xref: /linux/arch/x86/kvm/vmx/nested.c (revision e0c1b49f5b674cca7b10549c53b3791d0bbc90a8)
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 "hyperv.h"
11 #include "mmu.h"
12 #include "nested.h"
13 #include "pmu.h"
14 #include "sgx.h"
15 #include "trace.h"
16 #include "vmx.h"
17 #include "x86.h"
18 
19 static bool __read_mostly enable_shadow_vmcs = 1;
20 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
21 
22 static bool __read_mostly nested_early_check = 0;
23 module_param(nested_early_check, bool, S_IRUGO);
24 
25 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
26 
27 /*
28  * Hyper-V requires all of these, so mark them as supported even though
29  * they are just treated the same as all-context.
30  */
31 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
32 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
33 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
34 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
35 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
36 
37 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
38 
39 enum {
40 	VMX_VMREAD_BITMAP,
41 	VMX_VMWRITE_BITMAP,
42 	VMX_BITMAP_NR
43 };
44 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
45 
46 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
47 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
48 
49 struct shadow_vmcs_field {
50 	u16	encoding;
51 	u16	offset;
52 };
53 static struct shadow_vmcs_field shadow_read_only_fields[] = {
54 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
55 #include "vmcs_shadow_fields.h"
56 };
57 static int max_shadow_read_only_fields =
58 	ARRAY_SIZE(shadow_read_only_fields);
59 
60 static struct shadow_vmcs_field shadow_read_write_fields[] = {
61 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
62 #include "vmcs_shadow_fields.h"
63 };
64 static int max_shadow_read_write_fields =
65 	ARRAY_SIZE(shadow_read_write_fields);
66 
67 static void init_vmcs_shadow_fields(void)
68 {
69 	int i, j;
70 
71 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
72 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
73 
74 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
75 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
76 		u16 field = entry.encoding;
77 
78 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
79 		    (i + 1 == max_shadow_read_only_fields ||
80 		     shadow_read_only_fields[i + 1].encoding != field + 1))
81 			pr_err("Missing field from shadow_read_only_field %x\n",
82 			       field + 1);
83 
84 		clear_bit(field, vmx_vmread_bitmap);
85 		if (field & 1)
86 #ifdef CONFIG_X86_64
87 			continue;
88 #else
89 			entry.offset += sizeof(u32);
90 #endif
91 		shadow_read_only_fields[j++] = entry;
92 	}
93 	max_shadow_read_only_fields = j;
94 
95 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
96 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
97 		u16 field = entry.encoding;
98 
99 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
100 		    (i + 1 == max_shadow_read_write_fields ||
101 		     shadow_read_write_fields[i + 1].encoding != field + 1))
102 			pr_err("Missing field from shadow_read_write_field %x\n",
103 			       field + 1);
104 
105 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
106 			  field <= GUEST_TR_AR_BYTES,
107 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
108 
109 		/*
110 		 * PML and the preemption timer can be emulated, but the
111 		 * processor cannot vmwrite to fields that don't exist
112 		 * on bare metal.
113 		 */
114 		switch (field) {
115 		case GUEST_PML_INDEX:
116 			if (!cpu_has_vmx_pml())
117 				continue;
118 			break;
119 		case VMX_PREEMPTION_TIMER_VALUE:
120 			if (!cpu_has_vmx_preemption_timer())
121 				continue;
122 			break;
123 		case GUEST_INTR_STATUS:
124 			if (!cpu_has_vmx_apicv())
125 				continue;
126 			break;
127 		default:
128 			break;
129 		}
130 
131 		clear_bit(field, vmx_vmwrite_bitmap);
132 		clear_bit(field, vmx_vmread_bitmap);
133 		if (field & 1)
134 #ifdef CONFIG_X86_64
135 			continue;
136 #else
137 			entry.offset += sizeof(u32);
138 #endif
139 		shadow_read_write_fields[j++] = entry;
140 	}
141 	max_shadow_read_write_fields = j;
142 }
143 
144 /*
145  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
146  * set the success or error code of an emulated VMX instruction (as specified
147  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
148  * instruction.
149  */
150 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
151 {
152 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
153 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
154 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
155 	return kvm_skip_emulated_instruction(vcpu);
156 }
157 
158 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
159 {
160 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
161 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
162 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
163 			| X86_EFLAGS_CF);
164 	return kvm_skip_emulated_instruction(vcpu);
165 }
166 
167 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
168 				u32 vm_instruction_error)
169 {
170 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
171 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
172 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
173 			| X86_EFLAGS_ZF);
174 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
175 	/*
176 	 * We don't need to force sync to shadow VMCS because
177 	 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
178 	 * fields and thus must be synced.
179 	 */
180 	if (to_vmx(vcpu)->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
181 		to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
182 
183 	return kvm_skip_emulated_instruction(vcpu);
184 }
185 
186 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
187 {
188 	struct vcpu_vmx *vmx = to_vmx(vcpu);
189 
190 	/*
191 	 * failValid writes the error number to the current VMCS, which
192 	 * can't be done if there isn't a current VMCS.
193 	 */
194 	if (vmx->nested.current_vmptr == INVALID_GPA &&
195 	    !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
196 		return nested_vmx_failInvalid(vcpu);
197 
198 	return nested_vmx_failValid(vcpu, vm_instruction_error);
199 }
200 
201 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
202 {
203 	/* TODO: not to reset guest simply here. */
204 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
205 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
206 }
207 
208 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
209 {
210 	return fixed_bits_valid(control, low, high);
211 }
212 
213 static inline u64 vmx_control_msr(u32 low, u32 high)
214 {
215 	return low | ((u64)high << 32);
216 }
217 
218 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
219 {
220 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
221 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
222 	vmx->nested.need_vmcs12_to_shadow_sync = false;
223 }
224 
225 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
226 {
227 	struct vcpu_vmx *vmx = to_vmx(vcpu);
228 
229 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
230 		kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
231 		vmx->nested.hv_evmcs = NULL;
232 	}
233 
234 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
235 }
236 
237 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
238 				     struct loaded_vmcs *prev)
239 {
240 	struct vmcs_host_state *dest, *src;
241 
242 	if (unlikely(!vmx->guest_state_loaded))
243 		return;
244 
245 	src = &prev->host_state;
246 	dest = &vmx->loaded_vmcs->host_state;
247 
248 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
249 	dest->ldt_sel = src->ldt_sel;
250 #ifdef CONFIG_X86_64
251 	dest->ds_sel = src->ds_sel;
252 	dest->es_sel = src->es_sel;
253 #endif
254 }
255 
256 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
257 {
258 	struct vcpu_vmx *vmx = to_vmx(vcpu);
259 	struct loaded_vmcs *prev;
260 	int cpu;
261 
262 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
263 		return;
264 
265 	cpu = get_cpu();
266 	prev = vmx->loaded_vmcs;
267 	vmx->loaded_vmcs = vmcs;
268 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
269 	vmx_sync_vmcs_host_state(vmx, prev);
270 	put_cpu();
271 
272 	vmx_register_cache_reset(vcpu);
273 }
274 
275 /*
276  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
277  * just stops using VMX.
278  */
279 static void free_nested(struct kvm_vcpu *vcpu)
280 {
281 	struct vcpu_vmx *vmx = to_vmx(vcpu);
282 
283 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
284 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
285 
286 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
287 		return;
288 
289 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
290 
291 	vmx->nested.vmxon = false;
292 	vmx->nested.smm.vmxon = false;
293 	vmx->nested.vmxon_ptr = INVALID_GPA;
294 	free_vpid(vmx->nested.vpid02);
295 	vmx->nested.posted_intr_nv = -1;
296 	vmx->nested.current_vmptr = INVALID_GPA;
297 	if (enable_shadow_vmcs) {
298 		vmx_disable_shadow_vmcs(vmx);
299 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
300 		free_vmcs(vmx->vmcs01.shadow_vmcs);
301 		vmx->vmcs01.shadow_vmcs = NULL;
302 	}
303 	kfree(vmx->nested.cached_vmcs12);
304 	vmx->nested.cached_vmcs12 = NULL;
305 	kfree(vmx->nested.cached_shadow_vmcs12);
306 	vmx->nested.cached_shadow_vmcs12 = NULL;
307 	/* Unpin physical memory we referred to in the vmcs02 */
308 	if (vmx->nested.apic_access_page) {
309 		kvm_release_page_clean(vmx->nested.apic_access_page);
310 		vmx->nested.apic_access_page = NULL;
311 	}
312 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
313 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
314 	vmx->nested.pi_desc = NULL;
315 
316 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
317 
318 	nested_release_evmcs(vcpu);
319 
320 	free_loaded_vmcs(&vmx->nested.vmcs02);
321 }
322 
323 /*
324  * Ensure that the current vmcs of the logical processor is the
325  * vmcs01 of the vcpu before calling free_nested().
326  */
327 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
328 {
329 	vcpu_load(vcpu);
330 	vmx_leave_nested(vcpu);
331 	vcpu_put(vcpu);
332 }
333 
334 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
335 
336 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
337 {
338 	return VALID_PAGE(root_hpa) &&
339 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
340 }
341 
342 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
343 				       gpa_t addr)
344 {
345 	uint i;
346 	struct kvm_mmu_root_info *cached_root;
347 
348 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
349 
350 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
351 		cached_root = &vcpu->arch.mmu->prev_roots[i];
352 
353 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
354 					    eptp))
355 			vcpu->arch.mmu->invlpg(vcpu, addr, cached_root->hpa);
356 	}
357 }
358 
359 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
360 		struct x86_exception *fault)
361 {
362 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
363 	struct vcpu_vmx *vmx = to_vmx(vcpu);
364 	u32 vm_exit_reason;
365 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
366 
367 	if (vmx->nested.pml_full) {
368 		vm_exit_reason = EXIT_REASON_PML_FULL;
369 		vmx->nested.pml_full = false;
370 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
371 	} else {
372 		if (fault->error_code & PFERR_RSVD_MASK)
373 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
374 		else
375 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
376 
377 		/*
378 		 * Although the caller (kvm_inject_emulated_page_fault) would
379 		 * have already synced the faulting address in the shadow EPT
380 		 * tables for the current EPTP12, we also need to sync it for
381 		 * any other cached EPTP02s based on the same EP4TA, since the
382 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
383 		 */
384 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
385 					   fault->address);
386 	}
387 
388 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
389 	vmcs12->guest_physical_address = fault->address;
390 }
391 
392 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
393 {
394 	kvm_init_shadow_ept_mmu(vcpu,
395 				to_vmx(vcpu)->nested.msrs.ept_caps &
396 				VMX_EPT_EXECUTE_ONLY_BIT,
397 				nested_ept_ad_enabled(vcpu),
398 				nested_ept_get_eptp(vcpu));
399 }
400 
401 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
402 {
403 	WARN_ON(mmu_is_nested(vcpu));
404 
405 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
406 	nested_ept_new_eptp(vcpu);
407 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
408 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
409 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
410 
411 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
412 }
413 
414 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
415 {
416 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
417 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
418 }
419 
420 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
421 					    u16 error_code)
422 {
423 	bool inequality, bit;
424 
425 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
426 	inequality =
427 		(error_code & vmcs12->page_fault_error_code_mask) !=
428 		 vmcs12->page_fault_error_code_match;
429 	return inequality ^ bit;
430 }
431 
432 
433 /*
434  * KVM wants to inject page-faults which it got to the guest. This function
435  * checks whether in a nested guest, we need to inject them to L1 or L2.
