check, bool, S_IRUGO);
26 
27 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
28 
29 /*
30  * Hyper-V requires all of these, so mark them as supported even though
31  * they are just treated the same as all-context.
32  */
33 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
34 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
35 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
36 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
37 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
38 
39 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
40 
41 enum {
42 	VMX_VMREAD_BITMAP,
43 	VMX_VMWRITE_BITMAP,
44 	VMX_BITMAP_NR
45 };
46 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
47 
48 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
49 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
50 
51 struct shadow_vmcs_field {
52 	u16	encoding;
53 	u16	offset;
54 };
55 static struct shadow_vmcs_field shadow_read_only_fields[] = {
56 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
57 #include "vmcs_shadow_fields.h"
58 };
59 static int max_shadow_read_only_fields =
60 	ARRAY_SIZE(shadow_read_only_fields);
61 
62 static struct shadow_vmcs_field shadow_read_write_fields[] = {
63 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
64 #include "vmcs_shadow_fields.h"
65 };
66 static int max_shadow_read_write_fields =
67 	ARRAY_SIZE(shadow_read_write_fields);
68 
69 static void init_vmcs_shadow_fields(void)
70 {
71 	int i, j;
72 
73 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
74 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
75 
76 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
77 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
78 		u16 field = entry.encoding;
79 
80 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
81 		    (i + 1 == max_shadow_read_only_fields ||
82 		     shadow_read_only_fields[i + 1].encoding != field + 1))
83 			pr_err("Missing field from shadow_read_only_field %x\n",
84 			       field + 1);
85 
86 		clear_bit(field, vmx_vmread_bitmap);
87 		if (field & 1)
88 #ifdef CONFIG_X86_64
89 			continue;
90 #else
91 			entry.offset += sizeof(u32);
92 #endif
93 		shadow_read_only_fields[j++] = entry;
94 	}
95 	max_shadow_read_only_fields = j;
96 
97 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
98 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
99 		u16 field = entry.encoding;
100 
101 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
102 		    (i + 1 == max_shadow_read_write_fields ||
103 		     shadow_read_write_fields[i + 1].encoding != field + 1))
104 			pr_err("Missing field from shadow_read_write_field %x\n",
105 			       field + 1);
106 
107 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
108 			  field <= GUEST_TR_AR_BYTES,
109 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
110 
111 		/*
112 		 * PML and the preemption timer can be emulated, but the
113 		 * processor cannot vmwrite to fields that don't exist
114 		 * on bare metal.
115 		 */
116 		switch (field) {
117 		case GUEST_PML_INDEX:
118 			if (!cpu_has_vmx_pml())
119 				continue;
120 			break;
121 		case VMX_PREEMPTION_TIMER_VALUE:
122 			if (!cpu_has_vmx_preemption_timer())
123 				continue;
124 			break;
125 		case GUEST_INTR_STATUS:
126 			if (!cpu_has_vmx_apicv())
127 				continue;
128 			break;
129 		default:
130 			break;
131 		}
132 
133 		clear_bit(field, vmx_vmwrite_bitmap);
134 		clear_bit(field, vmx_vmread_bitmap);
135 		if (field & 1)
136 #ifdef CONFIG_X86_64
137 			continue;
138 #else
139 			entry.offset += sizeof(u32);
140 #endif
141 		shadow_read_write_fields[j++] = entry;
142 	}
143 	max_shadow_read_write_fields = j;
144 }
145 
146 /*
147  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
148  * set the success or error code of an emulated VMX instruction (as specified
149  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
150  * instruction.
151  */
152 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
153 {
154 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
155 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
156 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
157 	return kvm_skip_emulated_instruction(vcpu);
158 }
159 
160 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
161 {
162 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
163 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
164 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
165 			| X86_EFLAGS_CF);
166 	return kvm_skip_emulated_instruction(vcpu);
167 }
168 
169 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
170 				u32 vm_instruction_error)
171 {
172 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
173 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
174 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
175 			| X86_EFLAGS_ZF);
176 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
177 	/*
178 	 * We don't need to force sync to shadow VMCS because
179 	 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
180 	 * fields and thus must be synced.
181 	 */
182 	if (to_vmx(vcpu)->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
183 		to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
184 
185 	return kvm_skip_emulated_instruction(vcpu);
186 }
187 
188 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
189 {
190 	struct vcpu_vmx *vmx = to_vmx(vcpu);
191 
192 	/*
193 	 * failValid writes the error number to the current VMCS, which
194 	 * can't be done if there isn't a current VMCS.
195 	 */
196 	if (vmx->nested.current_vmptr == INVALID_GPA &&
197 	    !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
198 		return nested_vmx_failInvalid(vcpu);
199 
200 	return nested_vmx_failValid(vcpu, vm_instruction_error);
201 }
202 
203 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
204 {
205 	/* TODO: not to reset guest simply here. */
206 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
207 	pr_debug_ratelimited("nested vmx abort, indicator %d\n", indicator);
208 }
209 
210 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
211 {
212 	return fixed_bits_valid(control, low, high);
213 }
214 
215 static inline u64 vmx_control_msr(u32 low, u32 high)
216 {
217 	return low | ((u64)high << 32);
218 }
219 
220 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
221 {
222 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
223 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
224 	vmx->nested.need_vmcs12_to_shadow_sync = false;
225 }
226 
227 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
228 {
229 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
230 	struct vcpu_vmx *vmx = to_vmx(vcpu);
231 
232 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
233 		kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
234 		vmx->nested.hv_evmcs = NULL;
235 	}
236 
237 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
238 
239 	if (hv_vcpu) {
240 		hv_vcpu->nested.pa_page_gpa = INVALID_GPA;
241 		hv_vcpu->nested.vm_id = 0;
242 		hv_vcpu->nested.vp_id = 0;
243 	}
244 }
245 
246 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
247 				     struct loaded_vmcs *prev)
248 {
249 	struct vmcs_host_state *dest, *src;
250 
251 	if (unlikely(!vmx->guest_state_loaded))
252 		return;
253 
254 	src = &prev->host_state;
255 	dest = &vmx->loaded_vmcs->host_state;
256 
257 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
258 	dest->ldt_sel = src->ldt_sel;
259 #ifdef CONFIG_X86_64
260 	dest->ds_sel = src->ds_sel;
261 	dest->es_sel = src->es_sel;
262 #endif
263 }
264 
265 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
266 {
267 	struct vcpu_vmx *vmx = to_vmx(vcpu);
268 	struct loaded_vmcs *prev;
269 	int cpu;
270 
271 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
272 		return;
273 
274 	cpu = get_cpu();
275 	prev = vmx->loaded_vmcs;
276 	vmx->loaded_vmcs = vmcs;
277 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
278 	vmx_sync_vmcs_host_state(vmx, prev);
279 	put_cpu();
280 
281 	vcpu->arch.regs_avail = ~VMX_REGS_LAZY_LOAD_SET;
282 
283 	/*
284 	 * All lazily updated registers will be reloaded from VMCS12 on both
285 	 * vmentry and vmexit.
286 	 */
287 	vcpu->arch.regs_dirty = 0;
288 }
289 
290 /*
291  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
292  * just stops using VMX.
293  */
294 static void free_nested(struct kvm_vcpu *vcpu)
295 {
296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
297 
298 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
299 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
300 
301 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
302 		return;
303 
304 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
305 
306 	vmx->nested.vmxon = false;
307 	vmx->nested.smm.vmxon = false;
308 	vmx->nested.vmxon_ptr = INVALID_GPA;
309 	free_vpid(vmx->nested.vpid02);
310 	vmx->nested.posted_intr_nv = -1;
311 	vmx->nested.current_vmptr = INVALID_GPA;
312 	if (enable_shadow_vmcs) {
313 		vmx_disable_shadow_vmcs(vmx);
314 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
315 		free_vmcs(vmx->vmcs01.shadow_vmcs);
316 		vmx->vmcs01.shadow_vmcs = NULL;
317 	}
318 	kfree(vmx->nested.cached_vmcs12);
319 	vmx->nested.cached_vmcs12 = NULL;
320 	kfree(vmx->nested.cached_shadow_vmcs12);
321 	vmx->nested.cached_shadow_vmcs12 = NULL;
322 	/*
323 	 * Unpin physical memory we referred to in the vmcs02.  The APIC access
324 	 * page's backing page (yeah, confusing) shouldn't actually be accessed,
325 	 * and if it is written, the contents are irrelevant.
326 	 */
327 	kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
328 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
329 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
330 	vmx->nested.pi_desc = NULL;
331 
332 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
333 
334 	nested_release_evmcs(vcpu);
335 
336 	free_loaded_vmcs(&vmx->nested.vmcs02);
337 }
338 
339 /*
340  * Ensure that the current vmcs of the logical processor is the
341  * vmcs01 of the vcpu before calling free_nested().
342  */
343 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
344 {
345 	vcpu_load(vcpu);
346 	vmx_leave_nested(vcpu);
347 	vcpu_put(vcpu);
348 }
349 
350 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
351 
352 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
353 {
354 	return VALID_PAGE(root_hpa) &&
355 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
356 }
357 
358 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
359 				       gpa_t addr)
360 {
361 	uint i;
362 	struct kvm_mmu_root_info *cached_root;
363 
364 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
365 
366 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
367 		cached_root = &vcpu->arch.mmu->prev_roots[i];
368 
369 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
370 					    eptp))
371 			vcpu->arch.mmu->invlpg(vcpu, addr, cached_root->hpa);
372 	}
373 }
374 
375 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
376 		struct x86_exception *fault)
377 {
378 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
379 	struct vcpu_vmx *vmx = to_vmx(vcpu);
380 	u32 vm_exit_reason;
381 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
382 
383 	if (vmx->nested.pml_full) {
384 		vm_exit_reason = EXIT_REASON_PML_FULL;
385 		vmx->nested.pml_full = false;
386 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
387 	} else {
388 		if (fault->error_code & PFERR_RSVD_MASK)
389 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
390 		else
391 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
392 
393 		/*
394 		 * Although the caller (kvm_inject_emulated_page_fault) would
395 		 * have already synced the faulting address in the shadow EPT
396 		 * tables for the current EPTP12, we also need to sync it for
397 		 * any other cached EPTP02s based on the same EP4TA, since the
398 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
399 		 */
400 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
401 					   fault->address);
402 	}
403 
404 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
405 	vmcs12->guest_physical_address = fault->address;
406 }
407 
408 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
409 {
410 	struct vcpu_vmx *vmx = to_vmx(vcpu);
411 	bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
412 	int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
413 
414 	kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
415 				nested_ept_ad_enabled(vcpu),
416 				nested_ept_get_eptp(vcpu));
417 }
418 
419 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
420 {
421 	WARN_ON(mmu_is_nested(vcpu));
422 
423 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
424 	nested_ept_new_eptp(vcpu);
425 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
426 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
427 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
428 
429 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
430 }
431 
432 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
433 {
434 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
435 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
436 }
437 
438 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
439 					    u16 error_code)
440 {
441 	bool inequality, bit;
442 
443 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
444 	inequality =
445 		(error_code & vmcs12->page_fault_error_code_mask) !=
446 		 vmcs12->page_fault_error_code_match;
447 	return inequality ^ bit;
448 }
449 
450 static bool nested_vmx_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
451 					   u32 error_code)
452 {
453 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
454 
455 	/*
456 	 * Drop bits 31:16 of the error code when performing the #PF mask+match
457 	 * check.  All VMCS fields involved are 32 bits, but Intel CPUs never
458 	 * set bits 31:16 and VMX disallows setting bits 31:16 in the injected
459 	 * error code.  Including the to-be-dropped bits in the check might
460 	 * result in an "impossible" or missed exit from L1's perspective.
461 	 */
462 	if (vector == PF_VECTOR)
463 		return nested_vmx_is_page_fault_vmexit(vmcs12, (u16)error_code);
464 
465 	return (vmcs12->exception_bitmap & (1u << vector));
466 }
467 
468 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
469 					       struct vmcs12 *vmcs12)
470 {
471 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
472 		return 0;
473 
474 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
475 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
476 		return -EINVAL;
477 
478 	return 0;
479 }
480 
481 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
482 						struct vmcs12 *vmcs12)
483 {
484 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
485 		return 0;
486 
487 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
488 		return -EINVAL;
489 
490 	return 0;
491 }
492 
493 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
494 						struct vmcs12 *vmcs12)
495 {
496 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
497 		return 0;
498 
499 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
500 		return -EINVAL;
501 
502 	return 0;
503 }
504 
505 /*
506  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
507  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
508  * only the "disable intercept" case needs to be handled.
509  */
510 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
511 							unsigned long *msr_bitmap_l0,
512 							u32 msr, int type)
513 {
514 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
515 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
516 
517 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
518 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
519 }
520 
521 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
522 {
523 	int msr;
524 
525 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
526 		unsigned word = msr / BITS_PER_LONG;
527 
528 		msr_bitmap[word] = ~0;
529 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
530 	}
531 }
532 
533 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
534 static inline									\
535 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
536 					 unsigned long *msr_bitmap_l1,		\
537 					 unsigned long *msr_bitmap_l0, u32 msr)	\
538 {										\
539 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
540 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
541 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
542 	else									\
543 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
544 }
545 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
546 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
547 
548 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
549 						    unsigned long *msr_bitmap_l1,
550 						    unsigned long *msr_bitmap_l0,
551 						    u32 msr, int types)
552 {
553 	if (types & MSR_TYPE_R)
554 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
555 						  msr_bitmap_l0, msr);
556 	if (types & MSR_TYPE_W)
557 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
558 						   msr_bitmap_l0, msr);
559 }
560 
561 /*
562  * Merge L0's and L1's MSR bitmap, return false to indicate that
563  * we do not use the hardware.
564  */
565 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
566 						 struct vmcs12 *vmcs12)
567 {
568 	struct vcpu_vmx *vmx = to_vmx(vcpu);
569 	int msr;
570 	unsigned long *msr_bitmap_l1;
571 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
572 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
573 	struct kvm_host_map *map = &vmx->nested.msr_bitmap_map;
574 
575 	/* Nothing to do if the MSR bitmap is not in use.  */
576 	if (!cpu_has_vmx_msr_bitmap() ||
577 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
578 		return false;
579 
580 	/*
581 	 * MSR bitmap update can be skipped when:
582 	 * - MSR bitmap for L1 hasn't changed.
583 	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
584 	 *   before.
585 	 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
586 	 *   and tells KVM (L0) there were no changes in MSR bitmap for L2.
587 	 */
588 	if (!vmx->nested.force_msr_bitmap_recalc && evmcs &&
589 	    evmcs->hv_enlightenments_control.msr_bitmap &&
590 	    evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
591 		return true;
592 
593 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
594 		return false;
595 
596 	msr_bitmap_l1 = (unsigned long *)map->hva;
597 
598 	/*
599 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
600 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
601 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
602 	 */
603 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
604 
605 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
606 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
607 			/*
608 			 * L0 need not intercept reads for MSRs between 0x800
609 			 * and 0x8ff, it just lets the processor take the value
610 			 * from the virtual-APIC page; take those 256 bits
611 			 * directly from the L1 bitmap.
612 			 */
613 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
614 				unsigned word = msr / BITS_PER_LONG;
615 
616 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
617 			}
618 		}
619 
620 		nested_vmx_disable_intercept_for_x2apic_msr(
621 			msr_bitmap_l1, msr_bitmap_l0,
622 			X2APIC_MSR(APIC_TASKPRI),
623 			MSR_TYPE_R | MSR_TYPE_W);
624 
625 		if (nested_cpu_has_vid(vmcs12)) {
626 			nested_vmx_disable_intercept_for_x2apic_msr(
627 				msr_bitmap_l1, msr_bitmap_l0,
628 				X2APIC_MSR(APIC_EOI),
629 				MSR_TYPE_W);
630 			nested_vmx_disable_intercept_for_x2apic_msr(
631 				msr_bitmap_l1, msr_bitmap_l0,
632 				X2APIC_MSR(APIC_SELF_IPI),
633 				MSR_TYPE_W);
634 		}
635 	}
636 
637 	/*
638 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
639 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
640 	 */
641 #ifdef CONFIG_X86_64
642 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
643 					 MSR_FS_BASE, MSR_TYPE_RW);
644 
645 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
646 					 MSR_GS_BASE, MSR_TYPE_RW);
647 
648 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
649 					 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
650 #endif
651 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
652 					 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
653 
654 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
655 					 MSR_IA32_PRED_CMD, MSR_TYPE_W);
656 
657 	kvm_vcpu_unmap(vcpu, &vmx->nested.msr_bitmap_map, false);
658 
659 	vmx->nested.force_msr_bitmap_recalc = false;
660 
661 	return true;
662 }
663 
664 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
665 				       struct vmcs12 *vmcs12)
666 {
667 	struct vcpu_vmx *vmx = to_vmx(vcpu);
668 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
669 
670 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
671 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
672 		return;
673 
674 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
675 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
676 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
677 		return;
678 
679 	kvm_read_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
680 			      VMCS12_SIZE);
681 }
682 
683 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
684 					      struct vmcs12 *vmcs12)
685 {
686 	struct vcpu_vmx *vmx = to_vmx(vcpu);
687 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
688 
689 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
690 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
691 		return;
692 
693 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
694 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
695 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
696 		return;
697 
698 	kvm_write_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
699 			       VMCS12_SIZE);
700 }
701 
702 /*
703  * In nested virtualization, check if L1 has set
704  * VM_EXIT_ACK_INTR_ON_EXIT
705  */
706 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
707 {
708 	return get_vmcs12(vcpu)->vm_exit_controls &
709 		VM_EXIT_ACK_INTR_ON_EXIT;
710 }
711 
712 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
713 					  struct vmcs12 *vmcs12)
714 {
715 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
716 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
717 		return -EINVAL;
718 	else
719 		return 0;
720 }
721 
722 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
723 					   struct vmcs12 *vmcs12)
724 {
725 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
726 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
727 	    !nested_cpu_has_vid(vmcs12) &&
728 	    !nested_cpu_has_posted_intr(vmcs12))
729 		return 0;
730 
731 	/*
732 	 * If virtualize x2apic mode is enabled,
733 	 * virtualize apic access must be disabled.
734 	 */
735 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
736 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
737 		return -EINVAL;
738 
739 	/*
740 	 * If virtual interrupt delivery is enabled,
741 	 * we must exit on external interrupts.
742 	 */
743 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
744 		return -EINVAL;
745 
746 	/*
747 	 * bits 15:8 should be zero in posted_intr_nv,
748 	 * the descriptor address has been already checked
749 	 * in nested_get_vmcs12_pages.
750 	 *
751 	 * bits 5:0 of posted_intr_desc_addr should be zero.
752 	 */
753 	if (nested_cpu_has_posted_intr(vmcs12) &&
754 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
755 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
756 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
757 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
758 		return -EINVAL;
759 
760 	/* tpr shadow is needed by all apicv features. */
761 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
762 		return -EINVAL;
763 
764 	return 0;
765 }
766 
767 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
768 				       u32 count, u64 addr)
769 {
770 	if (count == 0)
771 		return 0;
772 
773 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
774 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
775 		return -EINVAL;
776 
777 	return 0;
778 }
779 
780 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
781 						     struct vmcs12 *vmcs12)
782 {
783 	if (CC(nested_vmx_check_msr_switch(vcpu,
784 					   vmcs12->vm_exit_msr_load_count,
785 					   vmcs12->vm_exit_msr_load_addr)) ||
786 	    CC(nested_vmx_check_msr_switch(vcpu,
787 					   vmcs12->vm_exit_msr_store_count,
788 					   vmcs12->vm_exit_msr_store_addr)))
789 		return -EINVAL;
790 
791 	return 0;
792 }
793 
794 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
795                                                       struct vmcs12 *vmcs12)
796 {
797 	if (CC(nested_vmx_check_msr_switch(vcpu,
798 					   vmcs12->vm_entry_msr_load_count,
799 					   vmcs12->vm_entry_msr_load_addr)))
800                 return -EINVAL;
801 
802 	return 0;
803 }
804 
805 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
806 					 struct vmcs12 *vmcs12)
807 {
808 	if (!nested_cpu_has_pml(vmcs12))
809 		return 0;
810 
811 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
812 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
813 		return -EINVAL;
814 
815 	return 0;
816 }
817 
818 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
819 							struct vmcs12 *vmcs12)
820 {
821 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
822 	       !nested_cpu_has_ept(vmcs12)))
823 		return -EINVAL;
824 	return 0;
825 }
826 
827 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
828 							 struct vmcs12 *vmcs12)
829 {
830 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
831 	       !nested_cpu_has_ept(vmcs12)))
832 		return -EINVAL;
833 	return 0;
834 }
835 
836 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
837 						 struct vmcs12 *vmcs12)
838 {
839 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
840 		return 0;
841 
842 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
843 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
844 		return -EINVAL;
845 
846 	return 0;
847 }
848 
849 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
850 				       struct vmx_msr_entry *e)
851 {
852 	/* x2APIC MSR accesses are not allowed */
853 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
854 		return -EINVAL;
855 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
856 	    CC(e->index == MSR_IA32_UCODE_REV))
857 		return -EINVAL;
858 	if (CC(e->reserved != 0))
859 		return -EINVAL;
860 	return 0;
861 }
862 
863 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
864 				     struct vmx_msr_entry *e)
865 {
866 	if (CC(e->index == MSR_FS_BASE) ||
867 	    CC(e->index == MSR_GS_BASE) ||
868 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
869 	    nested_vmx_msr_check_common(vcpu, e))
870 		return -EINVAL;
871 	return 0;
872 }
873 
874 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
875 				      struct vmx_msr_entry *e)
876 {
877 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
878 	    nested_vmx_msr_check_common(vcpu, e))
879 		return -EINVAL;
880 	return 0;
881 }
882 
883 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
884 {
885 	struct vcpu_vmx *vmx = to_vmx(vcpu);
886 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
887 				       vmx->nested.msrs.misc_high);
888 
889 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
890 }
891 
892 /*
893  * Load guest's/host's msr at nested entry/exit.
894  * return 0 for success, entry index for failure.
895  *
896  * One of the failure modes for MSR load/store is when a list exceeds the
897  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
898  * as possible, process all valid entries before failing rather than precheck
899  * for a capacity violation.
900  */
901 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
902 {
903 	u32 i;
904 	struct vmx_msr_entry e;
905 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
906 
907 	for (i = 0; i < count; i++) {
908 		if (unlikely(i >= max_msr_list_size))
909 			goto fail;
910 
911 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
912 					&e, sizeof(e))) {
913 			pr_debug_ratelimited(
914 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
915 				__func__, i, gpa + i * sizeof(e));
916 			goto fail;
917 		}
918 		if (nested_vmx_load_msr_check(vcpu, &e)) {
919 			pr_debug_ratelimited(
920 				"%s check failed (%u, 0x%x, 0x%x)\n",
921 				__func__, i, e.index, e.reserved);
922 			goto fail;
923 		}
924 		if (kvm_set_msr(vcpu, e.index, e.value)) {
925 			pr_debug_ratelimited(
926 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
927 				__func__, i, e.index, e.value);
928 			goto fail;
929 		}
930 	}
931 	return 0;
932 fail:
933 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
934 	return i + 1;
935 }
936 
937 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
938 					    u32 msr_index,
939 					    u64 *data)
940 {
941 	struct vcpu_vmx *vmx = to_vmx(vcpu);
942 
943 	/*
944 	 * If the L0 hypervisor stored a more accurate value for the TSC that
945 	 * does not include the time taken for emulation of the L2->L1
946 	 * VM-exit in L0, use the more accurate value.
