xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 462cd772)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/objtool.h>
4 #include <linux/percpu.h>
5 
6 #include <asm/debugreg.h>
7 #include <asm/mmu_context.h>
8 
9 #include "cpuid.h"
10 #include "evmcs.h"
11 #include "hyperv.h"
12 #include "mmu.h"
13 #include "nested.h"
14 #include "pmu.h"
15 #include "sgx.h"
16 #include "trace.h"
17 #include "vmx.h"
18 #include "x86.h"
19 
20 static bool __read_mostly enable_shadow_vmcs = 1;
21 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
22 
23 static bool __read_mostly nested_early_check = 0;
24 module_param(nested_early_check, bool, S_IRUGO);
25 
26 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
27 
28 /*
29  * Hyper-V requires all of these, so mark them as supported even though
30  * they are just treated the same as all-context.
31  */
32 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
33 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
34 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
35 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
36 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
37 
38 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
39 
40 enum {
41 	VMX_VMREAD_BITMAP,
42 	VMX_VMWRITE_BITMAP,
43 	VMX_BITMAP_NR
44 };
45 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
46 
47 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
48 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
49 
50 struct shadow_vmcs_field {
51 	u16	encoding;
52 	u16	offset;
53 };
54 static struct shadow_vmcs_field shadow_read_only_fields[] = {
55 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
56 #include "vmcs_shadow_fields.h"
57 };
58 static int max_shadow_read_only_fields =
59 	ARRAY_SIZE(shadow_read_only_fields);
60 
61 static struct shadow_vmcs_field shadow_read_write_fields[] = {
62 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
63 #include "vmcs_shadow_fields.h"
64 };
65 static int max_shadow_read_write_fields =
66 	ARRAY_SIZE(shadow_read_write_fields);
67 
68 static void init_vmcs_shadow_fields(void)
69 {
70 	int i, j;
71 
72 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
73 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
74 
75 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
76 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
77 		u16 field = entry.encoding;
78 
79 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
80 		    (i + 1 == max_shadow_read_only_fields ||
81 		     shadow_read_only_fields[i + 1].encoding != field + 1))
82 			pr_err("Missing field from shadow_read_only_field %x\n",
83 			       field + 1);
84 
85 		clear_bit(field, vmx_vmread_bitmap);
86 		if (field & 1)
87 #ifdef CONFIG_X86_64
88 			continue;
89 #else
90 			entry.offset += sizeof(u32);
91 #endif
92 		shadow_read_only_fields[j++] = entry;
93 	}
94 	max_shadow_read_only_fields = j;
95 
96 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
97 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
98 		u16 field = entry.encoding;
99 
100 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
101 		    (i + 1 == max_shadow_read_write_fields ||
102 		     shadow_read_write_fields[i + 1].encoding != field + 1))
103 			pr_err("Missing field from shadow_read_write_field %x\n",
104 			       field + 1);
105 
106 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
107 			  field <= GUEST_TR_AR_BYTES,
108 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
109 
110 		/*
111 		 * PML and the preemption timer can be emulated, but the
112 		 * processor cannot vmwrite to fields that don't exist
113 		 * on bare metal.
114 		 */
115 		switch (field) {
116 		case GUEST_PML_INDEX:
117 			if (!cpu_has_vmx_pml())
118 				continue;
119 			break;
120 		case VMX_PREEMPTION_TIMER_VALUE:
121 			if (!cpu_has_vmx_preemption_timer())
122 				continue;
123 			break;
124 		case GUEST_INTR_STATUS:
125 			if (!cpu_has_vmx_apicv())
126 				continue;
127 			break;
128 		default:
129 			break;
130 		}
131 
132 		clear_bit(field, vmx_vmwrite_bitmap);
133 		clear_bit(field, vmx_vmread_bitmap);
134 		if (field & 1)
135 #ifdef CONFIG_X86_64
136 			continue;
137 #else
138 			entry.offset += sizeof(u32);
139 #endif
140 		shadow_read_write_fields[j++] = entry;
141 	}
142 	max_shadow_read_write_fields = j;
143 }
144 
145 /*
146  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
147  * set the success or error code of an emulated VMX instruction (as specified
148  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
149  * instruction.
150  */
151 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
152 {
153 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
154 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
155 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
156 	return kvm_skip_emulated_instruction(vcpu);
157 }
158 
159 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
160 {
161 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
162 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
163 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
164 			| X86_EFLAGS_CF);
165 	return kvm_skip_emulated_instruction(vcpu);
166 }
167 
168 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
169 				u32 vm_instruction_error)
170 {
171 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
172 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
173 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
174 			| X86_EFLAGS_ZF);
175 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
176 	/*
177 	 * We don't need to force sync to shadow VMCS because
178 	 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
179 	 * fields and thus must be synced.
180 	 */
181 	if (to_vmx(vcpu)->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
182 		to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
183 
184 	return kvm_skip_emulated_instruction(vcpu);
185 }
186 
187 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
188 {
189 	struct vcpu_vmx *vmx = to_vmx(vcpu);
190 
191 	/*
192 	 * failValid writes the error number to the current VMCS, which
193 	 * can't be done if there isn't a current VMCS.
194 	 */
195 	if (vmx->nested.current_vmptr == INVALID_GPA &&
196 	    !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
197 		return nested_vmx_failInvalid(vcpu);
198 
199 	return nested_vmx_failValid(vcpu, vm_instruction_error);
200 }
201 
202 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
203 {
204 	/* TODO: not to reset guest simply here. */
205 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
206 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
207 }
208 
209 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
210 {
211 	return fixed_bits_valid(control, low, high);
212 }
213 
214 static inline u64 vmx_control_msr(u32 low, u32 high)
215 {
216 	return low | ((u64)high << 32);
217 }
218 
219 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
220 {
221 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
222 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
223 	vmx->nested.need_vmcs12_to_shadow_sync = false;
224 }
225 
226 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
227 {
228 	struct vcpu_vmx *vmx = to_vmx(vcpu);
229 
230 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
231 		kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
232 		vmx->nested.hv_evmcs = NULL;
233 	}
234 
235 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
236 }
237 
238 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
239 				     struct loaded_vmcs *prev)
240 {
241 	struct vmcs_host_state *dest, *src;
242 
243 	if (unlikely(!vmx->guest_state_loaded))
244 		return;
245 
246 	src = &prev->host_state;
247 	dest = &vmx->loaded_vmcs->host_state;
248 
249 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
250 	dest->ldt_sel = src->ldt_sel;
251 #ifdef CONFIG_X86_64
252 	dest->ds_sel = src->ds_sel;
253 	dest->es_sel = src->es_sel;
254 #endif
255 }
256 
257 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
258 {
259 	struct vcpu_vmx *vmx = to_vmx(vcpu);
260 	struct loaded_vmcs *prev;
261 	int cpu;
262 
263 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
264 		return;
265 
266 	cpu = get_cpu();
267 	prev = vmx->loaded_vmcs;
268 	vmx->loaded_vmcs = vmcs;
269 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
270 	vmx_sync_vmcs_host_state(vmx, prev);
271 	put_cpu();
272 
273 	vcpu->arch.regs_avail = ~VMX_REGS_LAZY_LOAD_SET;
274 
275 	/*
276 	 * All lazily updated registers will be reloaded from VMCS12 on both
277 	 * vmentry and vmexit.
278 	 */
279 	vcpu->arch.regs_dirty = 0;
280 }
281 
282 /*
283  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
284  * just stops using VMX.
285  */
286 static void free_nested(struct kvm_vcpu *vcpu)
287 {
288 	struct vcpu_vmx *vmx = to_vmx(vcpu);
289 
290 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
291 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
292 
293 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
294 		return;
295 
296 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
297 
298 	vmx->nested.vmxon = false;
299 	vmx->nested.smm.vmxon = false;
300 	vmx->nested.vmxon_ptr = INVALID_GPA;
301 	free_vpid(vmx->nested.vpid02);
302 	vmx->nested.posted_intr_nv = -1;
303 	vmx->nested.current_vmptr = INVALID_GPA;
304 	if (enable_shadow_vmcs) {
305 		vmx_disable_shadow_vmcs(vmx);
306 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
307 		free_vmcs(vmx->vmcs01.shadow_vmcs);
308 		vmx->vmcs01.shadow_vmcs = NULL;
309 	}
310 	kfree(vmx->nested.cached_vmcs12);
311 	vmx->nested.cached_vmcs12 = NULL;
312 	kfree(vmx->nested.cached_shadow_vmcs12);
313 	vmx->nested.cached_shadow_vmcs12 = NULL;
314 	/*
315 	 * Unpin physical memory we referred to in the vmcs02.  The APIC access
316 	 * page's backing page (yeah, confusing) shouldn't actually be accessed,
317 	 * and if it is written, the contents are irrelevant.
318 	 */
319 	kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
320 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
321 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
322 	vmx->nested.pi_desc = NULL;
323 
324 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
325 
326 	nested_release_evmcs(vcpu);
327 
328 	free_loaded_vmcs(&vmx->nested.vmcs02);
329 }
330 
331 /*
332  * Ensure that the current vmcs of the logical processor is the
333  * vmcs01 of the vcpu before calling free_nested().
334  */
335 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
336 {
337 	vcpu_load(vcpu);
338 	vmx_leave_nested(vcpu);
339 	vcpu_put(vcpu);
340 }
341 
342 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
343 
344 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
345 {
346 	return VALID_PAGE(root_hpa) &&
347 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
348 }
349 
350 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
351 				       gpa_t addr)
352 {
353 	uint i;
354 	struct kvm_mmu_root_info *cached_root;
355 
356 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
357 
358 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
359 		cached_root = &vcpu->arch.mmu->prev_roots[i];
360 
361 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
362 					    eptp))
363 			vcpu->arch.mmu->invlpg(vcpu, addr, cached_root->hpa);
364 	}
365 }
366 
367 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
368 		struct x86_exception *fault)
369 {
370 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
371 	struct vcpu_vmx *vmx = to_vmx(vcpu);
372 	u32 vm_exit_reason;
373 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
374 
375 	if (vmx->nested.pml_full) {
376 		vm_exit_reason = EXIT_REASON_PML_FULL;
377 		vmx->nested.pml_full = false;
378 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
379 	} else {
380 		if (fault->error_code & PFERR_RSVD_MASK)
381 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
382 		else
383 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
384 
385 		/*
386 		 * Although the caller (kvm_inject_emulated_page_fault) would
387 		 * have already synced the faulting address in the shadow EPT
388 		 * tables for the current EPTP12, we also need to sync it for
389 		 * any other cached EPTP02s based on the same EP4TA, since the
390 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
391 		 */
392 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
393 					   fault->address);
394 	}
395 
396 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
397 	vmcs12->guest_physical_address = fault->address;
398 }
399 
400 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
401 {
402 	struct vcpu_vmx *vmx = to_vmx(vcpu);
403 	bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
404 	int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
405 
406 	kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
407 				nested_ept_ad_enabled(vcpu),
408 				nested_ept_get_eptp(vcpu));
409 }
410 
411 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
412 {
413 	WARN_ON(mmu_is_nested(vcpu));
414 
415 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
416 	nested_ept_new_eptp(vcpu);
417 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
418 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
419 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
420 
421 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
422 }
423 
424 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
425 {
426 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
427 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
428 }
429 
430 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
431 					    u16 error_code)
432 {
433 	bool inequality, bit;
434 
435 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
436 	inequality =
437 		(error_code & vmcs12->page_fault_error_code_mask) !=
438 		 vmcs12->page_fault_error_code_match;
439 	return inequality ^ bit;
440 }
441 
442 
443 /*
444  * KVM wants to inject page-faults which it got to the guest. This function
445  * checks whether in a nested guest, we need to inject them to L1 or L2.
446  */
447 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
448 {
449 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
450 	unsigned int nr = vcpu->arch.exception.nr;
451 	bool has_payload = vcpu->arch.exception.has_payload;
452 	unsigned long payload = vcpu->arch.exception.payload;
453 
454 	if (nr == PF_VECTOR) {
455 		if (vcpu->arch.exception.nested_apf) {
456 			*exit_qual = vcpu->arch.apf.nested_apf_token;
457 			return 1;
458 		}
459 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
460 						    vcpu->arch.exception.error_code)) {
461 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
462 			return 1;
463 		}
464 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
465 		if (nr == DB_VECTOR) {
466 			if (!has_payload) {
467 				payload = vcpu->arch.dr6;
468 				payload &= ~DR6_BT;
469 				payload ^= DR6_ACTIVE_LOW;
470 			}
471 			*exit_qual = payload;
472 		} else
473 			*exit_qual = 0;
474 		return 1;
475 	}
476 
477 	return 0;
478 }
479 
480 static bool nested_vmx_handle_page_fault_workaround(struct kvm_vcpu *vcpu,
481 						    struct x86_exception *fault)
482 {
483 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
484 
485 	WARN_ON(!is_guest_mode(vcpu));
486 
487 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
488 	    !WARN_ON_ONCE(to_vmx(vcpu)->nested.nested_run_pending)) {
489 		vmcs12->vm_exit_intr_error_code = fault->error_code;
490 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
491 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
492 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
493 				  fault->address);
494 		return true;
495 	}
496 	return false;
497 }
498 
499 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
500 					       struct vmcs12 *vmcs12)
501 {
502 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
503 		return 0;
504 
505 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
506 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
507 		return -EINVAL;
508 
509 	return 0;
510 }
511 
512 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
513 						struct vmcs12 *vmcs12)
514 {
515 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
516 		return 0;
517 
518 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
519 		return -EINVAL;
520 
521 	return 0;
522 }
523 
524 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
525 						struct vmcs12 *vmcs12)
526 {
527 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
528 		return 0;
529 
530 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
531 		return -EINVAL;
532 
533 	return 0;
534 }
535 
536 /*
537  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
538  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
539  * only the "disable intercept" case needs to be handled.
540  */
541 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
542 							unsigned long *msr_bitmap_l0,
543 							u32 msr, int type)
544 {
545 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
546 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
547 
548 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
549 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
550 }
551 
552 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
553 {
554 	int msr;
555 
556 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
557 		unsigned word = msr / BITS_PER_LONG;
558 
559 		msr_bitmap[word] = ~0;
560 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
561 	}
562 }
563 
564 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
565 static inline									\
566 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
567 					 unsigned long *msr_bitmap_l1,		\
568 					 unsigned long *msr_bitmap_l0, u32 msr)	\
569 {										\
570 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
571 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
572 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
573 	else									\
574 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
575 }
576 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
577 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
578 
579 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
580 						    unsigned long *msr_bitmap_l1,
581 						    unsigned long *msr_bitmap_l0,
582 						    u32 msr, int types)
583 {
584 	if (types & MSR_TYPE_R)
585 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
586 						  msr_bitmap_l0, msr);
587 	if (types & MSR_TYPE_W)
588 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
589 						   msr_bitmap_l0, msr);
590 }
591 
592 /*
593  * Merge L0's and L1's MSR bitmap, return false to indicate that
594  * we do not use the hardware.
595  */
596 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
597 						 struct vmcs12 *vmcs12)
598 {
599 	struct vcpu_vmx *vmx = to_vmx(vcpu);
600 	int msr;
601 	unsigned long *msr_bitmap_l1;
602 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
603 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
604 	struct kvm_host_map *map = &vmx->nested.msr_bitmap_map;
605 
606 	/* Nothing to do if the MSR bitmap is not in use.  */
607 	if (!cpu_has_vmx_msr_bitmap() ||
608 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
609 		return false;
610 
611 	/*
612 	 * MSR bitmap update can be skipped when:
613 	 * - MSR bitmap for L1 hasn't changed.
614 	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
615 	 *   before.
616 	 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
617 	 *   and tells KVM (L0) there were no changes in MSR bitmap for L2.
618 	 */
619 	if (!vmx->nested.force_msr_bitmap_recalc && evmcs &&
620 	    evmcs->hv_enlightenments_control.msr_bitmap &&
621 	    evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
622 		return true;
623 
624 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
625 		return false;
626 
627 	msr_bitmap_l1 = (unsigned long *)map->hva;
628 
629 	/*
630 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
631 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
632 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
633 	 */
634 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
635 
636 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
637 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
638 			/*
639 			 * L0 need not intercept reads for MSRs between 0x800
640 			 * and 0x8ff, it just lets the processor take the value
641 			 * from the virtual-APIC page; take those 256 bits
642 			 * directly from the L1 bitmap.
643 			 */
644 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
645 				unsigned word = msr / BITS_PER_LONG;
646 
647 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
648 			}
649 		}
650 
651 		nested_vmx_disable_intercept_for_x2apic_msr(
652 			msr_bitmap_l1, msr_bitmap_l0,
653 			X2APIC_MSR(APIC_TASKPRI),
654 			MSR_TYPE_R | MSR_TYPE_W);
655 
656 		if (nested_cpu_has_vid(vmcs12)) {
657 			nested_vmx_disable_intercept_for_x2apic_msr(
658 				msr_bitmap_l1, msr_bitmap_l0,
659 				X2APIC_MSR(APIC_EOI),
660 				MSR_TYPE_W);
661 			nested_vmx_disable_intercept_for_x2apic_msr(
662 				msr_bitmap_l1, msr_bitmap_l0,
663 				X2APIC_MSR(APIC_SELF_IPI),
664 				MSR_TYPE_W);
665 		}
666 	}
667 
668 	/*
669 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
670 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
671 	 */
672 #ifdef CONFIG_X86_64
673 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
674 					 MSR_FS_BASE, MSR_TYPE_RW);
675 
676 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
677 					 MSR_GS_BASE, MSR_TYPE_RW);
678 
679 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
680 					 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
681 #endif
682 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
683 					 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
684 
685 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
686 					 MSR_IA32_PRED_CMD, MSR_TYPE_W);
687 
688 	kvm_vcpu_unmap(vcpu, &vmx->nested.msr_bitmap_map, false);
689 
690 	vmx->nested.force_msr_bitmap_recalc = false;
691 
692 	return true;
693 }
694 
695 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
696 				       struct vmcs12 *vmcs12)
697 {
698 	struct vcpu_vmx *vmx = to_vmx(vcpu);
699 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
700 
701 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
702 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
703 		return;
704 
705 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
706 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
707 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
708 		return;
709 
710 	kvm_read_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
711 			      VMCS12_SIZE);
712 }
713 
714 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
715 					      struct vmcs12 *vmcs12)
716 {
717 	struct vcpu_vmx *vmx = to_vmx(vcpu);
718 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
719 
720 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
721 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
722 		return;
723 
724 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
725 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
726 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
727 		return;
728 
729 	kvm_write_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
730 			       VMCS12_SIZE);
731 }
732 
733 /*
734  * In nested virtualization, check if L1 has set
735  * VM_EXIT_ACK_INTR_ON_EXIT
736  */
737 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
738 {
739 	return get_vmcs12(vcpu)->vm_exit_controls &
740 		VM_EXIT_ACK_INTR_ON_EXIT;
741 }
742 
743 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
744 					  struct vmcs12 *vmcs12)
745 {
746 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
747 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
748 		return -EINVAL;
749 	else
750 		return 0;
751 }
752 
753 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
754 					   struct vmcs12 *vmcs12)
755 {
756 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
757 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
758 	    !nested_cpu_has_vid(vmcs12) &&
759 	    !nested_cpu_has_posted_intr(vmcs12))
760 		return 0;
761 
762 	/*
763 	 * If virtualize x2apic mode is enabled,
764 	 * virtualize apic access must be disabled.
765 	 */
766 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
767 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
768 		return -EINVAL;
769 
770 	/*
771 	 * If virtual interrupt delivery is enabled,
772 	 * we must exit on external interrupts.
773 	 */
774 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
775 		return -EINVAL;
776 
777 	/*
778 	 * bits 15:8 should be zero in posted_intr_nv,
779 	 * the descriptor address has been already checked
780 	 * in nested_get_vmcs12_pages.
781 	 *
782 	 * bits 5:0 of posted_intr_desc_addr should be zero.
783 	 */
784 	if (nested_cpu_has_posted_intr(vmcs12) &&
785 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
786 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
787 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
788 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
789 		return -EINVAL;
790 
791 	/* tpr shadow is needed by all apicv features. */
792 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
793 		return -EINVAL;
794 
795 	return 0;
796 }
797 
798 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
799 				       u32 count, u64 addr)
800 {
801 	if (count == 0)
802 		return 0;
803 
804 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
805 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
806 		return -EINVAL;
807 
808 	return 0;
809 }
810 
811 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
812 						     struct vmcs12 *vmcs12)
813 {
814 	if (CC(nested_vmx_check_msr_switch(vcpu,
815 					   vmcs12->vm_exit_msr_load_count,
816 					   vmcs12->vm_exit_msr_load_addr)) ||
817 	    CC(nested_vmx_check_msr_switch(vcpu,
818 					   vmcs12->vm_exit_msr_store_count,
819 					   vmcs12->vm_exit_msr_store_addr)))
820 		return -EINVAL;
821 
822 	return 0;
823 }
824 
825 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
826                                                       struct vmcs12 *vmcs12)
827 {
828 	if (CC(nested_vmx_check_msr_switch(vcpu,
829 					   vmcs12->vm_entry_msr_load_count,
830 					   vmcs12->vm_entry_msr_load_addr)))
831                 return -EINVAL;
832 
833 	return 0;
834 }
835 
836 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
837 					 struct vmcs12 *vmcs12)
838 {
839 	if (!nested_cpu_has_pml(vmcs12))
840 		return 0;
841 
842 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
843 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
844 		return -EINVAL;
845 
846 	return 0;
847 }
848 
849 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
850 							struct vmcs12 *vmcs12)
851 {
852 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
853 	       !nested_cpu_has_ept(vmcs12)))
854 		return -EINVAL;
855 	return 0;
856 }
857 
858 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
859 							 struct vmcs12 *vmcs12)
860 {
861 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
862 	       !nested_cpu_has_ept(vmcs12)))
863 		return -EINVAL;
864 	return 0;
865 }
866 
867 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
868 						 struct vmcs12 *vmcs12)
869 {
870 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
871 		return 0;
872 
873 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
874 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
875 		return -EINVAL;
876 
877 	return 0;
878 }
879 
880 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
881 				       struct vmx_msr_entry *e)
882 {
883 	/* x2APIC MSR accesses are not allowed */
884 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
885 		return -EINVAL;
886 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
887 	    CC(e->index == MSR_IA32_UCODE_REV))
888 		return -EINVAL;
889 	if (CC(e->reserved != 0))
890 		return -EINVAL;
891 	return 0;
892 }
893 
894 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
895 				     struct vmx_msr_entry *e)
896 {
897 	if (CC(e->index == MSR_FS_BASE) ||
898 	    CC(e->index == MSR_GS_BASE) ||
899 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
900 	    nested_vmx_msr_check_common(vcpu, e))
901 		return -EINVAL;
902 	return 0;
903 }
904 
905 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
906 				      struct vmx_msr_entry *e)
907 {
908 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
909 	    nested_vmx_msr_check_common(vcpu, e))
910 		return -EINVAL;
911 	return 0;
912 }
913 
914 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
915 {
916 	struct vcpu_vmx *vmx = to_vmx(vcpu);
917 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
918 				       vmx->nested.msrs.misc_high);
919 
920 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
921 }
922 
923 /*
924  * Load guest's/host's msr at nested entry/exit.