436  */
437 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
438 {
439 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
440 	unsigned int nr = vcpu->arch.exception.nr;
441 	bool has_payload = vcpu->arch.exception.has_payload;
442 	unsigned long payload = vcpu->arch.exception.payload;
443 
444 	if (nr == PF_VECTOR) {
445 		if (vcpu->arch.exception.nested_apf) {
446 			*exit_qual = vcpu->arch.apf.nested_apf_token;
447 			return 1;
448 		}
449 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
450 						    vcpu->arch.exception.error_code)) {
451 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
452 			return 1;
453 		}
454 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
455 		if (nr == DB_VECTOR) {
456 			if (!has_payload) {
457 				payload = vcpu->arch.dr6;
458 				payload &= ~DR6_BT;
459 				payload ^= DR6_ACTIVE_LOW;
460 			}
461 			*exit_qual = payload;
462 		} else
463 			*exit_qual = 0;
464 		return 1;
465 	}
466 
467 	return 0;
468 }
469 
470 
471 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
472 		struct x86_exception *fault)
473 {
474 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
475 
476 	WARN_ON(!is_guest_mode(vcpu));
477 
478 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
479 		!to_vmx(vcpu)->nested.nested_run_pending) {
480 		vmcs12->vm_exit_intr_error_code = fault->error_code;
481 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
482 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
483 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
484 				  fault->address);
485 	} else {
486 		kvm_inject_page_fault(vcpu, fault);
487 	}
488 }
489 
490 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
491 					       struct vmcs12 *vmcs12)
492 {
493 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
494 		return 0;
495 
496 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
497 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
498 		return -EINVAL;
499 
500 	return 0;
501 }
502 
503 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
504 						struct vmcs12 *vmcs12)
505 {
506 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
507 		return 0;
508 
509 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
510 		return -EINVAL;
511 
512 	return 0;
513 }
514 
515 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
516 						struct vmcs12 *vmcs12)
517 {
518 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
519 		return 0;
520 
521 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
522 		return -EINVAL;
523 
524 	return 0;
525 }
526 
527 /*
528  * Check if MSR is intercepted for L01 MSR bitmap.
529  */
530 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
531 {
532 	unsigned long *msr_bitmap;
533 	int f = sizeof(unsigned long);
534 
535 	if (!cpu_has_vmx_msr_bitmap())
536 		return true;
537 
538 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
539 
540 	if (msr <= 0x1fff) {
541 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
542 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
543 		msr &= 0x1fff;
544 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
545 	}
546 
547 	return true;
548 }
549 
550 /*
551  * If a msr is allowed by L0, we should check whether it is allowed by L1.
552  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
553  */
554 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
555 					       unsigned long *msr_bitmap_nested,
556 					       u32 msr, int type)
557 {
558 	int f = sizeof(unsigned long);
559 
560 	/*
561 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
562 	 * have the write-low and read-high bitmap offsets the wrong way round.
563 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
564 	 */
565 	if (msr <= 0x1fff) {
566 		if (type & MSR_TYPE_R &&
567 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
568 			/* read-low */
569 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
570 
571 		if (type & MSR_TYPE_W &&
572 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
573 			/* write-low */
574 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
575 
576 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
577 		msr &= 0x1fff;
578 		if (type & MSR_TYPE_R &&
579 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
580 			/* read-high */
581 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
582 
583 		if (type & MSR_TYPE_W &&
584 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
585 			/* write-high */
586 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
587 
588 	}
589 }
590 
591 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
592 {
593 	int msr;
594 
595 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
596 		unsigned word = msr / BITS_PER_LONG;
597 
598 		msr_bitmap[word] = ~0;
599 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
600 	}
601 }
602 
603 /*
604  * Merge L0's and L1's MSR bitmap, return false to indicate that
605  * we do not use the hardware.
606  */
607 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
608 						 struct vmcs12 *vmcs12)
609 {
610 	int msr;
611 	unsigned long *msr_bitmap_l1;
612 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
613 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
614 
615 	/* Nothing to do if the MSR bitmap is not in use.  */
616 	if (!cpu_has_vmx_msr_bitmap() ||
617 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
618 		return false;
619 
620 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
621 		return false;
622 
623 	msr_bitmap_l1 = (unsigned long *)map->hva;
624 
625 	/*
626 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
627 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
628 	 * the x2APIC MSR range and selectively disable them below.
629 	 */
630 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
631 
632 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
633 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
634 			/*
635 			 * L0 need not intercept reads for MSRs between 0x800
636 			 * and 0x8ff, it just lets the processor take the value
637 			 * from the virtual-APIC page; take those 256 bits
638 			 * directly from the L1 bitmap.
639 			 */
640 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
641 				unsigned word = msr / BITS_PER_LONG;
642 
643 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
644 			}
645 		}
646 
647 		nested_vmx_disable_intercept_for_msr(
648 			msr_bitmap_l1, msr_bitmap_l0,
649 			X2APIC_MSR(APIC_TASKPRI),
650 			MSR_TYPE_R | MSR_TYPE_W);
651 
652 		if (nested_cpu_has_vid(vmcs12)) {
653 			nested_vmx_disable_intercept_for_msr(
654 				msr_bitmap_l1, msr_bitmap_l0,
655 				X2APIC_MSR(APIC_EOI),
656 				MSR_TYPE_W);
657 			nested_vmx_disable_intercept_for_msr(
658 				msr_bitmap_l1, msr_bitmap_l0,
659 				X2APIC_MSR(APIC_SELF_IPI),
660 				MSR_TYPE_W);
661 		}
662 	}
663 
664 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
665 #ifdef CONFIG_X86_64
666 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
667 					     MSR_FS_BASE, MSR_TYPE_RW);
668 
669 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
670 					     MSR_GS_BASE, MSR_TYPE_RW);
671 
672 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
673 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
674 #endif
675 
676 	/*
677 	 * Checking the L0->L1 bitmap is trying to verify two things:
678 	 *
679 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
680 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
681 	 *    from the L12 MSR bitmap that is too permissive.
682 	 * 2. That L1 or L2s have actually used the MSR. This avoids
683 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
684 	 *    works properly because we only update the L01 MSR bitmap lazily.
685 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
686 	 *    updated to reflect this when L1 (or its L2s) actually write to
687 	 *    the MSR.
688 	 */
689 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
690 		nested_vmx_disable_intercept_for_msr(
691 					msr_bitmap_l1, msr_bitmap_l0,
692 					MSR_IA32_SPEC_CTRL,
693 					MSR_TYPE_R | MSR_TYPE_W);
694 
695 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
696 		nested_vmx_disable_intercept_for_msr(
697 					msr_bitmap_l1, msr_bitmap_l0,
698 					MSR_IA32_PRED_CMD,
699 					MSR_TYPE_W);
700 
701 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
702 
703 	return true;
704 }
705 
706 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
707 				       struct vmcs12 *vmcs12)
708 {
709 	struct kvm_host_map map;
710 	struct vmcs12 *shadow;
711 
712 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
713 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
714 		return;
715 
716 	shadow = get_shadow_vmcs12(vcpu);
717 
718 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
719 		return;
720 
721 	memcpy(shadow, map.hva, VMCS12_SIZE);
722 	kvm_vcpu_unmap(vcpu, &map, false);
723 }
724 
725 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
726 					      struct vmcs12 *vmcs12)
727 {
728 	struct vcpu_vmx *vmx = to_vmx(vcpu);
729 
730 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
731 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
732 		return;
733 
734 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
735 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
736 }
737 
738 /*
739  * In nested virtualization, check if L1 has set
740  * VM_EXIT_ACK_INTR_ON_EXIT
741  */
742 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
743 {
744 	return get_vmcs12(vcpu)->vm_exit_controls &
745 		VM_EXIT_ACK_INTR_ON_EXIT;
746 }
747 
748 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
749 					  struct vmcs12 *vmcs12)
750 {
751 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
752 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
753 		return -EINVAL;
754 	else
755 		return 0;
756 }
757 
758 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
759 					   struct vmcs12 *vmcs12)
760 {
761 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
762 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
763 	    !nested_cpu_has_vid(vmcs12) &&
764 	    !nested_cpu_has_posted_intr(vmcs12))
765 		return 0;
766 
767 	/*
768 	 * If virtualize x2apic mode is enabled,
769 	 * virtualize apic access must be disabled.
770 	 */
771 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
772 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
773 		return -EINVAL;
774 
775 	/*
776 	 * If virtual interrupt delivery is enabled,
777 	 * we must exit on external interrupts.
778 	 */
779 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
780 		return -EINVAL;
781 
782 	/*
783 	 * bits 15:8 should be zero in posted_intr_nv,
784 	 * the descriptor address has been already checked
785 	 * in nested_get_vmcs12_pages.
786 	 *
787 	 * bits 5:0 of posted_intr_desc_addr should be zero.
788 	 */
789 	if (nested_cpu_has_posted_intr(vmcs12) &&
790 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
791 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
792 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
793 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
794 		return -EINVAL;
795 
796 	/* tpr shadow is needed by all apicv features. */
797 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
798 		return -EINVAL;
799 
800 	return 0;
801 }
802 
803 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
804 				       u32 count, u64 addr)
805 {
806 	if (count == 0)
807 		return 0;
808 
809 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
810 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
811 		return -EINVAL;
812 
813 	return 0;
814 }
815 
816 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
817 						     struct vmcs12 *vmcs12)
818 {
819 	if (CC(nested_vmx_check_msr_switch(vcpu,
820 					   vmcs12->vm_exit_msr_load_count,
821 					   vmcs12->vm_exit_msr_load_addr)) ||
822 	    CC(nested_vmx_check_msr_switch(vcpu,
823 					   vmcs12->vm_exit_msr_store_count,
824 					   vmcs12->vm_exit_msr_store_addr)))
825 		return -EINVAL;
826 
827 	return 0;
828 }
829 
830 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
831                                                       struct vmcs12 *vmcs12)
832 {
833 	if (CC(nested_vmx_check_msr_switch(vcpu,
834 					   vmcs12->vm_entry_msr_load_count,
835 					   vmcs12->vm_entry_msr_load_addr)))
836                 return -EINVAL;
837 
838 	return 0;
839 }
840 
841 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
842 					 struct vmcs12 *vmcs12)
843 {
844 	if (!nested_cpu_has_pml(vmcs12))
845 		return 0;
846 
847 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
848 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
849 		return -EINVAL;
850 
851 	return 0;
852 }
853 
854 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
855 							struct vmcs12 *vmcs12)
856 {
857 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
858 	       !nested_cpu_has_ept(vmcs12)))
859 		return -EINVAL;
860 	return 0;
861 }
862 
863 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
864 							 struct vmcs12 *vmcs12)
865 {
866 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
867 	       !nested_cpu_has_ept(vmcs12)))
868 		return -EINVAL;
869 	return 0;
870 }
871 
872 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
873 						 struct vmcs12 *vmcs12)
874 {
875 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
876 		return 0;
877 
878 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
879 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
880 		return -EINVAL;
881 
882 	return 0;
883 }
884 
885 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
886 				       struct vmx_msr_entry *e)
887 {
888 	/* x2APIC MSR accesses are not allowed */
889 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
890 		return -EINVAL;
891 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
892 	    CC(e->index == MSR_IA32_UCODE_REV))
893 		return -EINVAL;
894 	if (CC(e->reserved != 0))
895 		return -EINVAL;
896 	return 0;
897 }
898 
899 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
900 				     struct vmx_msr_entry *e)
901 {
902 	if (CC(e->index == MSR_FS_BASE) ||
903 	    CC(e->index == MSR_GS_BASE) ||
904 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
905 	    nested_vmx_msr_check_common(vcpu, e))
906 		return -EINVAL;
907 	return 0;
908 }
909 
910 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
911 				      struct vmx_msr_entry *e)
912 {
913 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
914 	    nested_vmx_msr_check_common(vcpu, e))
915 		return -EINVAL;
916 	return 0;
917 }
918 
919 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
920 {
921 	struct vcpu_vmx *vmx = to_vmx(vcpu);
922 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
923 				       vmx->nested.msrs.misc_high);
924 
925 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
926 }
927 
928 /*
929  * Load guest's/host's msr at nested entry/exit.
930  * return 0 for success, entry index for failure.
931  *
932  * One of the failure modes for MSR load/store is when a list exceeds the
933  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
934  * as possible, process all valid entries before failing rather than precheck
935  * for a capacity violation.
936  */
937 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
938 {
939 	u32 i;
940 	struct vmx_msr_entry e;
941 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
942 
943 	for (i = 0; i < count; i++) {
944 		if (unlikely(i >= max_msr_list_size))
945 			goto fail;
946 
947 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
948 					&e, sizeof(e))) {
949 			pr_debug_ratelimited(
950 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
951 				__func__, i, gpa + i * sizeof(e));
952 			goto fail;
953 		}
954 		if (nested_vmx_load_msr_check(vcpu, &e)) {
955 			pr_debug_ratelimited(
956 				"%s check failed (%u, 0x%x, 0x%x)\n",
957 				__func__, i, e.index, e.reserved);
958 			goto fail;
959 		}
960 		if (kvm_set_msr(vcpu, e.index, e.value)) {
961 			pr_debug_ratelimited(
962 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
963 				__func__, i, e.index, e.value);
964 			goto fail;
965 		}
966 	}
967 	return 0;
968 fail:
969 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
970 	return i + 1;
971 }
972 
973 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
974 					    u32 msr_index,
975 					    u64 *data)
976 {
977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
978 
979 	/*
980 	 * If the L0 hypervisor stored a more accurate value for the TSC that
981 	 * does not include the time taken for emulation of the L2->L1
982 	 * VM-exit in L0, use the more accurate value.