947 	 */
948 	if (msr_index == MSR_IA32_TSC) {
949 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
950 						    MSR_IA32_TSC);
951 
952 		if (i >= 0) {
953 			u64 val = vmx->msr_autostore.guest.val[i].value;
954 
955 			*data = kvm_read_l1_tsc(vcpu, val);
956 			return true;
957 		}
958 	}
959 
960 	if (kvm_get_msr(vcpu, msr_index, data)) {
961 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
962 			msr_index);
963 		return false;
964 	}
965 	return true;
966 }
967 
968 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
969 				     struct vmx_msr_entry *e)
970 {
971 	if (kvm_vcpu_read_guest(vcpu,
972 				gpa + i * sizeof(*e),
973 				e, 2 * sizeof(u32))) {
974 		pr_debug_ratelimited(
975 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
976 			__func__, i, gpa + i * sizeof(*e));
977 		return false;
978 	}
979 	if (nested_vmx_store_msr_check(vcpu, e)) {
980 		pr_debug_ratelimited(
981 			"%s check failed (%u, 0x%x, 0x%x)\n",
982 			__func__, i, e->index, e->reserved);
983 		return false;
984 	}
985 	return true;
986 }
987 
988 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
989 {
990 	u64 data;
991 	u32 i;
992 	struct vmx_msr_entry e;
993 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
994 
995 	for (i = 0; i < count; i++) {
996 		if (unlikely(i >= max_msr_list_size))
997 			return -EINVAL;
998 
999 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1000 			return -EINVAL;
1001 
1002 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1003 			return -EINVAL;
1004 
1005 		if (kvm_vcpu_write_guest(vcpu,
1006 					 gpa + i * sizeof(e) +
1007 					     offsetof(struct vmx_msr_entry, value),
1008 					 &data, sizeof(data))) {
1009 			pr_debug_ratelimited(
1010 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1011 				__func__, i, e.index, data);
1012 			return -EINVAL;
1013 		}
1014 	}
1015 	return 0;
1016 }
1017 
1018 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1019 {
1020 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1021 	u32 count = vmcs12->vm_exit_msr_store_count;
1022 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1023 	struct vmx_msr_entry e;
1024 	u32 i;
1025 
1026 	for (i = 0; i < count; i++) {
1027 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1028 			return false;
1029 
1030 		if (e.index == msr_index)
1031 			return true;
1032 	}
1033 	return false;
1034 }
1035 
1036 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1037 					   u32 msr_index)
1038 {
1039 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1040 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1041 	bool in_vmcs12_store_list;
1042 	int msr_autostore_slot;
1043 	bool in_autostore_list;
1044 	int last;
1045 
1046 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1047 	in_autostore_list = msr_autostore_slot >= 0;
1048 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1049 
1050 	if (in_vmcs12_store_list && !in_autostore_list) {
1051 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1052 			/*
1053 			 * Emulated VMEntry does not fail here.  Instead a less
1054 			 * accurate value will be returned by
1055 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1056 			 * instead of reading the value from the vmcs02 VMExit
1057 			 * MSR-store area.
1058 			 */
1059 			pr_warn_ratelimited(
1060 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1061 				msr_index);
1062 			return;
1063 		}
1064 		last = autostore->nr++;
1065 		autostore->val[last].index = msr_index;
1066 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1067 		last = --autostore->nr;
1068 		autostore->val[msr_autostore_slot] = autostore->val[last];
1069 	}
1070 }
1071 
1072 /*
1073  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1074  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1075  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1076  * @entry_failure_code.
1077  */
1078 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1079 			       bool nested_ept, bool reload_pdptrs,
1080 			       enum vm_entry_failure_code *entry_failure_code)
1081 {
1082 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1083 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1084 		return -EINVAL;
1085 	}
1086 
1087 	/*
1088 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1089 	 * must not be dereferenced.
1090 	 */
1091 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1092 	    CC(!load_pdptrs(vcpu, cr3))) {
1093 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1094 		return -EINVAL;
1095 	}
1096 
1097 	vcpu->arch.cr3 = cr3;
1098 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1099 
1100 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1101 	kvm_init_mmu(vcpu);
1102 
1103 	if (!nested_ept)
1104 		kvm_mmu_new_pgd(vcpu, cr3);
1105 
1106 	return 0;
1107 }
1108 
1109 /*
1110  * Returns if KVM is able to config CPU to tag TLB entries
1111  * populated by L2 differently than TLB entries populated
1112  * by L1.
1113  *
1114  * If L0 uses EPT, L1 and L2 run with different EPTP because
1115  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1116  * are tagged with different EPTP.
1117  *
1118  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1119  * with different VPID (L1 entries are tagged with vmx->vpid
1120  * while L2 entries are tagged with vmx->nested.vpid02).
1121  */
1122 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1123 {
1124 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1125 
1126 	return enable_ept ||
1127 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1128 }
1129 
1130 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1131 					    struct vmcs12 *vmcs12,
1132 					    bool is_vmenter)
1133 {
1134 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1135 
1136 	/*
1137 	 * KVM_REQ_HV_TLB_FLUSH flushes entries from either L1's VP_ID or
1138 	 * L2's VP_ID upon request from the guest. Make sure we check for
1139 	 * pending entries in the right FIFO upon L1/L2 transition as these
1140 	 * requests are put by other vCPUs asynchronously.
1141 	 */
1142 	if (to_hv_vcpu(vcpu) && enable_ept)
1143 		kvm_make_request(KVM_REQ_HV_TLB_FLUSH, vcpu);
1144 
1145 	/*
1146 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1147 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1148 	 * full TLB flush from the guest's perspective.  This is required even
1149 	 * if VPID is disabled in the host as KVM may need to synchronize the
1150 	 * MMU in response to the guest TLB flush.
1151 	 *
1152 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1153 	 * EPT is a special snowflake, as guest-physical mappings aren't
1154 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1155 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1156 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1157 	 * those mappings.
1158 	 */
1159 	if (!nested_cpu_has_vpid(vmcs12)) {
1160 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1161 		return;
1162 	}
1163 
1164 	/* L2 should never have a VPID if VPID is disabled. */
1165 	WARN_ON(!enable_vpid);
1166 
1167 	/*
1168 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1169 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1170 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1171 	 * that the new vpid12 has never been used and thus represents a new
1172 	 * guest ASID that cannot have entries in the TLB.
1173 	 */
1174 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1175 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1176 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1177 		return;
1178 	}
1179 
1180 	/*
1181 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1182 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1183 	 * KVM was unable to allocate a VPID for L2, flush the current context
1184 	 * as the effective ASID is common to both L1 and L2.
1185 	 */
1186 	if (!nested_has_guest_tlb_tag(vcpu))
1187 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1188 }
1189 
1190 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1191 {
1192 	superset &= mask;
1193 	subset &= mask;
1194 
1195 	return (superset | subset) == superset;
1196 }
1197 
1198 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1199 {
1200 	const u64 feature_and_reserved =
1201 		/* feature (except bit 48; see below) */
1202 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1203 		/* reserved */
1204 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1205 	u64 vmx_basic = vmcs_config.nested.basic;
1206 
1207 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1208 		return -EINVAL;
1209 
1210 	/*
1211 	 * KVM does not emulate a version of VMX that constrains physical
1212 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1213 	 */
1214 	if (data & BIT_ULL(48))
1215 		return -EINVAL;
1216 
1217 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1218 	    vmx_basic_vmcs_revision_id(data))
1219 		return -EINVAL;
1220 
1221 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1222 		return -EINVAL;
1223 
1224 	vmx->nested.msrs.basic = data;
1225 	return 0;
1226 }
1227 
1228 static void vmx_get_control_msr(struct nested_vmx_msrs *msrs, u32 msr_index,
1229 				u32 **low, u32 **high)
1230 {
1231 	switch (msr_index) {
1232 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1233 		*low = &msrs->pinbased_ctls_low;
1234 		*high = &msrs->pinbased_ctls_high;
1235 		break;
1236 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1237 		*low = &msrs->procbased_ctls_low;
1238 		*high = &msrs->procbased_ctls_high;
1239 		break;
1240 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1241 		*low = &msrs->exit_ctls_low;
1242 		*high = &msrs->exit_ctls_high;
1243 		break;
1244 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1245 		*low = &msrs->entry_ctls_low;
1246 		*high = &msrs->entry_ctls_high;
1247 		break;
1248 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1249 		*low = &msrs->secondary_ctls_low;
1250 		*high = &msrs->secondary_ctls_high;
1251 		break;
1252 	default:
1253 		BUG();
1254 	}
1255 }
1256 
1257 static int
1258 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1259 {
1260 	u32 *lowp, *highp;
1261 	u64 supported;
1262 
1263 	vmx_get_control_msr(&vmcs_config.nested, msr_index, &lowp, &highp);
1264 
1265 	supported = vmx_control_msr(*lowp, *highp);
1266 
1267 	/* Check must-be-1 bits are still 1. */
1268 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1269 		return -EINVAL;
1270 
1271 	/* Check must-be-0 bits are still 0. */
1272 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1273 		return -EINVAL;
1274 
1275 	vmx_get_control_msr(&vmx->nested.msrs, msr_index, &lowp, &highp);
1276 	*lowp = data;
1277 	*highp = data >> 32;
1278 	return 0;
1279 }
1280 
1281 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1282 {
1283 	const u64 feature_and_reserved_bits =
1284 		/* feature */
1285 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1286 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1287 		/* reserved */
1288 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1289 	u64 vmx_misc = vmx_control_msr(vmcs_config.nested.misc_low,
1290 				       vmcs_config.nested.misc_high);
1291 
1292 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1293 		return -EINVAL;
1294 
1295 	if ((vmx->nested.msrs.pinbased_ctls_high &
1296 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1297 	    vmx_misc_preemption_timer_rate(data) !=
1298 	    vmx_misc_preemption_timer_rate(vmx_misc))
1299 		return -EINVAL;
1300 
1301 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1302 		return -EINVAL;
1303 
1304 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1305 		return -EINVAL;
1306 
1307 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1308 		return -EINVAL;
1309 
1310 	vmx->nested.msrs.misc_low = data;
1311 	vmx->nested.msrs.misc_high = data >> 32;
1312 
1313 	return 0;
1314 }
1315 
1316 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1317 {
1318 	u64 vmx_ept_vpid_cap = vmx_control_msr(vmcs_config.nested.ept_caps,
1319 					       vmcs_config.nested.vpid_caps);
1320 
1321 	/* Every bit is either reserved or a feature bit. */
1322 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1323 		return -EINVAL;
1324 
1325 	vmx->nested.msrs.ept_caps = data;
1326 	vmx->nested.msrs.vpid_caps = data >> 32;
1327 	return 0;
1328 }
1329 
1330 static u64 *vmx_get_fixed0_msr(struct nested_vmx_msrs *msrs, u32 msr_index)
1331 {
1332 	switch (msr_index) {
1333 	case MSR_IA32_VMX_CR0_FIXED0:
1334 		return &msrs->cr0_fixed0;
1335 	case MSR_IA32_VMX_CR4_FIXED0:
1336 		return &msrs->cr4_fixed0;
1337 	default:
1338 		BUG();
1339 	}
1340 }
1341 
1342 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1343 {
1344 	const u64 *msr = vmx_get_fixed0_msr(&vmcs_config.nested, msr_index);
1345 
1346 	/*
1347 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1348 	 * must be 1 in the restored value.
1349 	 */
1350 	if (!is_bitwise_subset(data, *msr, -1ULL))
1351 		return -EINVAL;
1352 
1353 	*vmx_get_fixed0_msr(&vmx->nested.msrs, msr_index) = data;
1354 	return 0;
1355 }
1356 
1357 /*
1358  * Called when userspace is restoring VMX MSRs.
1359  *
1360  * Returns 0 on success, non-0 otherwise.
1361  */
1362 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1363 {
1364 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1365 
1366 	/*
1367 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1368 	 * is in VMX operation.
1369 	 */
1370 	if (vmx->nested.vmxon)
1371 		return -EBUSY;
1372 
1373 	switch (msr_index) {
1374 	case MSR_IA32_VMX_BASIC:
1375 		return vmx_restore_vmx_basic(vmx, data);
1376 	case MSR_IA32_VMX_PINBASED_CTLS:
1377 	case MSR_IA32_VMX_PROCBASED_CTLS:
1378 	case MSR_IA32_VMX_EXIT_CTLS:
1379 	case MSR_IA32_VMX_ENTRY_CTLS:
1380 		/*
1381 		 * The "non-true" VMX capability MSRs are generated from the
1382 		 * "true" MSRs, so we do not support restoring them directly.
1383 		 *
1384 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1385 		 * should restore the "true" MSRs with the must-be-1 bits
1386 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1387 		 * DEFAULT SETTINGS".
1388 		 */
1389 		return -EINVAL;
1390 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1391 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1392 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1393 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1394 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1395 		return vmx_restore_control_msr(vmx, msr_index, data);
1396 	case MSR_IA32_VMX_MISC:
1397 		return vmx_restore_vmx_misc(vmx, data);
1398 	case MSR_IA32_VMX_CR0_FIXED0:
1399 	case MSR_IA32_VMX_CR4_FIXED0:
1400 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1401 	case MSR_IA32_VMX_CR0_FIXED1:
1402 	case MSR_IA32_VMX_CR4_FIXED1:
1403 		/*
1404 		 * These MSRs are generated based on the vCPU's CPUID, so we
1405 		 * do not support restoring them directly.
1406 		 */
1407 		return -EINVAL;
1408 	case MSR_IA32_VMX_EPT_VPID_CAP:
1409 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1410 	case MSR_IA32_VMX_VMCS_ENUM:
1411 		vmx->nested.msrs.vmcs_enum = data;
1412 		return 0;
1413 	case MSR_IA32_VMX_VMFUNC:
1414 		if (data & ~vmcs_config.nested.vmfunc_controls)
1415 			return -EINVAL;
1416 		vmx->nested.msrs.vmfunc_controls = data;
1417 		return 0;
1418 	default:
1419 		/*
1420 		 * The rest of the VMX capability MSRs do not support restore.
1421 		 */
1422 		return -EINVAL;
1423 	}
1424 }
1425 
1426 /* Returns 0 on success, non-0 otherwise. */
1427 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1428 {
1429 	switch (msr_index) {
1430 	case MSR_IA32_VMX_BASIC:
1431 		*pdata = msrs->basic;
1432 		break;
1433 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1434 	case MSR_IA32_VMX_PINBASED_CTLS:
1435 		*pdata = vmx_control_msr(
1436 			msrs->pinbased_ctls_low,
1437 			msrs->pinbased_ctls_high);
1438 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1439 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1440 		break;
1441 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1442 	case MSR_IA32_VMX_PROCBASED_CTLS:
1443 		*pdata = vmx_control_msr(
1444 			msrs->procbased_ctls_low,
1445 			msrs->procbased_ctls_high);
1446 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1447 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1448 		break;
1449 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1450 	case MSR_IA32_VMX_EXIT_CTLS:
1451 		*pdata = vmx_control_msr(
1452 			msrs->exit_ctls_low,
1453 			msrs->exit_ctls_high);
1454 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1455 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1456 		break;
1457 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1458 	case MSR_IA32_VMX_ENTRY_CTLS:
1459 		*pdata = vmx_control_msr(
1460 			msrs->entry_ctls_low,
1461 			msrs->entry_ctls_high);
1462 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1463 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1464 		break;
1465 	case MSR_IA32_VMX_MISC:
1466 		*pdata = vmx_control_msr(
1467 			msrs->misc_low,
1468 			msrs->misc_high);
1469 		break;
1470 	case MSR_IA32_VMX_CR0_FIXED0:
1471 		*pdata = msrs->cr0_fixed0;
1472 		break;
1473 	case MSR_IA32_VMX_CR0_FIXED1:
1474 		*pdata = msrs->cr0_fixed1;
1475 		break;
1476 	case MSR_IA32_VMX_CR4_FIXED0:
1477 		*pdata = msrs->cr4_fixed0;
1478 		break;
1479 	case MSR_IA32_VMX_CR4_FIXED1:
1480 		*pdata = msrs->cr4_fixed1;
1481 		break;
1482 	case MSR_IA32_VMX_VMCS_ENUM:
1483 		*pdata = msrs->vmcs_enum;
1484 		break;
1485 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1486 		*pdata = vmx_control_msr(
1487 			msrs->secondary_ctls_low,
1488 			msrs->secondary_ctls_high);
1489 		break;
1490 	case MSR_IA32_VMX_EPT_VPID_CAP:
1491 		*pdata = msrs->ept_caps |
1492 			((u64)msrs->vpid_caps << 32);
1493 		break;
1494 	case MSR_IA32_VMX_VMFUNC:
1495 		*pdata = msrs->vmfunc_controls;
1496 		break;
1497 	default:
1498 		return 1;
1499 	}
1500 
1501 	return 0;
1502 }
1503 
1504 /*
1505  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1506  * been modified by the L1 guest.  Note, "writable" in this context means
1507  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1508  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1509  * VM-exit information fields (which are actually writable if the vCPU is
1510  * configured to support "VMWRITE to any supported field in the VMCS").
1511  */
1512 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1513 {
1514 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1515 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1516 	struct shadow_vmcs_field field;
1517 	unsigned long val;
1518 	int i;
1519 
1520 	if (WARN_ON(!shadow_vmcs))
1521 		return;
1522 
1523 	preempt_disable();
1524 
1525 	vmcs_load(shadow_vmcs);
1526 
1527 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1528 		field = shadow_read_write_fields[i];
1529 		val = __vmcs_readl(field.encoding);
1530 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1531 	}
1532 
1533 	vmcs_clear(shadow_vmcs);
1534 	vmcs_load(vmx->loaded_vmcs->vmcs);
1535 
1536 	preempt_enable();
1537 }
1538 
1539 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1540 {
1541 	const struct shadow_vmcs_field *fields[] = {
1542 		shadow_read_write_fields,
1543 		shadow_read_only_fields
1544 	};
1545 	const int max_fields[] = {
1546 		max_shadow_read_write_fields,
1547 		max_shadow_read_only_fields
1548 	};
1549 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1550 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1551 	struct shadow_vmcs_field field;
1552 	unsigned long val;
1553 	int i, q;
1554 
1555 	if (WARN_ON(!shadow_vmcs))
1556 		return;
1557 
1558 	vmcs_load(shadow_vmcs);
1559 
1560 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1561 		for (i = 0; i < max_fields[q]; i++) {
1562 			field = fields[q][i];
1563 			val = vmcs12_read_any(vmcs12, field.encoding,
1564 					      field.offset);
1565 			__vmcs_writel(field.encoding, val);
1566 		}
1567 	}
1568 
1569 	vmcs_clear(shadow_vmcs);
1570 	vmcs_load(vmx->loaded_vmcs->vmcs);
1571 }
1572 
1573 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1574 {
1575 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1576 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1577 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(&vmx->vcpu);
1578 
1579 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1580 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1581 	vmcs12->guest_rip = evmcs->guest_rip;
1582 
1583 	if (unlikely(!(hv_clean_fields &
1584 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_ENLIGHTENMENTSCONTROL))) {
1585 		hv_vcpu->nested.pa_page_gpa = evmcs->partition_assist_page;
1586 		hv_vcpu->nested.vm_id = evmcs->hv_vm_id;
1587 		hv_vcpu->nested.vp_id = evmcs->hv_vp_id;
1588 	}
1589 
1590 	if (unlikely(!(hv_clean_fields &
1591 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1592 		vmcs12->guest_rsp = evmcs->guest_rsp;
1593 		vmcs12->guest_rflags = evmcs->guest_rflags;
1594 		vmcs12->guest_interruptibility_info =
1595 			evmcs->guest_interruptibility_info;
1596 		/*
1597 		 * Not present in struct vmcs12:
1598 		 * vmcs12->guest_ssp = evmcs->guest_ssp;
1599 		 */
1600 	}
1601 
1602 	if (unlikely(!(hv_clean_fields &
1603 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1604 		vmcs12->cpu_based_vm_exec_control =
1605 			evmcs->cpu_based_vm_exec_control;
1606 	}
1607 
1608 	if (unlikely(!(hv_clean_fields &
1609 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1610 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1611 	}
1612 
1613 	if (unlikely(!(hv_clean_fields &
1614 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1615 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1616 	}
1617 
1618 	if (unlikely(!(hv_clean_fields &
1619 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1620 		vmcs12->vm_entry_intr_info_field =
1621 			evmcs->vm_entry_intr_info_field;
1622 		vmcs12->vm_entry_exception_error_code =
1623 			evmcs->vm_entry_exception_error_code;
1624 		vmcs12->vm_entry_instruction_len =
1625 			evmcs->vm_entry_instruction_len;
1626 	}
1627 
1628 	if (unlikely(!(hv_clean_fields &
1629 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1630 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1631 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1632 		vmcs12->host_cr0 = evmcs->host_cr0;
1633 		vmcs12->host_cr3 = evmcs->host_cr3;
1634 		vmcs12->host_cr4 = evmcs->host_cr4;
1635 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1636 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1637 		vmcs12->host_rip = evmcs->host_rip;
1638 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1639 		vmcs12->host_es_selector = evmcs->host_es_selector;
1640 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1641 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1642 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1643 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1644 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1645 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1646 		vmcs12->host_ia32_perf_global_ctrl = evmcs->host_ia32_perf_global_ctrl;
1647 		/*
1648 		 * Not present in struct vmcs12:
1649 		 * vmcs12->host_ia32_s_cet = evmcs->host_ia32_s_cet;
1650 		 * vmcs12->host_ssp = evmcs->host_ssp;
1651 		 * vmcs12->host_ia32_int_ssp_table_addr = evmcs->host_ia32_int_ssp_table_addr;
1652 		 */
1653 	}
1654 
1655 	if (unlikely(!(hv_clean_fields &
1656 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1657 		vmcs12->pin_based_vm_exec_control =
1658 			evmcs->pin_based_vm_exec_control;
1659 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1660 		vmcs12->secondary_vm_exec_control =
1661 			evmcs->secondary_vm_exec_control;
1662 	}
1663 
1664 	if (unlikely(!(hv_clean_fields &
1665 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1666 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1667 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1668 	}
1669 
1670 	if (unlikely(!(hv_clean_fields &
1671 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1672 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1673 	}
1674 
1675 	if (unlikely(!(hv_clean_fields &
1676 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1677 		vmcs12->guest_es_base = evmcs->guest_es_base;
1678 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1679 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1680 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1681 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1682 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1683 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1684 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1685 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1686 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1687 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1688 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1689 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1690 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1691 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1692 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1693 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1694 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1695 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1696 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1697 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1698 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1699 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1700 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1701 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1702 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1703 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1704 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1705 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1706 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1707 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1708 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1709 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1710 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1711 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1712 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1713 	}
1714 
1715 	if (unlikely(!(hv_clean_fields &
1716 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1717 		vmcs12->tsc_offset = evmcs->tsc_offset;
1718 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1719 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1720 		vmcs12->encls_exiting_bitmap = evmcs->encls_exiting_bitmap;
1721 		vmcs12->tsc_multiplier = evmcs->tsc_multiplier;
1722 	}
1723 
1724 	if (unlikely(!(hv_clean_fields &
1725 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1726 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1727 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1728 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1729 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1730 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1731 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1732 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1733 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1734 	}
1735 
1736 	if (unlikely(!(hv_clean_fields &
1737 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1738 		vmcs12->host_fs_base = evmcs->host_fs_base;
1739 		vmcs12->host_gs_base = evmcs->host_gs_base;
1740 		vmcs12->host_tr_base = evmcs->host_tr_base;
1741 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1742 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1743 		vmcs12->host_rsp = evmcs->host_rsp;
1744 	}
1745 
1746 	if (unlikely(!(hv_clean_fields &
1747 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1748 		vmcs12->ept_pointer = evmcs->ept_pointer;
1749 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1750 	}
1751 
1752 	if (unlikely(!(hv_clean_fields &
1753 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1754 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1755 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1756 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1757 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1758 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1759 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1760 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1761 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1762 		vmcs12->guest_pending_dbg_exceptions =
1763 			evmcs->guest_pending_dbg_exceptions;
1764 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1765 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1766 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1767 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1768 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1769 		vmcs12->guest_ia32_perf_global_ctrl = evmcs->guest_ia32_perf_global_ctrl;
1770 		/*
1771 		 * Not present in struct vmcs12:
1772 		 * vmcs12->guest_ia32_s_cet = evmcs->guest_ia32_s_cet;
1773 		 * vmcs12->guest_ia32_lbr_ctl = evmcs->guest_ia32_lbr_ctl;
1774 		 * vmcs12->guest_ia32_int_ssp_table_addr = evmcs->guest_ia32_int_ssp_table_addr;
1775 		 */
1776 	}
1777 
1778 	/*
1779 	 * Not used?