925  * return 0 for success, entry index for failure.
926  *
927  * One of the failure modes for MSR load/store is when a list exceeds the
928  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
929  * as possible, process all valid entries before failing rather than precheck
930  * for a capacity violation.
931  */
932 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
933 {
934 	u32 i;
935 	struct vmx_msr_entry e;
936 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
937 
938 	for (i = 0; i < count; i++) {
939 		if (unlikely(i >= max_msr_list_size))
940 			goto fail;
941 
942 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
943 					&e, sizeof(e))) {
944 			pr_debug_ratelimited(
945 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
946 				__func__, i, gpa + i * sizeof(e));
947 			goto fail;
948 		}
949 		if (nested_vmx_load_msr_check(vcpu, &e)) {
950 			pr_debug_ratelimited(
951 				"%s check failed (%u, 0x%x, 0x%x)\n",
952 				__func__, i, e.index, e.reserved);
953 			goto fail;
954 		}
955 		if (kvm_set_msr(vcpu, e.index, e.value)) {
956 			pr_debug_ratelimited(
957 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
958 				__func__, i, e.index, e.value);
959 			goto fail;
960 		}
961 	}
962 	return 0;
963 fail:
964 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
965 	return i + 1;
966 }
967 
968 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
969 					    u32 msr_index,
970 					    u64 *data)
971 {
972 	struct vcpu_vmx *vmx = to_vmx(vcpu);
973 
974 	/*
975 	 * If the L0 hypervisor stored a more accurate value for the TSC that
976 	 * does not include the time taken for emulation of the L2->L1
977 	 * VM-exit in L0, use the more accurate value.
978 	 */
979 	if (msr_index == MSR_IA32_TSC) {
980 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
981 						    MSR_IA32_TSC);
982 
983 		if (i >= 0) {
984 			u64 val = vmx->msr_autostore.guest.val[i].value;
985 
986 			*data = kvm_read_l1_tsc(vcpu, val);
987 			return true;
988 		}
989 	}
990 
991 	if (kvm_get_msr(vcpu, msr_index, data)) {
992 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
993 			msr_index);
994 		return false;
995 	}
996 	return true;
997 }
998 
999 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
1000 				     struct vmx_msr_entry *e)
1001 {
1002 	if (kvm_vcpu_read_guest(vcpu,
1003 				gpa + i * sizeof(*e),
1004 				e, 2 * sizeof(u32))) {
1005 		pr_debug_ratelimited(
1006 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
1007 			__func__, i, gpa + i * sizeof(*e));
1008 		return false;
1009 	}
1010 	if (nested_vmx_store_msr_check(vcpu, e)) {
1011 		pr_debug_ratelimited(
1012 			"%s check failed (%u, 0x%x, 0x%x)\n",
1013 			__func__, i, e->index, e->reserved);
1014 		return false;
1015 	}
1016 	return true;
1017 }
1018 
1019 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1020 {
1021 	u64 data;
1022 	u32 i;
1023 	struct vmx_msr_entry e;
1024 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1025 
1026 	for (i = 0; i < count; i++) {
1027 		if (unlikely(i >= max_msr_list_size))
1028 			return -EINVAL;
1029 
1030 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1031 			return -EINVAL;
1032 
1033 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1034 			return -EINVAL;
1035 
1036 		if (kvm_vcpu_write_guest(vcpu,
1037 					 gpa + i * sizeof(e) +
1038 					     offsetof(struct vmx_msr_entry, value),
1039 					 &data, sizeof(data))) {
1040 			pr_debug_ratelimited(
1041 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1042 				__func__, i, e.index, data);
1043 			return -EINVAL;
1044 		}
1045 	}
1046 	return 0;
1047 }
1048 
1049 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1050 {
1051 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1052 	u32 count = vmcs12->vm_exit_msr_store_count;
1053 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1054 	struct vmx_msr_entry e;
1055 	u32 i;
1056 
1057 	for (i = 0; i < count; i++) {
1058 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1059 			return false;
1060 
1061 		if (e.index == msr_index)
1062 			return true;
1063 	}
1064 	return false;
1065 }
1066 
1067 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1068 					   u32 msr_index)
1069 {
1070 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1071 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1072 	bool in_vmcs12_store_list;
1073 	int msr_autostore_slot;
1074 	bool in_autostore_list;
1075 	int last;
1076 
1077 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1078 	in_autostore_list = msr_autostore_slot >= 0;
1079 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1080 
1081 	if (in_vmcs12_store_list && !in_autostore_list) {
1082 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1083 			/*
1084 			 * Emulated VMEntry does not fail here.  Instead a less
1085 			 * accurate value will be returned by
1086 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1087 			 * instead of reading the value from the vmcs02 VMExit
1088 			 * MSR-store area.
1089 			 */
1090 			pr_warn_ratelimited(
1091 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1092 				msr_index);
1093 			return;
1094 		}
1095 		last = autostore->nr++;
1096 		autostore->val[last].index = msr_index;
1097 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1098 		last = --autostore->nr;
1099 		autostore->val[msr_autostore_slot] = autostore->val[last];
1100 	}
1101 }
1102 
1103 /*
1104  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1105  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1106  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1107  * @entry_failure_code.
1108  */
1109 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1110 			       bool nested_ept, bool reload_pdptrs,
1111 			       enum vm_entry_failure_code *entry_failure_code)
1112 {
1113 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1114 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1115 		return -EINVAL;
1116 	}
1117 
1118 	/*
1119 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1120 	 * must not be dereferenced.
1121 	 */
1122 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1123 	    CC(!load_pdptrs(vcpu, cr3))) {
1124 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1125 		return -EINVAL;
1126 	}
1127 
1128 	vcpu->arch.cr3 = cr3;
1129 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1130 
1131 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1132 	kvm_init_mmu(vcpu);
1133 
1134 	if (!nested_ept)
1135 		kvm_mmu_new_pgd(vcpu, cr3);
1136 
1137 	return 0;
1138 }
1139 
1140 /*
1141  * Returns if KVM is able to config CPU to tag TLB entries
1142  * populated by L2 differently than TLB entries populated
1143  * by L1.
1144  *
1145  * If L0 uses EPT, L1 and L2 run with different EPTP because
1146  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1147  * are tagged with different EPTP.
1148  *
1149  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1150  * with different VPID (L1 entries are tagged with vmx->vpid
1151  * while L2 entries are tagged with vmx->nested.vpid02).
1152  */
1153 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1154 {
1155 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1156 
1157 	return enable_ept ||
1158 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1159 }
1160 
1161 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1162 					    struct vmcs12 *vmcs12,
1163 					    bool is_vmenter)
1164 {
1165 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1166 
1167 	/*
1168 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1169 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1170 	 * full TLB flush from the guest's perspective.  This is required even
1171 	 * if VPID is disabled in the host as KVM may need to synchronize the
1172 	 * MMU in response to the guest TLB flush.
1173 	 *
1174 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1175 	 * EPT is a special snowflake, as guest-physical mappings aren't
1176 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1177 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1178 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1179 	 * those mappings.
1180 	 */
1181 	if (!nested_cpu_has_vpid(vmcs12)) {
1182 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1183 		return;
1184 	}
1185 
1186 	/* L2 should never have a VPID if VPID is disabled. */
1187 	WARN_ON(!enable_vpid);
1188 
1189 	/*
1190 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1191 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1192 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1193 	 * that the new vpid12 has never been used and thus represents a new
1194 	 * guest ASID that cannot have entries in the TLB.
1195 	 */
1196 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1197 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1198 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1199 		return;
1200 	}
1201 
1202 	/*
1203 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1204 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1205 	 * KVM was unable to allocate a VPID for L2, flush the current context
1206 	 * as the effective ASID is common to both L1 and L2.
1207 	 */
1208 	if (!nested_has_guest_tlb_tag(vcpu))
1209 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1210 }
1211 
1212 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1213 {
1214 	superset &= mask;
1215 	subset &= mask;
1216 
1217 	return (superset | subset) == superset;
1218 }
1219 
1220 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1221 {
1222 	const u64 feature_and_reserved =
1223 		/* feature (except bit 48; see below) */
1224 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1225 		/* reserved */
1226 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1227 	u64 vmx_basic = vmcs_config.nested.basic;
1228 
1229 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1230 		return -EINVAL;
1231 
1232 	/*
1233 	 * KVM does not emulate a version of VMX that constrains physical
1234 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1235 	 */
1236 	if (data & BIT_ULL(48))
1237 		return -EINVAL;
1238 
1239 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1240 	    vmx_basic_vmcs_revision_id(data))
1241 		return -EINVAL;
1242 
1243 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1244 		return -EINVAL;
1245 
1246 	vmx->nested.msrs.basic = data;
1247 	return 0;
1248 }
1249 
1250 static void vmx_get_control_msr(struct nested_vmx_msrs *msrs, u32 msr_index,
1251 				u32 **low, u32 **high)
1252 {
1253 	switch (msr_index) {
1254 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1255 		*low = &msrs->pinbased_ctls_low;
1256 		*high = &msrs->pinbased_ctls_high;
1257 		break;
1258 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1259 		*low = &msrs->procbased_ctls_low;
1260 		*high = &msrs->procbased_ctls_high;
1261 		break;
1262 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1263 		*low = &msrs->exit_ctls_low;
1264 		*high = &msrs->exit_ctls_high;
1265 		break;
1266 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1267 		*low = &msrs->entry_ctls_low;
1268 		*high = &msrs->entry_ctls_high;
1269 		break;
1270 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1271 		*low = &msrs->secondary_ctls_low;
1272 		*high = &msrs->secondary_ctls_high;
1273 		break;
1274 	default:
1275 		BUG();
1276 	}
1277 }
1278 
1279 static int
1280 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1281 {
1282 	u32 *lowp, *highp;
1283 	u64 supported;
1284 
1285 	vmx_get_control_msr(&vmcs_config.nested, msr_index, &lowp, &highp);
1286 
1287 	supported = vmx_control_msr(*lowp, *highp);
1288 
1289 	/* Check must-be-1 bits are still 1. */
1290 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1291 		return -EINVAL;
1292 
1293 	/* Check must-be-0 bits are still 0. */
1294 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1295 		return -EINVAL;
1296 
1297 	vmx_get_control_msr(&vmx->nested.msrs, msr_index, &lowp, &highp);
1298 	*lowp = data;
1299 	*highp = data >> 32;
1300 	return 0;
1301 }
1302 
1303 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1304 {
1305 	const u64 feature_and_reserved_bits =
1306 		/* feature */
1307 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1308 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1309 		/* reserved */
1310 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1311 	u64 vmx_misc = vmx_control_msr(vmcs_config.nested.misc_low,
1312 				       vmcs_config.nested.misc_high);
1313 
1314 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1315 		return -EINVAL;
1316 
1317 	if ((vmx->nested.msrs.pinbased_ctls_high &
1318 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1319 	    vmx_misc_preemption_timer_rate(data) !=
1320 	    vmx_misc_preemption_timer_rate(vmx_misc))
1321 		return -EINVAL;
1322 
1323 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1324 		return -EINVAL;
1325 
1326 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1327 		return -EINVAL;
1328 
1329 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1330 		return -EINVAL;
1331 
1332 	vmx->nested.msrs.misc_low = data;
1333 	vmx->nested.msrs.misc_high = data >> 32;
1334 
1335 	return 0;
1336 }
1337 
1338 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1339 {
1340 	u64 vmx_ept_vpid_cap = vmx_control_msr(vmcs_config.nested.ept_caps,
1341 					       vmcs_config.nested.vpid_caps);
1342 
1343 	/* Every bit is either reserved or a feature bit. */
1344 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1345 		return -EINVAL;
1346 
1347 	vmx->nested.msrs.ept_caps = data;
1348 	vmx->nested.msrs.vpid_caps = data >> 32;
1349 	return 0;
1350 }
1351 
1352 static u64 *vmx_get_fixed0_msr(struct nested_vmx_msrs *msrs, u32 msr_index)
1353 {
1354 	switch (msr_index) {
1355 	case MSR_IA32_VMX_CR0_FIXED0:
1356 		return &msrs->cr0_fixed0;
1357 	case MSR_IA32_VMX_CR4_FIXED0:
1358 		return &msrs->cr4_fixed0;
1359 	default:
1360 		BUG();
1361 	}
1362 }
1363 
1364 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1365 {
1366 	const u64 *msr = vmx_get_fixed0_msr(&vmcs_config.nested, msr_index);
1367 
1368 	/*
1369 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1370 	 * must be 1 in the restored value.
1371 	 */
1372 	if (!is_bitwise_subset(data, *msr, -1ULL))
1373 		return -EINVAL;
1374 
1375 	*vmx_get_fixed0_msr(&vmx->nested.msrs, msr_index) = data;
1376 	return 0;
1377 }
1378 
1379 /*
1380  * Called when userspace is restoring VMX MSRs.
1381  *
1382  * Returns 0 on success, non-0 otherwise.
1383  */
1384 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1385 {
1386 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1387 
1388 	/*
1389 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1390 	 * is in VMX operation.
1391 	 */
1392 	if (vmx->nested.vmxon)
1393 		return -EBUSY;
1394 
1395 	switch (msr_index) {
1396 	case MSR_IA32_VMX_BASIC:
1397 		return vmx_restore_vmx_basic(vmx, data);
1398 	case MSR_IA32_VMX_PINBASED_CTLS:
1399 	case MSR_IA32_VMX_PROCBASED_CTLS:
1400 	case MSR_IA32_VMX_EXIT_CTLS:
1401 	case MSR_IA32_VMX_ENTRY_CTLS:
1402 		/*
1403 		 * The "non-true" VMX capability MSRs are generated from the
1404 		 * "true" MSRs, so we do not support restoring them directly.
1405 		 *
1406 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1407 		 * should restore the "true" MSRs with the must-be-1 bits
1408 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1409 		 * DEFAULT SETTINGS".
1410 		 */
1411 		return -EINVAL;
1412 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1413 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1414 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1415 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1416 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1417 		return vmx_restore_control_msr(vmx, msr_index, data);
1418 	case MSR_IA32_VMX_MISC:
1419 		return vmx_restore_vmx_misc(vmx, data);
1420 	case MSR_IA32_VMX_CR0_FIXED0:
1421 	case MSR_IA32_VMX_CR4_FIXED0:
1422 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1423 	case MSR_IA32_VMX_CR0_FIXED1:
1424 	case MSR_IA32_VMX_CR4_FIXED1:
1425 		/*
1426 		 * These MSRs are generated based on the vCPU's CPUID, so we
1427 		 * do not support restoring them directly.
1428 		 */
1429 		return -EINVAL;
1430 	case MSR_IA32_VMX_EPT_VPID_CAP:
1431 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1432 	case MSR_IA32_VMX_VMCS_ENUM:
1433 		vmx->nested.msrs.vmcs_enum = data;
1434 		return 0;
1435 	case MSR_IA32_VMX_VMFUNC:
1436 		if (data & ~vmcs_config.nested.vmfunc_controls)
1437 			return -EINVAL;
1438 		vmx->nested.msrs.vmfunc_controls = data;
1439 		return 0;
1440 	default:
1441 		/*
1442 		 * The rest of the VMX capability MSRs do not support restore.
1443 		 */
1444 		return -EINVAL;
1445 	}
1446 }
1447 
1448 /* Returns 0 on success, non-0 otherwise. */
1449 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1450 {
1451 	switch (msr_index) {
1452 	case MSR_IA32_VMX_BASIC:
1453 		*pdata = msrs->basic;
1454 		break;
1455 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1456 	case MSR_IA32_VMX_PINBASED_CTLS:
1457 		*pdata = vmx_control_msr(
1458 			msrs->pinbased_ctls_low,
1459 			msrs->pinbased_ctls_high);
1460 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1461 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1462 		break;
1463 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1464 	case MSR_IA32_VMX_PROCBASED_CTLS:
1465 		*pdata = vmx_control_msr(
1466 			msrs->procbased_ctls_low,
1467 			msrs->procbased_ctls_high);
1468 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1469 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1470 		break;
1471 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1472 	case MSR_IA32_VMX_EXIT_CTLS:
1473 		*pdata = vmx_control_msr(
1474 			msrs->exit_ctls_low,
1475 			msrs->exit_ctls_high);
1476 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1477 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1478 		break;
1479 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1480 	case MSR_IA32_VMX_ENTRY_CTLS:
1481 		*pdata = vmx_control_msr(
1482 			msrs->entry_ctls_low,
1483 			msrs->entry_ctls_high);
1484 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1485 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1486 		break;
1487 	case MSR_IA32_VMX_MISC:
1488 		*pdata = vmx_control_msr(
1489 			msrs->misc_low,
1490 			msrs->misc_high);
1491 		break;
1492 	case MSR_IA32_VMX_CR0_FIXED0:
1493 		*pdata = msrs->cr0_fixed0;
1494 		break;
1495 	case MSR_IA32_VMX_CR0_FIXED1:
1496 		*pdata = msrs->cr0_fixed1;
1497 		break;
1498 	case MSR_IA32_VMX_CR4_FIXED0:
1499 		*pdata = msrs->cr4_fixed0;
1500 		break;
1501 	case MSR_IA32_VMX_CR4_FIXED1:
1502 		*pdata = msrs->cr4_fixed1;
1503 		break;
1504 	case MSR_IA32_VMX_VMCS_ENUM:
1505 		*pdata = msrs->vmcs_enum;
1506 		break;
1507 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1508 		*pdata = vmx_control_msr(
1509 			msrs->secondary_ctls_low,
1510 			msrs->secondary_ctls_high);
1511 		break;
1512 	case MSR_IA32_VMX_EPT_VPID_CAP:
1513 		*pdata = msrs->ept_caps |
1514 			((u64)msrs->vpid_caps << 32);
1515 		break;
1516 	case MSR_IA32_VMX_VMFUNC:
1517 		*pdata = msrs->vmfunc_controls;
1518 		break;
1519 	default:
1520 		return 1;
1521 	}
1522 
1523 	return 0;
1524 }
1525 
1526 /*
1527  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1528  * been modified by the L1 guest.  Note, "writable" in this context means
1529  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1530  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1531  * VM-exit information fields (which are actually writable if the vCPU is
1532  * configured to support "VMWRITE to any supported field in the VMCS").
1533  */
1534 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1535 {
1536 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1537 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1538 	struct shadow_vmcs_field field;
1539 	unsigned long val;
1540 	int i;
1541 
1542 	if (WARN_ON(!shadow_vmcs))
1543 		return;
1544 
1545 	preempt_disable();
1546 
1547 	vmcs_load(shadow_vmcs);
1548 
1549 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1550 		field = shadow_read_write_fields[i];
1551 		val = __vmcs_readl(field.encoding);
1552 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1553 	}
1554 
1555 	vmcs_clear(shadow_vmcs);
1556 	vmcs_load(vmx->loaded_vmcs->vmcs);
1557 
1558 	preempt_enable();
1559 }
1560 
1561 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1562 {
1563 	const struct shadow_vmcs_field *fields[] = {
1564 		shadow_read_write_fields,
1565 		shadow_read_only_fields
1566 	};
1567 	const int max_fields[] = {
1568 		max_shadow_read_write_fields,
1569 		max_shadow_read_only_fields
1570 	};
1571 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1572 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1573 	struct shadow_vmcs_field field;
1574 	unsigned long val;
1575 	int i, q;
1576 
1577 	if (WARN_ON(!shadow_vmcs))
1578 		return;
1579 
1580 	vmcs_load(shadow_vmcs);
1581 
1582 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1583 		for (i = 0; i < max_fields[q]; i++) {
1584 			field = fields[q][i];
1585 			val = vmcs12_read_any(vmcs12, field.encoding,
1586 					      field.offset);
1587 			__vmcs_writel(field.encoding, val);
1588 		}
1589 	}
1590 
1591 	vmcs_clear(shadow_vmcs);
1592 	vmcs_load(vmx->loaded_vmcs->vmcs);
1593 }
1594 
1595 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1596 {
1597 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1598 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1599 
1600 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1601 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1602 	vmcs12->guest_rip = evmcs->guest_rip;
1603 
1604 	if (unlikely(!(hv_clean_fields &
1605 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1606 		vmcs12->guest_rsp = evmcs->guest_rsp;
1607 		vmcs12->guest_rflags = evmcs->guest_rflags;
1608 		vmcs12->guest_interruptibility_info =
1609 			evmcs->guest_interruptibility_info;
1610 	}
1611 
1612 	if (unlikely(!(hv_clean_fields &
1613 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1614 		vmcs12->cpu_based_vm_exec_control =
1615 			evmcs->cpu_based_vm_exec_control;
1616 	}
1617 
1618 	if (unlikely(!(hv_clean_fields &
1619 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1620 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1621 	}
1622 
1623 	if (unlikely(!(hv_clean_fields &
1624 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1625 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1626 	}
1627 
1628 	if (unlikely(!(hv_clean_fields &
1629 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1630 		vmcs12->vm_entry_intr_info_field =
1631 			evmcs->vm_entry_intr_info_field;
1632 		vmcs12->vm_entry_exception_error_code =
1633 			evmcs->vm_entry_exception_error_code;
1634 		vmcs12->vm_entry_instruction_len =
1635 			evmcs->vm_entry_instruction_len;
1636 	}
1637 
1638 	if (unlikely(!(hv_clean_fields &
1639 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1640 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1641 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1642 		vmcs12->host_cr0 = evmcs->host_cr0;
1643 		vmcs12->host_cr3 = evmcs->host_cr3;
1644 		vmcs12->host_cr4 = evmcs->host_cr4;
1645 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1646 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1647 		vmcs12->host_rip = evmcs->host_rip;
1648 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1649 		vmcs12->host_es_selector = evmcs->host_es_selector;
1650 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1651 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1652 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1653 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1654 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1655 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1656 	}
1657 
1658 	if (unlikely(!(hv_clean_fields &
1659 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1660 		vmcs12->pin_based_vm_exec_control =
1661 			evmcs->pin_based_vm_exec_control;
1662 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1663 		vmcs12->secondary_vm_exec_control =
1664 			evmcs->secondary_vm_exec_control;
1665 	}
1666 
1667 	if (unlikely(!(hv_clean_fields &
1668 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1669 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1670 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1671 	}
1672 
1673 	if (unlikely(!(hv_clean_fields &
1674 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1675 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1676 	}
1677 
1678 	if (unlikely(!(hv_clean_fields &
1679 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1680 		vmcs12->guest_es_base = evmcs->guest_es_base;
1681 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1682 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1683 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1684 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1685 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1686 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1687 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1688 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1689 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1690 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1691 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1692 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1693 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1694 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1695 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1696 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1697 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1698 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1699 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1700 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1701 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1702 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1703 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1704 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1705 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1706 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1707 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1708 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1709 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1710 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1711 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1712 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1713 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1714 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1715 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1716 	}
1717 
1718 	if (unlikely(!(hv_clean_fields &
1719 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1720 		vmcs12->tsc_offset = evmcs->tsc_offset;
1721 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1722 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1723 	}
1724 
1725 	if (unlikely(!(hv_clean_fields &
1726 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1727 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1728 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1729 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1730 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1731 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1732 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1733 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1734 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1735 	}
1736 
1737 	if (unlikely(!(hv_clean_fields &
1738 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1739 		vmcs12->host_fs_base = evmcs->host_fs_base;
1740 		vmcs12->host_gs_base = evmcs->host_gs_base;
1741 		vmcs12->host_tr_base = evmcs->host_tr_base;
1742 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1743 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1744 		vmcs12->host_rsp = evmcs->host_rsp;
1745 	}
1746 
1747 	if (unlikely(!(hv_clean_fields &
1748 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1749 		vmcs12->ept_pointer = evmcs->ept_pointer;
1750 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1751 	}
1752 
1753 	if (unlikely(!(hv_clean_fields &
1754 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1755 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1756 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1757 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1758 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1759 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1760 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1761 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1762 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1763 		vmcs12->guest_pending_dbg_exceptions =
1764 			evmcs->guest_pending_dbg_exceptions;
1765 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1766 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1767 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1768 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1769 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1770 	}
1771 
1772 	/*
1773 	 * Not used?