983 	 */
984 	if (msr_index == MSR_IA32_TSC) {
985 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
986 						    MSR_IA32_TSC);
987 
988 		if (i >= 0) {
989 			u64 val = vmx->msr_autostore.guest.val[i].value;
990 
991 			*data = kvm_read_l1_tsc(vcpu, val);
992 			return true;
993 		}
994 	}
995 
996 	if (kvm_get_msr(vcpu, msr_index, data)) {
997 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
998 			msr_index);
999 		return false;
1000 	}
1001 	return true;
1002 }
1003 
1004 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
1005 				     struct vmx_msr_entry *e)
1006 {
1007 	if (kvm_vcpu_read_guest(vcpu,
1008 				gpa + i * sizeof(*e),
1009 				e, 2 * sizeof(u32))) {
1010 		pr_debug_ratelimited(
1011 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
1012 			__func__, i, gpa + i * sizeof(*e));
1013 		return false;
1014 	}
1015 	if (nested_vmx_store_msr_check(vcpu, e)) {
1016 		pr_debug_ratelimited(
1017 			"%s check failed (%u, 0x%x, 0x%x)\n",
1018 			__func__, i, e->index, e->reserved);
1019 		return false;
1020 	}
1021 	return true;
1022 }
1023 
1024 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1025 {
1026 	u64 data;
1027 	u32 i;
1028 	struct vmx_msr_entry e;
1029 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1030 
1031 	for (i = 0; i < count; i++) {
1032 		if (unlikely(i >= max_msr_list_size))
1033 			return -EINVAL;
1034 
1035 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1036 			return -EINVAL;
1037 
1038 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1039 			return -EINVAL;
1040 
1041 		if (kvm_vcpu_write_guest(vcpu,
1042 					 gpa + i * sizeof(e) +
1043 					     offsetof(struct vmx_msr_entry, value),
1044 					 &data, sizeof(data))) {
1045 			pr_debug_ratelimited(
1046 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1047 				__func__, i, e.index, data);
1048 			return -EINVAL;
1049 		}
1050 	}
1051 	return 0;
1052 }
1053 
1054 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1055 {
1056 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1057 	u32 count = vmcs12->vm_exit_msr_store_count;
1058 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1059 	struct vmx_msr_entry e;
1060 	u32 i;
1061 
1062 	for (i = 0; i < count; i++) {
1063 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1064 			return false;
1065 
1066 		if (e.index == msr_index)
1067 			return true;
1068 	}
1069 	return false;
1070 }
1071 
1072 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1073 					   u32 msr_index)
1074 {
1075 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1076 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1077 	bool in_vmcs12_store_list;
1078 	int msr_autostore_slot;
1079 	bool in_autostore_list;
1080 	int last;
1081 
1082 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1083 	in_autostore_list = msr_autostore_slot >= 0;
1084 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1085 
1086 	if (in_vmcs12_store_list && !in_autostore_list) {
1087 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1088 			/*
1089 			 * Emulated VMEntry does not fail here.  Instead a less
1090 			 * accurate value will be returned by
1091 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1092 			 * instead of reading the value from the vmcs02 VMExit
1093 			 * MSR-store area.
1094 			 */
1095 			pr_warn_ratelimited(
1096 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1097 				msr_index);
1098 			return;
1099 		}
1100 		last = autostore->nr++;
1101 		autostore->val[last].index = msr_index;
1102 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1103 		last = --autostore->nr;
1104 		autostore->val[msr_autostore_slot] = autostore->val[last];
1105 	}
1106 }
1107 
1108 /*
1109  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1110  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1111  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1112  * @entry_failure_code.
1113  */
1114 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1115 			       bool nested_ept, bool reload_pdptrs,
1116 			       enum vm_entry_failure_code *entry_failure_code)
1117 {
1118 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1119 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1120 		return -EINVAL;
1121 	}
1122 
1123 	/*
1124 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1125 	 * must not be dereferenced.
1126 	 */
1127 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1128 	    CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1129 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1130 		return -EINVAL;
1131 	}
1132 
1133 	if (!nested_ept)
1134 		kvm_mmu_new_pgd(vcpu, cr3);
1135 
1136 	vcpu->arch.cr3 = cr3;
1137 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1138 
1139 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1140 	kvm_init_mmu(vcpu);
1141 
1142 	return 0;
1143 }
1144 
1145 /*
1146  * Returns if KVM is able to config CPU to tag TLB entries
1147  * populated by L2 differently than TLB entries populated
1148  * by L1.
1149  *
1150  * If L0 uses EPT, L1 and L2 run with different EPTP because
1151  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1152  * are tagged with different EPTP.
1153  *
1154  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1155  * with different VPID (L1 entries are tagged with vmx->vpid
1156  * while L2 entries are tagged with vmx->nested.vpid02).
1157  */
1158 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1159 {
1160 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1161 
1162 	return enable_ept ||
1163 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1164 }
1165 
1166 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1167 					    struct vmcs12 *vmcs12,
1168 					    bool is_vmenter)
1169 {
1170 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1171 
1172 	/*
1173 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1174 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1175 	 * full TLB flush from the guest's perspective.  This is required even
1176 	 * if VPID is disabled in the host as KVM may need to synchronize the
1177 	 * MMU in response to the guest TLB flush.
1178 	 *
1179 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1180 	 * EPT is a special snowflake, as guest-physical mappings aren't
1181 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1182 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1183 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1184 	 * those mappings.
1185 	 */
1186 	if (!nested_cpu_has_vpid(vmcs12)) {
1187 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1188 		return;
1189 	}
1190 
1191 	/* L2 should never have a VPID if VPID is disabled. */
1192 	WARN_ON(!enable_vpid);
1193 
1194 	/*
1195 	 * If VPID is enabled and used by vmc12, but L2 does not have a unique
1196 	 * TLB tag (ASID), i.e. EPT is disabled and KVM was unable to allocate
1197 	 * a VPID for L2, flush the current context as the effective ASID is
1198 	 * common to both L1 and L2.
1199 	 *
1200 	 * Defer the flush so that it runs after vmcs02.EPTP has been set by
1201 	 * KVM_REQ_LOAD_MMU_PGD (if nested EPT is enabled) and to avoid
1202 	 * redundant flushes further down the nested pipeline.
1203 	 *
1204 	 * If a TLB flush isn't required due to any of the above, and vpid12 is
1205 	 * changing then the new "virtual" VPID (vpid12) will reuse the same
1206 	 * "real" VPID (vpid02), and so needs to be flushed.  There's no direct
1207 	 * mapping between vpid02 and vpid12, vpid02 is per-vCPU and reused for
1208 	 * all nested vCPUs.  Remember, a flush on VM-Enter does not invalidate
1209 	 * guest-physical mappings, so there is no need to sync the nEPT MMU.
1210 	 */
1211 	if (!nested_has_guest_tlb_tag(vcpu)) {
1212 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1213 	} else if (is_vmenter &&
1214 		   vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1215 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1216 		vpid_sync_context(nested_get_vpid02(vcpu));
1217 	}
1218 }
1219 
1220 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1221 {
1222 	superset &= mask;
1223 	subset &= mask;
1224 
1225 	return (superset | subset) == superset;
1226 }
1227 
1228 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1229 {
1230 	const u64 feature_and_reserved =
1231 		/* feature (except bit 48; see below) */
1232 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1233 		/* reserved */
1234 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1235 	u64 vmx_basic = vmx->nested.msrs.basic;
1236 
1237 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1238 		return -EINVAL;
1239 
1240 	/*
1241 	 * KVM does not emulate a version of VMX that constrains physical
1242 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1243 	 */
1244 	if (data & BIT_ULL(48))
1245 		return -EINVAL;
1246 
1247 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1248 	    vmx_basic_vmcs_revision_id(data))
1249 		return -EINVAL;
1250 
1251 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1252 		return -EINVAL;
1253 
1254 	vmx->nested.msrs.basic = data;
1255 	return 0;
1256 }
1257 
1258 static int
1259 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1260 {
1261 	u64 supported;
1262 	u32 *lowp, *highp;
1263 
1264 	switch (msr_index) {
1265 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1266 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1267 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1268 		break;
1269 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1270 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1271 		highp = &vmx->nested.msrs.procbased_ctls_high;
1272 		break;
1273 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1274 		lowp = &vmx->nested.msrs.exit_ctls_low;
1275 		highp = &vmx->nested.msrs.exit_ctls_high;
1276 		break;
1277 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1278 		lowp = &vmx->nested.msrs.entry_ctls_low;
1279 		highp = &vmx->nested.msrs.entry_ctls_high;
1280 		break;
1281 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1282 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1283 		highp = &vmx->nested.msrs.secondary_ctls_high;
1284 		break;
1285 	default:
1286 		BUG();
1287 	}
1288 
1289 	supported = vmx_control_msr(*lowp, *highp);
1290 
1291 	/* Check must-be-1 bits are still 1. */
1292 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1293 		return -EINVAL;
1294 
1295 	/* Check must-be-0 bits are still 0. */
1296 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1297 		return -EINVAL;
1298 
1299 	*lowp = data;
1300 	*highp = data >> 32;
1301 	return 0;
1302 }
1303 
1304 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1305 {
1306 	const u64 feature_and_reserved_bits =
1307 		/* feature */
1308 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1309 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1310 		/* reserved */
1311 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1312 	u64 vmx_misc;
1313 
1314 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1315 				   vmx->nested.msrs.misc_high);
1316 
1317 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1318 		return -EINVAL;
1319 
1320 	if ((vmx->nested.msrs.pinbased_ctls_high &
1321 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1322 	    vmx_misc_preemption_timer_rate(data) !=
1323 	    vmx_misc_preemption_timer_rate(vmx_misc))
1324 		return -EINVAL;
1325 
1326 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1327 		return -EINVAL;
1328 
1329 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1330 		return -EINVAL;
1331 
1332 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1333 		return -EINVAL;
1334 
1335 	vmx->nested.msrs.misc_low = data;
1336 	vmx->nested.msrs.misc_high = data >> 32;
1337 
1338 	return 0;
1339 }
1340 
1341 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1342 {
1343 	u64 vmx_ept_vpid_cap;
1344 
1345 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1346 					   vmx->nested.msrs.vpid_caps);
1347 
1348 	/* Every bit is either reserved or a feature bit. */
1349 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1350 		return -EINVAL;
1351 
1352 	vmx->nested.msrs.ept_caps = data;
1353 	vmx->nested.msrs.vpid_caps = data >> 32;
1354 	return 0;
1355 }
1356 
1357 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1358 {
1359 	u64 *msr;
1360 
1361 	switch (msr_index) {
1362 	case MSR_IA32_VMX_CR0_FIXED0:
1363 		msr = &vmx->nested.msrs.cr0_fixed0;
1364 		break;
1365 	case MSR_IA32_VMX_CR4_FIXED0:
1366 		msr = &vmx->nested.msrs.cr4_fixed0;
1367 		break;
1368 	default:
1369 		BUG();
1370 	}
1371 
1372 	/*
1373 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1374 	 * must be 1 in the restored value.
1375 	 */
1376 	if (!is_bitwise_subset(data, *msr, -1ULL))
1377 		return -EINVAL;
1378 
1379 	*msr = data;
1380 	return 0;
1381 }
1382 
1383 /*
1384  * Called when userspace is restoring VMX MSRs.
1385  *
1386  * Returns 0 on success, non-0 otherwise.
1387  */
1388 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1389 {
1390 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1391 
1392 	/*
1393 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1394 	 * is in VMX operation.
1395 	 */
1396 	if (vmx->nested.vmxon)
1397 		return -EBUSY;
1398 
1399 	switch (msr_index) {
1400 	case MSR_IA32_VMX_BASIC:
1401 		return vmx_restore_vmx_basic(vmx, data);
1402 	case MSR_IA32_VMX_PINBASED_CTLS:
1403 	case MSR_IA32_VMX_PROCBASED_CTLS:
1404 	case MSR_IA32_VMX_EXIT_CTLS:
1405 	case MSR_IA32_VMX_ENTRY_CTLS:
1406 		/*
1407 		 * The "non-true" VMX capability MSRs are generated from the
1408 		 * "true" MSRs, so we do not support restoring them directly.