1780 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1781 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1782 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1783 	 * vmcs12->page_fault_error_code_mask =
1784 	 *		evmcs->page_fault_error_code_mask;
1785 	 * vmcs12->page_fault_error_code_match =
1786 	 *		evmcs->page_fault_error_code_match;
1787 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1788 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1789 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1790 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1791 	 */
1792 
1793 	/*
1794 	 * Read only fields:
1795 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1796 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1797 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1798 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1799 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1800 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1801 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1802 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1803 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1804 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1805 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1806 	 *
1807 	 * Not present in struct vmcs12:
1808 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1809 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1810 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1811 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1812 	 */
1813 
1814 	return;
1815 }
1816 
1817 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1818 {
1819 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1820 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1821 
1822 	/*
1823 	 * Should not be changed by KVM:
1824 	 *
1825 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1826 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1827 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1828 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1829 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1830 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1831 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1832 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1833 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1834 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1835 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1836 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1837 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1838 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1839 	 * evmcs->host_rip = vmcs12->host_rip;
1840 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1841 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1842 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1843 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1844 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1845 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1846 	 * evmcs->host_rsp = vmcs12->host_rsp;
1847 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1848 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1849 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1850 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1851 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1852 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1853 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1854 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1855 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1856 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1857 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1858 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1859 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1860 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1861 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1862 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1863 	 * evmcs->page_fault_error_code_mask =
1864 	 *		vmcs12->page_fault_error_code_mask;
1865 	 * evmcs->page_fault_error_code_match =
1866 	 *		vmcs12->page_fault_error_code_match;
1867 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1868 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1869 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1870 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1871 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1872 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1873 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1874 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1875 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1876 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1877 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1878 	 * evmcs->guest_ia32_perf_global_ctrl = vmcs12->guest_ia32_perf_global_ctrl;
1879 	 * evmcs->host_ia32_perf_global_ctrl = vmcs12->host_ia32_perf_global_ctrl;
1880 	 * evmcs->encls_exiting_bitmap = vmcs12->encls_exiting_bitmap;
1881 	 * evmcs->tsc_multiplier = vmcs12->tsc_multiplier;
1882 	 *
1883 	 * Not present in struct vmcs12:
1884 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1885 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1886 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1887 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1888 	 * evmcs->host_ia32_s_cet = vmcs12->host_ia32_s_cet;
1889 	 * evmcs->host_ssp = vmcs12->host_ssp;
1890 	 * evmcs->host_ia32_int_ssp_table_addr = vmcs12->host_ia32_int_ssp_table_addr;
1891 	 * evmcs->guest_ia32_s_cet = vmcs12->guest_ia32_s_cet;
1892 	 * evmcs->guest_ia32_lbr_ctl = vmcs12->guest_ia32_lbr_ctl;
1893 	 * evmcs->guest_ia32_int_ssp_table_addr = vmcs12->guest_ia32_int_ssp_table_addr;
1894 	 * evmcs->guest_ssp = vmcs12->guest_ssp;
1895 	 */
1896 
1897 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1898 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1899 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1900 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1901 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1902 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1903 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1904 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1905 
1906 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1907 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1908 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1909 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1910 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1911 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1912 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1913 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1914 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1915 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1916 
1917 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1918 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1919 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1920 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1921 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1922 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1923 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1924 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1925 
1926 	evmcs->guest_es_base = vmcs12->guest_es_base;
1927 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1928 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1929 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1930 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1931 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1932 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1933 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1934 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1935 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1936 
1937 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1938 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1939 
1940 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1941 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1942 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1943 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1944 
1945 	evmcs->guest_pending_dbg_exceptions =
1946 		vmcs12->guest_pending_dbg_exceptions;
1947 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1948 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1949 
1950 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1951 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1952 
1953 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1954 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1955 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1956 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1957 
1958 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1959 
1960 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1961 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1962 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1963 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1964 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1965 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1966 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1967 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1968 
1969 	evmcs->exit_qualification = vmcs12->exit_qualification;
1970 
1971 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1972 	evmcs->guest_rsp = vmcs12->guest_rsp;
1973 	evmcs->guest_rflags = vmcs12->guest_rflags;
1974 
1975 	evmcs->guest_interruptibility_info =
1976 		vmcs12->guest_interruptibility_info;
1977 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1978 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1979 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1980 	evmcs->vm_entry_exception_error_code =
1981 		vmcs12->vm_entry_exception_error_code;
1982 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1983 
1984 	evmcs->guest_rip = vmcs12->guest_rip;
1985 
1986 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1987 
1988 	return;
1989 }
1990 
1991 /*
1992  * This is an equivalent of the nested hypervisor executing the vmptrld
1993  * instruction.
1994  */
1995 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1996 	struct kvm_vcpu *vcpu, bool from_launch)
1997 {
1998 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1999 	bool evmcs_gpa_changed = false;
2000 	u64 evmcs_gpa;
2001 
2002 	if (likely(!guest_cpuid_has_evmcs(vcpu)))
2003 		return EVMPTRLD_DISABLED;
2004 
2005 	evmcs_gpa = nested_get_evmptr(vcpu);
2006 	if (!evmptr_is_valid(evmcs_gpa)) {
2007 		nested_release_evmcs(vcpu);
2008 		return EVMPTRLD_DISABLED;
2009 	}
2010 
2011 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
2012 		vmx->nested.current_vmptr = INVALID_GPA;
2013 
2014 		nested_release_evmcs(vcpu);
2015 
2016 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2017 				 &vmx->nested.hv_evmcs_map))
2018 			return EVMPTRLD_ERROR;
2019 
2020 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2021 
2022 		/*
2023 		 * Currently, KVM only supports eVMCS version 1
2024 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2025 		 * value to first u32 field of eVMCS which should specify eVMCS
2026 		 * VersionNumber.
2027 		 *
2028 		 * Guest should be aware of supported eVMCS versions by host by
2029 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2030 		 * expected to set this CPUID leaf according to the value
2031 		 * returned in vmcs_version from nested_enable_evmcs().
2032 		 *
2033 		 * However, it turns out that Microsoft Hyper-V fails to comply
2034 		 * to their own invented interface: When Hyper-V use eVMCS, it
2035 		 * just sets first u32 field of eVMCS to revision_id specified
2036 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2037 		 * which is one of the supported versions specified in
2038 		 * CPUID.0x4000000A.EAX[0:15].
2039 		 *
2040 		 * To overcome Hyper-V bug, we accept here either a supported
2041 		 * eVMCS version or VMCS12 revision_id as valid values for first
2042 		 * u32 field of eVMCS.
2043 		 */
2044 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2045 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2046 			nested_release_evmcs(vcpu);
2047 			return EVMPTRLD_VMFAIL;
2048 		}
2049 
2050 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2051 
2052 		evmcs_gpa_changed = true;
2053 		/*
2054 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2055 		 * reloaded from guest's memory (read only fields, fields not
2056 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2057 		 * are no leftovers.
2058 		 */
2059 		if (from_launch) {
2060 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2061 			memset(vmcs12, 0, sizeof(*vmcs12));
2062 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2063 		}
2064 
2065 	}
2066 
2067 	/*
2068 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2069 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2070 	 */
2071 	if (from_launch || evmcs_gpa_changed) {
2072 		vmx->nested.hv_evmcs->hv_clean_fields &=
2073 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2074 
2075 		vmx->nested.force_msr_bitmap_recalc = true;
2076 	}
2077 
2078 	return EVMPTRLD_SUCCEEDED;
2079 }
2080 
2081 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2082 {
2083 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2084 
2085 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2086 		copy_vmcs12_to_enlightened(vmx);
2087 	else
2088 		copy_vmcs12_to_shadow(vmx);
2089 
2090 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2091 }
2092 
2093 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2094 {
2095 	struct vcpu_vmx *vmx =
2096 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2097 
2098 	vmx->nested.preemption_timer_expired = true;
2099 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2100 	kvm_vcpu_kick(&vmx->vcpu);
2101 
2102 	return HRTIMER_NORESTART;
2103 }
2104 
2105 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2106 {
2107 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2108 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2109 
2110 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2111 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2112 
2113 	if (!vmx->nested.has_preemption_timer_deadline) {
2114 		vmx->nested.preemption_timer_deadline =
2115 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2116 		vmx->nested.has_preemption_timer_deadline = true;
2117 	}
2118 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2119 }
2120 
2121 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2122 					u64 preemption_timeout)
2123 {
2124 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2125 
2126 	/*
2127 	 * A timer value of zero is architecturally guaranteed to cause
2128 	 * a VMExit prior to executing any instructions in the guest.
2129 	 */
2130 	if (preemption_timeout == 0) {
2131 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2132 		return;
2133 	}
2134 
2135 	if (vcpu->arch.virtual_tsc_khz == 0)
2136 		return;
2137 
2138 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2139 	preemption_timeout *= 1000000;
2140 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2141 	hrtimer_start(&vmx->nested.preemption_timer,
2142 		      ktime_add_ns(ktime_get(), preemption_timeout),
2143 		      HRTIMER_MODE_ABS_PINNED);
2144 }
2145 
2146 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2147 {
2148 	if (vmx->nested.nested_run_pending &&
2149 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2150 		return vmcs12->guest_ia32_efer;
2151 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2152 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2153 	else
2154 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2155 }
2156 
2157 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2158 {
2159 	struct kvm *kvm = vmx->vcpu.kvm;
2160 
2161 	/*
2162 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2163 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2164 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2165 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2166 	 */
2167 	if (vmx->nested.vmcs02_initialized)
2168 		return;
2169 	vmx->nested.vmcs02_initialized = true;
2170 
2171 	/*
2172 	 * We don't care what the EPTP value is we just need to guarantee
2173 	 * it's valid so we don't get a false positive when doing early
2174 	 * consistency checks.
2175 	 */
2176 	if (enable_ept && nested_early_check)
2177 		vmcs_write64(EPT_POINTER,
2178 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2179 
2180 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2181 	if (cpu_has_vmx_vmfunc())
2182 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2183 
2184 	if (cpu_has_vmx_posted_intr())
2185 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2186 
2187 	if (cpu_has_vmx_msr_bitmap())
2188 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2189 
2190 	/*
2191 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2192 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2193 	 * deal with filtering out L2 GPAs from the buffer.
2194 	 */
2195 	if (enable_pml) {
2196 		vmcs_write64(PML_ADDRESS, 0);
2197 		vmcs_write16(GUEST_PML_INDEX, -1);
2198 	}
2199 
2200 	if (cpu_has_vmx_encls_vmexit())
2201 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2202 
2203 	if (kvm_notify_vmexit_enabled(kvm))
2204 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
2205 
2206 	/*
2207 	 * Set the MSR load/store lists to match L0's settings.  Only the
2208 	 * addresses are constant (for vmcs02), the counts can change based
2209 	 * on L2's behavior, e.g. switching to/from long mode.
2210 	 */
2211 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2212 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2213 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2214 
2215 	vmx_set_constant_host_state(vmx);
2216 }
2217 
2218 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2219 				      struct vmcs12 *vmcs12)
2220 {
2221 	prepare_vmcs02_constant_state(vmx);
2222 
2223 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2224 
2225 	if (enable_vpid) {
2226 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2227 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2228 		else
2229 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2230 	}
2231 }
2232 
2233 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2234 				 struct vmcs12 *vmcs12)
2235 {
2236 	u32 exec_control;
2237 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2238 
2239 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2240 		prepare_vmcs02_early_rare(vmx, vmcs12);
2241 
2242 	/*
2243 	 * PIN CONTROLS
2244 	 */
2245 	exec_control = __pin_controls_get(vmcs01);
2246 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2247 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2248 
2249 	/* Posted interrupts setting is only taken from vmcs12.  */
2250 	vmx->nested.pi_pending = false;
2251 	if (nested_cpu_has_posted_intr(vmcs12))
2252 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2253 	else
2254 		exec_control &= ~PIN_BASED_POSTED_INTR;
2255 	pin_controls_set(vmx, exec_control);
2256 
2257 	/*
2258 	 * EXEC CONTROLS
2259 	 */
2260 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2261 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2262 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2263 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2264 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2265 
2266 	vmx->nested.l1_tpr_threshold = -1;
2267 	if (exec_control & CPU_BASED_TPR_SHADOW)
2268 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2269 #ifdef CONFIG_X86_64
2270 	else
2271 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2272 				CPU_BASED_CR8_STORE_EXITING;
2273 #endif
2274 
2275 	/*
2276 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2277 	 * for I/O port accesses.
2278 	 */
2279 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2280 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2281 
2282 	/*
2283 	 * This bit will be computed in nested_get_vmcs12_pages, because
2284 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2285 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2286 	 * only set/clear this bit.
2287 	 */
2288 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2289 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2290 
2291 	exec_controls_set(vmx, exec_control);
2292 
2293 	/*
2294 	 * SECONDARY EXEC CONTROLS
2295 	 */
2296 	if (cpu_has_secondary_exec_ctrls()) {
2297 		exec_control = __secondary_exec_controls_get(vmcs01);
2298 
2299 		/* Take the following fields only from vmcs12 */
2300 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2301 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2302 				  SECONDARY_EXEC_ENABLE_INVPCID |
2303 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2304 				  SECONDARY_EXEC_XSAVES |
2305 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2306 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2307 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2308 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2309 				  SECONDARY_EXEC_DESC);
2310 
2311 		if (nested_cpu_has(vmcs12,
2312 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2313 			exec_control |= vmcs12->secondary_vm_exec_control;
2314 
2315 		/* PML is emulated and never enabled in hardware for L2. */
2316 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2317 
2318 		/* VMCS shadowing for L2 is emulated for now */
2319 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2320 
2321 		/*
2322 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2323 		 * will not have to rewrite the controls just for this bit.
2324 		 */
2325 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2326 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2327 			exec_control |= SECONDARY_EXEC_DESC;
2328 
2329 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2330 			vmcs_write16(GUEST_INTR_STATUS,
2331 				vmcs12->guest_intr_status);
2332 
2333 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2334 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2335 
2336 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2337 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2338 
2339 		secondary_exec_controls_set(vmx, exec_control);
2340 	}
2341 
2342 	/*
2343 	 * ENTRY CONTROLS
2344 	 *
2345 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2346 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2347 	 * on the related bits (if supported by the CPU) in the hope that
2348 	 * we can avoid VMWrites during vmx_set_efer().
2349 	 *
2350 	 * Similarly, take vmcs01's PERF_GLOBAL_CTRL in the hope that if KVM is
2351 	 * loading PERF_GLOBAL_CTRL via the VMCS for L1, then KVM will want to
2352 	 * do the same for L2.
2353 	 */
2354 	exec_control = __vm_entry_controls_get(vmcs01);
2355 	exec_control |= (vmcs12->vm_entry_controls &
2356 			 ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
2357 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2358 	if (cpu_has_load_ia32_efer()) {
2359 		if (guest_efer & EFER_LMA)
2360 			exec_control |= VM_ENTRY_IA32E_MODE;
2361 		if (guest_efer != host_efer)
2362 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2363 	}
2364 	vm_entry_controls_set(vmx, exec_control);
2365 
2366 	/*
2367 	 * EXIT CONTROLS
2368 	 *
2369 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2370 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2371 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2372 	 */
2373 	exec_control = __vm_exit_controls_get(vmcs01);
2374 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2375 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2376 	else
2377 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2378 	vm_exit_controls_set(vmx, exec_control);
2379 
2380 	/*
2381 	 * Interrupt/Exception Fields
2382 	 */
2383 	if (vmx->nested.nested_run_pending) {
2384 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2385 			     vmcs12->vm_entry_intr_info_field);
2386 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2387 			     vmcs12->vm_entry_exception_error_code);
2388 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2389 			     vmcs12->vm_entry_instruction_len);
2390 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2391 			     vmcs12->guest_interruptibility_info);
2392 		vmx->loaded_vmcs->nmi_known_unmasked =
2393 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2394 	} else {
2395 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2396 	}
2397 }
2398 
2399 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2400 {
2401 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2402 
2403 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2404 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2405 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2406 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2407 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2408 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2409 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2410 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2411 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2412 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2413 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2414 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2415 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2416 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2417 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2418 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2419 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2420 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2421 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2422 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2423 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2424 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2425 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2426 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2427 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2428 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2429 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2430 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2431 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2432 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2433 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2434 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2435 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2436 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2437 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2438 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2439 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2440 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2441 
2442 		vmx->segment_cache.bitmask = 0;
2443 	}
2444 
2445 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2446 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2447 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2448 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2449 			    vmcs12->guest_pending_dbg_exceptions);
2450 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2451 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2452 
2453 		/*
2454 		 * L1 may access the L2's PDPTR, so save them to construct
2455 		 * vmcs12
2456 		 */
2457 		if (enable_ept) {
2458 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2459 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2460 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2461 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2462 		}
2463 
2464 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2465 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2466 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2467 	}
2468 
2469 	if (nested_cpu_has_xsaves(vmcs12))
2470 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2471 
2472 	/*
2473 	 * Whether page-faults are trapped is determined by a combination of
2474 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2475 	 * doesn't care about page faults then we should set all of these to
2476 	 * L1's desires. However, if L0 does care about (some) page faults, it
2477 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2478 	 * simply ask to exit on each and every L2 page fault. This is done by
2479 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2480 	 * Note that below we don't need special code to set EB.PF beyond the
2481 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2482 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2483 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2484 	 */
2485 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2486 		/*
2487 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2488 		 * go ahead and use it?
2489 		 */
2490 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2491 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2492 	} else {
2493 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2494 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2495 	}
2496 
2497 	if (cpu_has_vmx_apicv()) {
2498 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2499 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2500 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2501 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2502 	}
2503 
2504 	/*
2505 	 * Make sure the msr_autostore list is up to date before we set the
2506 	 * count in the vmcs02.
2507 	 */
2508 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2509 
2510 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2511 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2512 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2513 
2514 	set_cr4_guest_host_mask(vmx);
2515 }
2516 
2517 /*
2518  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2519  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2520  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2521  * guest in a way that will both be appropriate to L1's requests, and our
2522  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2523  * function also has additional necessary side-effects, like setting various
2524  * vcpu->arch fields.
2525  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2526  * is assigned to entry_failure_code on failure.
2527  */
2528 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2529 			  bool from_vmentry,
2530 			  enum vm_entry_failure_code *entry_failure_code)
2531 {
2532 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2533 	bool load_guest_pdptrs_vmcs12 = false;
2534 
2535 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2536 		prepare_vmcs02_rare(vmx, vmcs12);
2537 		vmx->nested.dirty_vmcs12 = false;
2538 
2539 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2540 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2541 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2542 	}
2543 
2544 	if (vmx->nested.nested_run_pending &&
2545 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2546 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2547 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2548 	} else {
2549 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2550 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.pre_vmenter_debugctl);
2551 	}
2552 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2553 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2554 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.pre_vmenter_bndcfgs);
2555 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2556 
2557 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2558 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2559 	 * trap. Note that CR0.TS also needs updating - we do this later.
2560 	 */
2561 	vmx_update_exception_bitmap(vcpu);
2562 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2563 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2564 
2565 	if (vmx->nested.nested_run_pending &&
2566 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2567 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2568 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2569 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2570 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2571 	}
2572 
2573 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2574 			vcpu->arch.l1_tsc_offset,
2575 			vmx_get_l2_tsc_offset(vcpu),
2576 			vmx_get_l2_tsc_multiplier(vcpu));
2577 
2578 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2579 			vcpu->arch.l1_tsc_scaling_ratio,
2580 			vmx_get_l2_tsc_multiplier(vcpu));
2581 
2582 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2583 	if (kvm_caps.has_tsc_control)
2584 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2585 
2586 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2587 
2588 	if (nested_cpu_has_ept(vmcs12))
2589 		nested_ept_init_mmu_context(vcpu);
2590 
2591 	/*
2592 	 * Override the CR0/CR4 read shadows after setting the effective guest
2593 	 * CR0/CR4.  The common helpers also set the shadows, but they don't
2594 	 * account for vmcs12's cr0/4_guest_host_mask.
2595 	 */
2596 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2597 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2598 
2599 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2600 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2601 
2602 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2603 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2604 	vmx_set_efer(vcpu, vcpu->arch.efer);
2605 
2606 	/*
2607 	 * Guest state is invalid and unrestricted guest is disabled,
2608 	 * which means L1 attempted VMEntry to L2 with invalid state.
2609 	 * Fail the VMEntry.
2610 	 *
2611 	 * However when force loading the guest state (SMM exit or
2612 	 * loading nested state after migration, it is possible to
2613 	 * have invalid guest state now, which will be later fixed by
2614 	 * restoring L2 register state
2615 	 */
2616 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2617 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2618 		return -EINVAL;
2619 	}
2620 
2621 	/* Shadow page tables on either EPT or shadow page tables. */
2622 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2623 				from_vmentry, entry_failure_code))
2624 		return -EINVAL;
2625 
2626 	/*
2627 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2628 	 * on nested VM-Exit, which can occur without actually running L2 and
2629 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2630 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2631 	 * transition to HLT instead of running L2.
2632 	 */
2633 	if (enable_ept)
2634 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2635 
2636 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2637 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2638 	    is_pae_paging(vcpu)) {
2639 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2640 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2641 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2642 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2643 	}
2644 
2645 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2646 	    intel_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)) &&
2647 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2648 				     vmcs12->guest_ia32_perf_global_ctrl))) {
2649 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2650 		return -EINVAL;
2651 	}
2652 
2653 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2654 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2655 
2656 	/*
2657 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2658 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2659 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2660 	 * here.
2661 	 */
2662 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2663 		vmx->nested.hv_evmcs->hv_clean_fields |=
2664 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2665 
2666 	return 0;
2667 }
2668 
2669 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2670 {
2671 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2672 	       nested_cpu_has_virtual_nmis(vmcs12)))
2673 		return -EINVAL;
2674 
2675 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2676 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2677 		return -EINVAL;
2678 
2679 	return 0;
2680 }
2681 
2682 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2683 {
2684 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2685 
2686 	/* Check for memory type validity */
2687 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2688 	case VMX_EPTP_MT_UC:
2689 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2690 			return false;
2691 		break;
2692 	case VMX_EPTP_MT_WB:
2693 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2694 			return false;
2695 		break;
2696 	default:
2697 		return false;
2698 	}
2699 
2700 	/* Page-walk levels validity. */
2701 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2702 	case VMX_EPTP_PWL_5:
2703 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2704 			return false;
2705 		break;
2706 	case VMX_EPTP_PWL_4:
2707 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2708 			return false;
2709 		break;
2710 	default:
2711 		return false;
2712 	}
2713 
2714 	/* Reserved bits should not be set */
2715 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2716 		return false;
2717 
2718 	/* AD, if set, should be supported */
2719 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2720 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2721 			return false;
2722 	}
2723 
2724 	return true;
2725 }
2726 
2727 /*
2728  * Checks related to VM-Execution Control Fields
2729  */
2730 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2731                                               struct vmcs12 *vmcs12)
2732 {
2733 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2734 
2735 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2736 				   vmx->nested.msrs.pinbased_ctls_low,
2737 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2738 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2739 				   vmx->nested.msrs.procbased_ctls_low,
2740 				   vmx->nested.msrs.procbased_ctls_high)))
2741 		return -EINVAL;
2742 
2743 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2744 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2745 				   vmx->nested.msrs.secondary_ctls_low,
2746 				   vmx->nested.msrs.secondary_ctls_high)))
2747 		return -EINVAL;
2748 
2749 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2750 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2751 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2752 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2753 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2754 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2755 	    nested_vmx_check_nmi_controls(vmcs12) ||
2756 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2757 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2758 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2759 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2760 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2761 		return -EINVAL;
2762 
2763 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2764 	    nested_cpu_has_save_preemption_timer(vmcs12))
2765 		return -EINVAL;
2766 
2767 	if (nested_cpu_has_ept(vmcs12) &&
2768 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2769 		return -EINVAL;
2770 
2771 	if (nested_cpu_has_vmfunc(vmcs12)) {
2772 		if (CC(vmcs12->vm_function_control &
2773 		       ~vmx->nested.msrs.vmfunc_controls))
2774 			return -EINVAL;
2775 
2776 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2777 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2778 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2779 				return -EINVAL;
2780 		}
2781 	}
2782 
2783 	return 0;
2784 }
2785 
2786 /*
2787  * Checks related to VM-Exit Control Fields
2788  */
2789 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2790                                          struct vmcs12 *vmcs12)
2791 {
2792 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2793 
2794 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2795 				    vmx->nested.msrs.exit_ctls_low,
2796 				    vmx->nested.msrs.exit_ctls_high)) ||
2797 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2798 		return -EINVAL;
2799 
2800 	return 0;
2801 }
2802 
2803 /*
2804  * Checks related to VM-Entry Control Fields
2805  */
2806 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2807 					  struct vmcs12 *vmcs12)
2808 {
2809 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2810 
2811 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2812 				    vmx->nested.msrs.entry_ctls_low,
2813 				    vmx->nested.msrs.entry_ctls_high)))
2814 		return -EINVAL;
2815 
2816 	/*
2817 	 * From the Intel SDM, volume 3:
2818 	 * Fields relevant to VM-entry event injection must be set properly.
2819 	 * These fields are the VM-entry interruption-information field, the
2820 	 * VM-entry exception error code, and the VM-entry instruction length.