1774 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1775 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1776 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1777 	 * vmcs12->page_fault_error_code_mask =
1778 	 *		evmcs->page_fault_error_code_mask;
1779 	 * vmcs12->page_fault_error_code_match =
1780 	 *		evmcs->page_fault_error_code_match;
1781 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1782 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1783 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1784 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1785 	 */
1786 
1787 	/*
1788 	 * Read only fields:
1789 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1790 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1791 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1792 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1793 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1794 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1795 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1796 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1797 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1798 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1799 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1800 	 *
1801 	 * Not present in struct vmcs12:
1802 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1803 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1804 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1805 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1806 	 */
1807 
1808 	return;
1809 }
1810 
1811 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1812 {
1813 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1814 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1815 
1816 	/*
1817 	 * Should not be changed by KVM:
1818 	 *
1819 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1820 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1821 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1822 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1823 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1824 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1825 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1826 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1827 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1828 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1829 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1830 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1831 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1832 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1833 	 * evmcs->host_rip = vmcs12->host_rip;
1834 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1835 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1836 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1837 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1838 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1839 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1840 	 * evmcs->host_rsp = vmcs12->host_rsp;
1841 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1842 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1843 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1844 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1845 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1846 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1847 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1848 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1849 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1850 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1851 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1852 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1853 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1854 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1855 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1856 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1857 	 * evmcs->page_fault_error_code_mask =
1858 	 *		vmcs12->page_fault_error_code_mask;
1859 	 * evmcs->page_fault_error_code_match =
1860 	 *		vmcs12->page_fault_error_code_match;
1861 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1862 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1863 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1864 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1865 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1866 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1867 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1868 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1869 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1870 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1871 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1872 	 *
1873 	 * Not present in struct vmcs12:
1874 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1875 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1876 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1877 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1878 	 */
1879 
1880 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1881 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1882 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1883 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1884 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1885 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1886 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1887 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1888 
1889 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1890 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1891 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1892 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1893 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1894 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1895 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1896 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1897 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1898 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1899 
1900 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1901 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1902 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1903 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1904 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1905 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1906 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1907 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1908 
1909 	evmcs->guest_es_base = vmcs12->guest_es_base;
1910 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1911 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1912 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1913 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1914 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1915 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1916 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1917 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1918 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1919 
1920 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1921 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1922 
1923 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1924 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1925 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1926 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1927 
1928 	evmcs->guest_pending_dbg_exceptions =
1929 		vmcs12->guest_pending_dbg_exceptions;
1930 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1931 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1932 
1933 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1934 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1935 
1936 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1937 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1938 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1939 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1940 
1941 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1942 
1943 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1944 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1945 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1946 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1947 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1948 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1949 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1950 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1951 
1952 	evmcs->exit_qualification = vmcs12->exit_qualification;
1953 
1954 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1955 	evmcs->guest_rsp = vmcs12->guest_rsp;
1956 	evmcs->guest_rflags = vmcs12->guest_rflags;
1957 
1958 	evmcs->guest_interruptibility_info =
1959 		vmcs12->guest_interruptibility_info;
1960 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1961 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1962 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1963 	evmcs->vm_entry_exception_error_code =
1964 		vmcs12->vm_entry_exception_error_code;
1965 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1966 
1967 	evmcs->guest_rip = vmcs12->guest_rip;
1968 
1969 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1970 
1971 	return;
1972 }
1973 
1974 /*
1975  * This is an equivalent of the nested hypervisor executing the vmptrld
1976  * instruction.
1977  */
1978 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1979 	struct kvm_vcpu *vcpu, bool from_launch)
1980 {
1981 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1982 	bool evmcs_gpa_changed = false;
1983 	u64 evmcs_gpa;
1984 
1985 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1986 		return EVMPTRLD_DISABLED;
1987 
1988 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa)) {
1989 		nested_release_evmcs(vcpu);
1990 		return EVMPTRLD_DISABLED;
1991 	}
1992 
1993 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1994 		vmx->nested.current_vmptr = INVALID_GPA;
1995 
1996 		nested_release_evmcs(vcpu);
1997 
1998 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1999 				 &vmx->nested.hv_evmcs_map))
2000 			return EVMPTRLD_ERROR;
2001 
2002 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2003 
2004 		/*
2005 		 * Currently, KVM only supports eVMCS version 1
2006 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2007 		 * value to first u32 field of eVMCS which should specify eVMCS
2008 		 * VersionNumber.
2009 		 *
2010 		 * Guest should be aware of supported eVMCS versions by host by
2011 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2012 		 * expected to set this CPUID leaf according to the value
2013 		 * returned in vmcs_version from nested_enable_evmcs().
2014 		 *
2015 		 * However, it turns out that Microsoft Hyper-V fails to comply
2016 		 * to their own invented interface: When Hyper-V use eVMCS, it
2017 		 * just sets first u32 field of eVMCS to revision_id specified
2018 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2019 		 * which is one of the supported versions specified in
2020 		 * CPUID.0x4000000A.EAX[0:15].
2021 		 *
2022 		 * To overcome Hyper-V bug, we accept here either a supported
2023 		 * eVMCS version or VMCS12 revision_id as valid values for first
2024 		 * u32 field of eVMCS.
2025 		 */
2026 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2027 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2028 			nested_release_evmcs(vcpu);
2029 			return EVMPTRLD_VMFAIL;
2030 		}
2031 
2032 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2033 
2034 		evmcs_gpa_changed = true;
2035 		/*
2036 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2037 		 * reloaded from guest's memory (read only fields, fields not
2038 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2039 		 * are no leftovers.
2040 		 */
2041 		if (from_launch) {
2042 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2043 			memset(vmcs12, 0, sizeof(*vmcs12));
2044 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2045 		}
2046 
2047 	}
2048 
2049 	/*
2050 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2051 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2052 	 */
2053 	if (from_launch || evmcs_gpa_changed) {
2054 		vmx->nested.hv_evmcs->hv_clean_fields &=
2055 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2056 
2057 		vmx->nested.force_msr_bitmap_recalc = true;
2058 	}
2059 
2060 	return EVMPTRLD_SUCCEEDED;
2061 }
2062 
2063 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2064 {
2065 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2066 
2067 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2068 		copy_vmcs12_to_enlightened(vmx);
2069 	else
2070 		copy_vmcs12_to_shadow(vmx);
2071 
2072 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2073 }
2074 
2075 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2076 {
2077 	struct vcpu_vmx *vmx =
2078 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2079 
2080 	vmx->nested.preemption_timer_expired = true;
2081 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2082 	kvm_vcpu_kick(&vmx->vcpu);
2083 
2084 	return HRTIMER_NORESTART;
2085 }
2086 
2087 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2088 {
2089 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2090 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2091 
2092 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2093 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2094 
2095 	if (!vmx->nested.has_preemption_timer_deadline) {
2096 		vmx->nested.preemption_timer_deadline =
2097 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2098 		vmx->nested.has_preemption_timer_deadline = true;
2099 	}
2100 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2101 }
2102 
2103 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2104 					u64 preemption_timeout)
2105 {
2106 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2107 
2108 	/*
2109 	 * A timer value of zero is architecturally guaranteed to cause
2110 	 * a VMExit prior to executing any instructions in the guest.
2111 	 */
2112 	if (preemption_timeout == 0) {
2113 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2114 		return;
2115 	}
2116 
2117 	if (vcpu->arch.virtual_tsc_khz == 0)
2118 		return;
2119 
2120 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2121 	preemption_timeout *= 1000000;
2122 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2123 	hrtimer_start(&vmx->nested.preemption_timer,
2124 		      ktime_add_ns(ktime_get(), preemption_timeout),
2125 		      HRTIMER_MODE_ABS_PINNED);
2126 }
2127 
2128 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2129 {
2130 	if (vmx->nested.nested_run_pending &&
2131 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2132 		return vmcs12->guest_ia32_efer;
2133 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2134 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2135 	else
2136 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2137 }
2138 
2139 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2140 {
2141 	struct kvm *kvm = vmx->vcpu.kvm;
2142 
2143 	/*
2144 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2145 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2146 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2147 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2148 	 */
2149 	if (vmx->nested.vmcs02_initialized)
2150 		return;
2151 	vmx->nested.vmcs02_initialized = true;
2152 
2153 	/*
2154 	 * We don't care what the EPTP value is we just need to guarantee
2155 	 * it's valid so we don't get a false positive when doing early
2156 	 * consistency checks.
2157 	 */
2158 	if (enable_ept && nested_early_check)
2159 		vmcs_write64(EPT_POINTER,
2160 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2161 
2162 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2163 	if (cpu_has_vmx_vmfunc())
2164 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2165 
2166 	if (cpu_has_vmx_posted_intr())
2167 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2168 
2169 	if (cpu_has_vmx_msr_bitmap())
2170 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2171 
2172 	/*
2173 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2174 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2175 	 * deal with filtering out L2 GPAs from the buffer.
2176 	 */
2177 	if (enable_pml) {
2178 		vmcs_write64(PML_ADDRESS, 0);
2179 		vmcs_write16(GUEST_PML_INDEX, -1);
2180 	}
2181 
2182 	if (cpu_has_vmx_encls_vmexit())
2183 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2184 
2185 	if (kvm_notify_vmexit_enabled(kvm))
2186 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
2187 
2188 	/*
2189 	 * Set the MSR load/store lists to match L0's settings.  Only the
2190 	 * addresses are constant (for vmcs02), the counts can change based
2191 	 * on L2's behavior, e.g. switching to/from long mode.
2192 	 */
2193 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2194 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2195 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2196 
2197 	vmx_set_constant_host_state(vmx);
2198 }
2199 
2200 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2201 				      struct vmcs12 *vmcs12)
2202 {
2203 	prepare_vmcs02_constant_state(vmx);
2204 
2205 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2206 
2207 	if (enable_vpid) {
2208 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2209 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2210 		else
2211 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2212 	}
2213 }
2214 
2215 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2216 				 struct vmcs12 *vmcs12)
2217 {
2218 	u32 exec_control;
2219 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2220 
2221 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2222 		prepare_vmcs02_early_rare(vmx, vmcs12);
2223 
2224 	/*
2225 	 * PIN CONTROLS
2226 	 */
2227 	exec_control = __pin_controls_get(vmcs01);
2228 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2229 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2230 
2231 	/* Posted interrupts setting is only taken from vmcs12.  */
2232 	vmx->nested.pi_pending = false;
2233 	if (nested_cpu_has_posted_intr(vmcs12))
2234 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2235 	else
2236 		exec_control &= ~PIN_BASED_POSTED_INTR;
2237 	pin_controls_set(vmx, exec_control);
2238 
2239 	/*
2240 	 * EXEC CONTROLS
2241 	 */
2242 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2243 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2244 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2245 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2246 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2247 
2248 	vmx->nested.l1_tpr_threshold = -1;
2249 	if (exec_control & CPU_BASED_TPR_SHADOW)
2250 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2251 #ifdef CONFIG_X86_64
2252 	else
2253 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2254 				CPU_BASED_CR8_STORE_EXITING;
2255 #endif
2256 
2257 	/*
2258 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2259 	 * for I/O port accesses.
2260 	 */
2261 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2262 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2263 
2264 	/*
2265 	 * This bit will be computed in nested_get_vmcs12_pages, because
2266 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2267 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2268 	 * only set/clear this bit.
2269 	 */
2270 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2271 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2272 
2273 	exec_controls_set(vmx, exec_control);
2274 
2275 	/*
2276 	 * SECONDARY EXEC CONTROLS
2277 	 */
2278 	if (cpu_has_secondary_exec_ctrls()) {
2279 		exec_control = __secondary_exec_controls_get(vmcs01);
2280 
2281 		/* Take the following fields only from vmcs12 */
2282 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2283 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2284 				  SECONDARY_EXEC_ENABLE_INVPCID |
2285 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2286 				  SECONDARY_EXEC_XSAVES |
2287 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2288 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2289 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2290 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2291 				  SECONDARY_EXEC_DESC);
2292 
2293 		if (nested_cpu_has(vmcs12,
2294 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2295 			exec_control |= vmcs12->secondary_vm_exec_control;
2296 
2297 		/* PML is emulated and never enabled in hardware for L2. */
2298 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2299 
2300 		/* VMCS shadowing for L2 is emulated for now */
2301 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2302 
2303 		/*
2304 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2305 		 * will not have to rewrite the controls just for this bit.
2306 		 */
2307 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2308 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2309 			exec_control |= SECONDARY_EXEC_DESC;
2310 
2311 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2312 			vmcs_write16(GUEST_INTR_STATUS,
2313 				vmcs12->guest_intr_status);
2314 
2315 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2316 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2317 
2318 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2319 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2320 
2321 		secondary_exec_controls_set(vmx, exec_control);
2322 	}
2323 
2324 	/*
2325 	 * ENTRY CONTROLS
2326 	 *
2327 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2328 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2329 	 * on the related bits (if supported by the CPU) in the hope that
2330 	 * we can avoid VMWrites during vmx_set_efer().
2331 	 */
2332 	exec_control = __vm_entry_controls_get(vmcs01);
2333 	exec_control |= vmcs12->vm_entry_controls;
2334 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2335 	if (cpu_has_load_ia32_efer()) {
2336 		if (guest_efer & EFER_LMA)
2337 			exec_control |= VM_ENTRY_IA32E_MODE;
2338 		if (guest_efer != host_efer)
2339 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2340 	}
2341 	vm_entry_controls_set(vmx, exec_control);
2342 
2343 	/*
2344 	 * EXIT CONTROLS
2345 	 *
2346 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2347 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2348 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2349 	 */
2350 	exec_control = __vm_exit_controls_get(vmcs01);
2351 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2352 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2353 	else
2354 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2355 	vm_exit_controls_set(vmx, exec_control);
2356 
2357 	/*
2358 	 * Interrupt/Exception Fields
2359 	 */
2360 	if (vmx->nested.nested_run_pending) {
2361 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2362 			     vmcs12->vm_entry_intr_info_field);
2363 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2364 			     vmcs12->vm_entry_exception_error_code);
2365 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2366 			     vmcs12->vm_entry_instruction_len);
2367 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2368 			     vmcs12->guest_interruptibility_info);
2369 		vmx->loaded_vmcs->nmi_known_unmasked =
2370 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2371 	} else {
2372 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2373 	}
2374 }
2375 
2376 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2377 {
2378 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2379 
2380 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2381 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2382 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2383 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2384 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2385 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2386 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2387 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2388 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2389 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2390 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2391 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2392 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2393 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2394 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2395 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2396 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2397 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2398 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2399 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2400 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2401 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2402 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2403 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2404 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2405 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2406 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2407 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2408 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2409 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2410 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2411 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2412 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2413 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2414 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2415 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2416 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2417 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2418 
2419 		vmx->segment_cache.bitmask = 0;
2420 	}
2421 
2422 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2423 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2424 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2425 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2426 			    vmcs12->guest_pending_dbg_exceptions);
2427 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2428 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2429 
2430 		/*
2431 		 * L1 may access the L2's PDPTR, so save them to construct
2432 		 * vmcs12
2433 		 */
2434 		if (enable_ept) {
2435 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2436 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2437 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2438 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2439 		}
2440 
2441 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2442 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2443 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2444 	}
2445 
2446 	if (nested_cpu_has_xsaves(vmcs12))
2447 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2448 
2449 	/*
2450 	 * Whether page-faults are trapped is determined by a combination of
2451 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2452 	 * doesn't care about page faults then we should set all of these to
2453 	 * L1's desires. However, if L0 does care about (some) page faults, it
2454 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2455 	 * simply ask to exit on each and every L2 page fault. This is done by
2456 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2457 	 * Note that below we don't need special code to set EB.PF beyond the
2458 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2459 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2460 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2461 	 */
2462 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2463 		/*
2464 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2465 		 * go ahead and use it?
2466 		 */
2467 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2468 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2469 	} else {
2470 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2471 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2472 	}
2473 
2474 	if (cpu_has_vmx_apicv()) {
2475 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2476 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2477 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2478 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2479 	}
2480 
2481 	/*
2482 	 * Make sure the msr_autostore list is up to date before we set the
2483 	 * count in the vmcs02.
2484 	 */
2485 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2486 
2487 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2488 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2489 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2490 
2491 	set_cr4_guest_host_mask(vmx);
2492 }
2493 
2494 /*
2495  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2496  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2497  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2498  * guest in a way that will both be appropriate to L1's requests, and our
2499  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2500  * function also has additional necessary side-effects, like setting various
2501  * vcpu->arch fields.
2502  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2503  * is assigned to entry_failure_code on failure.
2504  */
2505 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2506 			  bool from_vmentry,
2507 			  enum vm_entry_failure_code *entry_failure_code)
2508 {
2509 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2510 	bool load_guest_pdptrs_vmcs12 = false;
2511 
2512 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2513 		prepare_vmcs02_rare(vmx, vmcs12);
2514 		vmx->nested.dirty_vmcs12 = false;
2515 
2516 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2517 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2518 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2519 	}
2520 
2521 	if (vmx->nested.nested_run_pending &&
2522 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2523 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2524 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2525 	} else {
2526 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2527 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.pre_vmenter_debugctl);
2528 	}
2529 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2530 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2531 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.pre_vmenter_bndcfgs);
2532 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2533 
2534 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2535 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2536 	 * trap. Note that CR0.TS also needs updating - we do this later.
2537 	 */
2538 	vmx_update_exception_bitmap(vcpu);
2539 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2540 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2541 
2542 	if (vmx->nested.nested_run_pending &&
2543 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2544 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2545 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2546 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2547 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2548 	}
2549 
2550 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2551 			vcpu->arch.l1_tsc_offset,
2552 			vmx_get_l2_tsc_offset(vcpu),
2553 			vmx_get_l2_tsc_multiplier(vcpu));
2554 
2555 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2556 			vcpu->arch.l1_tsc_scaling_ratio,
2557 			vmx_get_l2_tsc_multiplier(vcpu));
2558 
2559 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2560 	if (kvm_caps.has_tsc_control)
2561 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2562 
2563 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2564 
2565 	if (nested_cpu_has_ept(vmcs12))
2566 		nested_ept_init_mmu_context(vcpu);
2567 
2568 	/*
2569 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2570 	 * bits which we consider mandatory enabled.
2571 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2572 	 * the specifications by L1; It's not enough to take
2573 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2574 	 * have more bits than L1 expected.
2575 	 */
2576 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2577 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2578 
2579 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2580 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2581 
2582 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2583 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2584 	vmx_set_efer(vcpu, vcpu->arch.efer);
2585 
2586 	/*
2587 	 * Guest state is invalid and unrestricted guest is disabled,
2588 	 * which means L1 attempted VMEntry to L2 with invalid state.
2589 	 * Fail the VMEntry.
2590 	 *
2591 	 * However when force loading the guest state (SMM exit or
2592 	 * loading nested state after migration, it is possible to
2593 	 * have invalid guest state now, which will be later fixed by
2594 	 * restoring L2 register state
2595 	 */
2596 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2597 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2598 		return -EINVAL;
2599 	}
2600 
2601 	/* Shadow page tables on either EPT or shadow page tables. */
2602 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2603 				from_vmentry, entry_failure_code))
2604 		return -EINVAL;
2605 
2606 	/*
2607 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2608 	 * on nested VM-Exit, which can occur without actually running L2 and
2609 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2610 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2611 	 * transition to HLT instead of running L2.
2612 	 */
2613 	if (enable_ept)
2614 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2615 
2616 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2617 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2618 	    is_pae_paging(vcpu)) {
2619 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2620 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2621 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2622 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2623 	}
2624 
2625 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2626 	    intel_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)) &&
2627 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2628 				     vmcs12->guest_ia32_perf_global_ctrl))) {
2629 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2630 		return -EINVAL;
2631 	}
2632 
2633 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2634 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2635 
2636 	/*
2637 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2638 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2639 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2640 	 * here.