1409 		 *
1410 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1411 		 * should restore the "true" MSRs with the must-be-1 bits
1412 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1413 		 * DEFAULT SETTINGS".
1414 		 */
1415 		return -EINVAL;
1416 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1417 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1418 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1419 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1420 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1421 		return vmx_restore_control_msr(vmx, msr_index, data);
1422 	case MSR_IA32_VMX_MISC:
1423 		return vmx_restore_vmx_misc(vmx, data);
1424 	case MSR_IA32_VMX_CR0_FIXED0:
1425 	case MSR_IA32_VMX_CR4_FIXED0:
1426 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1427 	case MSR_IA32_VMX_CR0_FIXED1:
1428 	case MSR_IA32_VMX_CR4_FIXED1:
1429 		/*
1430 		 * These MSRs are generated based on the vCPU's CPUID, so we
1431 		 * do not support restoring them directly.
1432 		 */
1433 		return -EINVAL;
1434 	case MSR_IA32_VMX_EPT_VPID_CAP:
1435 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1436 	case MSR_IA32_VMX_VMCS_ENUM:
1437 		vmx->nested.msrs.vmcs_enum = data;
1438 		return 0;
1439 	case MSR_IA32_VMX_VMFUNC:
1440 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1441 			return -EINVAL;
1442 		vmx->nested.msrs.vmfunc_controls = data;
1443 		return 0;
1444 	default:
1445 		/*
1446 		 * The rest of the VMX capability MSRs do not support restore.
1447 		 */
1448 		return -EINVAL;
1449 	}
1450 }
1451 
1452 /* Returns 0 on success, non-0 otherwise. */
1453 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1454 {
1455 	switch (msr_index) {
1456 	case MSR_IA32_VMX_BASIC:
1457 		*pdata = msrs->basic;
1458 		break;
1459 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1460 	case MSR_IA32_VMX_PINBASED_CTLS:
1461 		*pdata = vmx_control_msr(
1462 			msrs->pinbased_ctls_low,
1463 			msrs->pinbased_ctls_high);
1464 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1465 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1466 		break;
1467 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1468 	case MSR_IA32_VMX_PROCBASED_CTLS:
1469 		*pdata = vmx_control_msr(
1470 			msrs->procbased_ctls_low,
1471 			msrs->procbased_ctls_high);
1472 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1473 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1474 		break;
1475 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1476 	case MSR_IA32_VMX_EXIT_CTLS:
1477 		*pdata = vmx_control_msr(
1478 			msrs->exit_ctls_low,
1479 			msrs->exit_ctls_high);
1480 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1481 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1482 		break;
1483 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1484 	case MSR_IA32_VMX_ENTRY_CTLS:
1485 		*pdata = vmx_control_msr(
1486 			msrs->entry_ctls_low,
1487 			msrs->entry_ctls_high);
1488 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1489 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1490 		break;
1491 	case MSR_IA32_VMX_MISC:
1492 		*pdata = vmx_control_msr(
1493 			msrs->misc_low,
1494 			msrs->misc_high);
1495 		break;
1496 	case MSR_IA32_VMX_CR0_FIXED0:
1497 		*pdata = msrs->cr0_fixed0;
1498 		break;
1499 	case MSR_IA32_VMX_CR0_FIXED1:
1500 		*pdata = msrs->cr0_fixed1;
1501 		break;
1502 	case MSR_IA32_VMX_CR4_FIXED0:
1503 		*pdata = msrs->cr4_fixed0;
1504 		break;
1505 	case MSR_IA32_VMX_CR4_FIXED1:
1506 		*pdata = msrs->cr4_fixed1;
1507 		break;
1508 	case MSR_IA32_VMX_VMCS_ENUM:
1509 		*pdata = msrs->vmcs_enum;
1510 		break;
1511 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1512 		*pdata = vmx_control_msr(
1513 			msrs->secondary_ctls_low,
1514 			msrs->secondary_ctls_high);
1515 		break;
1516 	case MSR_IA32_VMX_EPT_VPID_CAP:
1517 		*pdata = msrs->ept_caps |
1518 			((u64)msrs->vpid_caps << 32);
1519 		break;
1520 	case MSR_IA32_VMX_VMFUNC:
1521 		*pdata = msrs->vmfunc_controls;
1522 		break;
1523 	default:
1524 		return 1;
1525 	}
1526 
1527 	return 0;
1528 }
1529 
1530 /*
1531  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1532  * been modified by the L1 guest.  Note, "writable" in this context means
1533  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1534  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1535  * VM-exit information fields (which are actually writable if the vCPU is
1536  * configured to support "VMWRITE to any supported field in the VMCS").
1537  */
1538 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1539 {
1540 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1541 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1542 	struct shadow_vmcs_field field;
1543 	unsigned long val;
1544 	int i;
1545 
1546 	if (WARN_ON(!shadow_vmcs))
1547 		return;
1548 
1549 	preempt_disable();
1550 
1551 	vmcs_load(shadow_vmcs);
1552 
1553 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1554 		field = shadow_read_write_fields[i];
1555 		val = __vmcs_readl(field.encoding);
1556 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1557 	}
1558 
1559 	vmcs_clear(shadow_vmcs);
1560 	vmcs_load(vmx->loaded_vmcs->vmcs);
1561 
1562 	preempt_enable();
1563 }
1564 
1565 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1566 {
1567 	const struct shadow_vmcs_field *fields[] = {
1568 		shadow_read_write_fields,
1569 		shadow_read_only_fields
1570 	};
1571 	const int max_fields[] = {
1572 		max_shadow_read_write_fields,
1573 		max_shadow_read_only_fields
1574 	};
1575 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1576 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1577 	struct shadow_vmcs_field field;
1578 	unsigned long val;
1579 	int i, q;
1580 
1581 	if (WARN_ON(!shadow_vmcs))
1582 		return;
1583 
1584 	vmcs_load(shadow_vmcs);
1585 
1586 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1587 		for (i = 0; i < max_fields[q]; i++) {
1588 			field = fields[q][i];
1589 			val = vmcs12_read_any(vmcs12, field.encoding,
1590 					      field.offset);
1591 			__vmcs_writel(field.encoding, val);
1592 		}
1593 	}
1594 
1595 	vmcs_clear(shadow_vmcs);
1596 	vmcs_load(vmx->loaded_vmcs->vmcs);
1597 }
1598 
1599 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1600 {
1601 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1602 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1603 
1604 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1605 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1606 	vmcs12->guest_rip = evmcs->guest_rip;
1607 
1608 	if (unlikely(!(hv_clean_fields &
1609 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1610 		vmcs12->guest_rsp = evmcs->guest_rsp;
1611 		vmcs12->guest_rflags = evmcs->guest_rflags;
1612 		vmcs12->guest_interruptibility_info =
1613 			evmcs->guest_interruptibility_info;
1614 	}
1615 
1616 	if (unlikely(!(hv_clean_fields &
1617 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1618 		vmcs12->cpu_based_vm_exec_control =
1619 			evmcs->cpu_based_vm_exec_control;
1620 	}
1621 
1622 	if (unlikely(!(hv_clean_fields &
1623 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1624 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1625 	}
1626 
1627 	if (unlikely(!(hv_clean_fields &
1628 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1629 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1630 	}
1631 
1632 	if (unlikely(!(hv_clean_fields &
1633 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1634 		vmcs12->vm_entry_intr_info_field =
1635 			evmcs->vm_entry_intr_info_field;
1636 		vmcs12->vm_entry_exception_error_code =
1637 			evmcs->vm_entry_exception_error_code;
1638 		vmcs12->vm_entry_instruction_len =
1639 			evmcs->vm_entry_instruction_len;
1640 	}
1641 
1642 	if (unlikely(!(hv_clean_fields &
1643 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1644 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1645 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1646 		vmcs12->host_cr0 = evmcs->host_cr0;
1647 		vmcs12->host_cr3 = evmcs->host_cr3;
1648 		vmcs12->host_cr4 = evmcs->host_cr4;
1649 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1650 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1651 		vmcs12->host_rip = evmcs->host_rip;
1652 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1653 		vmcs12->host_es_selector = evmcs->host_es_selector;
1654 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1655 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1656 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1657 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1658 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1659 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1660 	}
1661 
1662 	if (unlikely(!(hv_clean_fields &
1663 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1664 		vmcs12->pin_based_vm_exec_control =
1665 			evmcs->pin_based_vm_exec_control;
1666 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1667 		vmcs12->secondary_vm_exec_control =
1668 			evmcs->secondary_vm_exec_control;
1669 	}
1670 
1671 	if (unlikely(!(hv_clean_fields &
1672 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1673 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1674 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1675 	}
1676 
1677 	if (unlikely(!(hv_clean_fields &
1678 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1679 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1680 	}
1681 
1682 	if (unlikely(!(hv_clean_fields &
1683 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1684 		vmcs12->guest_es_base = evmcs->guest_es_base;
1685 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1686 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1687 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1688 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1689 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1690 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1691 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1692 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1693 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1694 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1695 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1696 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1697 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1698 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1699 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1700 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1701 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1702 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1703 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1704 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1705 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1706 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1707 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1708 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1709 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1710 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1711 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1712 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1713 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1714 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1715 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1716 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1717 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1718 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1719 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1720 	}
1721 
1722 	if (unlikely(!(hv_clean_fields &
1723 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1724 		vmcs12->tsc_offset = evmcs->tsc_offset;
1725 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1726 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1727 	}
1728 
1729 	if (unlikely(!(hv_clean_fields &
1730 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1731 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1732 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1733 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1734 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1735 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1736 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1737 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1738 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1739 	}
1740 
1741 	if (unlikely(!(hv_clean_fields &
1742 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1743 		vmcs12->host_fs_base = evmcs->host_fs_base;
1744 		vmcs12->host_gs_base = evmcs->host_gs_base;
1745 		vmcs12->host_tr_base = evmcs->host_tr_base;
1746 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1747 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1748 		vmcs12->host_rsp = evmcs->host_rsp;
1749 	}
1750 
1751 	if (unlikely(!(hv_clean_fields &
1752 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1753 		vmcs12->ept_pointer = evmcs->ept_pointer;
1754 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1755 	}
1756 
1757 	if (unlikely(!(hv_clean_fields &
1758 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1759 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1760 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1761 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1762 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1763 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1764 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1765 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1766 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1767 		vmcs12->guest_pending_dbg_exceptions =
1768 			evmcs->guest_pending_dbg_exceptions;
1769 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1770 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1771 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1772 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1773 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1774 	}
1775 
1776 	/*
1777 	 * Not used?