2821 	 */
2822 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2823 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2824 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2825 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2826 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2827 		bool should_have_error_code;
2828 		bool urg = nested_cpu_has2(vmcs12,
2829 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2830 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2831 
2832 		/* VM-entry interruption-info field: interruption type */
2833 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2834 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2835 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2836 			return -EINVAL;
2837 
2838 		/* VM-entry interruption-info field: vector */
2839 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2840 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2841 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2842 			return -EINVAL;
2843 
2844 		/* VM-entry interruption-info field: deliver error code */
2845 		should_have_error_code =
2846 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2847 			x86_exception_has_error_code(vector);
2848 		if (CC(has_error_code != should_have_error_code))
2849 			return -EINVAL;
2850 
2851 		/* VM-entry exception error code */
2852 		if (CC(has_error_code &&
2853 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2854 			return -EINVAL;
2855 
2856 		/* VM-entry interruption-info field: reserved bits */
2857 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2858 			return -EINVAL;
2859 
2860 		/* VM-entry instruction length */
2861 		switch (intr_type) {
2862 		case INTR_TYPE_SOFT_EXCEPTION:
2863 		case INTR_TYPE_SOFT_INTR:
2864 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2865 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2866 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2867 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2868 				return -EINVAL;
2869 		}
2870 	}
2871 
2872 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2873 		return -EINVAL;
2874 
2875 	return 0;
2876 }
2877 
2878 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2879 				     struct vmcs12 *vmcs12)
2880 {
2881 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2882 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2883 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2884 		return -EINVAL;
2885 
2886 	if (guest_cpuid_has_evmcs(vcpu))
2887 		return nested_evmcs_check_controls(vmcs12);
2888 
2889 	return 0;
2890 }
2891 
2892 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
2893 				       struct vmcs12 *vmcs12)
2894 {
2895 #ifdef CONFIG_X86_64
2896 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
2897 		!!(vcpu->arch.efer & EFER_LMA)))
2898 		return -EINVAL;
2899 #endif
2900 	return 0;
2901 }
2902 
2903 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2904 				       struct vmcs12 *vmcs12)
2905 {
2906 	bool ia32e;
2907 
2908 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2909 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2910 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2911 		return -EINVAL;
2912 
2913 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2914 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2915 		return -EINVAL;
2916 
2917 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2918 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2919 		return -EINVAL;
2920 
2921 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2922 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2923 					   vmcs12->host_ia32_perf_global_ctrl)))
2924 		return -EINVAL;
2925 
2926 #ifdef CONFIG_X86_64
2927 	ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
2928 #else
2929 	ia32e = false;
2930 #endif
2931 
2932 	if (ia32e) {
2933 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2934 			return -EINVAL;
2935 	} else {
2936 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2937 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2938 		    CC((vmcs12->host_rip) >> 32))
2939 			return -EINVAL;
2940 	}
2941 
2942 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2943 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2944 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2945 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2946 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2947 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2948 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2949 	    CC(vmcs12->host_cs_selector == 0) ||
2950 	    CC(vmcs12->host_tr_selector == 0) ||
2951 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2952 		return -EINVAL;
2953 
2954 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2955 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2956 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2957 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2958 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2959 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2960 		return -EINVAL;
2961 
2962 	/*
2963 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2964 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2965 	 * the values of the LMA and LME bits in the field must each be that of
2966 	 * the host address-space size VM-exit control.
2967 	 */
2968 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2969 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2970 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2971 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2972 			return -EINVAL;
2973 	}
2974 
2975 	return 0;
2976 }
2977 
2978 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2979 					  struct vmcs12 *vmcs12)
2980 {
2981 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2982 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
2983 	struct vmcs_hdr hdr;
2984 
2985 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
2986 		return 0;
2987 
2988 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2989 		return -EINVAL;
2990 
2991 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
2992 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
2993 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
2994                 return -EINVAL;
2995 
2996 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
2997 					    offsetof(struct vmcs12, hdr),
2998 					    sizeof(hdr))))
2999 		return -EINVAL;
3000 
3001 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
3002 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
3003 		return -EINVAL;
3004 
3005 	return 0;
3006 }
3007 
3008 /*
3009  * Checks related to Guest Non-register State
3010  */
3011 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
3012 {
3013 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
3014 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
3015 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
3016 		return -EINVAL;
3017 
3018 	return 0;
3019 }
3020 
3021 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
3022 					struct vmcs12 *vmcs12,
3023 					enum vm_entry_failure_code *entry_failure_code)
3024 {
3025 	bool ia32e;
3026 
3027 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
3028 
3029 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3030 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3031 		return -EINVAL;
3032 
3033 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
3034 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
3035 		return -EINVAL;
3036 
3037 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3038 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3039 		return -EINVAL;
3040 
3041 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3042 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3043 		return -EINVAL;
3044 	}
3045 
3046 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3047 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3048 					   vmcs12->guest_ia32_perf_global_ctrl)))
3049 		return -EINVAL;
3050 
3051 	/*
3052 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3053 	 * are performed on the field for the IA32_EFER MSR:
3054 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3055 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3056 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3057 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3058 	 *   CR0.PG) is 1.
3059 	 */
3060 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3061 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3062 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
3063 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3064 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3065 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3066 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3067 			return -EINVAL;
3068 	}
3069 
3070 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3071 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3072 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3073 		return -EINVAL;
3074 
3075 	if (nested_check_guest_non_reg_state(vmcs12))
3076 		return -EINVAL;
3077 
3078 	return 0;
3079 }
3080 
3081 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3082 {
3083 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3084 	unsigned long cr3, cr4;
3085 	bool vm_fail;
3086 
3087 	if (!nested_early_check)
3088 		return 0;
3089 
3090 	if (vmx->msr_autoload.host.nr)
3091 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3092 	if (vmx->msr_autoload.guest.nr)
3093 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3094 
3095 	preempt_disable();
3096 
3097 	vmx_prepare_switch_to_guest(vcpu);
3098 
3099 	/*
3100 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3101 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3102 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3103 	 * there is no need to preserve other bits or save/restore the field.
3104 	 */
3105 	vmcs_writel(GUEST_RFLAGS, 0);
3106 
3107 	cr3 = __get_current_cr3_fast();
3108 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3109 		vmcs_writel(HOST_CR3, cr3);
3110 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3111 	}
3112 
3113 	cr4 = cr4_read_shadow();
3114 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3115 		vmcs_writel(HOST_CR4, cr4);
3116 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3117 	}
3118 
3119 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3120 				 __vmx_vcpu_run_flags(vmx));
3121 
3122 	if (vmx->msr_autoload.host.nr)
3123 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3124 	if (vmx->msr_autoload.guest.nr)
3125 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3126 
3127 	if (vm_fail) {
3128 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3129 
3130 		preempt_enable();
3131 
3132 		trace_kvm_nested_vmenter_failed(
3133 			"early hardware check VM-instruction error: ", error);
3134 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3135 		return 1;
3136 	}
3137 
3138 	/*
3139 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3140 	 */
3141 	if (hw_breakpoint_active())
3142 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3143 	local_irq_enable();
3144 	preempt_enable();
3145 
3146 	/*
3147 	 * A non-failing VMEntry means we somehow entered guest mode with
3148 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3149 	 * is no telling what memory has been modified or what state has
3150 	 * been exposed to unknown code.  Hitting this all but guarantees
3151 	 * a (very critical) hardware issue.
3152 	 */
3153 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3154 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3155 
3156 	return 0;
3157 }
3158 
3159 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3160 {
3161 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3162 
3163 	/*
3164 	 * hv_evmcs may end up being not mapped after migration (when
3165 	 * L2 was running), map it here to make sure vmcs12 changes are
3166 	 * properly reflected.
3167 	 */
3168 	if (guest_cpuid_has_evmcs(vcpu) &&
3169 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3170 		enum nested_evmptrld_status evmptrld_status =
3171 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3172 
3173 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3174 		    evmptrld_status == EVMPTRLD_ERROR)
3175 			return false;
3176 
3177 		/*
3178 		 * Post migration VMCS12 always provides the most actual
3179 		 * information, copy it to eVMCS upon entry.
3180 		 */
3181 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3182 	}
3183 
3184 	return true;
3185 }
3186 
3187 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3188 {
3189 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3190 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3191 	struct kvm_host_map *map;
3192 
3193 	if (!vcpu->arch.pdptrs_from_userspace &&
3194 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3195 		/*
3196 		 * Reload the guest's PDPTRs since after a migration
3197 		 * the guest CR3 might be restored prior to setting the nested
3198 		 * state which can lead to a load of wrong PDPTRs.
3199 		 */
3200 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3201 			return false;
3202 	}
3203 
3204 
3205 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3206 		map = &vmx->nested.apic_access_page_map;
3207 
3208 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->apic_access_addr), map)) {
3209 			vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(map->pfn));
3210 		} else {
3211 			pr_debug_ratelimited("%s: no backing for APIC-access address in vmcs12\n",
3212 					     __func__);
3213 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3214 			vcpu->run->internal.suberror =
3215 				KVM_INTERNAL_ERROR_EMULATION;
3216 			vcpu->run->internal.ndata = 0;
3217 			return false;
3218 		}
3219 	}
3220 
3221 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3222 		map = &vmx->nested.virtual_apic_map;
3223 
3224 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3225 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3226 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3227 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3228 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3229 			/*
3230 			 * The processor will never use the TPR shadow, simply
3231 			 * clear the bit from the execution control.  Such a
3232 			 * configuration is useless, but it happens in tests.
3233 			 * For any other configuration, failing the vm entry is
3234 			 * _not_ what the processor does but it's basically the
3235 			 * only possibility we have.
3236 			 */
3237 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3238 		} else {
3239 			/*
3240 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3241 			 * force VM-Entry to fail.
3242 			 */
3243 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3244 		}
3245 	}
3246 
3247 	if (nested_cpu_has_posted_intr(vmcs12)) {
3248 		map = &vmx->nested.pi_desc_map;
3249 
3250 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3251 			vmx->nested.pi_desc =
3252 				(struct pi_desc *)(((void *)map->hva) +
3253 				offset_in_page(vmcs12->posted_intr_desc_addr));
3254 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3255 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3256 		} else {
3257 			/*
3258 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3259 			 * access the contents of the VMCS12 posted interrupt
3260 			 * descriptor. (Note that KVM may do this when it
3261 			 * should not, per the architectural specification.)
3262 			 */
3263 			vmx->nested.pi_desc = NULL;
3264 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3265 		}
3266 	}
3267 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3268 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3269 	else
3270 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3271 
3272 	return true;
3273 }
3274 
3275 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3276 {
3277 	/*
3278 	 * Note: nested_get_evmcs_page() also updates 'vp_assist_page' copy
3279 	 * in 'struct kvm_vcpu_hv' in case eVMCS is in use, this is mandatory
3280 	 * to make nested_evmcs_l2_tlb_flush_enabled() work correctly post
3281 	 * migration.
3282 	 */
3283 	if (!nested_get_evmcs_page(vcpu)) {
3284 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3285 				     __func__);
3286 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3287 		vcpu->run->internal.suberror =
3288 			KVM_INTERNAL_ERROR_EMULATION;
3289 		vcpu->run->internal.ndata = 0;
3290 
3291 		return false;
3292 	}
3293 
3294 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3295 		return false;
3296 
3297 	return true;
3298 }
3299 
3300 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3301 {
3302 	struct vmcs12 *vmcs12;
3303 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3304 	gpa_t dst;
3305 
3306 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3307 		return 0;
3308 
3309 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3310 		return 1;
3311 
3312 	/*
3313 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3314 	 * set is already checked as part of A/D emulation.
3315 	 */
3316 	vmcs12 = get_vmcs12(vcpu);
3317 	if (!nested_cpu_has_pml(vmcs12))
3318 		return 0;
3319 
3320 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3321 		vmx->nested.pml_full = true;
3322 		return 1;
3323 	}
3324 
3325 	gpa &= ~0xFFFull;
3326 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3327 
3328 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3329 				 offset_in_page(dst), sizeof(gpa)))
3330 		return 0;
3331 
3332 	vmcs12->guest_pml_index--;
3333 
3334 	return 0;
3335 }
3336 
3337 /*
3338  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3339  * for running VMX instructions (except VMXON, whose prerequisites are
3340  * slightly different). It also specifies what exception to inject otherwise.
3341  * Note that many of these exceptions have priority over VM exits, so they
3342  * don't have to be checked again here.
3343  */
3344 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3345 {
3346 	if (!to_vmx(vcpu)->nested.vmxon) {
3347 		kvm_queue_exception(vcpu, UD_VECTOR);
3348 		return 0;
3349 	}
3350 
3351 	if (vmx_get_cpl(vcpu)) {
3352 		kvm_inject_gp(vcpu, 0);
3353 		return 0;
3354 	}
3355 
3356 	return 1;
3357 }
3358 
3359 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3360 {
3361 	u8 rvi = vmx_get_rvi();
3362 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3363 
3364 	return ((rvi & 0xf0) > (vppr & 0xf0));
3365 }
3366 
3367 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3368 				   struct vmcs12 *vmcs12);
3369 
3370 /*
3371  * If from_vmentry is false, this is being called from state restore (either RSM
3372  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3373  *
3374  * Returns:
3375  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3376  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3377  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3378  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3379  */
3380 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3381 							bool from_vmentry)
3382 {
3383 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3384 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3385 	enum vm_entry_failure_code entry_failure_code;
3386 	bool evaluate_pending_interrupts;
3387 	union vmx_exit_reason exit_reason = {
3388 		.basic = EXIT_REASON_INVALID_STATE,
3389 		.failed_vmentry = 1,
3390 	};
3391 	u32 failed_index;
3392 
3393 	trace_kvm_nested_vmenter(kvm_rip_read(vcpu),
3394 				 vmx->nested.current_vmptr,
3395 				 vmcs12->guest_rip,
3396 				 vmcs12->guest_intr_status,
3397 				 vmcs12->vm_entry_intr_info_field,
3398 				 vmcs12->secondary_vm_exec_control & SECONDARY_EXEC_ENABLE_EPT,
3399 				 vmcs12->ept_pointer,
3400 				 vmcs12->guest_cr3,
3401 				 KVM_ISA_VMX);
3402 
3403 	kvm_service_local_tlb_flush_requests(vcpu);
3404 
3405 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3406 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3407 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3408 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3409 	if (!evaluate_pending_interrupts)
3410 		evaluate_pending_interrupts |= kvm_apic_has_pending_init_or_sipi(vcpu);
3411 
3412 	if (!vmx->nested.nested_run_pending ||
3413 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3414 		vmx->nested.pre_vmenter_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3415 	if (kvm_mpx_supported() &&
3416 	    (!vmx->nested.nested_run_pending ||
3417 	     !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
3418 		vmx->nested.pre_vmenter_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3419 
3420 	/*
3421 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3422 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3423 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3424 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3425 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3426 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3427 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3428 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3429 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3430 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3431 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3432 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3433 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3434 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3435 	 */
3436 	if (!enable_ept && !nested_early_check)
3437 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3438 
3439 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3440 
3441 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3442 
3443 	if (from_vmentry) {
3444 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3445 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3446 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3447 		}
3448 
3449 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3450 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3451 			return NVMX_VMENTRY_VMFAIL;
3452 		}
3453 
3454 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3455 						 &entry_failure_code)) {
3456 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3457 			vmcs12->exit_qualification = entry_failure_code;
3458 			goto vmentry_fail_vmexit;
3459 		}
3460 	}
3461 
3462 	enter_guest_mode(vcpu);
3463 
3464 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3465 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3466 		vmcs12->exit_qualification = entry_failure_code;
3467 		goto vmentry_fail_vmexit_guest_mode;
3468 	}
3469 
3470 	if (from_vmentry) {
3471 		failed_index = nested_vmx_load_msr(vcpu,
3472 						   vmcs12->vm_entry_msr_load_addr,
3473 						   vmcs12->vm_entry_msr_load_count);
3474 		if (failed_index) {
3475 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3476 			vmcs12->exit_qualification = failed_index;
3477 			goto vmentry_fail_vmexit_guest_mode;
3478 		}
3479 	} else {
3480 		/*
3481 		 * The MMU is not initialized to point at the right entities yet and
3482 		 * "get pages" would need to read data from the guest (i.e. we will
3483 		 * need to perform gpa to hpa translation). Request a call
3484 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3485 		 * have already been set at vmentry time and should not be reset.
3486 		 */
3487 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3488 	}
3489 
3490 	/*
3491 	 * Re-evaluate pending events if L1 had a pending IRQ/NMI/INIT/SIPI
3492 	 * when it executed VMLAUNCH/VMRESUME, as entering non-root mode can
3493 	 * effectively unblock various events, e.g. INIT/SIPI cause VM-Exit
3494 	 * unconditionally.
3495 	 */
3496 	if (unlikely(evaluate_pending_interrupts))
3497 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3498 
3499 	/*
3500 	 * Do not start the preemption timer hrtimer until after we know
3501 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3502 	 * the timer.
3503 	 */
3504 	vmx->nested.preemption_timer_expired = false;
3505 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3506 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3507 		vmx_start_preemption_timer(vcpu, timer_value);
3508 	}
3509 
3510 	/*
3511 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3512 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3513 	 * returned as far as L1 is concerned. It will only return (and set
3514 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3515 	 */
3516 	return NVMX_VMENTRY_SUCCESS;
3517 
3518 	/*
3519 	 * A failed consistency check that leads to a VMExit during L1's
3520 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3521 	 * 26.7 "VM-entry failures during or after loading guest state".
3522 	 */
3523 vmentry_fail_vmexit_guest_mode:
3524 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3525 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3526 	leave_guest_mode(vcpu);
3527 
3528 vmentry_fail_vmexit:
3529 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3530 
3531 	if (!from_vmentry)
3532 		return NVMX_VMENTRY_VMEXIT;
3533 
3534 	load_vmcs12_host_state(vcpu, vmcs12);
3535 	vmcs12->vm_exit_reason = exit_reason.full;
3536 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3537 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3538 	return NVMX_VMENTRY_VMEXIT;
3539 }
3540 
3541 /*
3542  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3543  * for running an L2 nested guest.
3544  */
3545 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3546 {
3547 	struct vmcs12 *vmcs12;
3548 	enum nvmx_vmentry_status status;
3549 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3550 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3551 	enum nested_evmptrld_status evmptrld_status;
3552 
3553 	if (!nested_vmx_check_permission(vcpu))
3554 		return 1;
3555 
3556 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3557 	if (evmptrld_status == EVMPTRLD_ERROR) {
3558 		kvm_queue_exception(vcpu, UD_VECTOR);
3559 		return 1;
3560 	}
3561 
3562 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
3563 
3564 	if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3565 		return nested_vmx_failInvalid(vcpu);
3566 
3567 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3568 	       vmx->nested.current_vmptr == INVALID_GPA))
3569 		return nested_vmx_failInvalid(vcpu);
3570 
3571 	vmcs12 = get_vmcs12(vcpu);
3572 
3573 	/*
3574 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3575 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3576 	 * rather than RFLAGS.ZF, and no error number is stored to the
3577 	 * VM-instruction error field.
3578 	 */
3579 	if (CC(vmcs12->hdr.shadow_vmcs))
3580 		return nested_vmx_failInvalid(vcpu);
3581 
3582 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3583 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3584 		/* Enlightened VMCS doesn't have launch state */
3585 		vmcs12->launch_state = !launch;
3586 	} else if (enable_shadow_vmcs) {
3587 		copy_shadow_to_vmcs12(vmx);
3588 	}
3589 
3590 	/*
3591 	 * The nested entry process starts with enforcing various prerequisites
3592 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3593 	 * they fail: As the SDM explains, some conditions should cause the
3594 	 * instruction to fail, while others will cause the instruction to seem
3595 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3596 	 * To speed up the normal (success) code path, we should avoid checking
3597 	 * for misconfigurations which will anyway be caught by the processor
3598 	 * when using the merged vmcs02.
3599 	 */
3600 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3601 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3602 
3603 	if (CC(vmcs12->launch_state == launch))
3604 		return nested_vmx_fail(vcpu,
3605 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3606 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3607 
3608 	if (nested_vmx_check_controls(vcpu, vmcs12))
3609 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3610 
3611 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3612 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3613 
3614 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3615 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3616 
3617 	/*
3618 	 * We're finally done with prerequisite checking, and can start with
3619 	 * the nested entry.
3620 	 */
3621 	vmx->nested.nested_run_pending = 1;
3622 	vmx->nested.has_preemption_timer_deadline = false;
3623 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3624 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3625 		goto vmentry_failed;
3626 
3627 	/* Emulate processing of posted interrupts on VM-Enter. */
3628 	if (nested_cpu_has_posted_intr(vmcs12) &&
3629 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3630 		vmx->nested.pi_pending = true;
3631 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3632 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3633 	}
3634 
3635 	/* Hide L1D cache contents from the nested guest.  */
3636 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3637 
3638 	/*
3639 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3640 	 * also be used as part of restoring nVMX state for
3641 	 * snapshot restore (migration).
3642 	 *
3643 	 * In this flow, it is assumed that vmcs12 cache was
3644 	 * transferred as part of captured nVMX state and should
3645 	 * therefore not be read from guest memory (which may not
3646 	 * exist on destination host yet).
3647 	 */
3648 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3649 
3650 	switch (vmcs12->guest_activity_state) {
3651 	case GUEST_ACTIVITY_HLT:
3652 		/*
3653 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3654 		 * awakened by event injection or by an NMI-window VM-exit or
3655 		 * by an interrupt-window VM-exit, halt the vcpu.
3656 		 */
3657 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3658 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3659 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3660 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3661 			vmx->nested.nested_run_pending = 0;
3662 			return kvm_emulate_halt_noskip(vcpu);
3663 		}
3664 		break;
3665 	case GUEST_ACTIVITY_WAIT_SIPI:
3666 		vmx->nested.nested_run_pending = 0;
3667 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3668 		break;
3669 	default:
3670 		break;
3671 	}
3672 
3673 	return 1;
3674 
3675 vmentry_failed:
3676 	vmx->nested.nested_run_pending = 0;
3677 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3678 		return 0;
3679 	if (status == NVMX_VMENTRY_VMEXIT)
3680 		return 1;
3681 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3682 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3683 }
3684 
3685 /*
3686  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3687  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3688  * This function returns the new value we should put in vmcs12.guest_cr0.
3689  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3690  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3691  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3692  *     didn't trap the bit, because if L1 did, so would L0).
3693  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3694  *     been modified by L2, and L1 knows it. So just leave the old value of
3695  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3696  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3697  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3698  *     changed these bits, and therefore they need to be updated, but L0
3699  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3700  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3701  */
3702 static inline unsigned long
3703 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3704 {
3705 	return
3706 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3707 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3708 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3709 			vcpu->arch.cr0_guest_owned_bits));
3710 }
3711 
3712 static inline unsigned long
3713 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3714 {
3715 	return
3716 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3717 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3718 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3719 			vcpu->arch.cr4_guest_owned_bits));
3720 }
3721 
3722 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3723 				      struct vmcs12 *vmcs12,
3724 				      u32 vm_exit_reason, u32 exit_intr_info)
3725 {
3726 	u32 idt_vectoring;
3727 	unsigned int nr;
3728 
3729 	/*
3730 	 * Per the SDM, VM-Exits due to double and triple faults are never
3731 	 * considered to occur during event delivery, even if the double/triple
3732 	 * fault is the result of an escalating vectoring issue.
3733 	 *
3734 	 * Note, the SDM qualifies the double fault behavior with "The original
3735 	 * event results in a double-fault exception".  It's unclear why the
3736 	 * qualification exists since exits due to double fault can occur only
3737 	 * while vectoring a different exception (injected events are never
3738 	 * subject to interception), i.e. there's _always_ an original event.
3739 	 *
3740 	 * The SDM also uses NMI as a confusing example for the "original event
3741 	 * causes the VM exit directly" clause.  NMI isn't special in any way,
3742 	 * the same rule applies to all events that cause an exit directly.
3743 	 * NMI is an odd choice for the example because NMIs can only occur on
3744 	 * instruction boundaries, i.e. they _can't_ occur during vectoring.
3745 	 */
3746 	if ((u16)vm_exit_reason == EXIT_REASON_TRIPLE_FAULT ||
3747 	    ((u16)vm_exit_reason == EXIT_REASON_EXCEPTION_NMI &&
3748 	     is_double_fault(exit_intr_info))) {
3749 		vmcs12->idt_vectoring_info_field = 0;
3750 	} else if (vcpu->arch.exception.injected) {
3751 		nr = vcpu->arch.exception.vector;
3752 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3753 
3754 		if (kvm_exception_is_soft(nr)) {
3755 			vmcs12->vm_exit_instruction_len =
3756 				vcpu->arch.event_exit_inst_len;
3757 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3758 		} else
3759 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3760 
3761 		if (vcpu->arch.exception.has_error_code) {
3762 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3763 			vmcs12->idt_vectoring_error_code =
3764 				vcpu->arch.exception.error_code;
3765 		}
3766 
3767 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3768 	} else if (vcpu->arch.nmi_injected) {
3769 		vmcs12->idt_vectoring_info_field =
3770 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3771 	} else if (vcpu->arch.interrupt.injected) {
3772 		nr = vcpu->arch.interrupt.nr;
3773 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3774 
3775 		if (vcpu->arch.interrupt.soft) {
3776 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3777 			vmcs12->vm_entry_instruction_len =
3778 				vcpu->arch.event_exit_inst_len;
3779 		} else
3780 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3781 
3782 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3783 	} else {
3784 		vmcs12->idt_vectoring_info_field = 0;
3785 	}
3786 }
3787 
3788 
3789 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3790 {
3791 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3792 	gfn_t gfn;
3793 
3794 	/*
3795 	 * Don't need to mark the APIC access page dirty; it is never
3796 	 * written to by the CPU during APIC virtualization.