2641 	 */
2642 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2643 		vmx->nested.hv_evmcs->hv_clean_fields |=
2644 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2645 
2646 	return 0;
2647 }
2648 
2649 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2650 {
2651 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2652 	       nested_cpu_has_virtual_nmis(vmcs12)))
2653 		return -EINVAL;
2654 
2655 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2656 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2657 		return -EINVAL;
2658 
2659 	return 0;
2660 }
2661 
2662 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2663 {
2664 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2665 
2666 	/* Check for memory type validity */
2667 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2668 	case VMX_EPTP_MT_UC:
2669 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2670 			return false;
2671 		break;
2672 	case VMX_EPTP_MT_WB:
2673 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2674 			return false;
2675 		break;
2676 	default:
2677 		return false;
2678 	}
2679 
2680 	/* Page-walk levels validity. */
2681 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2682 	case VMX_EPTP_PWL_5:
2683 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2684 			return false;
2685 		break;
2686 	case VMX_EPTP_PWL_4:
2687 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2688 			return false;
2689 		break;
2690 	default:
2691 		return false;
2692 	}
2693 
2694 	/* Reserved bits should not be set */
2695 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2696 		return false;
2697 
2698 	/* AD, if set, should be supported */
2699 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2700 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2701 			return false;
2702 	}
2703 
2704 	return true;
2705 }
2706 
2707 /*
2708  * Checks related to VM-Execution Control Fields
2709  */
2710 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2711                                               struct vmcs12 *vmcs12)
2712 {
2713 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2714 
2715 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2716 				   vmx->nested.msrs.pinbased_ctls_low,
2717 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2718 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2719 				   vmx->nested.msrs.procbased_ctls_low,
2720 				   vmx->nested.msrs.procbased_ctls_high)))
2721 		return -EINVAL;
2722 
2723 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2724 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2725 				   vmx->nested.msrs.secondary_ctls_low,
2726 				   vmx->nested.msrs.secondary_ctls_high)))
2727 		return -EINVAL;
2728 
2729 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2730 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2731 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2732 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2733 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2734 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2735 	    nested_vmx_check_nmi_controls(vmcs12) ||
2736 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2737 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2738 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2739 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2740 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2741 		return -EINVAL;
2742 
2743 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2744 	    nested_cpu_has_save_preemption_timer(vmcs12))
2745 		return -EINVAL;
2746 
2747 	if (nested_cpu_has_ept(vmcs12) &&
2748 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2749 		return -EINVAL;
2750 
2751 	if (nested_cpu_has_vmfunc(vmcs12)) {
2752 		if (CC(vmcs12->vm_function_control &
2753 		       ~vmx->nested.msrs.vmfunc_controls))
2754 			return -EINVAL;
2755 
2756 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2757 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2758 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2759 				return -EINVAL;
2760 		}
2761 	}
2762 
2763 	return 0;
2764 }
2765 
2766 /*
2767  * Checks related to VM-Exit Control Fields
2768  */
2769 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2770                                          struct vmcs12 *vmcs12)
2771 {
2772 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2773 
2774 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2775 				    vmx->nested.msrs.exit_ctls_low,
2776 				    vmx->nested.msrs.exit_ctls_high)) ||
2777 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2778 		return -EINVAL;
2779 
2780 	return 0;
2781 }
2782 
2783 /*
2784  * Checks related to VM-Entry Control Fields
2785  */
2786 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2787 					  struct vmcs12 *vmcs12)
2788 {
2789 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2790 
2791 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2792 				    vmx->nested.msrs.entry_ctls_low,
2793 				    vmx->nested.msrs.entry_ctls_high)))
2794 		return -EINVAL;
2795 
2796 	/*
2797 	 * From the Intel SDM, volume 3:
2798 	 * Fields relevant to VM-entry event injection must be set properly.
2799 	 * These fields are the VM-entry interruption-information field, the
2800 	 * VM-entry exception error code, and the VM-entry instruction length.
2801 	 */
2802 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2803 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2804 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2805 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2806 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2807 		bool should_have_error_code;
2808 		bool urg = nested_cpu_has2(vmcs12,
2809 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2810 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2811 
2812 		/* VM-entry interruption-info field: interruption type */
2813 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2814 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2815 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2816 			return -EINVAL;
2817 
2818 		/* VM-entry interruption-info field: vector */
2819 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2820 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2821 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2822 			return -EINVAL;
2823 
2824 		/* VM-entry interruption-info field: deliver error code */
2825 		should_have_error_code =
2826 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2827 			x86_exception_has_error_code(vector);
2828 		if (CC(has_error_code != should_have_error_code))
2829 			return -EINVAL;
2830 
2831 		/* VM-entry exception error code */
2832 		if (CC(has_error_code &&
2833 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2834 			return -EINVAL;
2835 
2836 		/* VM-entry interruption-info field: reserved bits */
2837 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2838 			return -EINVAL;
2839 
2840 		/* VM-entry instruction length */
2841 		switch (intr_type) {
2842 		case INTR_TYPE_SOFT_EXCEPTION:
2843 		case INTR_TYPE_SOFT_INTR:
2844 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2845 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2846 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2847 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2848 				return -EINVAL;
2849 		}
2850 	}
2851 
2852 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2853 		return -EINVAL;
2854 
2855 	return 0;
2856 }
2857 
2858 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2859 				     struct vmcs12 *vmcs12)
2860 {
2861 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2862 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2863 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2864 		return -EINVAL;
2865 
2866 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2867 		return nested_evmcs_check_controls(vmcs12);
2868 
2869 	return 0;
2870 }
2871 
2872 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
2873 				       struct vmcs12 *vmcs12)
2874 {
2875 #ifdef CONFIG_X86_64
2876 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
2877 		!!(vcpu->arch.efer & EFER_LMA)))
2878 		return -EINVAL;
2879 #endif
2880 	return 0;
2881 }
2882 
2883 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2884 				       struct vmcs12 *vmcs12)
2885 {
2886 	bool ia32e;
2887 
2888 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2889 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2890 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2891 		return -EINVAL;
2892 
2893 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2894 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2895 		return -EINVAL;
2896 
2897 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2898 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2899 		return -EINVAL;
2900 
2901 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2902 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2903 					   vmcs12->host_ia32_perf_global_ctrl)))
2904 		return -EINVAL;
2905 
2906 #ifdef CONFIG_X86_64
2907 	ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
2908 #else
2909 	ia32e = false;
2910 #endif
2911 
2912 	if (ia32e) {
2913 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2914 			return -EINVAL;
2915 	} else {
2916 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2917 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2918 		    CC((vmcs12->host_rip) >> 32))
2919 			return -EINVAL;
2920 	}
2921 
2922 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2923 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2924 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2925 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2926 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2927 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2928 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2929 	    CC(vmcs12->host_cs_selector == 0) ||
2930 	    CC(vmcs12->host_tr_selector == 0) ||
2931 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2932 		return -EINVAL;
2933 
2934 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2935 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2936 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2937 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2938 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2939 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2940 		return -EINVAL;
2941 
2942 	/*
2943 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2944 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2945 	 * the values of the LMA and LME bits in the field must each be that of
2946 	 * the host address-space size VM-exit control.
2947 	 */
2948 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2949 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2950 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2951 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2952 			return -EINVAL;
2953 	}
2954 
2955 	return 0;
2956 }
2957 
2958 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2959 					  struct vmcs12 *vmcs12)
2960 {
2961 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2962 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
2963 	struct vmcs_hdr hdr;
2964 
2965 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
2966 		return 0;
2967 
2968 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2969 		return -EINVAL;
2970 
2971 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
2972 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
2973 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
2974                 return -EINVAL;
2975 
2976 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
2977 					    offsetof(struct vmcs12, hdr),
2978 					    sizeof(hdr))))
2979 		return -EINVAL;
2980 
2981 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
2982 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2983 		return -EINVAL;
2984 
2985 	return 0;
2986 }
2987 
2988 /*
2989  * Checks related to Guest Non-register State
2990  */
2991 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2992 {
2993 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2994 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2995 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2996 		return -EINVAL;
2997 
2998 	return 0;
2999 }
3000 
3001 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
3002 					struct vmcs12 *vmcs12,
3003 					enum vm_entry_failure_code *entry_failure_code)
3004 {
3005 	bool ia32e;
3006 
3007 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
3008 
3009 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3010 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3011 		return -EINVAL;
3012 
3013 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
3014 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
3015 		return -EINVAL;
3016 
3017 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3018 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3019 		return -EINVAL;
3020 
3021 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3022 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3023 		return -EINVAL;
3024 	}
3025 
3026 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3027 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3028 					   vmcs12->guest_ia32_perf_global_ctrl)))
3029 		return -EINVAL;
3030 
3031 	/*
3032 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3033 	 * are performed on the field for the IA32_EFER MSR:
3034 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3035 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3036 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3037 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3038 	 *   CR0.PG) is 1.
3039 	 */
3040 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3041 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3042 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
3043 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3044 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3045 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3046 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3047 			return -EINVAL;
3048 	}
3049 
3050 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3051 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3052 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3053 		return -EINVAL;
3054 
3055 	if (nested_check_guest_non_reg_state(vmcs12))
3056 		return -EINVAL;
3057 
3058 	return 0;
3059 }
3060 
3061 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3062 {
3063 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3064 	unsigned long cr3, cr4;
3065 	bool vm_fail;
3066 
3067 	if (!nested_early_check)
3068 		return 0;
3069 
3070 	if (vmx->msr_autoload.host.nr)
3071 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3072 	if (vmx->msr_autoload.guest.nr)
3073 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3074 
3075 	preempt_disable();
3076 
3077 	vmx_prepare_switch_to_guest(vcpu);
3078 
3079 	/*
3080 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3081 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3082 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3083 	 * there is no need to preserve other bits or save/restore the field.
3084 	 */
3085 	vmcs_writel(GUEST_RFLAGS, 0);
3086 
3087 	cr3 = __get_current_cr3_fast();
3088 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3089 		vmcs_writel(HOST_CR3, cr3);
3090 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3091 	}
3092 
3093 	cr4 = cr4_read_shadow();
3094 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3095 		vmcs_writel(HOST_CR4, cr4);
3096 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3097 	}
3098 
3099 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3100 				 __vmx_vcpu_run_flags(vmx));
3101 
3102 	if (vmx->msr_autoload.host.nr)
3103 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3104 	if (vmx->msr_autoload.guest.nr)
3105 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3106 
3107 	if (vm_fail) {
3108 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3109 
3110 		preempt_enable();
3111 
3112 		trace_kvm_nested_vmenter_failed(
3113 			"early hardware check VM-instruction error: ", error);
3114 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3115 		return 1;
3116 	}
3117 
3118 	/*
3119 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3120 	 */
3121 	if (hw_breakpoint_active())
3122 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3123 	local_irq_enable();
3124 	preempt_enable();
3125 
3126 	/*
3127 	 * A non-failing VMEntry means we somehow entered guest mode with
3128 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3129 	 * is no telling what memory has been modified or what state has
3130 	 * been exposed to unknown code.  Hitting this all but guarantees
3131 	 * a (very critical) hardware issue.
3132 	 */
3133 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3134 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3135 
3136 	return 0;
3137 }
3138 
3139 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3140 {
3141 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3142 
3143 	/*
3144 	 * hv_evmcs may end up being not mapped after migration (when
3145 	 * L2 was running), map it here to make sure vmcs12 changes are
3146 	 * properly reflected.
3147 	 */
3148 	if (vmx->nested.enlightened_vmcs_enabled &&
3149 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3150 		enum nested_evmptrld_status evmptrld_status =
3151 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3152 
3153 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3154 		    evmptrld_status == EVMPTRLD_ERROR)
3155 			return false;
3156 
3157 		/*
3158 		 * Post migration VMCS12 always provides the most actual
3159 		 * information, copy it to eVMCS upon entry.
3160 		 */
3161 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3162 	}
3163 
3164 	return true;
3165 }
3166 
3167 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3168 {
3169 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3170 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3171 	struct kvm_host_map *map;
3172 
3173 	if (!vcpu->arch.pdptrs_from_userspace &&
3174 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3175 		/*
3176 		 * Reload the guest's PDPTRs since after a migration
3177 		 * the guest CR3 might be restored prior to setting the nested
3178 		 * state which can lead to a load of wrong PDPTRs.
3179 		 */
3180 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3181 			return false;
3182 	}
3183 
3184 
3185 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3186 		map = &vmx->nested.apic_access_page_map;
3187 
3188 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->apic_access_addr), map)) {
3189 			vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(map->pfn));
3190 		} else {
3191 			pr_debug_ratelimited("%s: no backing for APIC-access address in vmcs12\n",
3192 					     __func__);
3193 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3194 			vcpu->run->internal.suberror =
3195 				KVM_INTERNAL_ERROR_EMULATION;
3196 			vcpu->run->internal.ndata = 0;
3197 			return false;
3198 		}
3199 	}
3200 
3201 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3202 		map = &vmx->nested.virtual_apic_map;
3203 
3204 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3205 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3206 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3207 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3208 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3209 			/*
3210 			 * The processor will never use the TPR shadow, simply
3211 			 * clear the bit from the execution control.  Such a
3212 			 * configuration is useless, but it happens in tests.
3213 			 * For any other configuration, failing the vm entry is
3214 			 * _not_ what the processor does but it's basically the
3215 			 * only possibility we have.
3216 			 */
3217 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3218 		} else {
3219 			/*
3220 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3221 			 * force VM-Entry to fail.
3222 			 */
3223 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3224 		}
3225 	}
3226 
3227 	if (nested_cpu_has_posted_intr(vmcs12)) {
3228 		map = &vmx->nested.pi_desc_map;
3229 
3230 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3231 			vmx->nested.pi_desc =
3232 				(struct pi_desc *)(((void *)map->hva) +
3233 				offset_in_page(vmcs12->posted_intr_desc_addr));
3234 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3235 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3236 		} else {
3237 			/*
3238 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3239 			 * access the contents of the VMCS12 posted interrupt
3240 			 * descriptor. (Note that KVM may do this when it
3241 			 * should not, per the architectural specification.)
3242 			 */
3243 			vmx->nested.pi_desc = NULL;
3244 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3245 		}
3246 	}
3247 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3248 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3249 	else
3250 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3251 
3252 	return true;
3253 }
3254 
3255 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3256 {
3257 	if (!nested_get_evmcs_page(vcpu)) {
3258 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3259 				     __func__);
3260 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3261 		vcpu->run->internal.suberror =
3262 			KVM_INTERNAL_ERROR_EMULATION;
3263 		vcpu->run->internal.ndata = 0;
3264 
3265 		return false;
3266 	}
3267 
3268 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3269 		return false;
3270 
3271 	return true;
3272 }
3273 
3274 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3275 {
3276 	struct vmcs12 *vmcs12;
3277 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3278 	gpa_t dst;
3279 
3280 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3281 		return 0;
3282 
3283 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3284 		return 1;
3285 
3286 	/*
3287 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3288 	 * set is already checked as part of A/D emulation.
3289 	 */
3290 	vmcs12 = get_vmcs12(vcpu);
3291 	if (!nested_cpu_has_pml(vmcs12))
3292 		return 0;
3293 
3294 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3295 		vmx->nested.pml_full = true;
3296 		return 1;
3297 	}
3298 
3299 	gpa &= ~0xFFFull;
3300 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3301 
3302 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3303 				 offset_in_page(dst), sizeof(gpa)))
3304 		return 0;
3305 
3306 	vmcs12->guest_pml_index--;
3307 
3308 	return 0;
3309 }
3310 
3311 /*
3312  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3313  * for running VMX instructions (except VMXON, whose prerequisites are
3314  * slightly different). It also specifies what exception to inject otherwise.
3315  * Note that many of these exceptions have priority over VM exits, so they
3316  * don't have to be checked again here.
3317  */
3318 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3319 {
3320 	if (!to_vmx(vcpu)->nested.vmxon) {
3321 		kvm_queue_exception(vcpu, UD_VECTOR);
3322 		return 0;
3323 	}
3324 
3325 	if (vmx_get_cpl(vcpu)) {
3326 		kvm_inject_gp(vcpu, 0);
3327 		return 0;
3328 	}
3329 
3330 	return 1;
3331 }
3332 
3333 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3334 {
3335 	u8 rvi = vmx_get_rvi();
3336 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3337 
3338 	return ((rvi & 0xf0) > (vppr & 0xf0));
3339 }
3340 
3341 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3342 				   struct vmcs12 *vmcs12);
3343 
3344 /*
3345  * If from_vmentry is false, this is being called from state restore (either RSM
3346  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3347  *
3348  * Returns:
3349  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3350  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3351  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3352  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3353  */
3354 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3355 							bool from_vmentry)
3356 {
3357 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3358 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3359 	enum vm_entry_failure_code entry_failure_code;
3360 	bool evaluate_pending_interrupts;
3361 	union vmx_exit_reason exit_reason = {
3362 		.basic = EXIT_REASON_INVALID_STATE,
3363 		.failed_vmentry = 1,
3364 	};
3365 	u32 failed_index;
3366 
3367 	kvm_service_local_tlb_flush_requests(vcpu);
3368 
3369 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3370 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3371 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3372 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3373 
3374 	if (!vmx->nested.nested_run_pending ||
3375 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3376 		vmx->nested.pre_vmenter_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3377 	if (kvm_mpx_supported() &&
3378 	    (!vmx->nested.nested_run_pending ||
3379 	     !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
3380 		vmx->nested.pre_vmenter_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3381 
3382 	/*
3383 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3384 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3385 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3386 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3387 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3388 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3389 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3390 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3391 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3392 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3393 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3394 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3395 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3396 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3397 	 */
3398 	if (!enable_ept && !nested_early_check)
3399 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3400 
3401 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3402 
3403 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3404 
3405 	if (from_vmentry) {
3406 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3407 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3408 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3409 		}
3410 
3411 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3412 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3413 			return NVMX_VMENTRY_VMFAIL;
3414 		}
3415 
3416 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3417 						 &entry_failure_code)) {
3418 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3419 			vmcs12->exit_qualification = entry_failure_code;
3420 			goto vmentry_fail_vmexit;
3421 		}
3422 	}
3423 
3424 	enter_guest_mode(vcpu);
3425 
3426 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3427 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3428 		vmcs12->exit_qualification = entry_failure_code;
3429 		goto vmentry_fail_vmexit_guest_mode;
3430 	}
3431 
3432 	if (from_vmentry) {
3433 		failed_index = nested_vmx_load_msr(vcpu,
3434 						   vmcs12->vm_entry_msr_load_addr,
3435 						   vmcs12->vm_entry_msr_load_count);
3436 		if (failed_index) {
3437 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3438 			vmcs12->exit_qualification = failed_index;
3439 			goto vmentry_fail_vmexit_guest_mode;
3440 		}
3441 	} else {
3442 		/*
3443 		 * The MMU is not initialized to point at the right entities yet and
3444 		 * "get pages" would need to read data from the guest (i.e. we will
3445 		 * need to perform gpa to hpa translation). Request a call
3446 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3447 		 * have already been set at vmentry time and should not be reset.
3448 		 */
3449 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3450 	}
3451 
3452 	/*
3453 	 * If L1 had a pending IRQ/NMI until it executed
3454 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3455 	 * disallowed (e.g. interrupts disabled), L0 needs to
3456 	 * evaluate if this pending event should cause an exit from L2
3457 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3458 	 * intercept EXTERNAL_INTERRUPT).
3459 	 *
3460 	 * Usually this would be handled by the processor noticing an
3461 	 * IRQ/NMI window request, or checking RVI during evaluation of
3462 	 * pending virtual interrupts.  However, this setting was done
3463 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3464 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3465 	 */
3466 	if (unlikely(evaluate_pending_interrupts))
3467 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3468 
3469 	/*
3470 	 * Do not start the preemption timer hrtimer until after we know
3471 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3472 	 * the timer.
3473 	 */
3474 	vmx->nested.preemption_timer_expired = false;
3475 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3476 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3477 		vmx_start_preemption_timer(vcpu, timer_value);
3478 	}
3479 
3480 	/*
3481 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3482 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3483 	 * returned as far as L1 is concerned. It will only return (and set
3484 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3485 	 */
3486 	return NVMX_VMENTRY_SUCCESS;
3487 
3488 	/*
3489 	 * A failed consistency check that leads to a VMExit during L1's
3490 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3491 	 * 26.7 "VM-entry failures during or after loading guest state".
3492 	 */
3493 vmentry_fail_vmexit_guest_mode:
3494 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3495 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3496 	leave_guest_mode(vcpu);
3497 
3498 vmentry_fail_vmexit:
3499 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3500 
3501 	if (!from_vmentry)
3502 		return NVMX_VMENTRY_VMEXIT;
3503 
3504 	load_vmcs12_host_state(vcpu, vmcs12);
3505 	vmcs12->vm_exit_reason = exit_reason.full;
3506 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3507 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3508 	return NVMX_VMENTRY_VMEXIT;
3509 }
3510 
3511 /*
3512  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3513  * for running an L2 nested guest.
3514  */
3515 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3516 {
3517 	struct vmcs12 *vmcs12;
3518 	enum nvmx_vmentry_status status;
3519 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3520 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3521 	enum nested_evmptrld_status evmptrld_status;
3522 
3523 	if (!nested_vmx_check_permission(vcpu))
3524 		return 1;
3525 
3526 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3527 	if (evmptrld_status == EVMPTRLD_ERROR) {
3528 		kvm_queue_exception(vcpu, UD_VECTOR);
3529 		return 1;
3530 	}
3531 
3532 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
3533 
3534 	if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3535 		return nested_vmx_failInvalid(vcpu);
3536 
3537 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3538 	       vmx->nested.current_vmptr == INVALID_GPA))
3539 		return nested_vmx_failInvalid(vcpu);
3540 
3541 	vmcs12 = get_vmcs12(vcpu);
3542 
3543 	/*
3544 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3545 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3546 	 * rather than RFLAGS.ZF, and no error number is stored to the
3547 	 * VM-instruction error field.
3548 	 */
3549 	if (CC(vmcs12->hdr.shadow_vmcs))
3550 		return nested_vmx_failInvalid(vcpu);
3551 
3552 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3553 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3554 		/* Enlightened VMCS doesn't have launch state */
3555 		vmcs12->launch_state = !launch;
3556 	} else if (enable_shadow_vmcs) {
3557 		copy_shadow_to_vmcs12(vmx);
3558 	}
3559 
3560 	/*
3561 	 * The nested entry process starts with enforcing various prerequisites
3562 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3563 	 * they fail: As the SDM explains, some conditions should cause the
3564 	 * instruction to fail, while others will cause the instruction to seem
3565 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3566 	 * To speed up the normal (success) code path, we should avoid checking
3567 	 * for misconfigurations which will anyway be caught by the processor
3568 	 * when using the merged vmcs02.
3569 	 */
3570 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3571 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3572 
3573 	if (CC(vmcs12->launch_state == launch))
3574 		return nested_vmx_fail(vcpu,
3575 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3576 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3577 
3578 	if (nested_vmx_check_controls(vcpu, vmcs12))
3579 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3580 
3581 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3582 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3583 
3584 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3585 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3586 
3587 	/*
3588 	 * We're finally done with prerequisite checking, and can start with
3589 	 * the nested entry.
3590 	 */
3591 	vmx->nested.nested_run_pending = 1;
3592 	vmx->nested.has_preemption_timer_deadline = false;
3593 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3594 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3595 		goto vmentry_failed;
3596 
3597 	/* Emulate processing of posted interrupts on VM-Enter. */
3598 	if (nested_cpu_has_posted_intr(vmcs12) &&
3599 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3600 		vmx->nested.pi_pending = true;
3601 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3602 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3603 	}
3604 
3605 	/* Hide L1D cache contents from the nested guest.  */
3606 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3607 
3608 	/*
3609 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3610 	 * also be used as part of restoring nVMX state for
3611 	 * snapshot restore (migration).
3612 	 *
3613 	 * In this flow, it is assumed that vmcs12 cache was
3614 	 * transferred as part of captured nVMX state and should
3615 	 * therefore not be read from guest memory (which may not
3616 	 * exist on destination host yet).
3617 	 */
3618 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3619 
3620 	switch (vmcs12->guest_activity_state) {
3621 	case GUEST_ACTIVITY_HLT:
3622 		/*
3623 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3624 		 * awakened by event injection or by an NMI-window VM-exit or
3625 		 * by an interrupt-window VM-exit, halt the vcpu.