1778 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1779 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1780 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1781 	 * vmcs12->page_fault_error_code_mask =
1782 	 *		evmcs->page_fault_error_code_mask;
1783 	 * vmcs12->page_fault_error_code_match =
1784 	 *		evmcs->page_fault_error_code_match;
1785 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1786 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1787 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1788 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1789 	 */
1790 
1791 	/*
1792 	 * Read only fields:
1793 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1794 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1795 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1796 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1797 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1798 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1799 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1800 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1801 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1802 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1803 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1804 	 *
1805 	 * Not present in struct vmcs12:
1806 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1807 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1808 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1809 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1810 	 */
1811 
1812 	return;
1813 }
1814 
1815 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1816 {
1817 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1818 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1819 
1820 	/*
1821 	 * Should not be changed by KVM:
1822 	 *
1823 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1824 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1825 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1826 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1827 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1828 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1829 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1830 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1831 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1832 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1833 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1834 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1835 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1836 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1837 	 * evmcs->host_rip = vmcs12->host_rip;
1838 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1839 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1840 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1841 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1842 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1843 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1844 	 * evmcs->host_rsp = vmcs12->host_rsp;
1845 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1846 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1847 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1848 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1849 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1850 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1851 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1852 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1853 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1854 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1855 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1856 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1857 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1858 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1859 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1860 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1861 	 * evmcs->page_fault_error_code_mask =
1862 	 *		vmcs12->page_fault_error_code_mask;
1863 	 * evmcs->page_fault_error_code_match =
1864 	 *		vmcs12->page_fault_error_code_match;
1865 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1866 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1867 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1868 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1869 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1870 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1871 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1872 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1873 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1874 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1875 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1876 	 *
1877 	 * Not present in struct vmcs12:
1878 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1879 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1880 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1881 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1882 	 */
1883 
1884 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1885 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1886 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1887 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1888 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1889 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1890 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1891 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1892 
1893 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1894 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1895 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1896 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1897 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1898 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1899 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1900 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1901 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1902 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1903 
1904 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1905 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1906 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1907 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1908 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1909 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1910 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1911 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1912 
1913 	evmcs->guest_es_base = vmcs12->guest_es_base;
1914 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1915 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1916 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1917 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1918 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1919 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1920 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1921 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1922 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1923 
1924 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1925 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1926 
1927 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1928 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1929 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1930 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1931 
1932 	evmcs->guest_pending_dbg_exceptions =
1933 		vmcs12->guest_pending_dbg_exceptions;
1934 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1935 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1936 
1937 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1938 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1939 
1940 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1941 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1942 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1943 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1944 
1945 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1946 
1947 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1948 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1949 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1950 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1951 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1952 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1953 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1954 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1955 
1956 	evmcs->exit_qualification = vmcs12->exit_qualification;
1957 
1958 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1959 	evmcs->guest_rsp = vmcs12->guest_rsp;
1960 	evmcs->guest_rflags = vmcs12->guest_rflags;
1961 
1962 	evmcs->guest_interruptibility_info =
1963 		vmcs12->guest_interruptibility_info;
1964 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1965 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1966 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1967 	evmcs->vm_entry_exception_error_code =
1968 		vmcs12->vm_entry_exception_error_code;
1969 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1970 
1971 	evmcs->guest_rip = vmcs12->guest_rip;
1972 
1973 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1974 
1975 	return;
1976 }
1977 
1978 /*
1979  * This is an equivalent of the nested hypervisor executing the vmptrld
1980  * instruction.
1981  */
1982 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1983 	struct kvm_vcpu *vcpu, bool from_launch)
1984 {
1985 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1986 	bool evmcs_gpa_changed = false;
1987 	u64 evmcs_gpa;
1988 
1989 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1990 		return EVMPTRLD_DISABLED;
1991 
1992 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa)) {
1993 		nested_release_evmcs(vcpu);
1994 		return EVMPTRLD_DISABLED;
1995 	}
1996 
1997 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1998 		vmx->nested.current_vmptr = INVALID_GPA;
1999 
2000 		nested_release_evmcs(vcpu);
2001 
2002 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2003 				 &vmx->nested.hv_evmcs_map))
2004 			return EVMPTRLD_ERROR;
2005 
2006 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2007 
2008 		/*
2009 		 * Currently, KVM only supports eVMCS version 1
2010 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2011 		 * value to first u32 field of eVMCS which should specify eVMCS
2012 		 * VersionNumber.
2013 		 *
2014 		 * Guest should be aware of supported eVMCS versions by host by
2015 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2016 		 * expected to set this CPUID leaf according to the value
2017 		 * returned in vmcs_version from nested_enable_evmcs().
2018 		 *
2019 		 * However, it turns out that Microsoft Hyper-V fails to comply
2020 		 * to their own invented interface: When Hyper-V use eVMCS, it
2021 		 * just sets first u32 field of eVMCS to revision_id specified
2022 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2023 		 * which is one of the supported versions specified in
2024 		 * CPUID.0x4000000A.EAX[0:15].
2025 		 *
2026 		 * To overcome Hyper-V bug, we accept here either a supported
2027 		 * eVMCS version or VMCS12 revision_id as valid values for first
2028 		 * u32 field of eVMCS.
2029 		 */
2030 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2031 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2032 			nested_release_evmcs(vcpu);
2033 			return EVMPTRLD_VMFAIL;
2034 		}
2035 
2036 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2037 
2038 		evmcs_gpa_changed = true;
2039 		/*
2040 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2041 		 * reloaded from guest's memory (read only fields, fields not
2042 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2043 		 * are no leftovers.
2044 		 */
2045 		if (from_launch) {
2046 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2047 			memset(vmcs12, 0, sizeof(*vmcs12));
2048 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2049 		}
2050 
2051 	}
2052 
2053 	/*
2054 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2055 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2056 	 */
2057 	if (from_launch || evmcs_gpa_changed)
2058 		vmx->nested.hv_evmcs->hv_clean_fields &=
2059 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2060 
2061 	return EVMPTRLD_SUCCEEDED;
2062 }
2063 
2064 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2065 {
2066 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2067 
2068 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2069 		copy_vmcs12_to_enlightened(vmx);
2070 	else
2071 		copy_vmcs12_to_shadow(vmx);
2072 
2073 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2074 }
2075 
2076 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2077 {
2078 	struct vcpu_vmx *vmx =
2079 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2080 
2081 	vmx->nested.preemption_timer_expired = true;
2082 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2083 	kvm_vcpu_kick(&vmx->vcpu);
2084 
2085 	return HRTIMER_NORESTART;
2086 }
2087 
2088 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2089 {
2090 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2091 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2092 
2093 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2094 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2095 
2096 	if (!vmx->nested.has_preemption_timer_deadline) {
2097 		vmx->nested.preemption_timer_deadline =
2098 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2099 		vmx->nested.has_preemption_timer_deadline = true;
2100 	}
2101 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2102 }
2103 
2104 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2105 					u64 preemption_timeout)
2106 {
2107 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2108 
2109 	/*
2110 	 * A timer value of zero is architecturally guaranteed to cause
2111 	 * a VMExit prior to executing any instructions in the guest.
2112 	 */
2113 	if (preemption_timeout == 0) {
2114 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2115 		return;
2116 	}
2117 
2118 	if (vcpu->arch.virtual_tsc_khz == 0)
2119 		return;
2120 
2121 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2122 	preemption_timeout *= 1000000;
2123 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2124 	hrtimer_start(&vmx->nested.preemption_timer,
2125 		      ktime_add_ns(ktime_get(), preemption_timeout),
2126 		      HRTIMER_MODE_ABS_PINNED);
2127 }
2128 
2129 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2130 {
2131 	if (vmx->nested.nested_run_pending &&
2132 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2133 		return vmcs12->guest_ia32_efer;
2134 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2135 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2136 	else
2137 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2138 }
2139 
2140 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2141 {
2142 	/*
2143 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2144 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2145 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2146 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2147 	 */
2148 	if (vmx->nested.vmcs02_initialized)
2149 		return;
2150 	vmx->nested.vmcs02_initialized = true;
2151 
2152 	/*
2153 	 * We don't care what the EPTP value is we just need to guarantee
2154 	 * it's valid so we don't get a false positive when doing early
2155 	 * consistency checks.
2156 	 */
2157 	if (enable_ept && nested_early_check)
2158 		vmcs_write64(EPT_POINTER,
2159 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2160 
2161 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2162 	if (cpu_has_vmx_vmfunc())
2163 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2164 
2165 	if (cpu_has_vmx_posted_intr())
2166 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2167 
2168 	if (cpu_has_vmx_msr_bitmap())
2169 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2170 
2171 	/*
2172 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2173 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2174 	 * deal with filtering out L2 GPAs from the buffer.
2175 	 */
2176 	if (enable_pml) {
2177 		vmcs_write64(PML_ADDRESS, 0);
2178 		vmcs_write16(GUEST_PML_INDEX, -1);
2179 	}
2180 
2181 	if (cpu_has_vmx_encls_vmexit())
2182 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2183 
2184 	/*
2185 	 * Set the MSR load/store lists to match L0's settings.  Only the
2186 	 * addresses are constant (for vmcs02), the counts can change based
2187 	 * on L2's behavior, e.g. switching to/from long mode.
2188 	 */
2189 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2190 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2191 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2192 
2193 	vmx_set_constant_host_state(vmx);
2194 }
2195 
2196 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2197 				      struct vmcs12 *vmcs12)
2198 {
2199 	prepare_vmcs02_constant_state(vmx);
2200 
2201 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2202 
2203 	if (enable_vpid) {
2204 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2205 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2206 		else
2207 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2208 	}
2209 }
2210 
2211 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2212 				 struct vmcs12 *vmcs12)
2213 {
2214 	u32 exec_control;
2215 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2216 
2217 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2218 		prepare_vmcs02_early_rare(vmx, vmcs12);
2219 
2220 	/*
2221 	 * PIN CONTROLS
2222 	 */
2223 	exec_control = __pin_controls_get(vmcs01);
2224 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2225 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2226 
2227 	/* Posted interrupts setting is only taken from vmcs12.  */
2228 	vmx->nested.pi_pending = false;
2229 	if (nested_cpu_has_posted_intr(vmcs12))
2230 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2231 	else
2232 		exec_control &= ~PIN_BASED_POSTED_INTR;
2233 	pin_controls_set(vmx, exec_control);
2234 
2235 	/*
2236 	 * EXEC CONTROLS
2237 	 */
2238 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2239 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2240 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2241 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2242 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2243 
2244 	vmx->nested.l1_tpr_threshold = -1;
2245 	if (exec_control & CPU_BASED_TPR_SHADOW)
2246 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2247 #ifdef CONFIG_X86_64
2248 	else
2249 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2250 				CPU_BASED_CR8_STORE_EXITING;
2251 #endif
2252 
2253 	/*
2254 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2255 	 * for I/O port accesses.
2256 	 */
2257 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2258 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2259 
2260 	/*
2261 	 * This bit will be computed in nested_get_vmcs12_pages, because
2262 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2263 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2264 	 * only set/clear this bit.
2265 	 */
2266 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2267 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2268 
2269 	exec_controls_set(vmx, exec_control);
2270 
2271 	/*
2272 	 * SECONDARY EXEC CONTROLS
2273 	 */
2274 	if (cpu_has_secondary_exec_ctrls()) {
2275 		exec_control = __secondary_exec_controls_get(vmcs01);
2276 
2277 		/* Take the following fields only from vmcs12 */
2278 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2279 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2280 				  SECONDARY_EXEC_ENABLE_INVPCID |
2281 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2282 				  SECONDARY_EXEC_XSAVES |
2283 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2284 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2285 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2286 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2287 				  SECONDARY_EXEC_TSC_SCALING |
2288 				  SECONDARY_EXEC_DESC);
2289 
2290 		if (nested_cpu_has(vmcs12,
2291 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2292 			exec_control |= vmcs12->secondary_vm_exec_control;
2293 
2294 		/* PML is emulated and never enabled in hardware for L2. */
2295 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2296 
2297 		/* VMCS shadowing for L2 is emulated for now */
2298 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2299 
2300 		/*
2301 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2302 		 * will not have to rewrite the controls just for this bit.
2303 		 */
2304 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2305 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2306 			exec_control |= SECONDARY_EXEC_DESC;
2307 
2308 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2309 			vmcs_write16(GUEST_INTR_STATUS,
2310 				vmcs12->guest_intr_status);
2311 
2312 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2313 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2314 
2315 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2316 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2317 
2318 		secondary_exec_controls_set(vmx, exec_control);
2319 	}
2320 
2321 	/*
2322 	 * ENTRY CONTROLS
2323 	 *
2324 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2325 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2326 	 * on the related bits (if supported by the CPU) in the hope that
2327 	 * we can avoid VMWrites during vmx_set_efer().