3797 	 */
3798 
3799 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3800 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3801 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3802 	}
3803 
3804 	if (nested_cpu_has_posted_intr(vmcs12)) {
3805 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3806 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3807 	}
3808 }
3809 
3810 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3811 {
3812 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3813 	int max_irr;
3814 	void *vapic_page;
3815 	u16 status;
3816 
3817 	if (!vmx->nested.pi_pending)
3818 		return 0;
3819 
3820 	if (!vmx->nested.pi_desc)
3821 		goto mmio_needed;
3822 
3823 	vmx->nested.pi_pending = false;
3824 
3825 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3826 		return 0;
3827 
3828 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3829 	if (max_irr != 256) {
3830 		vapic_page = vmx->nested.virtual_apic_map.hva;
3831 		if (!vapic_page)
3832 			goto mmio_needed;
3833 
3834 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3835 			vapic_page, &max_irr);
3836 		status = vmcs_read16(GUEST_INTR_STATUS);
3837 		if ((u8)max_irr > ((u8)status & 0xff)) {
3838 			status &= ~0xff;
3839 			status |= (u8)max_irr;
3840 			vmcs_write16(GUEST_INTR_STATUS, status);
3841 		}
3842 	}
3843 
3844 	nested_mark_vmcs12_pages_dirty(vcpu);
3845 	return 0;
3846 
3847 mmio_needed:
3848 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3849 	return -ENXIO;
3850 }
3851 
3852 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu)
3853 {
3854 	struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
3855 	u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
3856 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3857 	unsigned long exit_qual;
3858 
3859 	if (ex->has_payload) {
3860 		exit_qual = ex->payload;
3861 	} else if (ex->vector == PF_VECTOR) {
3862 		exit_qual = vcpu->arch.cr2;
3863 	} else if (ex->vector == DB_VECTOR) {
3864 		exit_qual = vcpu->arch.dr6;
3865 		exit_qual &= ~DR6_BT;
3866 		exit_qual ^= DR6_ACTIVE_LOW;
3867 	} else {
3868 		exit_qual = 0;
3869 	}
3870 
3871 	if (ex->has_error_code) {
3872 		/*
3873 		 * Intel CPUs do not generate error codes with bits 31:16 set,
3874 		 * and more importantly VMX disallows setting bits 31:16 in the
3875 		 * injected error code for VM-Entry.  Drop the bits to mimic
3876 		 * hardware and avoid inducing failure on nested VM-Entry if L1
3877 		 * chooses to inject the exception back to L2.  AMD CPUs _do_
3878 		 * generate "full" 32-bit error codes, so KVM allows userspace
3879 		 * to inject exception error codes with bits 31:16 set.
3880 		 */
3881 		vmcs12->vm_exit_intr_error_code = (u16)ex->error_code;
3882 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3883 	}
3884 
3885 	if (kvm_exception_is_soft(ex->vector))
3886 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3887 	else
3888 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3889 
3890 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3891 	    vmx_get_nmi_mask(vcpu))
3892 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3893 
3894 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3895 }
3896 
3897 /*
3898  * Returns true if a debug trap is (likely) pending delivery.  Infer the class
3899  * of a #DB (trap-like vs. fault-like) from the exception payload (to-be-DR6).
3900  * Using the payload is flawed because code breakpoints (fault-like) and data
3901  * breakpoints (trap-like) set the same bits in DR6 (breakpoint detected), i.e.
3902  * this will return false positives if a to-be-injected code breakpoint #DB is
3903  * pending (from KVM's perspective, but not "pending" across an instruction
3904  * boundary).  ICEBP, a.k.a. INT1, is also not reflected here even though it
3905  * too is trap-like.
3906  *
3907  * KVM "works" despite these flaws as ICEBP isn't currently supported by the
3908  * emulator, Monitor Trap Flag is not marked pending on intercepted #DBs (the
3909  * #DB has already happened), and MTF isn't marked pending on code breakpoints
3910  * from the emulator (because such #DBs are fault-like and thus don't trigger
3911  * actions that fire on instruction retire).
3912  */
3913 static unsigned long vmx_get_pending_dbg_trap(struct kvm_queued_exception *ex)
3914 {
3915 	if (!ex->pending || ex->vector != DB_VECTOR)
3916 		return 0;
3917 
3918 	/* General Detect #DBs are always fault-like. */
3919 	return ex->payload & ~DR6_BD;
3920 }
3921 
3922 /*
3923  * Returns true if there's a pending #DB exception that is lower priority than
3924  * a pending Monitor Trap Flag VM-Exit.  TSS T-flag #DBs are not emulated by
3925  * KVM, but could theoretically be injected by userspace.  Note, this code is
3926  * imperfect, see above.
3927  */
3928 static bool vmx_is_low_priority_db_trap(struct kvm_queued_exception *ex)
3929 {
3930 	return vmx_get_pending_dbg_trap(ex) & ~DR6_BT;
3931 }
3932 
3933 /*
3934  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3935  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3936  * represents these debug traps with a payload that is said to be compatible
3937  * with the 'pending debug exceptions' field, write the payload to the VMCS
3938  * field if a VM-exit is delivered before the debug trap.
3939  */
3940 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3941 {
3942 	unsigned long pending_dbg;
3943 
3944 	pending_dbg = vmx_get_pending_dbg_trap(&vcpu->arch.exception);
3945 	if (pending_dbg)
3946 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, pending_dbg);
3947 }
3948 
3949 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3950 {
3951 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3952 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3953 }
3954 
3955 static bool vmx_has_nested_events(struct kvm_vcpu *vcpu)
3956 {
3957 	return nested_vmx_preemption_timer_pending(vcpu) ||
3958 	       to_vmx(vcpu)->nested.mtf_pending;
3959 }
3960 
3961 /*
3962  * Per the Intel SDM's table "Priority Among Concurrent Events", with minor
3963  * edits to fill in missing examples, e.g. #DB due to split-lock accesses,
3964  * and less minor edits to splice in the priority of VMX Non-Root specific
3965  * events, e.g. MTF and NMI/INTR-window exiting.
3966  *
3967  * 1 Hardware Reset and Machine Checks
3968  *	- RESET
3969  *	- Machine Check
3970  *
3971  * 2 Trap on Task Switch
3972  *	- T flag in TSS is set (on task switch)
3973  *
3974  * 3 External Hardware Interventions
3975  *	- FLUSH
3976  *	- STOPCLK
3977  *	- SMI
3978  *	- INIT
3979  *
3980  * 3.5 Monitor Trap Flag (MTF) VM-exit[1]
3981  *
3982  * 4 Traps on Previous Instruction
3983  *	- Breakpoints
3984  *	- Trap-class Debug Exceptions (#DB due to TF flag set, data/I-O
3985  *	  breakpoint, or #DB due to a split-lock access)
3986  *
3987  * 4.3	VMX-preemption timer expired VM-exit
3988  *
3989  * 4.6	NMI-window exiting VM-exit[2]
3990  *
3991  * 5 Nonmaskable Interrupts (NMI)
3992  *
3993  * 5.5 Interrupt-window exiting VM-exit and Virtual-interrupt delivery
3994  *
3995  * 6 Maskable Hardware Interrupts
3996  *
3997  * 7 Code Breakpoint Fault
3998  *
3999  * 8 Faults from Fetching Next Instruction
4000  *	- Code-Segment Limit Violation
4001  *	- Code Page Fault
4002  *	- Control protection exception (missing ENDBRANCH at target of indirect
4003  *					call or jump)
4004  *
4005  * 9 Faults from Decoding Next Instruction
4006  *	- Instruction length > 15 bytes
4007  *	- Invalid Opcode
4008  *	- Coprocessor Not Available
4009  *
4010  *10 Faults on Executing Instruction
4011  *	- Overflow
4012  *	- Bound error
4013  *	- Invalid TSS
4014  *	- Segment Not Present
4015  *	- Stack fault
4016  *	- General Protection
4017  *	- Data Page Fault
4018  *	- Alignment Check
4019  *	- x86 FPU Floating-point exception
4020  *	- SIMD floating-point exception
4021  *	- Virtualization exception
4022  *	- Control protection exception
4023  *
4024  * [1] Per the "Monitor Trap Flag" section: System-management interrupts (SMIs),
4025  *     INIT signals, and higher priority events take priority over MTF VM exits.
4026  *     MTF VM exits take priority over debug-trap exceptions and lower priority
4027  *     events.
4028  *
4029  * [2] Debug-trap exceptions and higher priority events take priority over VM exits
4030  *     caused by the VMX-preemption timer.  VM exits caused by the VMX-preemption
4031  *     timer take priority over VM exits caused by the "NMI-window exiting"
4032  *     VM-execution control and lower priority events.
4033  *
4034  * [3] Debug-trap exceptions and higher priority events take priority over VM exits
4035  *     caused by "NMI-window exiting".  VM exits caused by this control take
4036  *     priority over non-maskable interrupts (NMIs) and lower priority events.
4037  *
4038  * [4] Virtual-interrupt delivery has the same priority as that of VM exits due to
4039  *     the 1-setting of the "interrupt-window exiting" VM-execution control.  Thus,
4040  *     non-maskable interrupts (NMIs) and higher priority events take priority over
4041  *     delivery of a virtual interrupt; delivery of a virtual interrupt takes
4042  *     priority over external interrupts and lower priority events.
4043  */
4044 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
4045 {
4046 	struct kvm_lapic *apic = vcpu->arch.apic;
4047 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4048 	/*
4049 	 * Only a pending nested run blocks a pending exception.  If there is a
4050 	 * previously injected event, the pending exception occurred while said
4051 	 * event was being delivered and thus needs to be handled.
4052 	 */
4053 	bool block_nested_exceptions = vmx->nested.nested_run_pending;
4054 	/*
4055 	 * New events (not exceptions) are only recognized at instruction
4056 	 * boundaries.  If an event needs reinjection, then KVM is handling a
4057 	 * VM-Exit that occurred _during_ instruction execution; new events are
4058 	 * blocked until the instruction completes.
4059 	 */
4060 	bool block_nested_events = block_nested_exceptions ||
4061 				   kvm_event_needs_reinjection(vcpu);
4062 
4063 	if (lapic_in_kernel(vcpu) &&
4064 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
4065 		if (block_nested_events)
4066 			return -EBUSY;
4067 		nested_vmx_update_pending_dbg(vcpu);
4068 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
4069 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
4070 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
4071 
4072 		/* MTF is discarded if the vCPU is in WFS. */
4073 		vmx->nested.mtf_pending = false;
4074 		return 0;
4075 	}
4076 
4077 	if (lapic_in_kernel(vcpu) &&
4078 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
4079 		if (block_nested_events)
4080 			return -EBUSY;
4081 
4082 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
4083 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
4084 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
4085 						apic->sipi_vector & 0xFFUL);
4086 			return 0;
4087 		}
4088 		/* Fallthrough, the SIPI is completely ignored. */
4089 	}
4090 
4091 	/*
4092 	 * Process exceptions that are higher priority than Monitor Trap Flag:
4093 	 * fault-like exceptions, TSS T flag #DB (not emulated by KVM, but
4094 	 * could theoretically come in from userspace), and ICEBP (INT1).
4095 	 *
4096 	 * TODO: SMIs have higher priority than MTF and trap-like #DBs (except
4097 	 * for TSS T flag #DBs).  KVM also doesn't save/restore pending MTF
4098 	 * across SMI/RSM as it should; that needs to be addressed in order to
4099 	 * prioritize SMI over MTF and trap-like #DBs.
4100 	 */
4101 	if (vcpu->arch.exception_vmexit.pending &&
4102 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception_vmexit)) {
4103 		if (block_nested_exceptions)
4104 			return -EBUSY;
4105 
4106 		nested_vmx_inject_exception_vmexit(vcpu);
4107 		return 0;
4108 	}
4109 
4110 	if (vcpu->arch.exception.pending &&
4111 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception)) {
4112 		if (block_nested_exceptions)
4113 			return -EBUSY;
4114 		goto no_vmexit;
4115 	}
4116 
4117 	if (vmx->nested.mtf_pending) {
4118 		if (block_nested_events)
4119 			return -EBUSY;
4120 		nested_vmx_update_pending_dbg(vcpu);
4121 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
4122 		return 0;
4123 	}
4124 
4125 	if (vcpu->arch.exception_vmexit.pending) {
4126 		if (block_nested_exceptions)
4127 			return -EBUSY;
4128 
4129 		nested_vmx_inject_exception_vmexit(vcpu);
4130 		return 0;
4131 	}
4132 
4133 	if (vcpu->arch.exception.pending) {
4134 		if (block_nested_exceptions)
4135 			return -EBUSY;
4136 		goto no_vmexit;
4137 	}
4138 
4139 	if (nested_vmx_preemption_timer_pending(vcpu)) {
4140 		if (block_nested_events)
4141 			return -EBUSY;
4142 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
4143 		return 0;
4144 	}
4145 
4146 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
4147 		if (block_nested_events)
4148 			return -EBUSY;
4149 		goto no_vmexit;
4150 	}
4151 
4152 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
4153 		if (block_nested_events)
4154 			return -EBUSY;
4155 		if (!nested_exit_on_nmi(vcpu))
4156 			goto no_vmexit;
4157 
4158 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
4159 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
4160 				  INTR_INFO_VALID_MASK, 0);
4161 		/*
4162 		 * The NMI-triggered VM exit counts as injection:
4163 		 * clear this one and block further NMIs.
4164 		 */
4165 		vcpu->arch.nmi_pending = 0;
4166 		vmx_set_nmi_mask(vcpu, true);
4167 		return 0;
4168 	}
4169 
4170 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
4171 		if (block_nested_events)
4172 			return -EBUSY;
4173 		if (!nested_exit_on_intr(vcpu))
4174 			goto no_vmexit;
4175 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
4176 		return 0;
4177 	}
4178 
4179 no_vmexit:
4180 	return vmx_complete_nested_posted_interrupt(vcpu);
4181 }
4182 
4183 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
4184 {
4185 	ktime_t remaining =
4186 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
4187 	u64 value;
4188 
4189 	if (ktime_to_ns(remaining) <= 0)
4190 		return 0;
4191 
4192 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
4193 	do_div(value, 1000000);
4194 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
4195 }
4196 
4197 static bool is_vmcs12_ext_field(unsigned long field)
4198 {
4199 	switch (field) {
4200 	case GUEST_ES_SELECTOR:
4201 	case GUEST_CS_SELECTOR:
4202 	case GUEST_SS_SELECTOR:
4203 	case GUEST_DS_SELECTOR:
4204 	case GUEST_FS_SELECTOR:
4205 	case GUEST_GS_SELECTOR:
4206 	case GUEST_LDTR_SELECTOR:
4207 	case GUEST_TR_SELECTOR:
4208 	case GUEST_ES_LIMIT:
4209 	case GUEST_CS_LIMIT:
4210 	case GUEST_SS_LIMIT:
4211 	case GUEST_DS_LIMIT:
4212 	case GUEST_FS_LIMIT:
4213 	case GUEST_GS_LIMIT:
4214 	case GUEST_LDTR_LIMIT:
4215 	case GUEST_TR_LIMIT:
4216 	case GUEST_GDTR_LIMIT:
4217 	case GUEST_IDTR_LIMIT:
4218 	case GUEST_ES_AR_BYTES:
4219 	case GUEST_DS_AR_BYTES:
4220 	case GUEST_FS_AR_BYTES:
4221 	case GUEST_GS_AR_BYTES:
4222 	case GUEST_LDTR_AR_BYTES:
4223 	case GUEST_TR_AR_BYTES:
4224 	case GUEST_ES_BASE:
4225 	case GUEST_CS_BASE:
4226 	case GUEST_SS_BASE:
4227 	case GUEST_DS_BASE:
4228 	case GUEST_FS_BASE:
4229 	case GUEST_GS_BASE:
4230 	case GUEST_LDTR_BASE:
4231 	case GUEST_TR_BASE:
4232 	case GUEST_GDTR_BASE:
4233 	case GUEST_IDTR_BASE:
4234 	case GUEST_PENDING_DBG_EXCEPTIONS:
4235 	case GUEST_BNDCFGS:
4236 		return true;
4237 	default:
4238 		break;
4239 	}
4240 
4241 	return false;
4242 }
4243 
4244 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4245 				       struct vmcs12 *vmcs12)
4246 {
4247 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4248 
4249 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4250 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4251 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4252 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4253 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4254 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4255 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4256 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4257 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4258 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4259 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4260 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4261 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4262 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4263 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4264 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4265 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4266 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4267 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4268 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4269 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4270 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4271 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4272 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4273 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4274 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4275 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4276 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4277 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4278 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4279 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4280 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4281 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4282 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4283 	vmcs12->guest_pending_dbg_exceptions =
4284 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4285 
4286 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4287 }
4288 
4289 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4290 				       struct vmcs12 *vmcs12)
4291 {
4292 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4293 	int cpu;
4294 
4295 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4296 		return;
4297 
4298 
4299 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4300 
4301 	cpu = get_cpu();
4302 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4303 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4304 
4305 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4306 
4307 	vmx->loaded_vmcs = &vmx->vmcs01;
4308 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4309 	put_cpu();
4310 }
4311 
4312 /*
4313  * Update the guest state fields of vmcs12 to reflect changes that
4314  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4315  * VM-entry controls is also updated, since this is really a guest
4316  * state bit.)
4317  */
4318 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4319 {
4320 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4321 
4322 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4323 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4324 
4325 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4326 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4327 
4328 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4329 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4330 
4331 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4332 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4333 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4334 
4335 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4336 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4337 
4338 	vmcs12->guest_interruptibility_info =
4339 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4340 
4341 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4342 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4343 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4344 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4345 	else
4346 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4347 
4348 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4349 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4350 	    !vmx->nested.nested_run_pending)
4351 		vmcs12->vmx_preemption_timer_value =
4352 			vmx_get_preemption_timer_value(vcpu);
4353 
4354 	/*
4355 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4356 	 * own CR3 without exiting. If it has changed it, we must keep it.
4357 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4358 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4359 	 *
4360 	 * Additionally, restore L2's PDPTR to vmcs12.
4361 	 */
4362 	if (enable_ept) {
4363 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4364 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4365 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4366 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4367 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4368 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4369 		}
4370 	}
4371 
4372 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4373 
4374 	if (nested_cpu_has_vid(vmcs12))
4375 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4376 
4377 	vmcs12->vm_entry_controls =
4378 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4379 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4380 
4381 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4382 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4383 
4384 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4385 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4386 }
4387 
4388 /*
4389  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4390  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4391  * and this function updates it to reflect the changes to the guest state while
4392  * L2 was running (and perhaps made some exits which were handled directly by L0
4393  * without going back to L1), and to reflect the exit reason.
4394  * Note that we do not have to copy here all VMCS fields, just those that
4395  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4396  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4397  * which already writes to vmcs12 directly.
4398  */
4399 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4400 			   u32 vm_exit_reason, u32 exit_intr_info,
4401 			   unsigned long exit_qualification)
4402 {
4403 	/* update exit information fields: */
4404 	vmcs12->vm_exit_reason = vm_exit_reason;
4405 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4406 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4407 	vmcs12->exit_qualification = exit_qualification;
4408 
4409 	/*
4410 	 * On VM-Exit due to a failed VM-Entry, the VMCS isn't marked launched
4411 	 * and only EXIT_REASON and EXIT_QUALIFICATION are updated, all other
4412 	 * exit info fields are unmodified.
4413 	 */
4414 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4415 		vmcs12->launch_state = 1;
4416 
4417 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4418 		 * instead of reading the real value. */
4419 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4420 
4421 		/*
4422 		 * Transfer the event that L0 or L1 may wanted to inject into
4423 		 * L2 to IDT_VECTORING_INFO_FIELD.
4424 		 */
4425 		vmcs12_save_pending_event(vcpu, vmcs12,
4426 					  vm_exit_reason, exit_intr_info);
4427 
4428 		vmcs12->vm_exit_intr_info = exit_intr_info;
4429 		vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4430 		vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4431 
4432 		/*
4433 		 * According to spec, there's no need to store the guest's
4434 		 * MSRs if the exit is due to a VM-entry failure that occurs
4435 		 * during or after loading the guest state. Since this exit
4436 		 * does not fall in that category, we need to save the MSRs.
4437 		 */
4438 		if (nested_vmx_store_msr(vcpu,
4439 					 vmcs12->vm_exit_msr_store_addr,
4440 					 vmcs12->vm_exit_msr_store_count))
4441 			nested_vmx_abort(vcpu,
4442 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4443 	}
4444 }
4445 
4446 /*
4447  * A part of what we need to when the nested L2 guest exits and we want to
4448  * run its L1 parent, is to reset L1's guest state to the host state specified
4449  * in vmcs12.
4450  * This function is to be called not only on normal nested exit, but also on
4451  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4452  * Failures During or After Loading Guest State").
4453  * This function should be called when the active VMCS is L1's (vmcs01).
4454  */
4455 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4456 				   struct vmcs12 *vmcs12)
4457 {
4458 	enum vm_entry_failure_code ignored;
4459 	struct kvm_segment seg;
4460 
4461 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4462 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4463 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4464 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4465 	else
4466 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4467 	vmx_set_efer(vcpu, vcpu->arch.efer);
4468 
4469 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4470 	kvm_rip_write(vcpu, vmcs12->host_rip);
4471 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4472 	vmx_set_interrupt_shadow(vcpu, 0);
4473 
4474 	/*
4475 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4476 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4477 	 *
4478 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4479 	 * (KVM doesn't change it);
4480 	 */
4481 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4482 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4483 
4484 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4485 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4486 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4487 
4488 	nested_ept_uninit_mmu_context(vcpu);
4489 
4490 	/*
4491 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4492 	 * couldn't have changed.
4493 	 */
4494 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4495 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4496 
4497 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4498 
4499 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4500 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4501 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4502 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4503 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4504 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4505 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4506 
4507 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4508 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4509 		vmcs_write64(GUEST_BNDCFGS, 0);
4510 
4511 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4512 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4513 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4514 	}
4515 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
4516 	    intel_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)))
4517 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4518 					 vmcs12->host_ia32_perf_global_ctrl));
4519 
4520 	/* Set L1 segment info according to Intel SDM
4521 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4522 	seg = (struct kvm_segment) {
4523 		.base = 0,
4524 		.limit = 0xFFFFFFFF,
4525 		.selector = vmcs12->host_cs_selector,
4526 		.type = 11,
4527 		.present = 1,
4528 		.s = 1,
4529 		.g = 1
4530 	};
4531 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4532 		seg.l = 1;
4533 	else
4534 		seg.db = 1;
4535 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4536 	seg = (struct kvm_segment) {
4537 		.base = 0,
4538 		.limit = 0xFFFFFFFF,
4539 		.type = 3,
4540 		.present = 1,
4541 		.s = 1,
4542 		.db = 1,
4543 		.g = 1
4544 	};
4545 	seg.selector = vmcs12->host_ds_selector;
4546 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4547 	seg.selector = vmcs12->host_es_selector;
4548 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4549 	seg.selector = vmcs12->host_ss_selector;
4550 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4551 	seg.selector = vmcs12->host_fs_selector;
4552 	seg.base = vmcs12->host_fs_base;
4553 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4554 	seg.selector = vmcs12->host_gs_selector;
4555 	seg.base = vmcs12->host_gs_base;
4556 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4557 	seg = (struct kvm_segment) {
4558 		.base = vmcs12->host_tr_base,
4559 		.limit = 0x67,
4560 		.selector = vmcs12->host_tr_selector,
4561 		.type = 11,
4562 		.present = 1
4563 	};
4564 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4565 
4566 	memset(&seg, 0, sizeof(seg));
4567 	seg.unusable = 1;
4568 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4569 
4570 	kvm_set_dr(vcpu, 7, 0x400);
4571 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4572 
4573 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4574 				vmcs12->vm_exit_msr_load_count))
4575 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4576 
4577 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4578 }
4579 
4580 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4581 {
4582 	struct vmx_uret_msr *efer_msr;
4583 	unsigned int i;
4584 
4585 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4586 		return vmcs_read64(GUEST_IA32_EFER);
4587 
4588 	if (cpu_has_load_ia32_efer())
4589 		return host_efer;
4590 
4591 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4592 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4593 			return vmx->msr_autoload.guest.val[i].value;
4594 	}
4595 
4596 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4597 	if (efer_msr)
4598 		return efer_msr->data;
4599 
4600 	return host_efer;
4601 }
4602 
4603 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4604 {
4605 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4606 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4607 	struct vmx_msr_entry g, h;
4608 	gpa_t gpa;
4609 	u32 i, j;
4610 
4611 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4612 
4613 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4614 		/*
4615 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4616 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4617 		 * and vcpu->arch.dr7 is not squirreled away before the
4618 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4619 		 */
4620 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4621 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4622 		else
4623 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4624 	}
4625 
4626 	/*
4627 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4628 	 * handle a variety of side effects to KVM's software model.