3626 		 */
3627 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3628 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3629 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3630 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3631 			vmx->nested.nested_run_pending = 0;
3632 			return kvm_emulate_halt_noskip(vcpu);
3633 		}
3634 		break;
3635 	case GUEST_ACTIVITY_WAIT_SIPI:
3636 		vmx->nested.nested_run_pending = 0;
3637 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3638 		break;
3639 	default:
3640 		break;
3641 	}
3642 
3643 	return 1;
3644 
3645 vmentry_failed:
3646 	vmx->nested.nested_run_pending = 0;
3647 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3648 		return 0;
3649 	if (status == NVMX_VMENTRY_VMEXIT)
3650 		return 1;
3651 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3652 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3653 }
3654 
3655 /*
3656  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3657  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3658  * This function returns the new value we should put in vmcs12.guest_cr0.
3659  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3660  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3661  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3662  *     didn't trap the bit, because if L1 did, so would L0).
3663  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3664  *     been modified by L2, and L1 knows it. So just leave the old value of
3665  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3666  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3667  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3668  *     changed these bits, and therefore they need to be updated, but L0
3669  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3670  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3671  */
3672 static inline unsigned long
3673 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3674 {
3675 	return
3676 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3677 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3678 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3679 			vcpu->arch.cr0_guest_owned_bits));
3680 }
3681 
3682 static inline unsigned long
3683 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3684 {
3685 	return
3686 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3687 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3688 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3689 			vcpu->arch.cr4_guest_owned_bits));
3690 }
3691 
3692 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3693 				      struct vmcs12 *vmcs12,
3694 				      u32 vm_exit_reason, u32 exit_intr_info)
3695 {
3696 	u32 idt_vectoring;
3697 	unsigned int nr;
3698 
3699 	/*
3700 	 * Per the SDM, VM-Exits due to double and triple faults are never
3701 	 * considered to occur during event delivery, even if the double/triple
3702 	 * fault is the result of an escalating vectoring issue.
3703 	 *
3704 	 * Note, the SDM qualifies the double fault behavior with "The original
3705 	 * event results in a double-fault exception".  It's unclear why the
3706 	 * qualification exists since exits due to double fault can occur only
3707 	 * while vectoring a different exception (injected events are never
3708 	 * subject to interception), i.e. there's _always_ an original event.
3709 	 *
3710 	 * The SDM also uses NMI as a confusing example for the "original event
3711 	 * causes the VM exit directly" clause.  NMI isn't special in any way,
3712 	 * the same rule applies to all events that cause an exit directly.
3713 	 * NMI is an odd choice for the example because NMIs can only occur on
3714 	 * instruction boundaries, i.e. they _can't_ occur during vectoring.
3715 	 */
3716 	if ((u16)vm_exit_reason == EXIT_REASON_TRIPLE_FAULT ||
3717 	    ((u16)vm_exit_reason == EXIT_REASON_EXCEPTION_NMI &&
3718 	     is_double_fault(exit_intr_info))) {
3719 		vmcs12->idt_vectoring_info_field = 0;
3720 	} else if (vcpu->arch.exception.injected) {
3721 		nr = vcpu->arch.exception.nr;
3722 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3723 
3724 		if (kvm_exception_is_soft(nr)) {
3725 			vmcs12->vm_exit_instruction_len =
3726 				vcpu->arch.event_exit_inst_len;
3727 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3728 		} else
3729 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3730 
3731 		if (vcpu->arch.exception.has_error_code) {
3732 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3733 			vmcs12->idt_vectoring_error_code =
3734 				vcpu->arch.exception.error_code;
3735 		}
3736 
3737 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3738 	} else if (vcpu->arch.nmi_injected) {
3739 		vmcs12->idt_vectoring_info_field =
3740 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3741 	} else if (vcpu->arch.interrupt.injected) {
3742 		nr = vcpu->arch.interrupt.nr;
3743 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3744 
3745 		if (vcpu->arch.interrupt.soft) {
3746 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3747 			vmcs12->vm_entry_instruction_len =
3748 				vcpu->arch.event_exit_inst_len;
3749 		} else
3750 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3751 
3752 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3753 	} else {
3754 		vmcs12->idt_vectoring_info_field = 0;
3755 	}
3756 }
3757 
3758 
3759 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3760 {
3761 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3762 	gfn_t gfn;
3763 
3764 	/*
3765 	 * Don't need to mark the APIC access page dirty; it is never
3766 	 * written to by the CPU during APIC virtualization.
3767 	 */
3768 
3769 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3770 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3771 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3772 	}
3773 
3774 	if (nested_cpu_has_posted_intr(vmcs12)) {
3775 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3776 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3777 	}
3778 }
3779 
3780 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3781 {
3782 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3783 	int max_irr;
3784 	void *vapic_page;
3785 	u16 status;
3786 
3787 	if (!vmx->nested.pi_pending)
3788 		return 0;
3789 
3790 	if (!vmx->nested.pi_desc)
3791 		goto mmio_needed;
3792 
3793 	vmx->nested.pi_pending = false;
3794 
3795 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3796 		return 0;
3797 
3798 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3799 	if (max_irr != 256) {
3800 		vapic_page = vmx->nested.virtual_apic_map.hva;
3801 		if (!vapic_page)
3802 			goto mmio_needed;
3803 
3804 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3805 			vapic_page, &max_irr);
3806 		status = vmcs_read16(GUEST_INTR_STATUS);
3807 		if ((u8)max_irr > ((u8)status & 0xff)) {
3808 			status &= ~0xff;
3809 			status |= (u8)max_irr;
3810 			vmcs_write16(GUEST_INTR_STATUS, status);
3811 		}
3812 	}
3813 
3814 	nested_mark_vmcs12_pages_dirty(vcpu);
3815 	return 0;
3816 
3817 mmio_needed:
3818 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3819 	return -ENXIO;
3820 }
3821 
3822 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3823 					       unsigned long exit_qual)
3824 {
3825 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3826 	unsigned int nr = vcpu->arch.exception.nr;
3827 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3828 
3829 	if (vcpu->arch.exception.has_error_code) {
3830 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3831 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3832 	}
3833 
3834 	if (kvm_exception_is_soft(nr))
3835 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3836 	else
3837 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3838 
3839 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3840 	    vmx_get_nmi_mask(vcpu))
3841 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3842 
3843 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3844 }
3845 
3846 /*
3847  * Returns true if a debug trap is pending delivery.
3848  *
3849  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3850  * exception may be inferred from the presence of an exception payload.
3851  */
3852 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3853 {
3854 	return vcpu->arch.exception.pending &&
3855 			vcpu->arch.exception.nr == DB_VECTOR &&
3856 			vcpu->arch.exception.payload;
3857 }
3858 
3859 /*
3860  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3861  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3862  * represents these debug traps with a payload that is said to be compatible
3863  * with the 'pending debug exceptions' field, write the payload to the VMCS
3864  * field if a VM-exit is delivered before the debug trap.
3865  */
3866 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3867 {
3868 	if (vmx_pending_dbg_trap(vcpu))
3869 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3870 			    vcpu->arch.exception.payload);
3871 }
3872 
3873 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3874 {
3875 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3876 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3877 }
3878 
3879 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3880 {
3881 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3882 	unsigned long exit_qual;
3883 	bool block_nested_events =
3884 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3885 	bool mtf_pending = vmx->nested.mtf_pending;
3886 	struct kvm_lapic *apic = vcpu->arch.apic;
3887 
3888 	/*
3889 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3890 	 * this state is discarded.
3891 	 */
3892 	if (!block_nested_events)
3893 		vmx->nested.mtf_pending = false;
3894 
3895 	if (lapic_in_kernel(vcpu) &&
3896 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3897 		if (block_nested_events)
3898 			return -EBUSY;
3899 		nested_vmx_update_pending_dbg(vcpu);
3900 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3901 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3902 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3903 		return 0;
3904 	}
3905 
3906 	if (lapic_in_kernel(vcpu) &&
3907 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3908 		if (block_nested_events)
3909 			return -EBUSY;
3910 
3911 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3912 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3913 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3914 						apic->sipi_vector & 0xFFUL);
3915 		return 0;
3916 	}
3917 
3918 	/*
3919 	 * Process any exceptions that are not debug traps before MTF.
3920 	 *
3921 	 * Note that only a pending nested run can block a pending exception.
3922 	 * Otherwise an injected NMI/interrupt should either be
3923 	 * lost or delivered to the nested hypervisor in the IDT_VECTORING_INFO,
3924 	 * while delivering the pending exception.
3925 	 */
3926 
3927 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3928 		if (vmx->nested.nested_run_pending)
3929 			return -EBUSY;
3930 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3931 			goto no_vmexit;
3932 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3933 		return 0;
3934 	}
3935 
3936 	if (mtf_pending) {
3937 		if (block_nested_events)
3938 			return -EBUSY;
3939 		nested_vmx_update_pending_dbg(vcpu);
3940 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3941 		return 0;
3942 	}
3943 
3944 	if (vcpu->arch.exception.pending) {
3945 		if (vmx->nested.nested_run_pending)
3946 			return -EBUSY;
3947 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3948 			goto no_vmexit;
3949 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3950 		return 0;
3951 	}
3952 
3953 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3954 		if (block_nested_events)
3955 			return -EBUSY;
3956 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3957 		return 0;
3958 	}
3959 
3960 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3961 		if (block_nested_events)
3962 			return -EBUSY;
3963 		goto no_vmexit;
3964 	}
3965 
3966 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3967 		if (block_nested_events)
3968 			return -EBUSY;
3969 		if (!nested_exit_on_nmi(vcpu))
3970 			goto no_vmexit;
3971 
3972 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3973 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3974 				  INTR_INFO_VALID_MASK, 0);
3975 		/*
3976 		 * The NMI-triggered VM exit counts as injection:
3977 		 * clear this one and block further NMIs.
3978 		 */
3979 		vcpu->arch.nmi_pending = 0;
3980 		vmx_set_nmi_mask(vcpu, true);
3981 		return 0;
3982 	}
3983 
3984 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3985 		if (block_nested_events)
3986 			return -EBUSY;
3987 		if (!nested_exit_on_intr(vcpu))
3988 			goto no_vmexit;
3989 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3990 		return 0;
3991 	}
3992 
3993 no_vmexit:
3994 	return vmx_complete_nested_posted_interrupt(vcpu);
3995 }
3996 
3997 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3998 {
3999 	ktime_t remaining =
4000 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
4001 	u64 value;
4002 
4003 	if (ktime_to_ns(remaining) <= 0)
4004 		return 0;
4005 
4006 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
4007 	do_div(value, 1000000);
4008 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
4009 }
4010 
4011 static bool is_vmcs12_ext_field(unsigned long field)
4012 {
4013 	switch (field) {
4014 	case GUEST_ES_SELECTOR:
4015 	case GUEST_CS_SELECTOR:
4016 	case GUEST_SS_SELECTOR:
4017 	case GUEST_DS_SELECTOR:
4018 	case GUEST_FS_SELECTOR:
4019 	case GUEST_GS_SELECTOR:
4020 	case GUEST_LDTR_SELECTOR:
4021 	case GUEST_TR_SELECTOR:
4022 	case GUEST_ES_LIMIT:
4023 	case GUEST_CS_LIMIT:
4024 	case GUEST_SS_LIMIT:
4025 	case GUEST_DS_LIMIT:
4026 	case GUEST_FS_LIMIT:
4027 	case GUEST_GS_LIMIT:
4028 	case GUEST_LDTR_LIMIT:
4029 	case GUEST_TR_LIMIT:
4030 	case GUEST_GDTR_LIMIT:
4031 	case GUEST_IDTR_LIMIT:
4032 	case GUEST_ES_AR_BYTES:
4033 	case GUEST_DS_AR_BYTES:
4034 	case GUEST_FS_AR_BYTES:
4035 	case GUEST_GS_AR_BYTES:
4036 	case GUEST_LDTR_AR_BYTES:
4037 	case GUEST_TR_AR_BYTES:
4038 	case GUEST_ES_BASE:
4039 	case GUEST_CS_BASE:
4040 	case GUEST_SS_BASE:
4041 	case GUEST_DS_BASE:
4042 	case GUEST_FS_BASE:
4043 	case GUEST_GS_BASE:
4044 	case GUEST_LDTR_BASE:
4045 	case GUEST_TR_BASE:
4046 	case GUEST_GDTR_BASE:
4047 	case GUEST_IDTR_BASE:
4048 	case GUEST_PENDING_DBG_EXCEPTIONS:
4049 	case GUEST_BNDCFGS:
4050 		return true;
4051 	default:
4052 		break;
4053 	}
4054 
4055 	return false;
4056 }
4057 
4058 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4059 				       struct vmcs12 *vmcs12)
4060 {
4061 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4062 
4063 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4064 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4065 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4066 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4067 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4068 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4069 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4070 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4071 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4072 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4073 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4074 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4075 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4076 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4077 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4078 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4079 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4080 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4081 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4082 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4083 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4084 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4085 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4086 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4087 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4088 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4089 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4090 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4091 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4092 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4093 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4094 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4095 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4096 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4097 	vmcs12->guest_pending_dbg_exceptions =
4098 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4099 
4100 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4101 }
4102 
4103 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4104 				       struct vmcs12 *vmcs12)
4105 {
4106 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4107 	int cpu;
4108 
4109 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4110 		return;
4111 
4112 
4113 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4114 
4115 	cpu = get_cpu();
4116 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4117 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4118 
4119 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4120 
4121 	vmx->loaded_vmcs = &vmx->vmcs01;
4122 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4123 	put_cpu();
4124 }
4125 
4126 /*
4127  * Update the guest state fields of vmcs12 to reflect changes that
4128  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4129  * VM-entry controls is also updated, since this is really a guest
4130  * state bit.)
4131  */
4132 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4133 {
4134 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4135 
4136 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4137 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4138 
4139 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4140 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4141 
4142 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4143 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4144 
4145 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4146 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4147 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4148 
4149 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4150 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4151 
4152 	vmcs12->guest_interruptibility_info =
4153 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4154 
4155 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4156 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4157 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4158 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4159 	else
4160 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4161 
4162 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4163 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4164 	    !vmx->nested.nested_run_pending)
4165 		vmcs12->vmx_preemption_timer_value =
4166 			vmx_get_preemption_timer_value(vcpu);
4167 
4168 	/*
4169 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4170 	 * own CR3 without exiting. If it has changed it, we must keep it.
4171 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4172 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4173 	 *
4174 	 * Additionally, restore L2's PDPTR to vmcs12.
4175 	 */
4176 	if (enable_ept) {
4177 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4178 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4179 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4180 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4181 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4182 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4183 		}
4184 	}
4185 
4186 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4187 
4188 	if (nested_cpu_has_vid(vmcs12))
4189 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4190 
4191 	vmcs12->vm_entry_controls =
4192 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4193 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4194 
4195 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4196 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4197 
4198 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4199 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4200 }
4201 
4202 /*
4203  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4204  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4205  * and this function updates it to reflect the changes to the guest state while
4206  * L2 was running (and perhaps made some exits which were handled directly by L0
4207  * without going back to L1), and to reflect the exit reason.
4208  * Note that we do not have to copy here all VMCS fields, just those that
4209  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4210  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4211  * which already writes to vmcs12 directly.
4212  */
4213 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4214 			   u32 vm_exit_reason, u32 exit_intr_info,
4215 			   unsigned long exit_qualification)
4216 {
4217 	/* update exit information fields: */
4218 	vmcs12->vm_exit_reason = vm_exit_reason;
4219 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4220 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4221 	vmcs12->exit_qualification = exit_qualification;
4222 
4223 	/*
4224 	 * On VM-Exit due to a failed VM-Entry, the VMCS isn't marked launched
4225 	 * and only EXIT_REASON and EXIT_QUALIFICATION are updated, all other
4226 	 * exit info fields are unmodified.
4227 	 */
4228 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4229 		vmcs12->launch_state = 1;
4230 
4231 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4232 		 * instead of reading the real value. */
4233 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4234 
4235 		/*
4236 		 * Transfer the event that L0 or L1 may wanted to inject into
4237 		 * L2 to IDT_VECTORING_INFO_FIELD.
4238 		 */
4239 		vmcs12_save_pending_event(vcpu, vmcs12,
4240 					  vm_exit_reason, exit_intr_info);
4241 
4242 		vmcs12->vm_exit_intr_info = exit_intr_info;
4243 		vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4244 		vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4245 
4246 		/*
4247 		 * According to spec, there's no need to store the guest's
4248 		 * MSRs if the exit is due to a VM-entry failure that occurs
4249 		 * during or after loading the guest state. Since this exit
4250 		 * does not fall in that category, we need to save the MSRs.
4251 		 */
4252 		if (nested_vmx_store_msr(vcpu,
4253 					 vmcs12->vm_exit_msr_store_addr,
4254 					 vmcs12->vm_exit_msr_store_count))
4255 			nested_vmx_abort(vcpu,
4256 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4257 	}
4258 
4259 	/*
4260 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4261 	 * preserved above and would only end up incorrectly in L1.
4262 	 */
4263 	vcpu->arch.nmi_injected = false;
4264 	kvm_clear_exception_queue(vcpu);
4265 	kvm_clear_interrupt_queue(vcpu);
4266 }
4267 
4268 /*
4269  * A part of what we need to when the nested L2 guest exits and we want to
4270  * run its L1 parent, is to reset L1's guest state to the host state specified
4271  * in vmcs12.
4272  * This function is to be called not only on normal nested exit, but also on
4273  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4274  * Failures During or After Loading Guest State").
4275  * This function should be called when the active VMCS is L1's (vmcs01).
4276  */
4277 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4278 				   struct vmcs12 *vmcs12)
4279 {
4280 	enum vm_entry_failure_code ignored;
4281 	struct kvm_segment seg;
4282 
4283 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4284 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4285 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4286 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4287 	else
4288 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4289 	vmx_set_efer(vcpu, vcpu->arch.efer);
4290 
4291 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4292 	kvm_rip_write(vcpu, vmcs12->host_rip);
4293 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4294 	vmx_set_interrupt_shadow(vcpu, 0);
4295 
4296 	/*
4297 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4298 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4299 	 *
4300 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4301 	 * (KVM doesn't change it);
4302 	 */
4303 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4304 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4305 
4306 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4307 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4308 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4309 
4310 	nested_ept_uninit_mmu_context(vcpu);
4311 
4312 	/*
4313 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4314 	 * couldn't have changed.
4315 	 */
4316 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4317 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4318 
4319 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4320 
4321 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4322 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4323 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4324 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4325 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4326 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4327 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4328 
4329 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4330 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4331 		vmcs_write64(GUEST_BNDCFGS, 0);
4332 
4333 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4334 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4335 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4336 	}
4337 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
4338 	    intel_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)))
4339 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4340 					 vmcs12->host_ia32_perf_global_ctrl));
4341 
4342 	/* Set L1 segment info according to Intel SDM
4343 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4344 	seg = (struct kvm_segment) {
4345 		.base = 0,
4346 		.limit = 0xFFFFFFFF,
4347 		.selector = vmcs12->host_cs_selector,
4348 		.type = 11,
4349 		.present = 1,
4350 		.s = 1,
4351 		.g = 1
4352 	};
4353 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4354 		seg.l = 1;
4355 	else
4356 		seg.db = 1;
4357 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4358 	seg = (struct kvm_segment) {
4359 		.base = 0,
4360 		.limit = 0xFFFFFFFF,
4361 		.type = 3,
4362 		.present = 1,
4363 		.s = 1,
4364 		.db = 1,
4365 		.g = 1
4366 	};
4367 	seg.selector = vmcs12->host_ds_selector;
4368 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4369 	seg.selector = vmcs12->host_es_selector;
4370 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4371 	seg.selector = vmcs12->host_ss_selector;
4372 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4373 	seg.selector = vmcs12->host_fs_selector;
4374 	seg.base = vmcs12->host_fs_base;
4375 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4376 	seg.selector = vmcs12->host_gs_selector;
4377 	seg.base = vmcs12->host_gs_base;
4378 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4379 	seg = (struct kvm_segment) {
4380 		.base = vmcs12->host_tr_base,
4381 		.limit = 0x67,
4382 		.selector = vmcs12->host_tr_selector,
4383 		.type = 11,
4384 		.present = 1
4385 	};
4386 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4387 
4388 	memset(&seg, 0, sizeof(seg));
4389 	seg.unusable = 1;
4390 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4391 
4392 	kvm_set_dr(vcpu, 7, 0x400);
4393 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4394 
4395 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4396 				vmcs12->vm_exit_msr_load_count))
4397 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4398 
4399 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4400 }
4401 
4402 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4403 {
4404 	struct vmx_uret_msr *efer_msr;
4405 	unsigned int i;
4406 
4407 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4408 		return vmcs_read64(GUEST_IA32_EFER);
4409 
4410 	if (cpu_has_load_ia32_efer())
4411 		return host_efer;
4412 
4413 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4414 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4415 			return vmx->msr_autoload.guest.val[i].value;
4416 	}
4417 
4418 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4419 	if (efer_msr)
4420 		return efer_msr->data;
4421 
4422 	return host_efer;
4423 }
4424 
4425 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4426 {
4427 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4428 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4429 	struct vmx_msr_entry g, h;
4430 	gpa_t gpa;
4431 	u32 i, j;
4432 
4433 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4434 
4435 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4436 		/*
4437 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4438 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4439 		 * and vcpu->arch.dr7 is not squirreled away before the
4440 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4441 		 */
4442 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4443 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4444 		else
4445 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4446 	}
4447 
4448 	/*
4449 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4450 	 * handle a variety of side effects to KVM's software model.
4451 	 */
4452 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4453 
4454 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4455 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4456 
4457 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4458 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4459 
4460 	nested_ept_uninit_mmu_context(vcpu);
4461 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4462 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4463 
4464 	/*
4465 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4466 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4467 	 * VMFail, like everything else we just need to ensure our
4468 	 * software model is up-to-date.
4469 	 */
4470 	if (enable_ept && is_pae_paging(vcpu))
4471 		ept_save_pdptrs(vcpu);
4472 
4473 	kvm_mmu_reset_context(vcpu);
4474 
4475 	/*
4476 	 * This nasty bit of open coding is a compromise between blindly
4477 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4478 	 * of VMFail), leaving the nested VM's MSRs in the software model
4479 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4480 	 * expensive since the lists are unbound by hardware).  For each
4481 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4482 	 * list, reload it from the exit load list if it exists and differs
4483 	 * from the guest value.  The intent is to stuff host state as
4484 	 * silently as possible, not to fully process the exit load list.