2328 	 */
2329 	exec_control = __vm_entry_controls_get(vmcs01);
2330 	exec_control |= vmcs12->vm_entry_controls;
2331 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2332 	if (cpu_has_load_ia32_efer()) {
2333 		if (guest_efer & EFER_LMA)
2334 			exec_control |= VM_ENTRY_IA32E_MODE;
2335 		if (guest_efer != host_efer)
2336 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2337 	}
2338 	vm_entry_controls_set(vmx, exec_control);
2339 
2340 	/*
2341 	 * EXIT CONTROLS
2342 	 *
2343 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2344 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2345 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2346 	 */
2347 	exec_control = __vm_exit_controls_get(vmcs01);
2348 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2349 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2350 	else
2351 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2352 	vm_exit_controls_set(vmx, exec_control);
2353 
2354 	/*
2355 	 * Interrupt/Exception Fields
2356 	 */
2357 	if (vmx->nested.nested_run_pending) {
2358 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2359 			     vmcs12->vm_entry_intr_info_field);
2360 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2361 			     vmcs12->vm_entry_exception_error_code);
2362 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2363 			     vmcs12->vm_entry_instruction_len);
2364 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2365 			     vmcs12->guest_interruptibility_info);
2366 		vmx->loaded_vmcs->nmi_known_unmasked =
2367 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2368 	} else {
2369 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2370 	}
2371 }
2372 
2373 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2374 {
2375 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2376 
2377 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2378 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2379 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2380 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2381 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2382 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2383 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2384 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2385 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2386 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2387 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2388 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2389 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2390 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2391 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2392 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2393 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2394 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2395 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2396 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2397 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2398 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2399 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2400 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2401 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2402 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2403 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2404 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2405 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2406 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2407 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2408 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2409 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2410 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2411 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2412 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2413 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2414 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2415 
2416 		vmx->segment_cache.bitmask = 0;
2417 	}
2418 
2419 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2420 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2421 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2422 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2423 			    vmcs12->guest_pending_dbg_exceptions);
2424 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2425 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2426 
2427 		/*
2428 		 * L1 may access the L2's PDPTR, so save them to construct
2429 		 * vmcs12
2430 		 */
2431 		if (enable_ept) {
2432 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2433 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2434 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2435 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2436 		}
2437 
2438 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2439 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2440 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2441 	}
2442 
2443 	if (nested_cpu_has_xsaves(vmcs12))
2444 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2445 
2446 	/*
2447 	 * Whether page-faults are trapped is determined by a combination of
2448 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2449 	 * doesn't care about page faults then we should set all of these to
2450 	 * L1's desires. However, if L0 does care about (some) page faults, it
2451 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2452 	 * simply ask to exit on each and every L2 page fault. This is done by
2453 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2454 	 * Note that below we don't need special code to set EB.PF beyond the
2455 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2456 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2457 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2458 	 */
2459 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2460 		/*
2461 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2462 		 * go ahead and use it?
2463 		 */
2464 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2465 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2466 	} else {
2467 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2468 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2469 	}
2470 
2471 	if (cpu_has_vmx_apicv()) {
2472 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2473 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2474 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2475 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2476 	}
2477 
2478 	/*
2479 	 * Make sure the msr_autostore list is up to date before we set the
2480 	 * count in the vmcs02.
2481 	 */
2482 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2483 
2484 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2485 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2486 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2487 
2488 	set_cr4_guest_host_mask(vmx);
2489 }
2490 
2491 /*
2492  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2493  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2494  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2495  * guest in a way that will both be appropriate to L1's requests, and our
2496  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2497  * function also has additional necessary side-effects, like setting various
2498  * vcpu->arch fields.
2499  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2500  * is assigned to entry_failure_code on failure.
2501  */
2502 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2503 			  bool from_vmentry,
2504 			  enum vm_entry_failure_code *entry_failure_code)
2505 {
2506 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2507 	bool load_guest_pdptrs_vmcs12 = false;
2508 
2509 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2510 		prepare_vmcs02_rare(vmx, vmcs12);
2511 		vmx->nested.dirty_vmcs12 = false;
2512 
2513 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2514 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2515 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2516 	}
2517 
2518 	if (vmx->nested.nested_run_pending &&
2519 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2520 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2521 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2522 	} else {
2523 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2524 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2525 	}
2526 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2527 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2528 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2529 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2530 
2531 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2532 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2533 	 * trap. Note that CR0.TS also needs updating - we do this later.
2534 	 */
2535 	vmx_update_exception_bitmap(vcpu);
2536 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2537 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2538 
2539 	if (vmx->nested.nested_run_pending &&
2540 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2541 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2542 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2543 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2544 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2545 	}
2546 
2547 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2548 			vcpu->arch.l1_tsc_offset,
2549 			vmx_get_l2_tsc_offset(vcpu),
2550 			vmx_get_l2_tsc_multiplier(vcpu));
2551 
2552 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2553 			vcpu->arch.l1_tsc_scaling_ratio,
2554 			vmx_get_l2_tsc_multiplier(vcpu));
2555 
2556 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2557 	if (kvm_has_tsc_control)
2558 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2559 
2560 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2561 
2562 	if (nested_cpu_has_ept(vmcs12))
2563 		nested_ept_init_mmu_context(vcpu);
2564 
2565 	/*
2566 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2567 	 * bits which we consider mandatory enabled.
2568 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2569 	 * the specifications by L1; It's not enough to take
2570 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2571 	 * have more bits than L1 expected.
2572 	 */
2573 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2574 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2575 
2576 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2577 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2578 
2579 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2580 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2581 	vmx_set_efer(vcpu, vcpu->arch.efer);
2582 
2583 	/*
2584 	 * Guest state is invalid and unrestricted guest is disabled,
2585 	 * which means L1 attempted VMEntry to L2 with invalid state.
2586 	 * Fail the VMEntry.
2587 	 *
2588 	 * However when force loading the guest state (SMM exit or
2589 	 * loading nested state after migration, it is possible to
2590 	 * have invalid guest state now, which will be later fixed by
2591 	 * restoring L2 register state
2592 	 */
2593 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2594 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2595 		return -EINVAL;
2596 	}
2597 
2598 	/* Shadow page tables on either EPT or shadow page tables. */
2599 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2600 				from_vmentry, entry_failure_code))
2601 		return -EINVAL;
2602 
2603 	/*
2604 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2605 	 * on nested VM-Exit, which can occur without actually running L2 and
2606 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2607 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2608 	 * transition to HLT instead of running L2.
2609 	 */
2610 	if (enable_ept)
2611 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2612 
2613 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2614 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2615 	    is_pae_paging(vcpu)) {
2616 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2617 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2618 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2619 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2620 	}
2621 
2622 	if (!enable_ept)
2623 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2624 
2625 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2626 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2627 				     vmcs12->guest_ia32_perf_global_ctrl)))
2628 		return -EINVAL;
2629 
2630 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2631 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2632 
2633 	/*
2634 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2635 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2636 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2637 	 * here.
2638 	 */
2639 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2640 		vmx->nested.hv_evmcs->hv_clean_fields |=
2641 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2642 
2643 	return 0;
2644 }
2645 
2646 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2647 {
2648 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2649 	       nested_cpu_has_virtual_nmis(vmcs12)))
2650 		return -EINVAL;
2651 
2652 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2653 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2654 		return -EINVAL;
2655 
2656 	return 0;
2657 }
2658 
2659 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2660 {
2661 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2662 
2663 	/* Check for memory type validity */
2664 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2665 	case VMX_EPTP_MT_UC:
2666 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2667 			return false;
2668 		break;
2669 	case VMX_EPTP_MT_WB:
2670 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2671 			return false;
2672 		break;
2673 	default:
2674 		return false;
2675 	}
2676 
2677 	/* Page-walk levels validity. */
2678 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2679 	case VMX_EPTP_PWL_5:
2680 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2681 			return false;
2682 		break;
2683 	case VMX_EPTP_PWL_4:
2684 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2685 			return false;
2686 		break;
2687 	default:
2688 		return false;
2689 	}
2690 
2691 	/* Reserved bits should not be set */
2692 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2693 		return false;
2694 
2695 	/* AD, if set, should be supported */
2696 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2697 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2698 			return false;
2699 	}
2700 
2701 	return true;
2702 }
2703 
2704 /*
2705  * Checks related to VM-Execution Control Fields
2706  */
2707 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2708                                               struct vmcs12 *vmcs12)
2709 {
2710 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2711 
2712 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2713 				   vmx->nested.msrs.pinbased_ctls_low,
2714 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2715 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2716 				   vmx->nested.msrs.procbased_ctls_low,
2717 				   vmx->nested.msrs.procbased_ctls_high)))
2718 		return -EINVAL;
2719 
2720 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2721 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2722 				   vmx->nested.msrs.secondary_ctls_low,
2723 				   vmx->nested.msrs.secondary_ctls_high)))
2724 		return -EINVAL;
2725 
2726 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2727 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2728 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2729 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2730 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2731 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2732 	    nested_vmx_check_nmi_controls(vmcs12) ||
2733 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2734 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2735 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2736 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2737 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2738 		return -EINVAL;
2739 
2740 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2741 	    nested_cpu_has_save_preemption_timer(vmcs12))
2742 		return -EINVAL;
2743 
2744 	if (nested_cpu_has_ept(vmcs12) &&
2745 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2746 		return -EINVAL;
2747 
2748 	if (nested_cpu_has_vmfunc(vmcs12)) {
2749 		if (CC(vmcs12->vm_function_control &
2750 		       ~vmx->nested.msrs.vmfunc_controls))
2751 			return -EINVAL;
2752 
2753 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2754 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2755 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2756 				return -EINVAL;
2757 		}
2758 	}
2759 
2760 	return 0;
2761 }
2762 
2763 /*
2764  * Checks related to VM-Exit Control Fields
2765  */
2766 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2767                                          struct vmcs12 *vmcs12)
2768 {
2769 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2770 
2771 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2772 				    vmx->nested.msrs.exit_ctls_low,
2773 				    vmx->nested.msrs.exit_ctls_high)) ||
2774 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2775 		return -EINVAL;
2776 
2777 	return 0;
2778 }
2779 
2780 /*
2781  * Checks related to VM-Entry Control Fields
2782  */
2783 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2784 					  struct vmcs12 *vmcs12)
2785 {
2786 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2787 
2788 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2789 				    vmx->nested.msrs.entry_ctls_low,
2790 				    vmx->nested.msrs.entry_ctls_high)))
2791 		return -EINVAL;
2792 
2793 	/*
2794 	 * From the Intel SDM, volume 3:
2795 	 * Fields relevant to VM-entry event injection must be set properly.
2796 	 * These fields are the VM-entry interruption-information field, the
2797 	 * VM-entry exception error code, and the VM-entry instruction length.
2798 	 */
2799 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2800 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2801 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2802 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2803 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2804 		bool should_have_error_code;
2805 		bool urg = nested_cpu_has2(vmcs12,
2806 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2807 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2808 
2809 		/* VM-entry interruption-info field: interruption type */
2810 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2811 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2812 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2813 			return -EINVAL;
2814 
2815 		/* VM-entry interruption-info field: vector */
2816 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2817 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2818 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2819 			return -EINVAL;
2820 
2821 		/* VM-entry interruption-info field: deliver error code */
2822 		should_have_error_code =
2823 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2824 			x86_exception_has_error_code(vector);
2825 		if (CC(has_error_code != should_have_error_code))
2826 			return -EINVAL;
2827 
2828 		/* VM-entry exception error code */
2829 		if (CC(has_error_code &&
2830 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2831 			return -EINVAL;
2832 
2833 		/* VM-entry interruption-info field: reserved bits */
2834 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2835 			return -EINVAL;
2836 
2837 		/* VM-entry instruction length */
2838 		switch (intr_type) {
2839 		case INTR_TYPE_SOFT_EXCEPTION:
2840 		case INTR_TYPE_SOFT_INTR:
2841 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2842 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2843 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2844 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2845 				return -EINVAL;
2846 		}
2847 	}
2848 
2849 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2850 		return -EINVAL;
2851 
2852 	return 0;
2853 }
2854 
2855 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2856 				     struct vmcs12 *vmcs12)
2857 {
2858 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2859 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2860 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2861 		return -EINVAL;
2862 
2863 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2864 		return nested_evmcs_check_controls(vmcs12);
2865 
2866 	return 0;
2867 }
2868 
2869 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2870 				       struct vmcs12 *vmcs12)
2871 {
2872 	bool ia32e;
2873 
2874 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2875 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2876 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2877 		return -EINVAL;
2878 
2879 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2880 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2881 		return -EINVAL;
2882 
2883 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2884 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2885 		return -EINVAL;
2886 
2887 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2888 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2889 					   vmcs12->host_ia32_perf_global_ctrl)))
2890 		return -EINVAL;
2891 
2892 #ifdef CONFIG_X86_64
2893 	ia32e = !!(vcpu->arch.efer & EFER_LMA);
2894 #else
2895 	ia32e = false;
2896 #endif
2897 
2898 	if (ia32e) {
2899 		if (CC(!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)) ||
2900 		    CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2901 			return -EINVAL;
2902 	} else {
2903 		if (CC(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) ||
2904 		    CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2905 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2906 		    CC((vmcs12->host_rip) >> 32))
2907 			return -EINVAL;
2908 	}
2909 
2910 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2911 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2912 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2913 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2914 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2915 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2916 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2917 	    CC(vmcs12->host_cs_selector == 0) ||
2918 	    CC(vmcs12->host_tr_selector == 0) ||
2919 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2920 		return -EINVAL;
2921 
2922 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2923 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2924 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2925 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2926 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2927 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2928 		return -EINVAL;
2929 
2930 	/*
2931 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2932 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2933 	 * the values of the LMA and LME bits in the field must each be that of
2934 	 * the host address-space size VM-exit control.