4629 	 */
4630 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4631 
4632 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4633 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4634 
4635 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4636 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4637 
4638 	nested_ept_uninit_mmu_context(vcpu);
4639 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4640 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4641 
4642 	/*
4643 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4644 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4645 	 * VMFail, like everything else we just need to ensure our
4646 	 * software model is up-to-date.
4647 	 */
4648 	if (enable_ept && is_pae_paging(vcpu))
4649 		ept_save_pdptrs(vcpu);
4650 
4651 	kvm_mmu_reset_context(vcpu);
4652 
4653 	/*
4654 	 * This nasty bit of open coding is a compromise between blindly
4655 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4656 	 * of VMFail), leaving the nested VM's MSRs in the software model
4657 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4658 	 * expensive since the lists are unbound by hardware).  For each
4659 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4660 	 * list, reload it from the exit load list if it exists and differs
4661 	 * from the guest value.  The intent is to stuff host state as
4662 	 * silently as possible, not to fully process the exit load list.
4663 	 */
4664 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4665 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4666 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4667 			pr_debug_ratelimited(
4668 				"%s read MSR index failed (%u, 0x%08llx)\n",
4669 				__func__, i, gpa);
4670 			goto vmabort;
4671 		}
4672 
4673 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4674 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4675 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4676 				pr_debug_ratelimited(
4677 					"%s read MSR failed (%u, 0x%08llx)\n",
4678 					__func__, j, gpa);
4679 				goto vmabort;
4680 			}
4681 			if (h.index != g.index)
4682 				continue;
4683 			if (h.value == g.value)
4684 				break;
4685 
4686 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4687 				pr_debug_ratelimited(
4688 					"%s check failed (%u, 0x%x, 0x%x)\n",
4689 					__func__, j, h.index, h.reserved);
4690 				goto vmabort;
4691 			}
4692 
4693 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4694 				pr_debug_ratelimited(
4695 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4696 					__func__, j, h.index, h.value);
4697 				goto vmabort;
4698 			}
4699 		}
4700 	}
4701 
4702 	return;
4703 
4704 vmabort:
4705 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4706 }
4707 
4708 /*
4709  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4710  * and modify vmcs12 to make it see what it would expect to see there if
4711  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4712  */
4713 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4714 		       u32 exit_intr_info, unsigned long exit_qualification)
4715 {
4716 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4717 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4718 
4719 	/* Pending MTF traps are discarded on VM-Exit. */
4720 	vmx->nested.mtf_pending = false;
4721 
4722 	/* trying to cancel vmlaunch/vmresume is a bug */
4723 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4724 
4725 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4726 		/*
4727 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4728 		 * Enlightened VMCS after migration and we still need to
4729 		 * do that when something is forcing L2->L1 exit prior to
4730 		 * the first L2 run.
4731 		 */
4732 		(void)nested_get_evmcs_page(vcpu);
4733 	}
4734 
4735 	/* Service pending TLB flush requests for L2 before switching to L1. */
4736 	kvm_service_local_tlb_flush_requests(vcpu);
4737 
4738 	/*
4739 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4740 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4741 	 * up-to-date before switching to L1.
4742 	 */
4743 	if (enable_ept && is_pae_paging(vcpu))
4744 		vmx_ept_load_pdptrs(vcpu);
4745 
4746 	leave_guest_mode(vcpu);
4747 
4748 	if (nested_cpu_has_preemption_timer(vmcs12))
4749 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4750 
4751 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4752 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4753 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4754 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4755 	}
4756 
4757 	if (likely(!vmx->fail)) {
4758 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4759 
4760 		if (vm_exit_reason != -1)
4761 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4762 				       exit_intr_info, exit_qualification);
4763 
4764 		/*
4765 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4766 		 * also be used to capture vmcs12 cache as part of
4767 		 * capturing nVMX state for snapshot (migration).
4768 		 *
4769 		 * Otherwise, this flush will dirty guest memory at a
4770 		 * point it is already assumed by user-space to be
4771 		 * immutable.
4772 		 */
4773 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4774 	} else {
4775 		/*
4776 		 * The only expected VM-instruction error is "VM entry with
4777 		 * invalid control field(s)." Anything else indicates a
4778 		 * problem with L0.  And we should never get here with a
4779 		 * VMFail of any type if early consistency checks are enabled.
4780 		 */
4781 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4782 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4783 		WARN_ON_ONCE(nested_early_check);
4784 	}
4785 
4786 	/*
4787 	 * Drop events/exceptions that were queued for re-injection to L2
4788 	 * (picked up via vmx_complete_interrupts()), as well as exceptions
4789 	 * that were pending for L2.  Note, this must NOT be hoisted above
4790 	 * prepare_vmcs12(), events/exceptions queued for re-injection need to
4791 	 * be captured in vmcs12 (see vmcs12_save_pending_event()).
4792 	 */
4793 	vcpu->arch.nmi_injected = false;
4794 	kvm_clear_exception_queue(vcpu);
4795 	kvm_clear_interrupt_queue(vcpu);
4796 
4797 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4798 
4799 	/*
4800 	 * If IBRS is advertised to the vCPU, KVM must flush the indirect
4801 	 * branch predictors when transitioning from L2 to L1, as L1 expects
4802 	 * hardware (KVM in this case) to provide separate predictor modes.
4803 	 * Bare metal isolates VMX root (host) from VMX non-root (guest), but
4804 	 * doesn't isolate different VMCSs, i.e. in this case, doesn't provide
4805 	 * separate modes for L2 vs L1.
4806 	 */
4807 	if (guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
4808 		indirect_branch_prediction_barrier();
4809 
4810 	/* Update any VMCS fields that might have changed while L2 ran */
4811 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4812 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4813 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4814 	if (kvm_caps.has_tsc_control)
4815 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4816 
4817 	if (vmx->nested.l1_tpr_threshold != -1)
4818 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4819 
4820 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4821 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4822 		vmx_set_virtual_apic_mode(vcpu);
4823 	}
4824 
4825 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4826 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4827 		vmx_update_cpu_dirty_logging(vcpu);
4828 	}
4829 
4830 	/* Unpin physical memory we referred to in vmcs02 */
4831 	kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
4832 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4833 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4834 	vmx->nested.pi_desc = NULL;
4835 
4836 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4837 		vmx->nested.reload_vmcs01_apic_access_page = false;
4838 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4839 	}
4840 
4841 	if (vmx->nested.update_vmcs01_apicv_status) {
4842 		vmx->nested.update_vmcs01_apicv_status = false;
4843 		kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
4844 	}
4845 
4846 	if ((vm_exit_reason != -1) &&
4847 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4848 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4849 
4850 	/* in case we halted in L2 */
4851 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4852 
4853 	if (likely(!vmx->fail)) {
4854 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4855 		    nested_exit_intr_ack_set(vcpu)) {
4856 			int irq = kvm_cpu_get_interrupt(vcpu);
4857 			WARN_ON(irq < 0);
4858 			vmcs12->vm_exit_intr_info = irq |
4859 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4860 		}
4861 
4862 		if (vm_exit_reason != -1)
4863 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4864 						       vmcs12->exit_qualification,
4865 						       vmcs12->idt_vectoring_info_field,
4866 						       vmcs12->vm_exit_intr_info,
4867 						       vmcs12->vm_exit_intr_error_code,
4868 						       KVM_ISA_VMX);
4869 
4870 		load_vmcs12_host_state(vcpu, vmcs12);
4871 
4872 		return;
4873 	}
4874 
4875 	/*
4876 	 * After an early L2 VM-entry failure, we're now back
4877 	 * in L1 which thinks it just finished a VMLAUNCH or
4878 	 * VMRESUME instruction, so we need to set the failure
4879 	 * flag and the VM-instruction error field of the VMCS
4880 	 * accordingly, and skip the emulated instruction.
4881 	 */
4882 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4883 
4884 	/*
4885 	 * Restore L1's host state to KVM's software model.  We're here
4886 	 * because a consistency check was caught by hardware, which
4887 	 * means some amount of guest state has been propagated to KVM's
4888 	 * model and needs to be unwound to the host's state.
4889 	 */
4890 	nested_vmx_restore_host_state(vcpu);
4891 
4892 	vmx->fail = 0;
4893 }
4894 
4895 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4896 {
4897 	kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4898 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4899 }
4900 
4901 /*
4902  * Decode the memory-address operand of a vmx instruction, as recorded on an
4903  * exit caused by such an instruction (run by a guest hypervisor).
4904  * On success, returns 0. When the operand is invalid, returns 1 and throws
4905  * #UD, #GP, or #SS.
4906  */
4907 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4908 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4909 {
4910 	gva_t off;
4911 	bool exn;
4912 	struct kvm_segment s;
4913 
4914 	/*
4915 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4916 	 * Execution", on an exit, vmx_instruction_info holds most of the
4917 	 * addressing components of the operand. Only the displacement part
4918 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4919 	 * For how an actual address is calculated from all these components,
4920 	 * refer to Vol. 1, "Operand Addressing".
4921 	 */
4922 	int  scaling = vmx_instruction_info & 3;
4923 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4924 	bool is_reg = vmx_instruction_info & (1u << 10);
4925 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4926 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4927 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4928 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4929 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4930 
4931 	if (is_reg) {
4932 		kvm_queue_exception(vcpu, UD_VECTOR);
4933 		return 1;
4934 	}
4935 
4936 	/* Addr = segment_base + offset */
4937 	/* offset = base + [index * scale] + displacement */
4938 	off = exit_qualification; /* holds the displacement */
4939 	if (addr_size == 1)
4940 		off = (gva_t)sign_extend64(off, 31);
4941 	else if (addr_size == 0)
4942 		off = (gva_t)sign_extend64(off, 15);
4943 	if (base_is_valid)
4944 		off += kvm_register_read(vcpu, base_reg);
4945 	if (index_is_valid)
4946 		off += kvm_register_read(vcpu, index_reg) << scaling;
4947 	vmx_get_segment(vcpu, &s, seg_reg);
4948 
4949 	/*
4950 	 * The effective address, i.e. @off, of a memory operand is truncated
4951 	 * based on the address size of the instruction.  Note that this is
4952 	 * the *effective address*, i.e. the address prior to accounting for
4953 	 * the segment's base.
4954 	 */
4955 	if (addr_size == 1) /* 32 bit */
4956 		off &= 0xffffffff;
4957 	else if (addr_size == 0) /* 16 bit */
4958 		off &= 0xffff;
4959 
4960 	/* Checks for #GP/#SS exceptions. */
4961 	exn = false;
4962 	if (is_long_mode(vcpu)) {
4963 		/*
4964 		 * The virtual/linear address is never truncated in 64-bit
4965 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4966 		 * address when using FS/GS with a non-zero base.
4967 		 */
4968 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4969 			*ret = s.base + off;
4970 		else
4971 			*ret = off;
4972 
4973 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4974 		 * non-canonical form. This is the only check on the memory
4975 		 * destination for long mode!
4976 		 */
4977 		exn = is_noncanonical_address(*ret, vcpu);
4978 	} else {
4979 		/*
4980 		 * When not in long mode, the virtual/linear address is
4981 		 * unconditionally truncated to 32 bits regardless of the
4982 		 * address size.
4983 		 */
4984 		*ret = (s.base + off) & 0xffffffff;
4985 
4986 		/* Protected mode: apply checks for segment validity in the
4987 		 * following order:
4988 		 * - segment type check (#GP(0) may be thrown)
4989 		 * - usability check (#GP(0)/#SS(0))
4990 		 * - limit check (#GP(0)/#SS(0))
4991 		 */
4992 		if (wr)
4993 			/* #GP(0) if the destination operand is located in a
4994 			 * read-only data segment or any code segment.
4995 			 */
4996 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4997 		else
4998 			/* #GP(0) if the source operand is located in an
4999 			 * execute-only code segment
5000 			 */
5001 			exn = ((s.type & 0xa) == 8);
5002 		if (exn) {
5003 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5004 			return 1;
5005 		}
5006 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
5007 		 */
5008 		exn = (s.unusable != 0);
5009 
5010 		/*
5011 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
5012 		 * outside the segment limit.  All CPUs that support VMX ignore
5013 		 * limit checks for flat segments, i.e. segments with base==0,
5014 		 * limit==0xffffffff and of type expand-up data or code.
5015 		 */
5016 		if (!(s.base == 0 && s.limit == 0xffffffff &&
5017 		     ((s.type & 8) || !(s.type & 4))))
5018 			exn = exn || ((u64)off + len - 1 > s.limit);
5019 	}
5020 	if (exn) {
5021 		kvm_queue_exception_e(vcpu,
5022 				      seg_reg == VCPU_SREG_SS ?
5023 						SS_VECTOR : GP_VECTOR,
5024 				      0);
5025 		return 1;
5026 	}
5027 
5028 	return 0;
5029 }
5030 
5031 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
5032 				int *ret)
5033 {
5034 	gva_t gva;
5035 	struct x86_exception e;
5036 	int r;
5037 
5038 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5039 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
5040 				sizeof(*vmpointer), &gva)) {
5041 		*ret = 1;
5042 		return -EINVAL;
5043 	}
5044 
5045 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
5046 	if (r != X86EMUL_CONTINUE) {
5047 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
5048 		return -EINVAL;
5049 	}
5050 
5051 	return 0;
5052 }
5053 
5054 /*
5055  * Allocate a shadow VMCS and associate it with the currently loaded
5056  * VMCS, unless such a shadow VMCS already exists. The newly allocated
5057  * VMCS is also VMCLEARed, so that it is ready for use.
5058  */
5059 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
5060 {
5061 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5062 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
5063 
5064 	/*
5065 	 * KVM allocates a shadow VMCS only when L1 executes VMXON and frees it
5066 	 * when L1 executes VMXOFF or the vCPU is forced out of nested
5067 	 * operation.  VMXON faults if the CPU is already post-VMXON, so it
5068 	 * should be impossible to already have an allocated shadow VMCS.  KVM
5069 	 * doesn't support virtualization of VMCS shadowing, so vmcs01 should
5070 	 * always be the loaded VMCS.
5071 	 */
5072 	if (WARN_ON(loaded_vmcs != &vmx->vmcs01 || loaded_vmcs->shadow_vmcs))
5073 		return loaded_vmcs->shadow_vmcs;
5074 
5075 	loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
5076 	if (loaded_vmcs->shadow_vmcs)
5077 		vmcs_clear(loaded_vmcs->shadow_vmcs);
5078 
5079 	return loaded_vmcs->shadow_vmcs;
5080 }
5081 
5082 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
5083 {
5084 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5085 	int r;
5086 
5087 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
5088 	if (r < 0)
5089 		goto out_vmcs02;
5090 
5091 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5092 	if (!vmx->nested.cached_vmcs12)
5093 		goto out_cached_vmcs12;
5094 
5095 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
5096 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5097 	if (!vmx->nested.cached_shadow_vmcs12)
5098 		goto out_cached_shadow_vmcs12;
5099 
5100 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
5101 		goto out_shadow_vmcs;
5102 
5103 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
5104 		     HRTIMER_MODE_ABS_PINNED);
5105 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
5106 
5107 	vmx->nested.vpid02 = allocate_vpid();
5108 
5109 	vmx->nested.vmcs02_initialized = false;
5110 	vmx->nested.vmxon = true;
5111 
5112 	if (vmx_pt_mode_is_host_guest()) {
5113 		vmx->pt_desc.guest.ctl = 0;
5114 		pt_update_intercept_for_msr(vcpu);
5115 	}
5116 
5117 	return 0;
5118 
5119 out_shadow_vmcs:
5120 	kfree(vmx->nested.cached_shadow_vmcs12);
5121 
5122 out_cached_shadow_vmcs12:
5123 	kfree(vmx->nested.cached_vmcs12);
5124 
5125 out_cached_vmcs12:
5126 	free_loaded_vmcs(&vmx->nested.vmcs02);
5127 
5128 out_vmcs02:
5129 	return -ENOMEM;
5130 }
5131 
5132 /* Emulate the VMXON instruction. */
5133 static int handle_vmxon(struct kvm_vcpu *vcpu)
5134 {
5135 	int ret;
5136 	gpa_t vmptr;
5137 	uint32_t revision;
5138 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5139 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
5140 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
5141 
5142 	/*
5143 	 * Manually check CR4.VMXE checks, KVM must force CR4.VMXE=1 to enter
5144 	 * the guest and so cannot rely on hardware to perform the check,
5145 	 * which has higher priority than VM-Exit (see Intel SDM's pseudocode
5146 	 * for VMXON).
5147 	 *
5148 	 * Rely on hardware for the other pre-VM-Exit checks, CR0.PE=1, !VM86
5149 	 * and !COMPATIBILITY modes.  For an unrestricted guest, KVM doesn't
5150 	 * force any of the relevant guest state.  For a restricted guest, KVM
5151 	 * does force CR0.PE=1, but only to also force VM86 in order to emulate
5152 	 * Real Mode, and so there's no need to check CR0.PE manually.
5153 	 */
5154 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
5155 		kvm_queue_exception(vcpu, UD_VECTOR);
5156 		return 1;
5157 	}
5158 
5159 	/*
5160 	 * The CPL is checked for "not in VMX operation" and for "in VMX root",
5161 	 * and has higher priority than the VM-Fail due to being post-VMXON,
5162 	 * i.e. VMXON #GPs outside of VMX non-root if CPL!=0.  In VMX non-root,
5163 	 * VMXON causes VM-Exit and KVM unconditionally forwards VMXON VM-Exits
5164 	 * from L2 to L1, i.e. there's no need to check for the vCPU being in
5165 	 * VMX non-root.
5166 	 *
5167 	 * Forwarding the VM-Exit unconditionally, i.e. without performing the
5168 	 * #UD checks (see above), is functionally ok because KVM doesn't allow
5169 	 * L1 to run L2 without CR4.VMXE=0, and because KVM never modifies L2's
5170 	 * CR0 or CR4, i.e. it's L2's responsibility to emulate #UDs that are
5171 	 * missed by hardware due to shadowing CR0 and/or CR4.
5172 	 */
5173 	if (vmx_get_cpl(vcpu)) {
5174 		kvm_inject_gp(vcpu, 0);
5175 		return 1;
5176 	}
5177 
5178 	if (vmx->nested.vmxon)
5179 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5180 
5181 	/*
5182 	 * Invalid CR0/CR4 generates #GP.  These checks are performed if and
5183 	 * only if the vCPU isn't already in VMX operation, i.e. effectively
5184 	 * have lower priority than the VM-Fail above.
5185 	 */
5186 	if (!nested_host_cr0_valid(vcpu, kvm_read_cr0(vcpu)) ||
5187 	    !nested_host_cr4_valid(vcpu, kvm_read_cr4(vcpu))) {
5188 		kvm_inject_gp(vcpu, 0);
5189 		return 1;
5190 	}
5191 
5192 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5193 			!= VMXON_NEEDED_FEATURES) {
5194 		kvm_inject_gp(vcpu, 0);
5195 		return 1;
5196 	}
5197 
5198 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
5199 		return ret;
5200 
5201 	/*
5202 	 * SDM 3: 24.11.5
5203 	 * The first 4 bytes of VMXON region contain the supported
5204 	 * VMCS revision identifier
5205 	 *
5206 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
5207 	 * which replaces physical address width with 32
5208 	 */
5209 	if (!page_address_valid(vcpu, vmptr))
5210 		return nested_vmx_failInvalid(vcpu);
5211 
5212 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
5213 	    revision != VMCS12_REVISION)
5214 		return nested_vmx_failInvalid(vcpu);
5215 
5216 	vmx->nested.vmxon_ptr = vmptr;
5217 	ret = enter_vmx_operation(vcpu);
5218 	if (ret)
5219 		return ret;
5220 
5221 	return nested_vmx_succeed(vcpu);
5222 }
5223 
5224 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
5225 {
5226 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5227 
5228 	if (vmx->nested.current_vmptr == INVALID_GPA)
5229 		return;
5230 
5231 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
5232 
5233 	if (enable_shadow_vmcs) {
5234 		/* copy to memory all shadowed fields in case
5235 		   they were modified */
5236 		copy_shadow_to_vmcs12(vmx);
5237 		vmx_disable_shadow_vmcs(vmx);
5238 	}
5239 	vmx->nested.posted_intr_nv = -1;
5240 
5241 	/* Flush VMCS12 to guest memory */
5242 	kvm_vcpu_write_guest_page(vcpu,
5243 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
5244 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5245 
5246 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5247 
5248 	vmx->nested.current_vmptr = INVALID_GPA;
5249 }
5250 
5251 /* Emulate the VMXOFF instruction */
5252 static int handle_vmxoff(struct kvm_vcpu *vcpu)
5253 {
5254 	if (!nested_vmx_check_permission(vcpu))
5255 		return 1;
5256 
5257 	free_nested(vcpu);
5258 
5259 	if (kvm_apic_has_pending_init_or_sipi(vcpu))
5260 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5261 
5262 	return nested_vmx_succeed(vcpu);
5263 }
5264 
5265 /* Emulate the VMCLEAR instruction */
5266 static int handle_vmclear(struct kvm_vcpu *vcpu)
5267 {
5268 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5269 	u32 zero = 0;
5270 	gpa_t vmptr;
5271 	int r;
5272 
5273 	if (!nested_vmx_check_permission(vcpu))
5274 		return 1;
5275 
5276 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5277 		return r;
5278 
5279 	if (!page_address_valid(vcpu, vmptr))
5280 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5281 
5282 	if (vmptr == vmx->nested.vmxon_ptr)
5283 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5284 
5285 	/*
5286 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
5287 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
5288 	 * way to distinguish it from VMCS12) and we must not corrupt it by
5289 	 * writing to the non-existent 'launch_state' field. The area doesn't
5290 	 * have to be the currently active EVMCS on the calling CPU and there's
5291 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5292 	 * state. It is possible that the area will stay mapped as
5293 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5294 	 */
5295 	if (likely(!guest_cpuid_has_evmcs(vcpu) ||
5296 		   !evmptr_is_valid(nested_get_evmptr(vcpu)))) {
5297 		if (vmptr == vmx->nested.current_vmptr)
5298 			nested_release_vmcs12(vcpu);
5299 
5300 		/*
5301 		 * Silently ignore memory errors on VMCLEAR, Intel's pseudocode
5302 		 * for VMCLEAR includes a "ensure that data for VMCS referenced
5303 		 * by the operand is in memory" clause that guards writes to
5304 		 * memory, i.e. doing nothing for I/O is architecturally valid.
5305 		 *
5306 		 * FIXME: Suppress failures if and only if no memslot is found,
5307 		 * i.e. exit to userspace if __copy_to_user() fails.
5308 		 */
5309 		(void)kvm_vcpu_write_guest(vcpu,
5310 					   vmptr + offsetof(struct vmcs12,
5311 							    launch_state),
5312 					   &zero, sizeof(zero));
5313 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5314 		nested_release_evmcs(vcpu);
5315 	}
5316 
5317 	return nested_vmx_succeed(vcpu);
5318 }
5319 
5320 /* Emulate the VMLAUNCH instruction */
5321 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5322 {
5323 	return nested_vmx_run(vcpu, true);
5324 }
5325 
5326 /* Emulate the VMRESUME instruction */
5327 static int handle_vmresume(struct kvm_vcpu *vcpu)
5328 {
5329 
5330 	return nested_vmx_run(vcpu, false);
5331 }
5332 
5333 static int handle_vmread(struct kvm_vcpu *vcpu)
5334 {
5335 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5336 						    : get_vmcs12(vcpu);
5337 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5338 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5339 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5340 	struct x86_exception e;
5341 	unsigned long field;
5342 	u64 value;
5343 	gva_t gva = 0;
5344 	short offset;
5345 	int len, r;
5346 
5347 	if (!nested_vmx_check_permission(vcpu))
5348 		return 1;
5349 
5350 	/* Decode instruction info and find the field to read */
5351 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5352 
5353 	if (!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
5354 		/*
5355 		 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5356 		 * any VMREAD sets the ALU flags for VMfailInvalid.
5357 		 */
5358 		if (vmx->nested.current_vmptr == INVALID_GPA ||
5359 		    (is_guest_mode(vcpu) &&
5360 		     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5361 			return nested_vmx_failInvalid(vcpu);
5362 
5363 		offset = get_vmcs12_field_offset(field);
5364 		if (offset < 0)
5365 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5366 
5367 		if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5368 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5369 
5370 		/* Read the field, zero-extended to a u64 value */
5371 		value = vmcs12_read_any(vmcs12, field, offset);
5372 	} else {
5373 		/*
5374 		 * Hyper-V TLFS (as of 6.0b) explicitly states, that while an
5375 		 * enlightened VMCS is active VMREAD/VMWRITE instructions are
5376 		 * unsupported. Unfortunately, certain versions of Windows 11
5377 		 * don't comply with this requirement which is not enforced in
5378 		 * genuine Hyper-V. Allow VMREAD from an enlightened VMCS as a
5379 		 * workaround, as misbehaving guests will panic on VM-Fail.