4485 	 */
4486 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4487 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4488 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4489 			pr_debug_ratelimited(
4490 				"%s read MSR index failed (%u, 0x%08llx)\n",
4491 				__func__, i, gpa);
4492 			goto vmabort;
4493 		}
4494 
4495 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4496 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4497 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4498 				pr_debug_ratelimited(
4499 					"%s read MSR failed (%u, 0x%08llx)\n",
4500 					__func__, j, gpa);
4501 				goto vmabort;
4502 			}
4503 			if (h.index != g.index)
4504 				continue;
4505 			if (h.value == g.value)
4506 				break;
4507 
4508 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4509 				pr_debug_ratelimited(
4510 					"%s check failed (%u, 0x%x, 0x%x)\n",
4511 					__func__, j, h.index, h.reserved);
4512 				goto vmabort;
4513 			}
4514 
4515 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4516 				pr_debug_ratelimited(
4517 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4518 					__func__, j, h.index, h.value);
4519 				goto vmabort;
4520 			}
4521 		}
4522 	}
4523 
4524 	return;
4525 
4526 vmabort:
4527 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4528 }
4529 
4530 /*
4531  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4532  * and modify vmcs12 to make it see what it would expect to see there if
4533  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4534  */
4535 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4536 		       u32 exit_intr_info, unsigned long exit_qualification)
4537 {
4538 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4539 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4540 
4541 	/* trying to cancel vmlaunch/vmresume is a bug */
4542 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4543 
4544 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4545 		/*
4546 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4547 		 * Enlightened VMCS after migration and we still need to
4548 		 * do that when something is forcing L2->L1 exit prior to
4549 		 * the first L2 run.
4550 		 */
4551 		(void)nested_get_evmcs_page(vcpu);
4552 	}
4553 
4554 	/* Service pending TLB flush requests for L2 before switching to L1. */
4555 	kvm_service_local_tlb_flush_requests(vcpu);
4556 
4557 	/*
4558 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4559 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4560 	 * up-to-date before switching to L1.
4561 	 */
4562 	if (enable_ept && is_pae_paging(vcpu))
4563 		vmx_ept_load_pdptrs(vcpu);
4564 
4565 	leave_guest_mode(vcpu);
4566 
4567 	if (nested_cpu_has_preemption_timer(vmcs12))
4568 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4569 
4570 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4571 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4572 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4573 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4574 	}
4575 
4576 	if (likely(!vmx->fail)) {
4577 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4578 
4579 		if (vm_exit_reason != -1)
4580 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4581 				       exit_intr_info, exit_qualification);
4582 
4583 		/*
4584 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4585 		 * also be used to capture vmcs12 cache as part of
4586 		 * capturing nVMX state for snapshot (migration).
4587 		 *
4588 		 * Otherwise, this flush will dirty guest memory at a
4589 		 * point it is already assumed by user-space to be
4590 		 * immutable.
4591 		 */
4592 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4593 	} else {
4594 		/*
4595 		 * The only expected VM-instruction error is "VM entry with
4596 		 * invalid control field(s)." Anything else indicates a
4597 		 * problem with L0.  And we should never get here with a
4598 		 * VMFail of any type if early consistency checks are enabled.
4599 		 */
4600 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4601 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4602 		WARN_ON_ONCE(nested_early_check);
4603 	}
4604 
4605 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4606 
4607 	/* Update any VMCS fields that might have changed while L2 ran */
4608 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4609 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4610 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4611 	if (kvm_caps.has_tsc_control)
4612 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4613 
4614 	if (vmx->nested.l1_tpr_threshold != -1)
4615 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4616 
4617 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4618 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4619 		vmx_set_virtual_apic_mode(vcpu);
4620 	}
4621 
4622 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4623 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4624 		vmx_update_cpu_dirty_logging(vcpu);
4625 	}
4626 
4627 	/* Unpin physical memory we referred to in vmcs02 */
4628 	kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
4629 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4630 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4631 	vmx->nested.pi_desc = NULL;
4632 
4633 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4634 		vmx->nested.reload_vmcs01_apic_access_page = false;
4635 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4636 	}
4637 
4638 	if (vmx->nested.update_vmcs01_apicv_status) {
4639 		vmx->nested.update_vmcs01_apicv_status = false;
4640 		kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
4641 	}
4642 
4643 	if ((vm_exit_reason != -1) &&
4644 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4645 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4646 
4647 	/* in case we halted in L2 */
4648 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4649 
4650 	if (likely(!vmx->fail)) {
4651 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4652 		    nested_exit_intr_ack_set(vcpu)) {
4653 			int irq = kvm_cpu_get_interrupt(vcpu);
4654 			WARN_ON(irq < 0);
4655 			vmcs12->vm_exit_intr_info = irq |
4656 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4657 		}
4658 
4659 		if (vm_exit_reason != -1)
4660 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4661 						       vmcs12->exit_qualification,
4662 						       vmcs12->idt_vectoring_info_field,
4663 						       vmcs12->vm_exit_intr_info,
4664 						       vmcs12->vm_exit_intr_error_code,
4665 						       KVM_ISA_VMX);
4666 
4667 		load_vmcs12_host_state(vcpu, vmcs12);
4668 
4669 		return;
4670 	}
4671 
4672 	/*
4673 	 * After an early L2 VM-entry failure, we're now back
4674 	 * in L1 which thinks it just finished a VMLAUNCH or
4675 	 * VMRESUME instruction, so we need to set the failure
4676 	 * flag and the VM-instruction error field of the VMCS
4677 	 * accordingly, and skip the emulated instruction.
4678 	 */
4679 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4680 
4681 	/*
4682 	 * Restore L1's host state to KVM's software model.  We're here
4683 	 * because a consistency check was caught by hardware, which
4684 	 * means some amount of guest state has been propagated to KVM's
4685 	 * model and needs to be unwound to the host's state.
4686 	 */
4687 	nested_vmx_restore_host_state(vcpu);
4688 
4689 	vmx->fail = 0;
4690 }
4691 
4692 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4693 {
4694 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4695 }
4696 
4697 /*
4698  * Decode the memory-address operand of a vmx instruction, as recorded on an
4699  * exit caused by such an instruction (run by a guest hypervisor).
4700  * On success, returns 0. When the operand is invalid, returns 1 and throws
4701  * #UD, #GP, or #SS.
4702  */
4703 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4704 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4705 {
4706 	gva_t off;
4707 	bool exn;
4708 	struct kvm_segment s;
4709 
4710 	/*
4711 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4712 	 * Execution", on an exit, vmx_instruction_info holds most of the
4713 	 * addressing components of the operand. Only the displacement part
4714 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4715 	 * For how an actual address is calculated from all these components,
4716 	 * refer to Vol. 1, "Operand Addressing".
4717 	 */
4718 	int  scaling = vmx_instruction_info & 3;
4719 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4720 	bool is_reg = vmx_instruction_info & (1u << 10);
4721 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4722 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4723 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4724 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4725 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4726 
4727 	if (is_reg) {
4728 		kvm_queue_exception(vcpu, UD_VECTOR);
4729 		return 1;
4730 	}
4731 
4732 	/* Addr = segment_base + offset */
4733 	/* offset = base + [index * scale] + displacement */
4734 	off = exit_qualification; /* holds the displacement */
4735 	if (addr_size == 1)
4736 		off = (gva_t)sign_extend64(off, 31);
4737 	else if (addr_size == 0)
4738 		off = (gva_t)sign_extend64(off, 15);
4739 	if (base_is_valid)
4740 		off += kvm_register_read(vcpu, base_reg);
4741 	if (index_is_valid)
4742 		off += kvm_register_read(vcpu, index_reg) << scaling;
4743 	vmx_get_segment(vcpu, &s, seg_reg);
4744 
4745 	/*
4746 	 * The effective address, i.e. @off, of a memory operand is truncated
4747 	 * based on the address size of the instruction.  Note that this is
4748 	 * the *effective address*, i.e. the address prior to accounting for
4749 	 * the segment's base.
4750 	 */
4751 	if (addr_size == 1) /* 32 bit */
4752 		off &= 0xffffffff;
4753 	else if (addr_size == 0) /* 16 bit */
4754 		off &= 0xffff;
4755 
4756 	/* Checks for #GP/#SS exceptions. */
4757 	exn = false;
4758 	if (is_long_mode(vcpu)) {
4759 		/*
4760 		 * The virtual/linear address is never truncated in 64-bit
4761 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4762 		 * address when using FS/GS with a non-zero base.
4763 		 */
4764 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4765 			*ret = s.base + off;
4766 		else
4767 			*ret = off;
4768 
4769 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4770 		 * non-canonical form. This is the only check on the memory
4771 		 * destination for long mode!
4772 		 */
4773 		exn = is_noncanonical_address(*ret, vcpu);
4774 	} else {
4775 		/*
4776 		 * When not in long mode, the virtual/linear address is
4777 		 * unconditionally truncated to 32 bits regardless of the
4778 		 * address size.
4779 		 */
4780 		*ret = (s.base + off) & 0xffffffff;
4781 
4782 		/* Protected mode: apply checks for segment validity in the
4783 		 * following order:
4784 		 * - segment type check (#GP(0) may be thrown)
4785 		 * - usability check (#GP(0)/#SS(0))
4786 		 * - limit check (#GP(0)/#SS(0))
4787 		 */
4788 		if (wr)
4789 			/* #GP(0) if the destination operand is located in a
4790 			 * read-only data segment or any code segment.
4791 			 */
4792 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4793 		else
4794 			/* #GP(0) if the source operand is located in an
4795 			 * execute-only code segment
4796 			 */
4797 			exn = ((s.type & 0xa) == 8);
4798 		if (exn) {
4799 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4800 			return 1;
4801 		}
4802 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4803 		 */
4804 		exn = (s.unusable != 0);
4805 
4806 		/*
4807 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4808 		 * outside the segment limit.  All CPUs that support VMX ignore
4809 		 * limit checks for flat segments, i.e. segments with base==0,
4810 		 * limit==0xffffffff and of type expand-up data or code.
4811 		 */
4812 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4813 		     ((s.type & 8) || !(s.type & 4))))
4814 			exn = exn || ((u64)off + len - 1 > s.limit);
4815 	}
4816 	if (exn) {
4817 		kvm_queue_exception_e(vcpu,
4818 				      seg_reg == VCPU_SREG_SS ?
4819 						SS_VECTOR : GP_VECTOR,
4820 				      0);
4821 		return 1;
4822 	}
4823 
4824 	return 0;
4825 }
4826 
4827 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4828 				int *ret)
4829 {
4830 	gva_t gva;
4831 	struct x86_exception e;
4832 	int r;
4833 
4834 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4835 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4836 				sizeof(*vmpointer), &gva)) {
4837 		*ret = 1;
4838 		return -EINVAL;
4839 	}
4840 
4841 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4842 	if (r != X86EMUL_CONTINUE) {
4843 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4844 		return -EINVAL;
4845 	}
4846 
4847 	return 0;
4848 }
4849 
4850 /*
4851  * Allocate a shadow VMCS and associate it with the currently loaded
4852  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4853  * VMCS is also VMCLEARed, so that it is ready for use.
4854  */
4855 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4856 {
4857 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4858 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4859 
4860 	/*
4861 	 * KVM allocates a shadow VMCS only when L1 executes VMXON and frees it
4862 	 * when L1 executes VMXOFF or the vCPU is forced out of nested
4863 	 * operation.  VMXON faults if the CPU is already post-VMXON, so it
4864 	 * should be impossible to already have an allocated shadow VMCS.  KVM
4865 	 * doesn't support virtualization of VMCS shadowing, so vmcs01 should
4866 	 * always be the loaded VMCS.
4867 	 */
4868 	if (WARN_ON(loaded_vmcs != &vmx->vmcs01 || loaded_vmcs->shadow_vmcs))
4869 		return loaded_vmcs->shadow_vmcs;
4870 
4871 	loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4872 	if (loaded_vmcs->shadow_vmcs)
4873 		vmcs_clear(loaded_vmcs->shadow_vmcs);
4874 
4875 	return loaded_vmcs->shadow_vmcs;
4876 }
4877 
4878 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4879 {
4880 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4881 	int r;
4882 
4883 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4884 	if (r < 0)
4885 		goto out_vmcs02;
4886 
4887 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4888 	if (!vmx->nested.cached_vmcs12)
4889 		goto out_cached_vmcs12;
4890 
4891 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
4892 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4893 	if (!vmx->nested.cached_shadow_vmcs12)
4894 		goto out_cached_shadow_vmcs12;
4895 
4896 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4897 		goto out_shadow_vmcs;
4898 
4899 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4900 		     HRTIMER_MODE_ABS_PINNED);
4901 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4902 
4903 	vmx->nested.vpid02 = allocate_vpid();
4904 
4905 	vmx->nested.vmcs02_initialized = false;
4906 	vmx->nested.vmxon = true;
4907 
4908 	if (vmx_pt_mode_is_host_guest()) {
4909 		vmx->pt_desc.guest.ctl = 0;
4910 		pt_update_intercept_for_msr(vcpu);
4911 	}
4912 
4913 	return 0;
4914 
4915 out_shadow_vmcs:
4916 	kfree(vmx->nested.cached_shadow_vmcs12);
4917 
4918 out_cached_shadow_vmcs12:
4919 	kfree(vmx->nested.cached_vmcs12);
4920 
4921 out_cached_vmcs12:
4922 	free_loaded_vmcs(&vmx->nested.vmcs02);
4923 
4924 out_vmcs02:
4925 	return -ENOMEM;
4926 }
4927 
4928 /* Emulate the VMXON instruction. */
4929 static int handle_vmxon(struct kvm_vcpu *vcpu)
4930 {
4931 	int ret;
4932 	gpa_t vmptr;
4933 	uint32_t revision;
4934 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4935 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4936 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4937 
4938 	/*
4939 	 * Note, KVM cannot rely on hardware to perform the CR0/CR4 #UD checks
4940 	 * that have higher priority than VM-Exit (see Intel SDM's pseudocode
4941 	 * for VMXON), as KVM must load valid CR0/CR4 values into hardware while
4942 	 * running the guest, i.e. KVM needs to check the _guest_ values.
4943 	 *
4944 	 * Rely on hardware for the other two pre-VM-Exit checks, !VM86 and
4945 	 * !COMPATIBILITY modes.  KVM may run the guest in VM86 to emulate Real
4946 	 * Mode, but KVM will never take the guest out of those modes.
4947 	 */
4948 	if (!nested_host_cr0_valid(vcpu, kvm_read_cr0(vcpu)) ||
4949 	    !nested_host_cr4_valid(vcpu, kvm_read_cr4(vcpu))) {
4950 		kvm_queue_exception(vcpu, UD_VECTOR);
4951 		return 1;
4952 	}
4953 
4954 	/*
4955 	 * CPL=0 and all other checks that are lower priority than VM-Exit must
4956 	 * be checked manually.
4957 	 */
4958 	if (vmx_get_cpl(vcpu)) {
4959 		kvm_inject_gp(vcpu, 0);
4960 		return 1;
4961 	}
4962 
4963 	if (vmx->nested.vmxon)
4964 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4965 
4966 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4967 			!= VMXON_NEEDED_FEATURES) {
4968 		kvm_inject_gp(vcpu, 0);
4969 		return 1;
4970 	}
4971 
4972 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4973 		return ret;
4974 
4975 	/*
4976 	 * SDM 3: 24.11.5
4977 	 * The first 4 bytes of VMXON region contain the supported
4978 	 * VMCS revision identifier
4979 	 *
4980 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4981 	 * which replaces physical address width with 32
4982 	 */
4983 	if (!page_address_valid(vcpu, vmptr))
4984 		return nested_vmx_failInvalid(vcpu);
4985 
4986 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4987 	    revision != VMCS12_REVISION)
4988 		return nested_vmx_failInvalid(vcpu);
4989 
4990 	vmx->nested.vmxon_ptr = vmptr;
4991 	ret = enter_vmx_operation(vcpu);
4992 	if (ret)
4993 		return ret;
4994 
4995 	return nested_vmx_succeed(vcpu);
4996 }
4997 
4998 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4999 {
5000 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5001 
5002 	if (vmx->nested.current_vmptr == INVALID_GPA)
5003 		return;
5004 
5005 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
5006 
5007 	if (enable_shadow_vmcs) {
5008 		/* copy to memory all shadowed fields in case
5009 		   they were modified */
5010 		copy_shadow_to_vmcs12(vmx);
5011 		vmx_disable_shadow_vmcs(vmx);
5012 	}
5013 	vmx->nested.posted_intr_nv = -1;
5014 
5015 	/* Flush VMCS12 to guest memory */
5016 	kvm_vcpu_write_guest_page(vcpu,
5017 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
5018 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5019 
5020 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5021 
5022 	vmx->nested.current_vmptr = INVALID_GPA;
5023 }
5024 
5025 /* Emulate the VMXOFF instruction */
5026 static int handle_vmxoff(struct kvm_vcpu *vcpu)
5027 {
5028 	if (!nested_vmx_check_permission(vcpu))
5029 		return 1;
5030 
5031 	free_nested(vcpu);
5032 
5033 	/* Process a latched INIT during time CPU was in VMX operation */
5034 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5035 
5036 	return nested_vmx_succeed(vcpu);
5037 }
5038 
5039 /* Emulate the VMCLEAR instruction */
5040 static int handle_vmclear(struct kvm_vcpu *vcpu)
5041 {
5042 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5043 	u32 zero = 0;
5044 	gpa_t vmptr;
5045 	u64 evmcs_gpa;
5046 	int r;
5047 
5048 	if (!nested_vmx_check_permission(vcpu))
5049 		return 1;
5050 
5051 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5052 		return r;
5053 
5054 	if (!page_address_valid(vcpu, vmptr))
5055 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5056 
5057 	if (vmptr == vmx->nested.vmxon_ptr)
5058 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5059 
5060 	/*
5061 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
5062 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
5063 	 * way to distinguish it from VMCS12) and we must not corrupt it by
5064 	 * writing to the non-existent 'launch_state' field. The area doesn't
5065 	 * have to be the currently active EVMCS on the calling CPU and there's
5066 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5067 	 * state. It is possible that the area will stay mapped as
5068 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5069 	 */
5070 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
5071 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
5072 		if (vmptr == vmx->nested.current_vmptr)
5073 			nested_release_vmcs12(vcpu);
5074 
5075 		kvm_vcpu_write_guest(vcpu,
5076 				     vmptr + offsetof(struct vmcs12,
5077 						      launch_state),
5078 				     &zero, sizeof(zero));
5079 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5080 		nested_release_evmcs(vcpu);
5081 	}
5082 
5083 	return nested_vmx_succeed(vcpu);
5084 }
5085 
5086 /* Emulate the VMLAUNCH instruction */
5087 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5088 {
5089 	return nested_vmx_run(vcpu, true);
5090 }
5091 
5092 /* Emulate the VMRESUME instruction */
5093 static int handle_vmresume(struct kvm_vcpu *vcpu)
5094 {
5095 
5096 	return nested_vmx_run(vcpu, false);
5097 }
5098 
5099 static int handle_vmread(struct kvm_vcpu *vcpu)
5100 {
5101 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5102 						    : get_vmcs12(vcpu);
5103 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5104 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5105 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5106 	struct x86_exception e;
5107 	unsigned long field;
5108 	u64 value;
5109 	gva_t gva = 0;
5110 	short offset;
5111 	int len, r;
5112 
5113 	if (!nested_vmx_check_permission(vcpu))
5114 		return 1;
5115 
5116 	/* Decode instruction info and find the field to read */
5117 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5118 
5119 	if (!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
5120 		/*
5121 		 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5122 		 * any VMREAD sets the ALU flags for VMfailInvalid.
5123 		 */
5124 		if (vmx->nested.current_vmptr == INVALID_GPA ||
5125 		    (is_guest_mode(vcpu) &&
5126 		     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5127 			return nested_vmx_failInvalid(vcpu);
5128 
5129 		offset = get_vmcs12_field_offset(field);
5130 		if (offset < 0)
5131 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5132 
5133 		if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5134 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5135 
5136 		/* Read the field, zero-extended to a u64 value */
5137 		value = vmcs12_read_any(vmcs12, field, offset);
5138 	} else {
5139 		/*
5140 		 * Hyper-V TLFS (as of 6.0b) explicitly states, that while an
5141 		 * enlightened VMCS is active VMREAD/VMWRITE instructions are
5142 		 * unsupported. Unfortunately, certain versions of Windows 11
5143 		 * don't comply with this requirement which is not enforced in
5144 		 * genuine Hyper-V. Allow VMREAD from an enlightened VMCS as a
5145 		 * workaround, as misbehaving guests will panic on VM-Fail.
5146 		 * Note, enlightened VMCS is incompatible with shadow VMCS so
5147 		 * all VMREADs from L2 should go to L1.
5148 		 */
5149 		if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5150 			return nested_vmx_failInvalid(vcpu);
5151 
5152 		offset = evmcs_field_offset(field, NULL);
5153 		if (offset < 0)
5154 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5155 
5156 		/* Read the field, zero-extended to a u64 value */
5157 		value = evmcs_read_any(vmx->nested.hv_evmcs, field, offset);
5158 	}
5159 
5160 	/*
5161 	 * Now copy part of this value to register or memory, as requested.
5162 	 * Note that the number of bits actually copied is 32 or 64 depending
5163 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5164 	 */
5165 	if (instr_info & BIT(10)) {
5166 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5167 	} else {
5168 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5169 		if (get_vmx_mem_address(vcpu, exit_qualification,
5170 					instr_info, true, len, &gva))
5171 			return 1;
5172 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5173 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5174 		if (r != X86EMUL_CONTINUE)
5175 			return kvm_handle_memory_failure(vcpu, r, &e);
5176 	}
5177 
5178 	return nested_vmx_succeed(vcpu);
5179 }
5180 
5181 static bool is_shadow_field_rw(unsigned long field)
5182 {
5183 	switch (field) {
5184 #define SHADOW_FIELD_RW(x, y) case x:
5185 #include "vmcs_shadow_fields.h"
5186 		return true;
5187 	default:
5188 		break;
5189 	}
5190 	return false;
5191 }
5192 
5193 static bool is_shadow_field_ro(unsigned long field)
5194 {
5195 	switch (field) {
5196 #define SHADOW_FIELD_RO(x, y) case x:
5197 #include "vmcs_shadow_fields.h"
5198 		return true;
5199 	default:
5200 		break;
5201 	}
5202 	return false;
5203 }
5204 
5205 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5206 {
5207 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5208 						    : get_vmcs12(vcpu);
5209 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5210 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5211 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5212 	struct x86_exception e;
5213 	unsigned long field;
5214 	short offset;
5215 	gva_t gva;
5216 	int len, r;
5217 
5218 	/*
5219 	 * The value to write might be 32 or 64 bits, depending on L1's long
5220 	 * mode, and eventually we need to write that into a field of several
5221 	 * possible lengths. The code below first zero-extends the value to 64
5222 	 * bit (value), and then copies only the appropriate number of
5223 	 * bits into the vmcs12 field.