2935 	 */
2936 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2937 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2938 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2939 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2940 			return -EINVAL;
2941 	}
2942 
2943 	return 0;
2944 }
2945 
2946 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2947 					  struct vmcs12 *vmcs12)
2948 {
2949 	int r = 0;
2950 	struct vmcs12 *shadow;
2951 	struct kvm_host_map map;
2952 
2953 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
2954 		return 0;
2955 
2956 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2957 		return -EINVAL;
2958 
2959 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2960 		return -EINVAL;
2961 
2962 	shadow = map.hva;
2963 
2964 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2965 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2966 		r = -EINVAL;
2967 
2968 	kvm_vcpu_unmap(vcpu, &map, false);
2969 	return r;
2970 }
2971 
2972 /*
2973  * Checks related to Guest Non-register State
2974  */
2975 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2976 {
2977 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2978 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2979 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2980 		return -EINVAL;
2981 
2982 	return 0;
2983 }
2984 
2985 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2986 					struct vmcs12 *vmcs12,
2987 					enum vm_entry_failure_code *entry_failure_code)
2988 {
2989 	bool ia32e;
2990 
2991 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
2992 
2993 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2994 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2995 		return -EINVAL;
2996 
2997 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2998 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2999 		return -EINVAL;
3000 
3001 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3002 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3003 		return -EINVAL;
3004 
3005 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3006 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3007 		return -EINVAL;
3008 	}
3009 
3010 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3011 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3012 					   vmcs12->guest_ia32_perf_global_ctrl)))
3013 		return -EINVAL;
3014 
3015 	/*
3016 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3017 	 * are performed on the field for the IA32_EFER MSR:
3018 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3019 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3020 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3021 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3022 	 *   CR0.PG) is 1.
3023 	 */
3024 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3025 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3026 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
3027 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3028 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3029 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3030 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3031 			return -EINVAL;
3032 	}
3033 
3034 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3035 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3036 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3037 		return -EINVAL;
3038 
3039 	if (nested_check_guest_non_reg_state(vmcs12))
3040 		return -EINVAL;
3041 
3042 	return 0;
3043 }
3044 
3045 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3046 {
3047 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3048 	unsigned long cr3, cr4;
3049 	bool vm_fail;
3050 
3051 	if (!nested_early_check)
3052 		return 0;
3053 
3054 	if (vmx->msr_autoload.host.nr)
3055 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3056 	if (vmx->msr_autoload.guest.nr)
3057 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3058 
3059 	preempt_disable();
3060 
3061 	vmx_prepare_switch_to_guest(vcpu);
3062 
3063 	/*
3064 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3065 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3066 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3067 	 * there is no need to preserve other bits or save/restore the field.
3068 	 */
3069 	vmcs_writel(GUEST_RFLAGS, 0);
3070 
3071 	cr3 = __get_current_cr3_fast();
3072 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3073 		vmcs_writel(HOST_CR3, cr3);
3074 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3075 	}
3076 
3077 	cr4 = cr4_read_shadow();
3078 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3079 		vmcs_writel(HOST_CR4, cr4);
3080 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3081 	}
3082 
3083 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3084 				 vmx->loaded_vmcs->launched);
3085 
3086 	if (vmx->msr_autoload.host.nr)
3087 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3088 	if (vmx->msr_autoload.guest.nr)
3089 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3090 
3091 	if (vm_fail) {
3092 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3093 
3094 		preempt_enable();
3095 
3096 		trace_kvm_nested_vmenter_failed(
3097 			"early hardware check VM-instruction error: ", error);
3098 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3099 		return 1;
3100 	}
3101 
3102 	/*
3103 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3104 	 */
3105 	if (hw_breakpoint_active())
3106 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3107 	local_irq_enable();
3108 	preempt_enable();
3109 
3110 	/*
3111 	 * A non-failing VMEntry means we somehow entered guest mode with
3112 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3113 	 * is no telling what memory has been modified or what state has
3114 	 * been exposed to unknown code.  Hitting this all but guarantees
3115 	 * a (very critical) hardware issue.
3116 	 */
3117 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3118 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3119 
3120 	return 0;
3121 }
3122 
3123 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3124 {
3125 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3126 
3127 	/*
3128 	 * hv_evmcs may end up being not mapped after migration (when
3129 	 * L2 was running), map it here to make sure vmcs12 changes are
3130 	 * properly reflected.
3131 	 */
3132 	if (vmx->nested.enlightened_vmcs_enabled &&
3133 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3134 		enum nested_evmptrld_status evmptrld_status =
3135 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3136 
3137 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3138 		    evmptrld_status == EVMPTRLD_ERROR)
3139 			return false;
3140 
3141 		/*
3142 		 * Post migration VMCS12 always provides the most actual
3143 		 * information, copy it to eVMCS upon entry.
3144 		 */
3145 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3146 	}
3147 
3148 	return true;
3149 }
3150 
3151 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3152 {
3153 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3154 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3155 	struct kvm_host_map *map;
3156 	struct page *page;
3157 	u64 hpa;
3158 
3159 	if (!vcpu->arch.pdptrs_from_userspace &&
3160 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3161 		/*
3162 		 * Reload the guest's PDPTRs since after a migration
3163 		 * the guest CR3 might be restored prior to setting the nested
3164 		 * state which can lead to a load of wrong PDPTRs.
3165 		 */
3166 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3)))
3167 			return false;
3168 	}
3169 
3170 
3171 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3172 		/*
3173 		 * Translate L1 physical address to host physical
3174 		 * address for vmcs02. Keep the page pinned, so this
3175 		 * physical address remains valid. We keep a reference
3176 		 * to it so we can release it later.
3177 		 */
3178 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3179 			kvm_release_page_clean(vmx->nested.apic_access_page);
3180 			vmx->nested.apic_access_page = NULL;
3181 		}
3182 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3183 		if (!is_error_page(page)) {
3184 			vmx->nested.apic_access_page = page;
3185 			hpa = page_to_phys(vmx->nested.apic_access_page);
3186 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3187 		} else {
3188 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3189 					     __func__);
3190 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3191 			vcpu->run->internal.suberror =
3192 				KVM_INTERNAL_ERROR_EMULATION;
3193 			vcpu->run->internal.ndata = 0;
3194 			return false;
3195 		}
3196 	}
3197 
3198 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3199 		map = &vmx->nested.virtual_apic_map;
3200 
3201 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3202 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3203 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3204 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3205 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3206 			/*
3207 			 * The processor will never use the TPR shadow, simply
3208 			 * clear the bit from the execution control.  Such a
3209 			 * configuration is useless, but it happens in tests.
3210 			 * For any other configuration, failing the vm entry is
3211 			 * _not_ what the processor does but it's basically the
3212 			 * only possibility we have.
3213 			 */
3214 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3215 		} else {
3216 			/*
3217 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3218 			 * force VM-Entry to fail.
3219 			 */
3220 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3221 		}
3222 	}
3223 
3224 	if (nested_cpu_has_posted_intr(vmcs12)) {
3225 		map = &vmx->nested.pi_desc_map;
3226 
3227 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3228 			vmx->nested.pi_desc =
3229 				(struct pi_desc *)(((void *)map->hva) +
3230 				offset_in_page(vmcs12->posted_intr_desc_addr));
3231 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3232 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3233 		} else {
3234 			/*
3235 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3236 			 * access the contents of the VMCS12 posted interrupt
3237 			 * descriptor. (Note that KVM may do this when it
3238 			 * should not, per the architectural specification.)
3239 			 */
3240 			vmx->nested.pi_desc = NULL;
3241 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3242 		}
3243 	}
3244 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3245 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3246 	else
3247 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3248 
3249 	return true;
3250 }
3251 
3252 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3253 {
3254 	if (!nested_get_evmcs_page(vcpu)) {
3255 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3256 				     __func__);
3257 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3258 		vcpu->run->internal.suberror =
3259 			KVM_INTERNAL_ERROR_EMULATION;
3260 		vcpu->run->internal.ndata = 0;
3261 
3262 		return false;
3263 	}
3264 
3265 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3266 		return false;
3267 
3268 	return true;
3269 }
3270 
3271 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3272 {
3273 	struct vmcs12 *vmcs12;
3274 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3275 	gpa_t dst;
3276 
3277 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3278 		return 0;
3279 
3280 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3281 		return 1;
3282 
3283 	/*
3284 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3285 	 * set is already checked as part of A/D emulation.
3286 	 */
3287 	vmcs12 = get_vmcs12(vcpu);
3288 	if (!nested_cpu_has_pml(vmcs12))
3289 		return 0;
3290 
3291 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3292 		vmx->nested.pml_full = true;
3293 		return 1;
3294 	}
3295 
3296 	gpa &= ~0xFFFull;
3297 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3298 
3299 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3300 				 offset_in_page(dst), sizeof(gpa)))
3301 		return 0;
3302 
3303 	vmcs12->guest_pml_index--;
3304 
3305 	return 0;
3306 }
3307 
3308 /*
3309  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3310  * for running VMX instructions (except VMXON, whose prerequisites are
3311  * slightly different). It also specifies what exception to inject otherwise.
3312  * Note that many of these exceptions have priority over VM exits, so they
3313  * don't have to be checked again here.
3314  */
3315 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3316 {
3317 	if (!to_vmx(vcpu)->nested.vmxon) {
3318 		kvm_queue_exception(vcpu, UD_VECTOR);
3319 		return 0;
3320 	}
3321 
3322 	if (vmx_get_cpl(vcpu)) {
3323 		kvm_inject_gp(vcpu, 0);
3324 		return 0;
3325 	}
3326 
3327 	return 1;
3328 }
3329 
3330 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3331 {
3332 	u8 rvi = vmx_get_rvi();
3333 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3334 
3335 	return ((rvi & 0xf0) > (vppr & 0xf0));
3336 }
3337 
3338 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3339 				   struct vmcs12 *vmcs12);
3340 
3341 /*
3342  * If from_vmentry is false, this is being called from state restore (either RSM
3343  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3344  *
3345  * Returns:
3346  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3347  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3348  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3349  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3350  */
3351 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3352 							bool from_vmentry)
3353 {
3354 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3355 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3356 	enum vm_entry_failure_code entry_failure_code;
3357 	bool evaluate_pending_interrupts;
3358 	union vmx_exit_reason exit_reason = {
3359 		.basic = EXIT_REASON_INVALID_STATE,
3360 		.failed_vmentry = 1,
3361 	};
3362 	u32 failed_index;
3363 
3364 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3365 		kvm_vcpu_flush_tlb_current(vcpu);
3366 
3367 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3368 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3369 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3370 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3371 
3372 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3373 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3374 	if (kvm_mpx_supported() &&
3375 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3376 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3377 
3378 	/*
3379 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3380 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3381 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3382 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3383 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3384 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3385 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3386 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3387 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3388 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3389 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3390 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3391 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3392 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3393 	 */
3394 	if (!enable_ept && !nested_early_check)
3395 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3396 
3397 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3398 
3399 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3400 
3401 	if (from_vmentry) {
3402 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3403 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3404 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3405 		}
3406 
3407 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3408 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3409 			return NVMX_VMENTRY_VMFAIL;
3410 		}
3411 
3412 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3413 						 &entry_failure_code)) {
3414 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3415 			vmcs12->exit_qualification = entry_failure_code;
3416 			goto vmentry_fail_vmexit;
3417 		}
3418 	}
3419 
3420 	enter_guest_mode(vcpu);
3421 
3422 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3423 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3424 		vmcs12->exit_qualification = entry_failure_code;
3425 		goto vmentry_fail_vmexit_guest_mode;
3426 	}
3427 
3428 	if (from_vmentry) {
3429 		failed_index = nested_vmx_load_msr(vcpu,
3430 						   vmcs12->vm_entry_msr_load_addr,
3431 						   vmcs12->vm_entry_msr_load_count);
3432 		if (failed_index) {
3433 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3434 			vmcs12->exit_qualification = failed_index;
3435 			goto vmentry_fail_vmexit_guest_mode;
3436 		}
3437 	} else {
3438 		/*
3439 		 * The MMU is not initialized to point at the right entities yet and
3440 		 * "get pages" would need to read data from the guest (i.e. we will
3441 		 * need to perform gpa to hpa translation). Request a call
3442 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3443 		 * have already been set at vmentry time and should not be reset.