5380 		 * Note, enlightened VMCS is incompatible with shadow VMCS so
5381 		 * all VMREADs from L2 should go to L1.
5382 		 */
5383 		if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5384 			return nested_vmx_failInvalid(vcpu);
5385 
5386 		offset = evmcs_field_offset(field, NULL);
5387 		if (offset < 0)
5388 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5389 
5390 		/* Read the field, zero-extended to a u64 value */
5391 		value = evmcs_read_any(vmx->nested.hv_evmcs, field, offset);
5392 	}
5393 
5394 	/*
5395 	 * Now copy part of this value to register or memory, as requested.
5396 	 * Note that the number of bits actually copied is 32 or 64 depending
5397 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5398 	 */
5399 	if (instr_info & BIT(10)) {
5400 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5401 	} else {
5402 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5403 		if (get_vmx_mem_address(vcpu, exit_qualification,
5404 					instr_info, true, len, &gva))
5405 			return 1;
5406 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5407 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5408 		if (r != X86EMUL_CONTINUE)
5409 			return kvm_handle_memory_failure(vcpu, r, &e);
5410 	}
5411 
5412 	return nested_vmx_succeed(vcpu);
5413 }
5414 
5415 static bool is_shadow_field_rw(unsigned long field)
5416 {
5417 	switch (field) {
5418 #define SHADOW_FIELD_RW(x, y) case x:
5419 #include "vmcs_shadow_fields.h"
5420 		return true;
5421 	default:
5422 		break;
5423 	}
5424 	return false;
5425 }
5426 
5427 static bool is_shadow_field_ro(unsigned long field)
5428 {
5429 	switch (field) {
5430 #define SHADOW_FIELD_RO(x, y) case x:
5431 #include "vmcs_shadow_fields.h"
5432 		return true;
5433 	default:
5434 		break;
5435 	}
5436 	return false;
5437 }
5438 
5439 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5440 {
5441 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5442 						    : get_vmcs12(vcpu);
5443 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5444 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5445 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5446 	struct x86_exception e;
5447 	unsigned long field;
5448 	short offset;
5449 	gva_t gva;
5450 	int len, r;
5451 
5452 	/*
5453 	 * The value to write might be 32 or 64 bits, depending on L1's long
5454 	 * mode, and eventually we need to write that into a field of several
5455 	 * possible lengths. The code below first zero-extends the value to 64
5456 	 * bit (value), and then copies only the appropriate number of
5457 	 * bits into the vmcs12 field.
5458 	 */
5459 	u64 value = 0;
5460 
5461 	if (!nested_vmx_check_permission(vcpu))
5462 		return 1;
5463 
5464 	/*
5465 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5466 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5467 	 */
5468 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5469 	    (is_guest_mode(vcpu) &&
5470 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5471 		return nested_vmx_failInvalid(vcpu);
5472 
5473 	if (instr_info & BIT(10))
5474 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5475 	else {
5476 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5477 		if (get_vmx_mem_address(vcpu, exit_qualification,
5478 					instr_info, false, len, &gva))
5479 			return 1;
5480 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5481 		if (r != X86EMUL_CONTINUE)
5482 			return kvm_handle_memory_failure(vcpu, r, &e);
5483 	}
5484 
5485 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5486 
5487 	offset = get_vmcs12_field_offset(field);
5488 	if (offset < 0)
5489 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5490 
5491 	/*
5492 	 * If the vCPU supports "VMWRITE to any supported field in the
5493 	 * VMCS," then the "read-only" fields are actually read/write.
5494 	 */
5495 	if (vmcs_field_readonly(field) &&
5496 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5497 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5498 
5499 	/*
5500 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5501 	 * vmcs12, else we may crush a field or consume a stale value.
5502 	 */
5503 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5504 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5505 
5506 	/*
5507 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5508 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5509 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5510 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5511 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5512 	 * the stripped down value, L2 sees the full value as stored by KVM).
5513 	 */
5514 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5515 		value &= 0x1f0ff;
5516 
5517 	vmcs12_write_any(vmcs12, field, offset, value);
5518 
5519 	/*
5520 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5521 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5522 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5523 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5524 	 */
5525 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5526 		/*
5527 		 * L1 can read these fields without exiting, ensure the
5528 		 * shadow VMCS is up-to-date.
5529 		 */
5530 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5531 			preempt_disable();
5532 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5533 
5534 			__vmcs_writel(field, value);
5535 
5536 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5537 			vmcs_load(vmx->loaded_vmcs->vmcs);
5538 			preempt_enable();
5539 		}
5540 		vmx->nested.dirty_vmcs12 = true;
5541 	}
5542 
5543 	return nested_vmx_succeed(vcpu);
5544 }
5545 
5546 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5547 {
5548 	vmx->nested.current_vmptr = vmptr;
5549 	if (enable_shadow_vmcs) {
5550 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5551 		vmcs_write64(VMCS_LINK_POINTER,
5552 			     __pa(vmx->vmcs01.shadow_vmcs));
5553 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5554 	}
5555 	vmx->nested.dirty_vmcs12 = true;
5556 	vmx->nested.force_msr_bitmap_recalc = true;
5557 }
5558 
5559 /* Emulate the VMPTRLD instruction */
5560 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5561 {
5562 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5563 	gpa_t vmptr;
5564 	int r;
5565 
5566 	if (!nested_vmx_check_permission(vcpu))
5567 		return 1;
5568 
5569 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5570 		return r;
5571 
5572 	if (!page_address_valid(vcpu, vmptr))
5573 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5574 
5575 	if (vmptr == vmx->nested.vmxon_ptr)
5576 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5577 
5578 	/* Forbid normal VMPTRLD if Enlightened version was used */
5579 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5580 		return 1;
5581 
5582 	if (vmx->nested.current_vmptr != vmptr) {
5583 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5584 		struct vmcs_hdr hdr;
5585 
5586 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5587 			/*
5588 			 * Reads from an unbacked page return all 1s,
5589 			 * which means that the 32 bits located at the
5590 			 * given physical address won't match the required
5591 			 * VMCS12_REVISION identifier.
5592 			 */
5593 			return nested_vmx_fail(vcpu,
5594 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5595 		}
5596 
5597 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5598 						 offsetof(struct vmcs12, hdr),
5599 						 sizeof(hdr))) {
5600 			return nested_vmx_fail(vcpu,
5601 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5602 		}
5603 
5604 		if (hdr.revision_id != VMCS12_REVISION ||
5605 		    (hdr.shadow_vmcs &&
5606 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5607 			return nested_vmx_fail(vcpu,
5608 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5609 		}
5610 
5611 		nested_release_vmcs12(vcpu);
5612 
5613 		/*
5614 		 * Load VMCS12 from guest memory since it is not already
5615 		 * cached.
5616 		 */
5617 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5618 					  VMCS12_SIZE)) {
5619 			return nested_vmx_fail(vcpu,
5620 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5621 		}
5622 
5623 		set_current_vmptr(vmx, vmptr);
5624 	}
5625 
5626 	return nested_vmx_succeed(vcpu);
5627 }
5628 
5629 /* Emulate the VMPTRST instruction */
5630 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5631 {
5632 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5633 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5634 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5635 	struct x86_exception e;
5636 	gva_t gva;
5637 	int r;
5638 
5639 	if (!nested_vmx_check_permission(vcpu))
5640 		return 1;
5641 
5642 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5643 		return 1;
5644 
5645 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5646 				true, sizeof(gpa_t), &gva))
5647 		return 1;
5648 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5649 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5650 					sizeof(gpa_t), &e);
5651 	if (r != X86EMUL_CONTINUE)
5652 		return kvm_handle_memory_failure(vcpu, r, &e);
5653 
5654 	return nested_vmx_succeed(vcpu);
5655 }
5656 
5657 /* Emulate the INVEPT instruction */
5658 static int handle_invept(struct kvm_vcpu *vcpu)
5659 {
5660 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5661 	u32 vmx_instruction_info, types;
5662 	unsigned long type, roots_to_free;
5663 	struct kvm_mmu *mmu;
5664 	gva_t gva;
5665 	struct x86_exception e;
5666 	struct {
5667 		u64 eptp, gpa;
5668 	} operand;
5669 	int i, r, gpr_index;
5670 
5671 	if (!(vmx->nested.msrs.secondary_ctls_high &
5672 	      SECONDARY_EXEC_ENABLE_EPT) ||
5673 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5674 		kvm_queue_exception(vcpu, UD_VECTOR);
5675 		return 1;
5676 	}
5677 
5678 	if (!nested_vmx_check_permission(vcpu))
5679 		return 1;
5680 
5681 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5682 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5683 	type = kvm_register_read(vcpu, gpr_index);
5684 
5685 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5686 
5687 	if (type >= 32 || !(types & (1 << type)))
5688 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5689 
5690 	/* According to the Intel VMX instruction reference, the memory
5691 	 * operand is read even if it isn't needed (e.g., for type==global)
5692 	 */
5693 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5694 			vmx_instruction_info, false, sizeof(operand), &gva))
5695 		return 1;
5696 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5697 	if (r != X86EMUL_CONTINUE)
5698 		return kvm_handle_memory_failure(vcpu, r, &e);
5699 
5700 	/*
5701 	 * Nested EPT roots are always held through guest_mmu,
5702 	 * not root_mmu.
5703 	 */
5704 	mmu = &vcpu->arch.guest_mmu;
5705 
5706 	switch (type) {
5707 	case VMX_EPT_EXTENT_CONTEXT:
5708 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5709 			return nested_vmx_fail(vcpu,
5710 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5711 
5712 		roots_to_free = 0;
5713 		if (nested_ept_root_matches(mmu->root.hpa, mmu->root.pgd,
5714 					    operand.eptp))
5715 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5716 
5717 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5718 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5719 						    mmu->prev_roots[i].pgd,
5720 						    operand.eptp))
5721 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5722 		}
5723 		break;
5724 	case VMX_EPT_EXTENT_GLOBAL:
5725 		roots_to_free = KVM_MMU_ROOTS_ALL;
5726 		break;
5727 	default:
5728 		BUG();
5729 		break;
5730 	}
5731 
5732 	if (roots_to_free)
5733 		kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
5734 
5735 	return nested_vmx_succeed(vcpu);
5736 }
5737 
5738 static int handle_invvpid(struct kvm_vcpu *vcpu)
5739 {
5740 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5741 	u32 vmx_instruction_info;
5742 	unsigned long type, types;
5743 	gva_t gva;
5744 	struct x86_exception e;
5745 	struct {
5746 		u64 vpid;
5747 		u64 gla;
5748 	} operand;
5749 	u16 vpid02;
5750 	int r, gpr_index;
5751 
5752 	if (!(vmx->nested.msrs.secondary_ctls_high &
5753 	      SECONDARY_EXEC_ENABLE_VPID) ||
5754 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5755 		kvm_queue_exception(vcpu, UD_VECTOR);
5756 		return 1;
5757 	}
5758 
5759 	if (!nested_vmx_check_permission(vcpu))
5760 		return 1;
5761 
5762 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5763 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5764 	type = kvm_register_read(vcpu, gpr_index);
5765 
5766 	types = (vmx->nested.msrs.vpid_caps &
5767 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5768 
5769 	if (type >= 32 || !(types & (1 << type)))
5770 		return nested_vmx_fail(vcpu,
5771 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5772 
5773 	/* according to the intel vmx instruction reference, the memory
5774 	 * operand is read even if it isn't needed (e.g., for type==global)
5775 	 */
5776 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5777 			vmx_instruction_info, false, sizeof(operand), &gva))
5778 		return 1;
5779 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5780 	if (r != X86EMUL_CONTINUE)
5781 		return kvm_handle_memory_failure(vcpu, r, &e);
5782 
5783 	if (operand.vpid >> 16)
5784 		return nested_vmx_fail(vcpu,
5785 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5786 
5787 	vpid02 = nested_get_vpid02(vcpu);
5788 	switch (type) {
5789 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5790 		if (!operand.vpid ||
5791 		    is_noncanonical_address(operand.gla, vcpu))
5792 			return nested_vmx_fail(vcpu,
5793 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5794 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5795 		break;
5796 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5797 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5798 		if (!operand.vpid)
5799 			return nested_vmx_fail(vcpu,
5800 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5801 		vpid_sync_context(vpid02);
5802 		break;
5803 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5804 		vpid_sync_context(vpid02);
5805 		break;
5806 	default:
5807 		WARN_ON_ONCE(1);
5808 		return kvm_skip_emulated_instruction(vcpu);
5809 	}
5810 
5811 	/*
5812 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5813 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5814 	 * roots as VPIDs are not tracked in the MMU role.
5815 	 *
5816 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5817 	 * an MMU when EPT is disabled.
5818 	 *
5819 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5820 	 */
5821 	if (!enable_ept)
5822 		kvm_mmu_free_guest_mode_roots(vcpu->kvm, &vcpu->arch.root_mmu);
5823 
5824 	return nested_vmx_succeed(vcpu);
5825 }
5826 
5827 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5828 				     struct vmcs12 *vmcs12)
5829 {
5830 	u32 index = kvm_rcx_read(vcpu);
5831 	u64 new_eptp;
5832 
5833 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5834 		return 1;
5835 	if (index >= VMFUNC_EPTP_ENTRIES)
5836 		return 1;
5837 
5838 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5839 				     &new_eptp, index * 8, 8))
5840 		return 1;
5841 
5842 	/*
5843 	 * If the (L2) guest does a vmfunc to the currently
5844 	 * active ept pointer, we don't have to do anything else
5845 	 */
5846 	if (vmcs12->ept_pointer != new_eptp) {
5847 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5848 			return 1;
5849 
5850 		vmcs12->ept_pointer = new_eptp;
5851 		nested_ept_new_eptp(vcpu);
5852 
5853 		if (!nested_cpu_has_vpid(vmcs12))
5854 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5855 	}
5856 
5857 	return 0;
5858 }
5859 
5860 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5861 {
5862 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5863 	struct vmcs12 *vmcs12;
5864 	u32 function = kvm_rax_read(vcpu);
5865 
5866 	/*
5867 	 * VMFUNC should never execute cleanly while L1 is active; KVM supports
5868 	 * VMFUNC for nested VMs, but not for L1.
5869 	 */
5870 	if (WARN_ON_ONCE(!is_guest_mode(vcpu))) {
5871 		kvm_queue_exception(vcpu, UD_VECTOR);
5872 		return 1;
5873 	}
5874 
5875 	vmcs12 = get_vmcs12(vcpu);
5876 
5877 	/*
5878 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5879 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5880 	 */
5881 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5882 		kvm_queue_exception(vcpu, UD_VECTOR);
5883 		return 1;
5884 	}
5885 
5886 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5887 		goto fail;
5888 
5889 	switch (function) {
5890 	case 0:
5891 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5892 			goto fail;
5893 		break;
5894 	default:
5895 		goto fail;
5896 	}
5897 	return kvm_skip_emulated_instruction(vcpu);
5898 
5899 fail:
5900 	/*
5901 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5902 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5903 	 * EXIT_REASON_VMFUNC as the exit reason.
5904 	 */
5905 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5906 			  vmx_get_intr_info(vcpu),
5907 			  vmx_get_exit_qual(vcpu));
5908 	return 1;
5909 }
5910 
5911 /*
5912  * Return true if an IO instruction with the specified port and size should cause
5913  * a VM-exit into L1.
5914  */
5915 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5916 				 int size)
5917 {
5918 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5919 	gpa_t bitmap, last_bitmap;
5920 	u8 b;
5921 
5922 	last_bitmap = INVALID_GPA;
5923 	b = -1;
5924 
5925 	while (size > 0) {
5926 		if (port < 0x8000)
5927 			bitmap = vmcs12->io_bitmap_a;
5928 		else if (port < 0x10000)
5929 			bitmap = vmcs12->io_bitmap_b;
5930 		else
5931 			return true;
5932 		bitmap += (port & 0x7fff) / 8;
5933 
5934 		if (last_bitmap != bitmap)
5935 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5936 				return true;
5937 		if (b & (1 << (port & 7)))
5938 			return true;
5939 
5940 		port++;
5941 		size--;
5942 		last_bitmap = bitmap;
5943 	}
5944 
5945 	return false;
5946 }
5947 
5948 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5949 				       struct vmcs12 *vmcs12)
5950 {
5951 	unsigned long exit_qualification;
5952 	unsigned short port;
5953 	int size;
5954 
5955 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5956 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5957 
5958 	exit_qualification = vmx_get_exit_qual(vcpu);
5959 
5960 	port = exit_qualification >> 16;
5961 	size = (exit_qualification & 7) + 1;
5962 
5963 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5964 }
5965 
5966 /*
5967  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5968  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5969  * disinterest in the current event (read or write a specific MSR) by using an
5970  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5971  */
5972 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5973 					struct vmcs12 *vmcs12,
5974 					union vmx_exit_reason exit_reason)
5975 {
5976 	u32 msr_index = kvm_rcx_read(vcpu);
5977 	gpa_t bitmap;
5978 
5979 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5980 		return true;
5981 
5982 	/*
5983 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5984 	 * for the four combinations of read/write and low/high MSR numbers.
5985 	 * First we need to figure out which of the four to use:
5986 	 */
5987 	bitmap = vmcs12->msr_bitmap;
5988 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5989 		bitmap += 2048;
5990 	if (msr_index >= 0xc0000000) {
5991 		msr_index -= 0xc0000000;
5992 		bitmap += 1024;
5993 	}
5994 
5995 	/* Then read the msr_index'th bit from this bitmap: */
5996 	if (msr_index < 1024*8) {
5997 		unsigned char b;
5998 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5999 			return true;
6000 		return 1 & (b >> (msr_index & 7));
6001 	} else
6002 		return true; /* let L1 handle the wrong parameter */
6003 }
6004 
6005 /*
6006  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6007  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6008  * intercept (via guest_host_mask etc.) the current event.
6009  */
6010 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6011 	struct vmcs12 *vmcs12)
6012 {
6013 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
6014 	int cr = exit_qualification & 15;
6015 	int reg;
6016 	unsigned long val;
6017 
6018 	switch ((exit_qualification >> 4) & 3) {
6019 	case 0: /* mov to cr */
6020 		reg = (exit_qualification >> 8) & 15;
6021 		val = kvm_register_read(vcpu, reg);
6022 		switch (cr) {
6023 		case 0:
6024 			if (vmcs12->cr0_guest_host_mask &
6025 			    (val ^ vmcs12->cr0_read_shadow))
6026 				return true;
6027 			break;
6028 		case 3:
6029 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6030 				return true;
6031 			break;
6032 		case 4:
6033 			if (vmcs12->cr4_guest_host_mask &
6034 			    (vmcs12->cr4_read_shadow ^ val))
6035 				return true;
6036 			break;
6037 		case 8:
6038 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6039 				return true;
6040 			break;
6041 		}
6042 		break;
6043 	case 2: /* clts */
6044 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6045 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
6046 			return true;
6047 		break;
6048 	case 1: /* mov from cr */
6049 		switch (cr) {
6050 		case 3:
6051 			if (vmcs12->cpu_based_vm_exec_control &
6052 			    CPU_BASED_CR3_STORE_EXITING)
6053 				return true;
6054 			break;
6055 		case 8:
6056 			if (vmcs12->cpu_based_vm_exec_control &
6057 			    CPU_BASED_CR8_STORE_EXITING)
6058 				return true;
6059 			break;
6060 		}
6061 		break;
6062 	case 3: /* lmsw */
6063 		/*
6064 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6065 		 * cr0. Other attempted changes are ignored, with no exit.
6066 		 */
6067 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6068 		if (vmcs12->cr0_guest_host_mask & 0xe &
6069 		    (val ^ vmcs12->cr0_read_shadow))
6070 			return true;
6071 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6072 		    !(vmcs12->cr0_read_shadow & 0x1) &&
6073 		    (val & 0x1))
6074 			return true;
6075 		break;
6076 	}
6077 	return false;
6078 }
6079 
6080 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
6081 					  struct vmcs12 *vmcs12)
6082 {
6083 	u32 encls_leaf;
6084 
6085 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
6086 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
6087 		return false;
6088 
6089 	encls_leaf = kvm_rax_read(vcpu);
6090 	if (encls_leaf > 62)
6091 		encls_leaf = 63;
6092 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
6093 }
6094 
6095 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
6096 	struct vmcs12 *vmcs12, gpa_t bitmap)
6097 {
6098 	u32 vmx_instruction_info;
6099 	unsigned long field;
6100 	u8 b;
6101 
6102 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
6103 		return true;
6104 
6105 	/* Decode instruction info and find the field to access */
6106 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6107 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6108 
6109 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
6110 	if (field >> 15)
6111 		return true;
6112 
6113 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
6114 		return true;
6115 
6116 	return 1 & (b >> (field & 7));
6117 }
6118 
6119 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
6120 {
6121 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
6122 
6123 	if (nested_cpu_has_mtf(vmcs12))
6124 		return true;
6125 
6126 	/*
6127 	 * An MTF VM-exit may be injected into the guest by setting the
6128 	 * interruption-type to 7 (other event) and the vector field to 0. Such
6129 	 * is the case regardless of the 'monitor trap flag' VM-execution
6130 	 * control.
6131 	 */
6132 	return entry_intr_info == (INTR_INFO_VALID_MASK
6133 				   | INTR_TYPE_OTHER_EVENT);
6134 }
6135 
6136 /*
6137  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
6138  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
6139  */
6140 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
6141 				     union vmx_exit_reason exit_reason)
6142 {
6143 	u32 intr_info;
6144 
6145 	switch ((u16)exit_reason.basic) {
6146 	case EXIT_REASON_EXCEPTION_NMI:
6147 		intr_info = vmx_get_intr_info(vcpu);
6148 		if (is_nmi(intr_info))
6149 			return true;
6150 		else if (is_page_fault(intr_info))
6151 			return vcpu->arch.apf.host_apf_flags ||
6152 			       vmx_need_pf_intercept(vcpu);
6153 		else if (is_debug(intr_info) &&
6154 			 vcpu->guest_debug &
6155 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6156 			return true;
6157 		else if (is_breakpoint(intr_info) &&
6158 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6159 			return true;
6160 		else if (is_alignment_check(intr_info) &&
6161 			 !vmx_guest_inject_ac(vcpu))
6162 			return true;
6163 		return false;
6164 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6165 		return true;
6166 	case EXIT_REASON_MCE_DURING_VMENTRY:
6167 		return true;
6168 	case EXIT_REASON_EPT_VIOLATION:
6169 		/*
6170 		 * L0 always deals with the EPT violation. If nested EPT is
6171 		 * used, and the nested mmu code discovers that the address is
6172 		 * missing in the guest EPT table (EPT12), the EPT violation
6173 		 * will be injected with nested_ept_inject_page_fault()
6174 		 */
6175 		return true;
6176 	case EXIT_REASON_EPT_MISCONFIG:
6177 		/*
6178 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
6179 		 * table (shadow on EPT) or a merged EPT table that L0 built
6180 		 * (EPT on EPT). So any problems with the structure of the
6181 		 * table is L0's fault.
6182 		 */
6183 		return true;
6184 	case EXIT_REASON_PREEMPTION_TIMER:
6185 		return true;
6186 	case EXIT_REASON_PML_FULL:
6187 		/*
6188 		 * PML is emulated for an L1 VMM and should never be enabled in
6189 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
6190 		 */
6191 		return true;
6192 	case EXIT_REASON_VMFUNC:
6193 		/* VM functions are emulated through L2->L0 vmexits. */
6194 		return true;
6195 	case EXIT_REASON_BUS_LOCK:
6196 		/*
6197 		 * At present, bus lock VM exit is never exposed to L1.
6198 		 * Handle L2's bus locks in L0 directly.
6199 		 */
6200 		return true;
6201 	case EXIT_REASON_VMCALL:
6202 		/* Hyper-V L2 TLB flush hypercall is handled by L0 */
6203 		return guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
6204 			nested_evmcs_l2_tlb_flush_enabled(vcpu) &&
6205 			kvm_hv_is_tlb_flush_hcall(vcpu);
6206 	default:
6207 		break;
6208 	}
6209 	return false;
6210 }
6211 
6212 /*
6213  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
6214  * is_guest_mode (L2).