5224 	 */
5225 	u64 value = 0;
5226 
5227 	if (!nested_vmx_check_permission(vcpu))
5228 		return 1;
5229 
5230 	/*
5231 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5232 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5233 	 */
5234 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5235 	    (is_guest_mode(vcpu) &&
5236 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5237 		return nested_vmx_failInvalid(vcpu);
5238 
5239 	if (instr_info & BIT(10))
5240 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5241 	else {
5242 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5243 		if (get_vmx_mem_address(vcpu, exit_qualification,
5244 					instr_info, false, len, &gva))
5245 			return 1;
5246 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5247 		if (r != X86EMUL_CONTINUE)
5248 			return kvm_handle_memory_failure(vcpu, r, &e);
5249 	}
5250 
5251 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5252 
5253 	offset = get_vmcs12_field_offset(field);
5254 	if (offset < 0)
5255 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5256 
5257 	/*
5258 	 * If the vCPU supports "VMWRITE to any supported field in the
5259 	 * VMCS," then the "read-only" fields are actually read/write.
5260 	 */
5261 	if (vmcs_field_readonly(field) &&
5262 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5263 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5264 
5265 	/*
5266 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5267 	 * vmcs12, else we may crush a field or consume a stale value.
5268 	 */
5269 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5270 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5271 
5272 	/*
5273 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5274 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5275 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5276 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5277 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5278 	 * the stripped down value, L2 sees the full value as stored by KVM).
5279 	 */
5280 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5281 		value &= 0x1f0ff;
5282 
5283 	vmcs12_write_any(vmcs12, field, offset, value);
5284 
5285 	/*
5286 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5287 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5288 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5289 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5290 	 */
5291 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5292 		/*
5293 		 * L1 can read these fields without exiting, ensure the
5294 		 * shadow VMCS is up-to-date.
5295 		 */
5296 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5297 			preempt_disable();
5298 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5299 
5300 			__vmcs_writel(field, value);
5301 
5302 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5303 			vmcs_load(vmx->loaded_vmcs->vmcs);
5304 			preempt_enable();
5305 		}
5306 		vmx->nested.dirty_vmcs12 = true;
5307 	}
5308 
5309 	return nested_vmx_succeed(vcpu);
5310 }
5311 
5312 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5313 {
5314 	vmx->nested.current_vmptr = vmptr;
5315 	if (enable_shadow_vmcs) {
5316 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5317 		vmcs_write64(VMCS_LINK_POINTER,
5318 			     __pa(vmx->vmcs01.shadow_vmcs));
5319 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5320 	}
5321 	vmx->nested.dirty_vmcs12 = true;
5322 	vmx->nested.force_msr_bitmap_recalc = true;
5323 }
5324 
5325 /* Emulate the VMPTRLD instruction */
5326 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5327 {
5328 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5329 	gpa_t vmptr;
5330 	int r;
5331 
5332 	if (!nested_vmx_check_permission(vcpu))
5333 		return 1;
5334 
5335 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5336 		return r;
5337 
5338 	if (!page_address_valid(vcpu, vmptr))
5339 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5340 
5341 	if (vmptr == vmx->nested.vmxon_ptr)
5342 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5343 
5344 	/* Forbid normal VMPTRLD if Enlightened version was used */
5345 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5346 		return 1;
5347 
5348 	if (vmx->nested.current_vmptr != vmptr) {
5349 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5350 		struct vmcs_hdr hdr;
5351 
5352 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5353 			/*
5354 			 * Reads from an unbacked page return all 1s,
5355 			 * which means that the 32 bits located at the
5356 			 * given physical address won't match the required
5357 			 * VMCS12_REVISION identifier.
5358 			 */
5359 			return nested_vmx_fail(vcpu,
5360 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5361 		}
5362 
5363 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5364 						 offsetof(struct vmcs12, hdr),
5365 						 sizeof(hdr))) {
5366 			return nested_vmx_fail(vcpu,
5367 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5368 		}
5369 
5370 		if (hdr.revision_id != VMCS12_REVISION ||
5371 		    (hdr.shadow_vmcs &&
5372 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5373 			return nested_vmx_fail(vcpu,
5374 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5375 		}
5376 
5377 		nested_release_vmcs12(vcpu);
5378 
5379 		/*
5380 		 * Load VMCS12 from guest memory since it is not already
5381 		 * cached.
5382 		 */
5383 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5384 					  VMCS12_SIZE)) {
5385 			return nested_vmx_fail(vcpu,
5386 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5387 		}
5388 
5389 		set_current_vmptr(vmx, vmptr);
5390 	}
5391 
5392 	return nested_vmx_succeed(vcpu);
5393 }
5394 
5395 /* Emulate the VMPTRST instruction */
5396 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5397 {
5398 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5399 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5400 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5401 	struct x86_exception e;
5402 	gva_t gva;
5403 	int r;
5404 
5405 	if (!nested_vmx_check_permission(vcpu))
5406 		return 1;
5407 
5408 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5409 		return 1;
5410 
5411 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5412 				true, sizeof(gpa_t), &gva))
5413 		return 1;
5414 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5415 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5416 					sizeof(gpa_t), &e);
5417 	if (r != X86EMUL_CONTINUE)
5418 		return kvm_handle_memory_failure(vcpu, r, &e);
5419 
5420 	return nested_vmx_succeed(vcpu);
5421 }
5422 
5423 /* Emulate the INVEPT instruction */
5424 static int handle_invept(struct kvm_vcpu *vcpu)
5425 {
5426 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5427 	u32 vmx_instruction_info, types;
5428 	unsigned long type, roots_to_free;
5429 	struct kvm_mmu *mmu;
5430 	gva_t gva;
5431 	struct x86_exception e;
5432 	struct {
5433 		u64 eptp, gpa;
5434 	} operand;
5435 	int i, r, gpr_index;
5436 
5437 	if (!(vmx->nested.msrs.secondary_ctls_high &
5438 	      SECONDARY_EXEC_ENABLE_EPT) ||
5439 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5440 		kvm_queue_exception(vcpu, UD_VECTOR);
5441 		return 1;
5442 	}
5443 
5444 	if (!nested_vmx_check_permission(vcpu))
5445 		return 1;
5446 
5447 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5448 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5449 	type = kvm_register_read(vcpu, gpr_index);
5450 
5451 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5452 
5453 	if (type >= 32 || !(types & (1 << type)))
5454 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5455 
5456 	/* According to the Intel VMX instruction reference, the memory
5457 	 * operand is read even if it isn't needed (e.g., for type==global)
5458 	 */
5459 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5460 			vmx_instruction_info, false, sizeof(operand), &gva))
5461 		return 1;
5462 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5463 	if (r != X86EMUL_CONTINUE)
5464 		return kvm_handle_memory_failure(vcpu, r, &e);
5465 
5466 	/*
5467 	 * Nested EPT roots are always held through guest_mmu,
5468 	 * not root_mmu.
5469 	 */
5470 	mmu = &vcpu->arch.guest_mmu;
5471 
5472 	switch (type) {
5473 	case VMX_EPT_EXTENT_CONTEXT:
5474 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5475 			return nested_vmx_fail(vcpu,
5476 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5477 
5478 		roots_to_free = 0;
5479 		if (nested_ept_root_matches(mmu->root.hpa, mmu->root.pgd,
5480 					    operand.eptp))
5481 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5482 
5483 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5484 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5485 						    mmu->prev_roots[i].pgd,
5486 						    operand.eptp))
5487 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5488 		}
5489 		break;
5490 	case VMX_EPT_EXTENT_GLOBAL:
5491 		roots_to_free = KVM_MMU_ROOTS_ALL;
5492 		break;
5493 	default:
5494 		BUG();
5495 		break;
5496 	}
5497 
5498 	if (roots_to_free)
5499 		kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
5500 
5501 	return nested_vmx_succeed(vcpu);
5502 }
5503 
5504 static int handle_invvpid(struct kvm_vcpu *vcpu)
5505 {
5506 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5507 	u32 vmx_instruction_info;
5508 	unsigned long type, types;
5509 	gva_t gva;
5510 	struct x86_exception e;
5511 	struct {
5512 		u64 vpid;
5513 		u64 gla;
5514 	} operand;
5515 	u16 vpid02;
5516 	int r, gpr_index;
5517 
5518 	if (!(vmx->nested.msrs.secondary_ctls_high &
5519 	      SECONDARY_EXEC_ENABLE_VPID) ||
5520 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5521 		kvm_queue_exception(vcpu, UD_VECTOR);
5522 		return 1;
5523 	}
5524 
5525 	if (!nested_vmx_check_permission(vcpu))
5526 		return 1;
5527 
5528 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5529 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5530 	type = kvm_register_read(vcpu, gpr_index);
5531 
5532 	types = (vmx->nested.msrs.vpid_caps &
5533 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5534 
5535 	if (type >= 32 || !(types & (1 << type)))
5536 		return nested_vmx_fail(vcpu,
5537 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5538 
5539 	/* according to the intel vmx instruction reference, the memory
5540 	 * operand is read even if it isn't needed (e.g., for type==global)
5541 	 */
5542 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5543 			vmx_instruction_info, false, sizeof(operand), &gva))
5544 		return 1;
5545 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5546 	if (r != X86EMUL_CONTINUE)
5547 		return kvm_handle_memory_failure(vcpu, r, &e);
5548 
5549 	if (operand.vpid >> 16)
5550 		return nested_vmx_fail(vcpu,
5551 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5552 
5553 	vpid02 = nested_get_vpid02(vcpu);
5554 	switch (type) {
5555 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5556 		if (!operand.vpid ||
5557 		    is_noncanonical_address(operand.gla, vcpu))
5558 			return nested_vmx_fail(vcpu,
5559 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5560 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5561 		break;
5562 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5563 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5564 		if (!operand.vpid)
5565 			return nested_vmx_fail(vcpu,
5566 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5567 		vpid_sync_context(vpid02);
5568 		break;
5569 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5570 		vpid_sync_context(vpid02);
5571 		break;
5572 	default:
5573 		WARN_ON_ONCE(1);
5574 		return kvm_skip_emulated_instruction(vcpu);
5575 	}
5576 
5577 	/*
5578 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5579 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5580 	 * roots as VPIDs are not tracked in the MMU role.
5581 	 *
5582 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5583 	 * an MMU when EPT is disabled.
5584 	 *
5585 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5586 	 */
5587 	if (!enable_ept)
5588 		kvm_mmu_free_guest_mode_roots(vcpu->kvm, &vcpu->arch.root_mmu);
5589 
5590 	return nested_vmx_succeed(vcpu);
5591 }
5592 
5593 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5594 				     struct vmcs12 *vmcs12)
5595 {
5596 	u32 index = kvm_rcx_read(vcpu);
5597 	u64 new_eptp;
5598 
5599 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5600 		return 1;
5601 	if (index >= VMFUNC_EPTP_ENTRIES)
5602 		return 1;
5603 
5604 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5605 				     &new_eptp, index * 8, 8))
5606 		return 1;
5607 
5608 	/*
5609 	 * If the (L2) guest does a vmfunc to the currently
5610 	 * active ept pointer, we don't have to do anything else
5611 	 */
5612 	if (vmcs12->ept_pointer != new_eptp) {
5613 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5614 			return 1;
5615 
5616 		vmcs12->ept_pointer = new_eptp;
5617 		nested_ept_new_eptp(vcpu);
5618 
5619 		if (!nested_cpu_has_vpid(vmcs12))
5620 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5621 	}
5622 
5623 	return 0;
5624 }
5625 
5626 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5627 {
5628 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5629 	struct vmcs12 *vmcs12;
5630 	u32 function = kvm_rax_read(vcpu);
5631 
5632 	/*
5633 	 * VMFUNC is only supported for nested guests, but we always enable the
5634 	 * secondary control for simplicity; for non-nested mode, fake that we
5635 	 * didn't by injecting #UD.
5636 	 */
5637 	if (!is_guest_mode(vcpu)) {
5638 		kvm_queue_exception(vcpu, UD_VECTOR);
5639 		return 1;
5640 	}
5641 
5642 	vmcs12 = get_vmcs12(vcpu);
5643 
5644 	/*
5645 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5646 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5647 	 */
5648 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5649 		kvm_queue_exception(vcpu, UD_VECTOR);
5650 		return 1;
5651 	}
5652 
5653 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5654 		goto fail;
5655 
5656 	switch (function) {
5657 	case 0:
5658 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5659 			goto fail;
5660 		break;
5661 	default:
5662 		goto fail;
5663 	}
5664 	return kvm_skip_emulated_instruction(vcpu);
5665 
5666 fail:
5667 	/*
5668 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5669 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5670 	 * EXIT_REASON_VMFUNC as the exit reason.
5671 	 */
5672 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5673 			  vmx_get_intr_info(vcpu),
5674 			  vmx_get_exit_qual(vcpu));
5675 	return 1;
5676 }
5677 
5678 /*
5679  * Return true if an IO instruction with the specified port and size should cause
5680  * a VM-exit into L1.
5681  */
5682 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5683 				 int size)
5684 {
5685 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5686 	gpa_t bitmap, last_bitmap;
5687 	u8 b;
5688 
5689 	last_bitmap = INVALID_GPA;
5690 	b = -1;
5691 
5692 	while (size > 0) {
5693 		if (port < 0x8000)
5694 			bitmap = vmcs12->io_bitmap_a;
5695 		else if (port < 0x10000)
5696 			bitmap = vmcs12->io_bitmap_b;
5697 		else
5698 			return true;
5699 		bitmap += (port & 0x7fff) / 8;
5700 
5701 		if (last_bitmap != bitmap)
5702 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5703 				return true;
5704 		if (b & (1 << (port & 7)))
5705 			return true;
5706 
5707 		port++;
5708 		size--;
5709 		last_bitmap = bitmap;
5710 	}
5711 
5712 	return false;
5713 }
5714 
5715 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5716 				       struct vmcs12 *vmcs12)
5717 {
5718 	unsigned long exit_qualification;
5719 	unsigned short port;
5720 	int size;
5721 
5722 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5723 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5724 
5725 	exit_qualification = vmx_get_exit_qual(vcpu);
5726 
5727 	port = exit_qualification >> 16;
5728 	size = (exit_qualification & 7) + 1;
5729 
5730 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5731 }
5732 
5733 /*
5734  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5735  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5736  * disinterest in the current event (read or write a specific MSR) by using an
5737  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5738  */
5739 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5740 					struct vmcs12 *vmcs12,
5741 					union vmx_exit_reason exit_reason)
5742 {
5743 	u32 msr_index = kvm_rcx_read(vcpu);
5744 	gpa_t bitmap;
5745 
5746 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5747 		return true;
5748 
5749 	/*
5750 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5751 	 * for the four combinations of read/write and low/high MSR numbers.
5752 	 * First we need to figure out which of the four to use:
5753 	 */
5754 	bitmap = vmcs12->msr_bitmap;
5755 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5756 		bitmap += 2048;
5757 	if (msr_index >= 0xc0000000) {
5758 		msr_index -= 0xc0000000;
5759 		bitmap += 1024;
5760 	}
5761 
5762 	/* Then read the msr_index'th bit from this bitmap: */
5763 	if (msr_index < 1024*8) {
5764 		unsigned char b;
5765 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5766 			return true;
5767 		return 1 & (b >> (msr_index & 7));
5768 	} else
5769 		return true; /* let L1 handle the wrong parameter */
5770 }
5771 
5772 /*
5773  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5774  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5775  * intercept (via guest_host_mask etc.) the current event.
5776  */
5777 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5778 	struct vmcs12 *vmcs12)
5779 {
5780 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5781 	int cr = exit_qualification & 15;
5782 	int reg;
5783 	unsigned long val;
5784 
5785 	switch ((exit_qualification >> 4) & 3) {
5786 	case 0: /* mov to cr */
5787 		reg = (exit_qualification >> 8) & 15;
5788 		val = kvm_register_read(vcpu, reg);
5789 		switch (cr) {
5790 		case 0:
5791 			if (vmcs12->cr0_guest_host_mask &
5792 			    (val ^ vmcs12->cr0_read_shadow))
5793 				return true;
5794 			break;
5795 		case 3:
5796 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5797 				return true;
5798 			break;
5799 		case 4:
5800 			if (vmcs12->cr4_guest_host_mask &
5801 			    (vmcs12->cr4_read_shadow ^ val))
5802 				return true;
5803 			break;
5804 		case 8:
5805 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5806 				return true;
5807 			break;
5808 		}
5809 		break;
5810 	case 2: /* clts */
5811 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5812 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5813 			return true;
5814 		break;
5815 	case 1: /* mov from cr */
5816 		switch (cr) {
5817 		case 3:
5818 			if (vmcs12->cpu_based_vm_exec_control &
5819 			    CPU_BASED_CR3_STORE_EXITING)
5820 				return true;
5821 			break;
5822 		case 8:
5823 			if (vmcs12->cpu_based_vm_exec_control &
5824 			    CPU_BASED_CR8_STORE_EXITING)
5825 				return true;
5826 			break;
5827 		}
5828 		break;
5829 	case 3: /* lmsw */
5830 		/*
5831 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5832 		 * cr0. Other attempted changes are ignored, with no exit.
5833 		 */
5834 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5835 		if (vmcs12->cr0_guest_host_mask & 0xe &
5836 		    (val ^ vmcs12->cr0_read_shadow))
5837 			return true;
5838 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5839 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5840 		    (val & 0x1))
5841 			return true;
5842 		break;
5843 	}
5844 	return false;
5845 }
5846 
5847 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
5848 					  struct vmcs12 *vmcs12)
5849 {
5850 	u32 encls_leaf;
5851 
5852 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
5853 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
5854 		return false;
5855 
5856 	encls_leaf = kvm_rax_read(vcpu);
5857 	if (encls_leaf > 62)
5858 		encls_leaf = 63;
5859 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
5860 }
5861 
5862 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5863 	struct vmcs12 *vmcs12, gpa_t bitmap)
5864 {
5865 	u32 vmx_instruction_info;
5866 	unsigned long field;
5867 	u8 b;
5868 
5869 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5870 		return true;
5871 
5872 	/* Decode instruction info and find the field to access */
5873 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5874 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5875 
5876 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5877 	if (field >> 15)
5878 		return true;
5879 
5880 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5881 		return true;
5882 
5883 	return 1 & (b >> (field & 7));
5884 }
5885 
5886 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5887 {
5888 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5889 
5890 	if (nested_cpu_has_mtf(vmcs12))
5891 		return true;
5892 
5893 	/*
5894 	 * An MTF VM-exit may be injected into the guest by setting the
5895 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5896 	 * is the case regardless of the 'monitor trap flag' VM-execution
5897 	 * control.
5898 	 */
5899 	return entry_intr_info == (INTR_INFO_VALID_MASK
5900 				   | INTR_TYPE_OTHER_EVENT);
5901 }
5902 
5903 /*
5904  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5905  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5906  */
5907 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
5908 				     union vmx_exit_reason exit_reason)
5909 {
5910 	u32 intr_info;
5911 
5912 	switch ((u16)exit_reason.basic) {
5913 	case EXIT_REASON_EXCEPTION_NMI:
5914 		intr_info = vmx_get_intr_info(vcpu);
5915 		if (is_nmi(intr_info))
5916 			return true;
5917 		else if (is_page_fault(intr_info))
5918 			return vcpu->arch.apf.host_apf_flags ||
5919 			       vmx_need_pf_intercept(vcpu);
5920 		else if (is_debug(intr_info) &&
5921 			 vcpu->guest_debug &
5922 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5923 			return true;
5924 		else if (is_breakpoint(intr_info) &&
5925 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5926 			return true;
5927 		else if (is_alignment_check(intr_info) &&
5928 			 !vmx_guest_inject_ac(vcpu))
5929 			return true;
5930 		return false;
5931 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5932 		return true;
5933 	case EXIT_REASON_MCE_DURING_VMENTRY:
5934 		return true;
5935 	case EXIT_REASON_EPT_VIOLATION:
5936 		/*
5937 		 * L0 always deals with the EPT violation. If nested EPT is
5938 		 * used, and the nested mmu code discovers that the address is
5939 		 * missing in the guest EPT table (EPT12), the EPT violation
5940 		 * will be injected with nested_ept_inject_page_fault()
5941 		 */
5942 		return true;
5943 	case EXIT_REASON_EPT_MISCONFIG:
5944 		/*
5945 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5946 		 * table (shadow on EPT) or a merged EPT table that L0 built
5947 		 * (EPT on EPT). So any problems with the structure of the
5948 		 * table is L0's fault.
5949 		 */
5950 		return true;
5951 	case EXIT_REASON_PREEMPTION_TIMER:
5952 		return true;
5953 	case EXIT_REASON_PML_FULL:
5954 		/*
5955 		 * PML is emulated for an L1 VMM and should never be enabled in
5956 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
5957 		 */
5958 		return true;
5959 	case EXIT_REASON_VMFUNC:
5960 		/* VM functions are emulated through L2->L0 vmexits. */
5961 		return true;
5962 	case EXIT_REASON_BUS_LOCK:
5963 		/*
5964 		 * At present, bus lock VM exit is never exposed to L1.
5965 		 * Handle L2's bus locks in L0 directly.
5966 		 */
5967 		return true;
5968 	default:
5969 		break;
5970 	}
5971 	return false;
5972 }
5973 
5974 /*
5975  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5976  * is_guest_mode (L2).
5977  */
5978 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
5979 				     union vmx_exit_reason exit_reason)
5980 {
5981 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5982 	u32 intr_info;
5983 
5984 	switch ((u16)exit_reason.basic) {
5985 	case EXIT_REASON_EXCEPTION_NMI:
5986 		intr_info = vmx_get_intr_info(vcpu);
5987 		if (is_nmi(intr_info))
5988 			return true;
5989 		else if (is_page_fault(intr_info))
5990 			return true;
5991 		return vmcs12->exception_bitmap &
5992 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5993 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5994 		return nested_exit_on_intr(vcpu);
5995 	case EXIT_REASON_TRIPLE_FAULT:
5996 		return true;
5997 	case EXIT_REASON_INTERRUPT_WINDOW:
5998 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5999 	case EXIT_REASON_NMI_WINDOW:
6000 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
6001 	case EXIT_REASON_TASK_SWITCH:
6002 		return true;
6003 	case EXIT_REASON_CPUID:
6004 		return true;
6005 	case EXIT_REASON_HLT:
6006 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6007 	case EXIT_REASON_INVD:
6008 		return true;
6009 	case EXIT_REASON_INVLPG:
6010 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6011 	case EXIT_REASON_RDPMC:
6012 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6013 	case EXIT_REASON_RDRAND:
6014 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
6015 	case EXIT_REASON_RDSEED:
6016 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
6017 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
6018 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6019 	case EXIT_REASON_VMREAD:
6020 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6021 			vmcs12->vmread_bitmap);
6022 	case EXIT_REASON_VMWRITE:
6023 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6024 			vmcs12->vmwrite_bitmap);
6025 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6026 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6027 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
6028 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6029 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
6030 		/*
6031 		 * VMX instructions trap unconditionally. This allows L1 to
6032 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6033 		 */
6034 		return true;
6035 	case EXIT_REASON_CR_ACCESS:
6036 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6037 	case EXIT_REASON_DR_ACCESS:
6038 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6039 	case EXIT_REASON_IO_INSTRUCTION:
6040 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
6041 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6042 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6043 	case EXIT_REASON_MSR_READ:
6044 	case EXIT_REASON_MSR_WRITE:
6045 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6046 	case EXIT_REASON_INVALID_STATE:
6047 		return true;
6048 	case EXIT_REASON_MWAIT_INSTRUCTION:
6049 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6050 	case EXIT_REASON_MONITOR_TRAP_FLAG:
6051 		return nested_vmx_exit_handled_mtf(vmcs12);
6052 	case EXIT_REASON_MONITOR_INSTRUCTION:
6053 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6054 	case EXIT_REASON_PAUSE_INSTRUCTION:
6055 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6056 			nested_cpu_has2(vmcs12,
6057 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6058 	case EXIT_REASON_MCE_DURING_VMENTRY:
6059 		return true;
6060 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6061 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6062 	case EXIT_REASON_APIC_ACCESS:
6063 	case EXIT_REASON_APIC_WRITE:
6064 	case EXIT_REASON_EOI_INDUCED:
6065 		/*
6066 		 * The controls for "virtualize APIC accesses," "APIC-
6067 		 * register virtualization," and "virtual-interrupt
6068 		 * delivery" only come from vmcs12.
6069 		 */
6070 		return true;
6071 	case EXIT_REASON_INVPCID:
6072 		return
6073 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6074 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6075 	case EXIT_REASON_WBINVD:
6076 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6077 	case EXIT_REASON_XSETBV:
6078 		return true;
6079 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6080 		/*
6081 		 * This should never happen, since it is not possible to
6082 		 * set XSS to a non-zero value---neither in L1 nor in L2.
6083 		 * If if it were, XSS would have to be checked against
6084 		 * the XSS exit bitmap in vmcs12.
6085 		 */
6086 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
6087 	case EXIT_REASON_UMWAIT:
6088 	case EXIT_REASON_TPAUSE:
6089 		return nested_cpu_has2(vmcs12,
6090 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6091 	case EXIT_REASON_ENCLS:
6092 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6093 	case EXIT_REASON_NOTIFY:
6094 		/* Notify VM exit is not exposed to L1 */
6095 		return false;
6096 	default:
6097 		return true;
6098 	}
6099 }
6100 
6101 /*
6102  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6103  * reflected into L1.
6104  */
6105 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6106 {
6107 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6108 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6109 	unsigned long exit_qual;
6110 	u32 exit_intr_info;
6111 
6112 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6113 
6114 	/*
6115 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6116 	 * has already loaded L2's state.
6117 	 */
6118 	if (unlikely(vmx->fail)) {
6119 		trace_kvm_nested_vmenter_failed(
6120 			"hardware VM-instruction error: ",
6121 			vmcs_read32(VM_INSTRUCTION_ERROR));
6122 		exit_intr_info = 0;
6123 		exit_qual = 0;
6124 		goto reflect_vmexit;
6125 	}
6126 
6127 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6128 
6129 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6130 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6131 		return false;
6132 
6133 	/* If L1 doesn't want the exit, handle it in L0. */
6134 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6135 		return false;
6136 
6137 	/*
6138 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6139 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6140 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6141 	 * interrupts are never reflected to L1 so it's a non-issue.
6142 	 */
6143 	exit_intr_info = vmx_get_intr_info(vcpu);
6144 	if (is_exception_with_error_code(exit_intr_info)) {
6145 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6146 
6147 		vmcs12->vm_exit_intr_error_code =
6148 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6149 	}
6150 	exit_qual = vmx_get_exit_qual(vcpu);
6151 
6152 reflect_vmexit:
6153 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6154 	return true;
6155 }
6156 
6157 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6158 				struct kvm_nested_state __user *user_kvm_nested_state,
6159 				u32 user_data_size)
6160 {
6161 	struct vcpu_vmx *vmx;
6162 	struct vmcs12 *vmcs12;
6163 	struct kvm_nested_state kvm_state = {
6164 		.flags = 0,
6165 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6166 		.size = sizeof(kvm_state),
6167 		.hdr.vmx.flags = 0,
6168 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6169 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6170 		.hdr.vmx.preemption_timer_deadline = 0,
6171 	};
6172 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6173 		&user_kvm_nested_state->data.vmx[0];
6174 
6175 	if (!vcpu)
6176 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6177 
6178 	vmx = to_vmx(vcpu);
6179 	vmcs12 = get_vmcs12(vcpu);
6180 
6181 	if (nested_vmx_allowed(vcpu) &&
6182 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6183 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6184 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6185 
6186 		if (vmx_has_valid_vmcs12(vcpu)) {
6187 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6188 
6189 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6190 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6191 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6192 
6193 			if (is_guest_mode(vcpu) &&
6194 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6195 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6196 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6197 		}
6198 
6199 		if (vmx->nested.smm.vmxon)
6200 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6201 
6202 		if (vmx->nested.smm.guest_mode)
6203 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6204 
6205 		if (is_guest_mode(vcpu)) {
6206 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6207 
6208 			if (vmx->nested.nested_run_pending)
6209 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6210 
6211 			if (vmx->nested.mtf_pending)
6212 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6213 
6214 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6215 			    vmx->nested.has_preemption_timer_deadline) {
6216 				kvm_state.hdr.vmx.flags |=
6217 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6218 				kvm_state.hdr.vmx.preemption_timer_deadline =
6219 					vmx->nested.preemption_timer_deadline;
6220 			}
6221 		}
6222 	}
6223 
6224 	if (user_data_size < kvm_state.size)
6225 		goto out;
6226 
6227 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6228 		return -EFAULT;
6229 
6230 	if (!vmx_has_valid_vmcs12(vcpu))
6231 		goto out;
6232 
6233 	/*
6234 	 * When running L2, the authoritative vmcs12 state is in the
6235 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6236 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6237 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6238 	 * vmcs12 state is in the vmcs12 already.
6239 	 */
6240 	if (is_guest_mode(vcpu)) {
6241 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6242 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6243 	} else  {
6244 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6245 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6246 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6247 				/*
6248 				 * L1 hypervisor is not obliged to keep eVMCS
6249 				 * clean fields data always up-to-date while
6250 				 * not in guest mode, 'hv_clean_fields' is only
6251 				 * supposed to be actual upon vmentry so we need
6252 				 * to ignore it here and do full copy.
6253 				 */
6254 				copy_enlightened_to_vmcs12(vmx, 0);
6255 			else if (enable_shadow_vmcs)
6256 				copy_shadow_to_vmcs12(vmx);
6257 		}
6258 	}
6259 
6260 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6261 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6262 
6263 	/*
6264 	 * Copy over the full allocated size of vmcs12 rather than just the size
6265 	 * of the struct.
6266 	 */
6267 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6268 		return -EFAULT;
6269 
6270 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6271 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6272 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6273 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6274 			return -EFAULT;
6275 	}
6276 out:
6277 	return kvm_state.size;
6278 }
6279 
6280 /*
6281  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6282  */
6283 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6284 {
6285 	if (is_guest_mode(vcpu)) {
6286 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6287 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6288 	}
6289 	free_nested(vcpu);
6290 }
6291 
6292 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6293 				struct kvm_nested_state __user *user_kvm_nested_state,
6294 				struct kvm_nested_state *kvm_state)
6295 {
6296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6297 	struct vmcs12 *vmcs12;
6298 	enum vm_entry_failure_code ignored;
6299 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6300 		&user_kvm_nested_state->data.vmx[0];
6301 	int ret;
6302 
6303 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6304 		return -EINVAL;
6305 
6306 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6307 		if (kvm_state->hdr.vmx.smm.flags)
6308 			return -EINVAL;
6309 
6310 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6311 			return -EINVAL;
6312 
6313 		/*
6314 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6315 		 * enable eVMCS capability on vCPU. However, since then
6316 		 * code was changed such that flag signals vmcs12 should
6317 		 * be copied into eVMCS in guest memory.
6318 		 *
6319 		 * To preserve backwards compatability, allow user
6320 		 * to set this flag even when there is no VMXON region.
6321 		 */
6322 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6323 			return -EINVAL;
6324 	} else {
6325 		if (!nested_vmx_allowed(vcpu))
6326 			return -EINVAL;
6327 
6328 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6329 			return -EINVAL;
6330 	}
6331 
6332 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6333 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6334 		return -EINVAL;
6335 
6336 	if (kvm_state->hdr.vmx.smm.flags &
6337 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6338 		return -EINVAL;
6339 
6340 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6341 		return -EINVAL;
6342 
6343 	/*
6344 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6345 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6346 	 * must be zero.
6347 	 */
6348 	if (is_smm(vcpu) ?
6349 		(kvm_state->flags &
6350 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6351 		: kvm_state->hdr.vmx.smm.flags)
6352 		return -EINVAL;
6353 
6354 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6355 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6356 		return -EINVAL;
6357 
6358 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6359 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6360 			return -EINVAL;
6361 
6362 	vmx_leave_nested(vcpu);
6363 
6364 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6365 		return 0;
6366 
6367 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6368 	ret = enter_vmx_operation(vcpu);
6369 	if (ret)
6370 		return ret;
6371 
6372 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6373 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6374 		/* See vmx_has_valid_vmcs12.  */
6375 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6376 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6377 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6378 			return -EINVAL;
6379 		else
6380 			return 0;
6381 	}
6382 
6383 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6384 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6385 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6386 			return -EINVAL;
6387 
6388 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6389 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6390 		/*
6391 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6392 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6393 		 * restored yet. EVMCS will be mapped from
6394 		 * nested_get_vmcs12_pages().
6395 		 */
6396 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6397 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6398 	} else {
6399 		return -EINVAL;
6400 	}
6401 
6402 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6403 		vmx->nested.smm.vmxon = true;
6404 		vmx->nested.vmxon = false;
6405 
6406 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6407 			vmx->nested.smm.guest_mode = true;
6408 	}
6409 
6410 	vmcs12 = get_vmcs12(vcpu);
6411 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6412 		return -EFAULT;
6413 
6414 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6415 		return -EINVAL;
6416 
6417 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6418 		return 0;
6419 
6420 	vmx->nested.nested_run_pending =
6421 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6422 
6423 	vmx->nested.mtf_pending =
6424 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6425 
6426 	ret = -EINVAL;
6427 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6428 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6429 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6430 
6431 		if (kvm_state->size <
6432 		    sizeof(*kvm_state) +
6433 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6434 			goto error_guest_mode;
6435 
6436 		if (copy_from_user(shadow_vmcs12,
6437 				   user_vmx_nested_state->shadow_vmcs12,
6438 				   sizeof(*shadow_vmcs12))) {
6439 			ret = -EFAULT;
6440 			goto error_guest_mode;
6441 		}
6442 
6443 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6444 		    !shadow_vmcs12->hdr.shadow_vmcs)
6445 			goto error_guest_mode;
6446 	}
6447 
6448 	vmx->nested.has_preemption_timer_deadline = false;
6449 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6450 		vmx->nested.has_preemption_timer_deadline = true;
6451 		vmx->nested.preemption_timer_deadline =
6452 			kvm_state->hdr.vmx.preemption_timer_deadline;
6453 	}
6454 
6455 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6456 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6457 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6458 		goto error_guest_mode;
6459 
6460 	vmx->nested.dirty_vmcs12 = true;
6461 	vmx->nested.force_msr_bitmap_recalc = true;
6462 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6463 	if (ret)
6464 		goto error_guest_mode;
6465 
6466 	return 0;
6467 
6468 error_guest_mode:
6469 	vmx->nested.nested_run_pending = 0;
6470 	return ret;
6471 }
6472 
6473 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6474 {
6475 	if (enable_shadow_vmcs) {
6476 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6477 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6478 	}
6479 }
6480 
6481 /*
6482  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6483  * that madness to get the encoding for comparison.
6484  */
6485 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6486 
6487 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6488 {
6489 	/*
6490 	 * Note these are the so called "index" of the VMCS field encoding, not
6491 	 * the index into vmcs12.
6492 	 */
6493 	unsigned int max_idx, idx;
6494 	int i;
6495 
6496 	/*
6497 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6498 	 * vmcs12, regardless of whether or not the associated feature is
6499 	 * exposed to L1.  Simply find the field with the highest index.
6500 	 */
6501 	max_idx = 0;
6502 	for (i = 0; i < nr_vmcs12_fields; i++) {
6503 		/* The vmcs12 table is very, very sparsely populated. */
6504 		if (!vmcs12_field_offsets[i])
6505 			continue;
6506 
6507 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6508 		if (idx > max_idx)
6509 			max_idx = idx;
6510 	}
6511 
6512 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6513 }
6514 
6515 /*
6516  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6517  * returned for the various VMX controls MSRs when nested VMX is enabled.
6518  * The same values should also be used to verify that vmcs12 control fields are
6519  * valid during nested entry from L1 to L2.
6520  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6521  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6522  * bit in the high half is on if the corresponding bit in the control field
6523  * may be on. See also vmx_control_verify().
6524  */
6525 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6526 {
6527 	/*
6528 	 * Note that as a general rule, the high half of the MSRs (bits in
6529 	 * the control fields which may be 1) should be initialized by the
6530 	 * intersection of the underlying hardware's MSR (i.e., features which
6531 	 * can be supported) and the list of features we want to expose -
6532 	 * because they are known to be properly supported in our code.
6533 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6534 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6535 	 * reason is that if one of these bits is necessary, it will appear
6536 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6537 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6538 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6539 	 * These rules have exceptions below.
6540 	 */
6541 
6542 	/* pin-based controls */
6543 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6544 		msrs->pinbased_ctls_low,
6545 		msrs->pinbased_ctls_high);
6546 	msrs->pinbased_ctls_low |=
6547 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6548 	msrs->pinbased_ctls_high &=
6549 		PIN_BASED_EXT_INTR_MASK |
6550 		PIN_BASED_NMI_EXITING |
6551 		PIN_BASED_VIRTUAL_NMIS |
6552 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6553 	msrs->pinbased_ctls_high |=
6554 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6555 		PIN_BASED_VMX_PREEMPTION_TIMER;
6556 
6557 	/* exit controls */
6558 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6559 		msrs->exit_ctls_low,
6560 		msrs->exit_ctls_high);
6561 	msrs->exit_ctls_low =
6562 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6563 
6564 	msrs->exit_ctls_high &=
6565 #ifdef CONFIG_X86_64
6566 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6567 #endif
6568 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6569 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6570 	msrs->exit_ctls_high |=
6571 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6572 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6573 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6574 
6575 	/* We support free control of debug control saving. */
6576 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6577 
6578 	/* entry controls */
6579 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6580 		msrs->entry_ctls_low,
6581 		msrs->entry_ctls_high);
6582 	msrs->entry_ctls_low =
6583 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6584 	msrs->entry_ctls_high &=
6585 #ifdef CONFIG_X86_64
6586 		VM_ENTRY_IA32E_MODE |
6587 #endif
6588 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6589 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6590 	msrs->entry_ctls_high |=
6591 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6592 
6593 	/* We support free control of debug control loading. */
6594 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6595 
6596 	/* cpu-based controls */
6597 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6598 		msrs->procbased_ctls_low,
6599 		msrs->procbased_ctls_high);
6600 	msrs->procbased_ctls_low =
6601 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6602 	msrs->procbased_ctls_high &=
6603 		CPU_BASED_INTR_WINDOW_EXITING |
6604 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6605 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6606 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6607 		CPU_BASED_CR3_STORE_EXITING |
6608 #ifdef CONFIG_X86_64
6609 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6610 #endif
6611 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6612 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6613 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6614 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6615 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6616 	/*
6617 	 * We can allow some features even when not supported by the
6618 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6619 	 * can use it to avoid exits to L1 - even when L0 runs L2
6620 	 * without MSR bitmaps.
6621 	 */
6622 	msrs->procbased_ctls_high |=
6623 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6624 		CPU_BASED_USE_MSR_BITMAPS;
6625 
6626 	/* We support free control of CR3 access interception. */
6627 	msrs->procbased_ctls_low &=
6628 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6629 
6630 	/*
6631 	 * secondary cpu-based controls.  Do not include those that
6632 	 * depend on CPUID bits, they are added later by
6633 	 * vmx_vcpu_after_set_cpuid.
6634 	 */
6635 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6636 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6637 		      msrs->secondary_ctls_low,
6638 		      msrs->secondary_ctls_high);
6639 
6640 	msrs->secondary_ctls_low = 0;
6641 	msrs->secondary_ctls_high &=
6642 		SECONDARY_EXEC_DESC |
6643 		SECONDARY_EXEC_ENABLE_RDTSCP |
6644 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6645 		SECONDARY_EXEC_WBINVD_EXITING |
6646 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6647 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6648 		SECONDARY_EXEC_RDRAND_EXITING |
6649 		SECONDARY_EXEC_ENABLE_INVPCID |
6650 		SECONDARY_EXEC_RDSEED_EXITING |
6651 		SECONDARY_EXEC_XSAVES |
6652 		SECONDARY_EXEC_TSC_SCALING;
6653 
6654 	/*
6655 	 * We can emulate "VMCS shadowing," even if the hardware
6656 	 * doesn't support it.
6657 	 */
6658 	msrs->secondary_ctls_high |=
6659 		SECONDARY_EXEC_SHADOW_VMCS;
6660 
6661 	if (enable_ept) {
6662 		/* nested EPT: emulate EPT also to L1 */
6663 		msrs->secondary_ctls_high |=
6664 			SECONDARY_EXEC_ENABLE_EPT;
6665 		msrs->ept_caps =
6666 			VMX_EPT_PAGE_WALK_4_BIT |
6667 			VMX_EPT_PAGE_WALK_5_BIT |
6668 			VMX_EPTP_WB_BIT |
6669 			VMX_EPT_INVEPT_BIT |
6670 			VMX_EPT_EXECUTE_ONLY_BIT;
6671 
6672 		msrs->ept_caps &= ept_caps;
6673 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6674 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6675 			VMX_EPT_1GB_PAGE_BIT;
6676 		if (enable_ept_ad_bits) {
6677 			msrs->secondary_ctls_high |=
6678 				SECONDARY_EXEC_ENABLE_PML;
6679 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6680 		}
6681 	}
6682 
6683 	if (cpu_has_vmx_vmfunc()) {
6684 		msrs->secondary_ctls_high |=
6685 			SECONDARY_EXEC_ENABLE_VMFUNC;
6686 		/*
6687 		 * Advertise EPTP switching unconditionally
6688 		 * since we emulate it
6689 		 */
6690 		if (enable_ept)
6691 			msrs->vmfunc_controls =
6692 				VMX_VMFUNC_EPTP_SWITCHING;
6693 	}
6694 
6695 	/*
6696 	 * Old versions of KVM use the single-context version without
6697 	 * checking for support, so declare that it is supported even
6698 	 * though it is treated as global context.  The alternative is
6699 	 * not failing the single-context invvpid, and it is worse.
6700 	 */
6701 	if (enable_vpid) {
6702 		msrs->secondary_ctls_high |=
6703 			SECONDARY_EXEC_ENABLE_VPID;
6704 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6705 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6706 	}
6707 
6708 	if (enable_unrestricted_guest)
6709 		msrs->secondary_ctls_high |=
6710 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6711 
6712 	if (flexpriority_enabled)
6713 		msrs->secondary_ctls_high |=
6714 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6715 
6716 	if (enable_sgx)
6717 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6718 
6719 	/* miscellaneous data */
6720 	rdmsr(MSR_IA32_VMX_MISC,
6721 		msrs->misc_low,
6722 		msrs->misc_high);
6723 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6724 	msrs->misc_low |=
6725 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6726 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6727 		VMX_MISC_ACTIVITY_HLT |
6728 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6729 	msrs->misc_high = 0;
6730 
6731 	/*
6732 	 * This MSR reports some information about VMX support. We
6733 	 * should return information about the VMX we emulate for the
6734 	 * guest, and the VMCS structure we give it - not about the
6735 	 * VMX support of the underlying hardware.
6736 	 */
6737 	msrs->basic =
6738 		VMCS12_REVISION |
6739 		VMX_BASIC_TRUE_CTLS |
6740 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6741 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6742 
6743 	if (cpu_has_vmx_basic_inout())
6744 		msrs->basic |= VMX_BASIC_INOUT;
6745 
6746 	/*
6747 	 * These MSRs specify bits which the guest must keep fixed on
6748 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6749 	 * We picked the standard core2 setting.
6750 	 */
6751 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6752 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6753 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6754 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6755 
6756 	/* These MSRs specify bits which the guest must keep fixed off. */
6757 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6758 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6759 
6760 	if (vmx_umip_emulated())
6761 		msrs->cr4_fixed1 |= X86_CR4_UMIP;
6762 
6763 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
6764 }
6765 
6766 void nested_vmx_hardware_unsetup(void)
6767 {
6768 	int i;
6769 
6770 	if (enable_shadow_vmcs) {
6771 		for (i = 0; i < VMX_BITMAP_NR; i++)
6772 			free_page((unsigned long)vmx_bitmap[i]);
6773 	}
6774 }
6775 
6776 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6777 {
6778 	int i;
6779 
6780 	if (!cpu_has_vmx_shadow_vmcs())
6781 		enable_shadow_vmcs = 0;
6782 	if (enable_shadow_vmcs) {
6783 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6784 			/*
6785 			 * The vmx_bitmap is not tied to a VM and so should
6786 			 * not be charged to a memcg.
6787 			 */
6788 			vmx_bitmap[i] = (unsigned long *)
6789 				__get_free_page(GFP_KERNEL);
6790 			if (!vmx_bitmap[i]) {
6791 				nested_vmx_hardware_unsetup();
6792 				return -ENOMEM;
6793 			}
6794 		}
6795 
6796 		init_vmcs_shadow_fields();
6797 	}
6798 
6799 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6800 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6801 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6802 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6803 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6804 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6805 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6806 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmxoff;
6807 	exit_handlers[EXIT_REASON_VMON]		= handle_vmxon;
6808 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6809 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6810 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6811 
6812 	return 0;
6813 }
6814 
6815 struct kvm_x86_nested_ops vmx_nested_ops = {
6816 	.leave_nested = vmx_leave_nested,
6817 	.check_events = vmx_check_nested_events,
6818 	.handle_page_fault_workaround = nested_vmx_handle_page_fault_workaround,
6819 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6820 	.triple_fault = nested_vmx_triple_fault,
6821 	.get_state = vmx_get_nested_state,
6822 	.set_state = vmx_set_nested_state,
6823 	.get_nested_state_pages = vmx_get_nested_state_pages,
6824 	.write_log_dirty = nested_vmx_write_pml_buffer,
6825 	.enable_evmcs = nested_enable_evmcs,
6826 	.get_evmcs_version = nested_get_evmcs_version,
6827 };
6828