6215  */
6216 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
6217 				     union vmx_exit_reason exit_reason)
6218 {
6219 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6220 	u32 intr_info;
6221 
6222 	switch ((u16)exit_reason.basic) {
6223 	case EXIT_REASON_EXCEPTION_NMI:
6224 		intr_info = vmx_get_intr_info(vcpu);
6225 		if (is_nmi(intr_info))
6226 			return true;
6227 		else if (is_page_fault(intr_info))
6228 			return true;
6229 		return vmcs12->exception_bitmap &
6230 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
6231 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6232 		return nested_exit_on_intr(vcpu);
6233 	case EXIT_REASON_TRIPLE_FAULT:
6234 		return true;
6235 	case EXIT_REASON_INTERRUPT_WINDOW:
6236 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
6237 	case EXIT_REASON_NMI_WINDOW:
6238 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
6239 	case EXIT_REASON_TASK_SWITCH:
6240 		return true;
6241 	case EXIT_REASON_CPUID:
6242 		return true;
6243 	case EXIT_REASON_HLT:
6244 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6245 	case EXIT_REASON_INVD:
6246 		return true;
6247 	case EXIT_REASON_INVLPG:
6248 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6249 	case EXIT_REASON_RDPMC:
6250 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6251 	case EXIT_REASON_RDRAND:
6252 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
6253 	case EXIT_REASON_RDSEED:
6254 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
6255 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
6256 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6257 	case EXIT_REASON_VMREAD:
6258 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6259 			vmcs12->vmread_bitmap);
6260 	case EXIT_REASON_VMWRITE:
6261 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6262 			vmcs12->vmwrite_bitmap);
6263 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6264 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6265 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
6266 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6267 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
6268 		/*
6269 		 * VMX instructions trap unconditionally. This allows L1 to
6270 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6271 		 */
6272 		return true;
6273 	case EXIT_REASON_CR_ACCESS:
6274 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6275 	case EXIT_REASON_DR_ACCESS:
6276 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6277 	case EXIT_REASON_IO_INSTRUCTION:
6278 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
6279 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6280 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6281 	case EXIT_REASON_MSR_READ:
6282 	case EXIT_REASON_MSR_WRITE:
6283 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6284 	case EXIT_REASON_INVALID_STATE:
6285 		return true;
6286 	case EXIT_REASON_MWAIT_INSTRUCTION:
6287 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6288 	case EXIT_REASON_MONITOR_TRAP_FLAG:
6289 		return nested_vmx_exit_handled_mtf(vmcs12);
6290 	case EXIT_REASON_MONITOR_INSTRUCTION:
6291 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6292 	case EXIT_REASON_PAUSE_INSTRUCTION:
6293 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6294 			nested_cpu_has2(vmcs12,
6295 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6296 	case EXIT_REASON_MCE_DURING_VMENTRY:
6297 		return true;
6298 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6299 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6300 	case EXIT_REASON_APIC_ACCESS:
6301 	case EXIT_REASON_APIC_WRITE:
6302 	case EXIT_REASON_EOI_INDUCED:
6303 		/*
6304 		 * The controls for "virtualize APIC accesses," "APIC-
6305 		 * register virtualization," and "virtual-interrupt
6306 		 * delivery" only come from vmcs12.
6307 		 */
6308 		return true;
6309 	case EXIT_REASON_INVPCID:
6310 		return
6311 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6312 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6313 	case EXIT_REASON_WBINVD:
6314 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6315 	case EXIT_REASON_XSETBV:
6316 		return true;
6317 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6318 		/*
6319 		 * This should never happen, since it is not possible to
6320 		 * set XSS to a non-zero value---neither in L1 nor in L2.
6321 		 * If if it were, XSS would have to be checked against
6322 		 * the XSS exit bitmap in vmcs12.
6323 		 */
6324 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
6325 	case EXIT_REASON_UMWAIT:
6326 	case EXIT_REASON_TPAUSE:
6327 		return nested_cpu_has2(vmcs12,
6328 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6329 	case EXIT_REASON_ENCLS:
6330 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6331 	case EXIT_REASON_NOTIFY:
6332 		/* Notify VM exit is not exposed to L1 */
6333 		return false;
6334 	default:
6335 		return true;
6336 	}
6337 }
6338 
6339 /*
6340  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6341  * reflected into L1.
6342  */
6343 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6344 {
6345 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6346 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6347 	unsigned long exit_qual;
6348 	u32 exit_intr_info;
6349 
6350 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6351 
6352 	/*
6353 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6354 	 * has already loaded L2's state.
6355 	 */
6356 	if (unlikely(vmx->fail)) {
6357 		trace_kvm_nested_vmenter_failed(
6358 			"hardware VM-instruction error: ",
6359 			vmcs_read32(VM_INSTRUCTION_ERROR));
6360 		exit_intr_info = 0;
6361 		exit_qual = 0;
6362 		goto reflect_vmexit;
6363 	}
6364 
6365 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6366 
6367 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6368 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6369 		return false;
6370 
6371 	/* If L1 doesn't want the exit, handle it in L0. */
6372 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6373 		return false;
6374 
6375 	/*
6376 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6377 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6378 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6379 	 * interrupts are never reflected to L1 so it's a non-issue.
6380 	 */
6381 	exit_intr_info = vmx_get_intr_info(vcpu);
6382 	if (is_exception_with_error_code(exit_intr_info)) {
6383 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6384 
6385 		vmcs12->vm_exit_intr_error_code =
6386 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6387 	}
6388 	exit_qual = vmx_get_exit_qual(vcpu);
6389 
6390 reflect_vmexit:
6391 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6392 	return true;
6393 }
6394 
6395 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6396 				struct kvm_nested_state __user *user_kvm_nested_state,
6397 				u32 user_data_size)
6398 {
6399 	struct vcpu_vmx *vmx;
6400 	struct vmcs12 *vmcs12;
6401 	struct kvm_nested_state kvm_state = {
6402 		.flags = 0,
6403 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6404 		.size = sizeof(kvm_state),
6405 		.hdr.vmx.flags = 0,
6406 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6407 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6408 		.hdr.vmx.preemption_timer_deadline = 0,
6409 	};
6410 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6411 		&user_kvm_nested_state->data.vmx[0];
6412 
6413 	if (!vcpu)
6414 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6415 
6416 	vmx = to_vmx(vcpu);
6417 	vmcs12 = get_vmcs12(vcpu);
6418 
6419 	if (nested_vmx_allowed(vcpu) &&
6420 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6421 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6422 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6423 
6424 		if (vmx_has_valid_vmcs12(vcpu)) {
6425 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6426 
6427 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6428 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6429 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6430 
6431 			if (is_guest_mode(vcpu) &&
6432 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6433 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6434 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6435 		}
6436 
6437 		if (vmx->nested.smm.vmxon)
6438 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6439 
6440 		if (vmx->nested.smm.guest_mode)
6441 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6442 
6443 		if (is_guest_mode(vcpu)) {
6444 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6445 
6446 			if (vmx->nested.nested_run_pending)
6447 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6448 
6449 			if (vmx->nested.mtf_pending)
6450 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6451 
6452 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6453 			    vmx->nested.has_preemption_timer_deadline) {
6454 				kvm_state.hdr.vmx.flags |=
6455 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6456 				kvm_state.hdr.vmx.preemption_timer_deadline =
6457 					vmx->nested.preemption_timer_deadline;
6458 			}
6459 		}
6460 	}
6461 
6462 	if (user_data_size < kvm_state.size)
6463 		goto out;
6464 
6465 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6466 		return -EFAULT;
6467 
6468 	if (!vmx_has_valid_vmcs12(vcpu))
6469 		goto out;
6470 
6471 	/*
6472 	 * When running L2, the authoritative vmcs12 state is in the
6473 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6474 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6475 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6476 	 * vmcs12 state is in the vmcs12 already.
6477 	 */
6478 	if (is_guest_mode(vcpu)) {
6479 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6480 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6481 	} else  {
6482 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6483 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6484 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6485 				/*
6486 				 * L1 hypervisor is not obliged to keep eVMCS
6487 				 * clean fields data always up-to-date while
6488 				 * not in guest mode, 'hv_clean_fields' is only
6489 				 * supposed to be actual upon vmentry so we need
6490 				 * to ignore it here and do full copy.
6491 				 */
6492 				copy_enlightened_to_vmcs12(vmx, 0);
6493 			else if (enable_shadow_vmcs)
6494 				copy_shadow_to_vmcs12(vmx);
6495 		}
6496 	}
6497 
6498 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6499 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6500 
6501 	/*
6502 	 * Copy over the full allocated size of vmcs12 rather than just the size
6503 	 * of the struct.
6504 	 */
6505 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6506 		return -EFAULT;
6507 
6508 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6509 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6510 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6511 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6512 			return -EFAULT;
6513 	}
6514 out:
6515 	return kvm_state.size;
6516 }
6517 
6518 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6519 {
6520 	if (is_guest_mode(vcpu)) {
6521 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6522 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6523 	}
6524 	free_nested(vcpu);
6525 }
6526 
6527 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6528 				struct kvm_nested_state __user *user_kvm_nested_state,
6529 				struct kvm_nested_state *kvm_state)
6530 {
6531 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6532 	struct vmcs12 *vmcs12;
6533 	enum vm_entry_failure_code ignored;
6534 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6535 		&user_kvm_nested_state->data.vmx[0];
6536 	int ret;
6537 
6538 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6539 		return -EINVAL;
6540 
6541 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6542 		if (kvm_state->hdr.vmx.smm.flags)
6543 			return -EINVAL;
6544 
6545 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6546 			return -EINVAL;
6547 
6548 		/*
6549 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6550 		 * enable eVMCS capability on vCPU. However, since then
6551 		 * code was changed such that flag signals vmcs12 should
6552 		 * be copied into eVMCS in guest memory.
6553 		 *
6554 		 * To preserve backwards compatability, allow user
6555 		 * to set this flag even when there is no VMXON region.
6556 		 */
6557 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6558 			return -EINVAL;
6559 	} else {
6560 		if (!nested_vmx_allowed(vcpu))
6561 			return -EINVAL;
6562 
6563 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6564 			return -EINVAL;
6565 	}
6566 
6567 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6568 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6569 		return -EINVAL;
6570 
6571 	if (kvm_state->hdr.vmx.smm.flags &
6572 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6573 		return -EINVAL;
6574 
6575 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6576 		return -EINVAL;
6577 
6578 	/*
6579 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6580 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6581 	 * must be zero.
6582 	 */
6583 	if (is_smm(vcpu) ?
6584 		(kvm_state->flags &
6585 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6586 		: kvm_state->hdr.vmx.smm.flags)
6587 		return -EINVAL;
6588 
6589 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6590 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6591 		return -EINVAL;
6592 
6593 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6594 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6595 			return -EINVAL;
6596 
6597 	vmx_leave_nested(vcpu);
6598 
6599 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6600 		return 0;
6601 
6602 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6603 	ret = enter_vmx_operation(vcpu);
6604 	if (ret)
6605 		return ret;
6606 
6607 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6608 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6609 		/* See vmx_has_valid_vmcs12.  */
6610 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6611 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6612 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6613 			return -EINVAL;
6614 		else
6615 			return 0;
6616 	}
6617 
6618 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6619 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6620 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6621 			return -EINVAL;
6622 
6623 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6624 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6625 		/*
6626 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6627 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6628 		 * restored yet. EVMCS will be mapped from
6629 		 * nested_get_vmcs12_pages().
6630 		 */
6631 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6632 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6633 	} else {
6634 		return -EINVAL;
6635 	}
6636 
6637 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6638 		vmx->nested.smm.vmxon = true;
6639 		vmx->nested.vmxon = false;
6640 
6641 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6642 			vmx->nested.smm.guest_mode = true;
6643 	}
6644 
6645 	vmcs12 = get_vmcs12(vcpu);
6646 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6647 		return -EFAULT;
6648 
6649 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6650 		return -EINVAL;
6651 
6652 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6653 		return 0;
6654 
6655 	vmx->nested.nested_run_pending =
6656 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6657 
6658 	vmx->nested.mtf_pending =
6659 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6660 
6661 	ret = -EINVAL;
6662 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6663 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6664 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6665 
6666 		if (kvm_state->size <
6667 		    sizeof(*kvm_state) +
6668 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6669 			goto error_guest_mode;
6670 
6671 		if (copy_from_user(shadow_vmcs12,
6672 				   user_vmx_nested_state->shadow_vmcs12,
6673 				   sizeof(*shadow_vmcs12))) {
6674 			ret = -EFAULT;
6675 			goto error_guest_mode;
6676 		}
6677 
6678 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6679 		    !shadow_vmcs12->hdr.shadow_vmcs)
6680 			goto error_guest_mode;
6681 	}
6682 
6683 	vmx->nested.has_preemption_timer_deadline = false;
6684 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6685 		vmx->nested.has_preemption_timer_deadline = true;
6686 		vmx->nested.preemption_timer_deadline =
6687 			kvm_state->hdr.vmx.preemption_timer_deadline;
6688 	}
6689 
6690 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6691 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6692 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6693 		goto error_guest_mode;
6694 
6695 	vmx->nested.dirty_vmcs12 = true;
6696 	vmx->nested.force_msr_bitmap_recalc = true;
6697 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6698 	if (ret)
6699 		goto error_guest_mode;
6700 
6701 	if (vmx->nested.mtf_pending)
6702 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6703 
6704 	return 0;
6705 
6706 error_guest_mode:
6707 	vmx->nested.nested_run_pending = 0;
6708 	return ret;
6709 }
6710 
6711 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6712 {
6713 	if (enable_shadow_vmcs) {
6714 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6715 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6716 	}
6717 }
6718 
6719 /*
6720  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6721  * that madness to get the encoding for comparison.
6722  */
6723 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6724 
6725 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6726 {
6727 	/*
6728 	 * Note these are the so called "index" of the VMCS field encoding, not
6729 	 * the index into vmcs12.
6730 	 */
6731 	unsigned int max_idx, idx;
6732 	int i;
6733 
6734 	/*
6735 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6736 	 * vmcs12, regardless of whether or not the associated feature is
6737 	 * exposed to L1.  Simply find the field with the highest index.
6738 	 */
6739 	max_idx = 0;
6740 	for (i = 0; i < nr_vmcs12_fields; i++) {
6741 		/* The vmcs12 table is very, very sparsely populated. */
6742 		if (!vmcs12_field_offsets[i])
6743 			continue;
6744 
6745 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6746 		if (idx > max_idx)
6747 			max_idx = idx;
6748 	}
6749 
6750 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6751 }
6752 
6753 /*
6754  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6755  * returned for the various VMX controls MSRs when nested VMX is enabled.
6756  * The same values should also be used to verify that vmcs12 control fields are
6757  * valid during nested entry from L1 to L2.
6758  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6759  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6760  * bit in the high half is on if the corresponding bit in the control field
6761  * may be on. See also vmx_control_verify().
6762  */
6763 void nested_vmx_setup_ctls_msrs(struct vmcs_config *vmcs_conf, u32 ept_caps)
6764 {
6765 	struct nested_vmx_msrs *msrs = &vmcs_conf->nested;
6766 
6767 	/*
6768 	 * Note that as a general rule, the high half of the MSRs (bits in
6769 	 * the control fields which may be 1) should be initialized by the
6770 	 * intersection of the underlying hardware's MSR (i.e., features which
6771 	 * can be supported) and the list of features we want to expose -
6772 	 * because they are known to be properly supported in our code.
6773 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6774 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6775 	 * reason is that if one of these bits is necessary, it will appear
6776 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6777 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6778 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6779 	 * These rules have exceptions below.
6780 	 */
6781 
6782 	/* pin-based controls */
6783 	msrs->pinbased_ctls_low =
6784 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6785 
6786 	msrs->pinbased_ctls_high = vmcs_conf->pin_based_exec_ctrl;
6787 	msrs->pinbased_ctls_high &=
6788 		PIN_BASED_EXT_INTR_MASK |
6789 		PIN_BASED_NMI_EXITING |
6790 		PIN_BASED_VIRTUAL_NMIS |
6791 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6792 	msrs->pinbased_ctls_high |=
6793 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6794 		PIN_BASED_VMX_PREEMPTION_TIMER;
6795 
6796 	/* exit controls */
6797 	msrs->exit_ctls_low =
6798 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6799 
6800 	msrs->exit_ctls_high = vmcs_conf->vmexit_ctrl;
6801 	msrs->exit_ctls_high &=
6802 #ifdef CONFIG_X86_64
6803 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6804 #endif
6805 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6806 		VM_EXIT_CLEAR_BNDCFGS;
6807 	msrs->exit_ctls_high |=
6808 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6809 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6810 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT |
6811 		VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6812 
6813 	/* We support free control of debug control saving. */
6814 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6815 
6816 	/* entry controls */
6817 	msrs->entry_ctls_low =
6818 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6819 
6820 	msrs->entry_ctls_high = vmcs_conf->vmentry_ctrl;
6821 	msrs->entry_ctls_high &=
6822 #ifdef CONFIG_X86_64
6823 		VM_ENTRY_IA32E_MODE |
6824 #endif
6825 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
6826 	msrs->entry_ctls_high |=
6827 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER |
6828 		 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
6829 
6830 	/* We support free control of debug control loading. */
6831 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6832 
6833 	/* cpu-based controls */
6834 	msrs->procbased_ctls_low =
6835 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6836 
6837 	msrs->procbased_ctls_high = vmcs_conf->cpu_based_exec_ctrl;
6838 	msrs->procbased_ctls_high &=
6839 		CPU_BASED_INTR_WINDOW_EXITING |
6840 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6841 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6842 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6843 		CPU_BASED_CR3_STORE_EXITING |
6844 #ifdef CONFIG_X86_64
6845 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6846 #endif
6847 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6848 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6849 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6850 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6851 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6852 	/*
6853 	 * We can allow some features even when not supported by the
6854 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6855 	 * can use it to avoid exits to L1 - even when L0 runs L2
6856 	 * without MSR bitmaps.
6857 	 */
6858 	msrs->procbased_ctls_high |=
6859 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6860 		CPU_BASED_USE_MSR_BITMAPS;
6861 
6862 	/* We support free control of CR3 access interception. */
6863 	msrs->procbased_ctls_low &=
6864 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6865 
6866 	/*
6867 	 * secondary cpu-based controls.  Do not include those that
6868 	 * depend on CPUID bits, they are added later by
6869 	 * vmx_vcpu_after_set_cpuid.
6870 	 */
6871 	msrs->secondary_ctls_low = 0;
6872 
6873 	msrs->secondary_ctls_high = vmcs_conf->cpu_based_2nd_exec_ctrl;
6874 	msrs->secondary_ctls_high &=
6875 		SECONDARY_EXEC_DESC |
6876 		SECONDARY_EXEC_ENABLE_RDTSCP |
6877 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6878 		SECONDARY_EXEC_WBINVD_EXITING |
6879 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6880 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6881 		SECONDARY_EXEC_RDRAND_EXITING |
6882 		SECONDARY_EXEC_ENABLE_INVPCID |
6883 		SECONDARY_EXEC_ENABLE_VMFUNC |
6884 		SECONDARY_EXEC_RDSEED_EXITING |
6885 		SECONDARY_EXEC_XSAVES |
6886 		SECONDARY_EXEC_TSC_SCALING |
6887 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
6888 
6889 	/*
6890 	 * We can emulate "VMCS shadowing," even if the hardware
6891 	 * doesn't support it.
6892 	 */
6893 	msrs->secondary_ctls_high |=
6894 		SECONDARY_EXEC_SHADOW_VMCS;
6895 
6896 	if (enable_ept) {
6897 		/* nested EPT: emulate EPT also to L1 */
6898 		msrs->secondary_ctls_high |=
6899 			SECONDARY_EXEC_ENABLE_EPT;
6900 		msrs->ept_caps =
6901 			VMX_EPT_PAGE_WALK_4_BIT |
6902 			VMX_EPT_PAGE_WALK_5_BIT |
6903 			VMX_EPTP_WB_BIT |
6904 			VMX_EPT_INVEPT_BIT |
6905 			VMX_EPT_EXECUTE_ONLY_BIT;
6906 
6907 		msrs->ept_caps &= ept_caps;
6908 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6909 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6910 			VMX_EPT_1GB_PAGE_BIT;
6911 		if (enable_ept_ad_bits) {
6912 			msrs->secondary_ctls_high |=
6913 				SECONDARY_EXEC_ENABLE_PML;
6914 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6915 		}
6916 
6917 		/*
6918 		 * Advertise EPTP switching irrespective of hardware support,
6919 		 * KVM emulates it in software so long as VMFUNC is supported.
6920 		 */
6921 		if (cpu_has_vmx_vmfunc())
6922 			msrs->vmfunc_controls = VMX_VMFUNC_EPTP_SWITCHING;
6923 	}
6924 
6925 	/*
6926 	 * Old versions of KVM use the single-context version without
6927 	 * checking for support, so declare that it is supported even
6928 	 * though it is treated as global context.  The alternative is
6929 	 * not failing the single-context invvpid, and it is worse.
6930 	 */
6931 	if (enable_vpid) {
6932 		msrs->secondary_ctls_high |=
6933 			SECONDARY_EXEC_ENABLE_VPID;
6934 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6935 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6936 	}
6937 
6938 	if (enable_unrestricted_guest)
6939 		msrs->secondary_ctls_high |=
6940 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6941 
6942 	if (flexpriority_enabled)
6943 		msrs->secondary_ctls_high |=
6944 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6945 
6946 	if (enable_sgx)
6947 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6948 
6949 	/* miscellaneous data */
6950 	msrs->misc_low = (u32)vmcs_conf->misc & VMX_MISC_SAVE_EFER_LMA;
6951 	msrs->misc_low |=
6952 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6953 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6954 		VMX_MISC_ACTIVITY_HLT |
6955 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6956 	msrs->misc_high = 0;
6957 
6958 	/*
6959 	 * This MSR reports some information about VMX support. We
6960 	 * should return information about the VMX we emulate for the
6961 	 * guest, and the VMCS structure we give it - not about the
6962 	 * VMX support of the underlying hardware.
6963 	 */
6964 	msrs->basic =
6965 		VMCS12_REVISION |
6966 		VMX_BASIC_TRUE_CTLS |
6967 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6968 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6969 
6970 	if (cpu_has_vmx_basic_inout())
6971 		msrs->basic |= VMX_BASIC_INOUT;
6972 
6973 	/*
6974 	 * These MSRs specify bits which the guest must keep fixed on
6975 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6976 	 * We picked the standard core2 setting.
6977 	 */
6978 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6979 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6980 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6981 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6982 
6983 	/* These MSRs specify bits which the guest must keep fixed off. */
6984 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6985 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6986 
6987 	if (vmx_umip_emulated())
6988 		msrs->cr4_fixed1 |= X86_CR4_UMIP;
6989 
6990 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
6991 }
6992 
6993 void nested_vmx_hardware_unsetup(void)
6994 {
6995 	int i;
6996 
6997 	if (enable_shadow_vmcs) {
6998 		for (i = 0; i < VMX_BITMAP_NR; i++)
6999 			free_page((unsigned long)vmx_bitmap[i]);
7000 	}
7001 }
7002 
7003 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
7004 {
7005 	int i;
7006 
7007 	if (!cpu_has_vmx_shadow_vmcs())
7008 		enable_shadow_vmcs = 0;
7009 	if (enable_shadow_vmcs) {
7010 		for (i = 0; i < VMX_BITMAP_NR; i++) {
7011 			/*
7012 			 * The vmx_bitmap is not tied to a VM and so should
7013 			 * not be charged to a memcg.
7014 			 */
7015 			vmx_bitmap[i] = (unsigned long *)
7016 				__get_free_page(GFP_KERNEL);
7017 			if (!vmx_bitmap[i]) {
7018 				nested_vmx_hardware_unsetup();
7019 				return -ENOMEM;
7020 			}
7021 		}
7022 
7023 		init_vmcs_shadow_fields();
7024 	}
7025 
7026 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
7027 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
7028 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
7029 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
7030 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
7031 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
7032 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
7033 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmxoff;
7034 	exit_handlers[EXIT_REASON_VMON]		= handle_vmxon;
7035 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
7036 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
7037 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
7038 
7039 	return 0;
7040 }
7041 
7042 struct kvm_x86_nested_ops vmx_nested_ops = {
7043 	.leave_nested = vmx_leave_nested,
7044 	.is_exception_vmexit = nested_vmx_is_exception_vmexit,
7045 	.check_events = vmx_check_nested_events,
7046 	.has_events = vmx_has_nested_events,
7047 	.triple_fault = nested_vmx_triple_fault,
7048 	.get_state = vmx_get_nested_state,
7049 	.set_state = vmx_set_nested_state,
7050 	.get_nested_state_pages = vmx_get_nested_state_pages,
7051 	.write_log_dirty = nested_vmx_write_pml_buffer,
7052 	.enable_evmcs = nested_enable_evmcs,
7053 	.get_evmcs_version = nested_get_evmcs_version,
7054 	.hv_inject_synthetic_vmexit_post_tlb_flush = vmx_hv_inject_synthetic_vmexit_post_tlb_flush,
7055 };
7056