3444 		 */
3445 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3446 	}
3447 
3448 	/*
3449 	 * If L1 had a pending IRQ/NMI until it executed
3450 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3451 	 * disallowed (e.g. interrupts disabled), L0 needs to
3452 	 * evaluate if this pending event should cause an exit from L2
3453 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3454 	 * intercept EXTERNAL_INTERRUPT).
3455 	 *
3456 	 * Usually this would be handled by the processor noticing an
3457 	 * IRQ/NMI window request, or checking RVI during evaluation of
3458 	 * pending virtual interrupts.  However, this setting was done
3459 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3460 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3461 	 */
3462 	if (unlikely(evaluate_pending_interrupts))
3463 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3464 
3465 	/*
3466 	 * Do not start the preemption timer hrtimer until after we know
3467 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3468 	 * the timer.
3469 	 */
3470 	vmx->nested.preemption_timer_expired = false;
3471 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3472 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3473 		vmx_start_preemption_timer(vcpu, timer_value);
3474 	}
3475 
3476 	/*
3477 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3478 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3479 	 * returned as far as L1 is concerned. It will only return (and set
3480 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3481 	 */
3482 	return NVMX_VMENTRY_SUCCESS;
3483 
3484 	/*
3485 	 * A failed consistency check that leads to a VMExit during L1's
3486 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3487 	 * 26.7 "VM-entry failures during or after loading guest state".
3488 	 */
3489 vmentry_fail_vmexit_guest_mode:
3490 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3491 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3492 	leave_guest_mode(vcpu);
3493 
3494 vmentry_fail_vmexit:
3495 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3496 
3497 	if (!from_vmentry)
3498 		return NVMX_VMENTRY_VMEXIT;
3499 
3500 	load_vmcs12_host_state(vcpu, vmcs12);
3501 	vmcs12->vm_exit_reason = exit_reason.full;
3502 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3503 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3504 	return NVMX_VMENTRY_VMEXIT;
3505 }
3506 
3507 /*
3508  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3509  * for running an L2 nested guest.
3510  */
3511 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3512 {
3513 	struct vmcs12 *vmcs12;
3514 	enum nvmx_vmentry_status status;
3515 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3516 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3517 	enum nested_evmptrld_status evmptrld_status;
3518 
3519 	if (!nested_vmx_check_permission(vcpu))
3520 		return 1;
3521 
3522 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3523 	if (evmptrld_status == EVMPTRLD_ERROR) {
3524 		kvm_queue_exception(vcpu, UD_VECTOR);
3525 		return 1;
3526 	} else if (CC(evmptrld_status == EVMPTRLD_VMFAIL)) {
3527 		return nested_vmx_failInvalid(vcpu);
3528 	}
3529 
3530 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3531 	       vmx->nested.current_vmptr == INVALID_GPA))
3532 		return nested_vmx_failInvalid(vcpu);
3533 
3534 	vmcs12 = get_vmcs12(vcpu);
3535 
3536 	/*
3537 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3538 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3539 	 * rather than RFLAGS.ZF, and no error number is stored to the
3540 	 * VM-instruction error field.
3541 	 */
3542 	if (CC(vmcs12->hdr.shadow_vmcs))
3543 		return nested_vmx_failInvalid(vcpu);
3544 
3545 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3546 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3547 		/* Enlightened VMCS doesn't have launch state */
3548 		vmcs12->launch_state = !launch;
3549 	} else if (enable_shadow_vmcs) {
3550 		copy_shadow_to_vmcs12(vmx);
3551 	}
3552 
3553 	/*
3554 	 * The nested entry process starts with enforcing various prerequisites
3555 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3556 	 * they fail: As the SDM explains, some conditions should cause the
3557 	 * instruction to fail, while others will cause the instruction to seem
3558 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3559 	 * To speed up the normal (success) code path, we should avoid checking
3560 	 * for misconfigurations which will anyway be caught by the processor
3561 	 * when using the merged vmcs02.
3562 	 */
3563 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3564 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3565 
3566 	if (CC(vmcs12->launch_state == launch))
3567 		return nested_vmx_fail(vcpu,
3568 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3569 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3570 
3571 	if (nested_vmx_check_controls(vcpu, vmcs12))
3572 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3573 
3574 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3575 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3576 
3577 	/*
3578 	 * We're finally done with prerequisite checking, and can start with
3579 	 * the nested entry.
3580 	 */
3581 	vmx->nested.nested_run_pending = 1;
3582 	vmx->nested.has_preemption_timer_deadline = false;
3583 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3584 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3585 		goto vmentry_failed;
3586 
3587 	/* Emulate processing of posted interrupts on VM-Enter. */
3588 	if (nested_cpu_has_posted_intr(vmcs12) &&
3589 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3590 		vmx->nested.pi_pending = true;
3591 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3592 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3593 	}
3594 
3595 	/* Hide L1D cache contents from the nested guest.  */
3596 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3597 
3598 	/*
3599 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3600 	 * also be used as part of restoring nVMX state for
3601 	 * snapshot restore (migration).
3602 	 *
3603 	 * In this flow, it is assumed that vmcs12 cache was
3604 	 * transferred as part of captured nVMX state and should
3605 	 * therefore not be read from guest memory (which may not
3606 	 * exist on destination host yet).
3607 	 */
3608 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3609 
3610 	switch (vmcs12->guest_activity_state) {
3611 	case GUEST_ACTIVITY_HLT:
3612 		/*
3613 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3614 		 * awakened by event injection or by an NMI-window VM-exit or
3615 		 * by an interrupt-window VM-exit, halt the vcpu.
3616 		 */
3617 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3618 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3619 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3620 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3621 			vmx->nested.nested_run_pending = 0;
3622 			return kvm_vcpu_halt(vcpu);
3623 		}
3624 		break;
3625 	case GUEST_ACTIVITY_WAIT_SIPI:
3626 		vmx->nested.nested_run_pending = 0;
3627 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3628 		break;
3629 	default:
3630 		break;
3631 	}
3632 
3633 	return 1;
3634 
3635 vmentry_failed:
3636 	vmx->nested.nested_run_pending = 0;
3637 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3638 		return 0;
3639 	if (status == NVMX_VMENTRY_VMEXIT)
3640 		return 1;
3641 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3642 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3643 }
3644 
3645 /*
3646  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3647  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3648  * This function returns the new value we should put in vmcs12.guest_cr0.
3649  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3650  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3651  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3652  *     didn't trap the bit, because if L1 did, so would L0).
3653  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3654  *     been modified by L2, and L1 knows it. So just leave the old value of
3655  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3656  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3657  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3658  *     changed these bits, and therefore they need to be updated, but L0
3659  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3660  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3661  */
3662 static inline unsigned long
3663 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3664 {
3665 	return
3666 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3667 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3668 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3669 			vcpu->arch.cr0_guest_owned_bits));
3670 }
3671 
3672 static inline unsigned long
3673 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3674 {
3675 	return
3676 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3677 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3678 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3679 			vcpu->arch.cr4_guest_owned_bits));
3680 }
3681 
3682 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3683 				      struct vmcs12 *vmcs12)
3684 {
3685 	u32 idt_vectoring;
3686 	unsigned int nr;
3687 
3688 	if (vcpu->arch.exception.injected) {
3689 		nr = vcpu->arch.exception.nr;
3690 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3691 
3692 		if (kvm_exception_is_soft(nr)) {
3693 			vmcs12->vm_exit_instruction_len =
3694 				vcpu->arch.event_exit_inst_len;
3695 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3696 		} else
3697 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3698 
3699 		if (vcpu->arch.exception.has_error_code) {
3700 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3701 			vmcs12->idt_vectoring_error_code =
3702 				vcpu->arch.exception.error_code;
3703 		}
3704 
3705 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3706 	} else if (vcpu->arch.nmi_injected) {
3707 		vmcs12->idt_vectoring_info_field =
3708 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3709 	} else if (vcpu->arch.interrupt.injected) {
3710 		nr = vcpu->arch.interrupt.nr;
3711 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3712 
3713 		if (vcpu->arch.interrupt.soft) {
3714 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3715 			vmcs12->vm_entry_instruction_len =
3716 				vcpu->arch.event_exit_inst_len;
3717 		} else
3718 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3719 
3720 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3721 	}
3722 }
3723 
3724 
3725 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3726 {
3727 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3728 	gfn_t gfn;
3729 
3730 	/*
3731 	 * Don't need to mark the APIC access page dirty; it is never
3732 	 * written to by the CPU during APIC virtualization.
3733 	 */
3734 
3735 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3736 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3737 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3738 	}
3739 
3740 	if (nested_cpu_has_posted_intr(vmcs12)) {
3741 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3742 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3743 	}
3744 }
3745 
3746 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3747 {
3748 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3749 	int max_irr;
3750 	void *vapic_page;
3751 	u16 status;
3752 
3753 	if (!vmx->nested.pi_pending)
3754 		return 0;
3755 
3756 	if (!vmx->nested.pi_desc)
3757 		goto mmio_needed;
3758 
3759 	vmx->nested.pi_pending = false;
3760 
3761 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3762 		return 0;
3763 
3764 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3765 	if (max_irr != 256) {
3766 		vapic_page = vmx->nested.virtual_apic_map.hva;
3767 		if (!vapic_page)
3768 			goto mmio_needed;
3769 
3770 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3771 			vapic_page, &max_irr);
3772 		status = vmcs_read16(GUEST_INTR_STATUS);
3773 		if ((u8)max_irr > ((u8)status & 0xff)) {
3774 			status &= ~0xff;
3775 			status |= (u8)max_irr;
3776 			vmcs_write16(GUEST_INTR_STATUS, status);
3777 		}
3778 	}
3779 
3780 	nested_mark_vmcs12_pages_dirty(vcpu);
3781 	return 0;
3782 
3783 mmio_needed:
3784 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3785 	return -ENXIO;
3786 }
3787 
3788 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3789 					       unsigned long exit_qual)
3790 {
3791 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3792 	unsigned int nr = vcpu->arch.exception.nr;
3793 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3794 
3795 	if (vcpu->arch.exception.has_error_code) {
3796 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3797 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3798 	}
3799 
3800 	if (kvm_exception_is_soft(nr))
3801 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3802 	else
3803 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3804 
3805 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3806 	    vmx_get_nmi_mask(vcpu))
3807 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3808 
3809 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3810 }
3811 
3812 /*
3813  * Returns true if a debug trap is pending delivery.
3814  *
3815  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3816  * exception may be inferred from the presence of an exception payload.
3817  */
3818 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3819 {
3820 	return vcpu->arch.exception.pending &&
3821 			vcpu->arch.exception.nr == DB_VECTOR &&
3822 			vcpu->arch.exception.payload;
3823 }
3824 
3825 /*
3826  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3827  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3828  * represents these debug traps with a payload that is said to be compatible
3829  * with the 'pending debug exceptions' field, write the payload to the VMCS
3830  * field if a VM-exit is delivered before the debug trap.
3831  */
3832 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3833 {
3834 	if (vmx_pending_dbg_trap(vcpu))
3835 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3836 			    vcpu->arch.exception.payload);
3837 }
3838 
3839 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3840 {
3841 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3842 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3843 }
3844 
3845 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3846 {
3847 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3848 	unsigned long exit_qual;
3849 	bool block_nested_events =
3850 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3851 	bool mtf_pending = vmx->nested.mtf_pending;
3852 	struct kvm_lapic *apic = vcpu->arch.apic;
3853 
3854 	/*
3855 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3856 	 * this state is discarded.
3857 	 */
3858 	if (!block_nested_events)
3859 		vmx->nested.mtf_pending = false;
3860 
3861 	if (lapic_in_kernel(vcpu) &&
3862 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3863 		if (block_nested_events)
3864 			return -EBUSY;
3865 		nested_vmx_update_pending_dbg(vcpu);
3866 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3867 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3868 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3869 		return 0;
3870 	}
3871 
3872 	if (lapic_in_kernel(vcpu) &&
3873 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3874 		if (block_nested_events)
3875 			return -EBUSY;
3876 
3877 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3878 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3879 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3880 						apic->sipi_vector & 0xFFUL);
3881 		return 0;
3882 	}
3883 
3884 	/*
3885 	 * Process any exceptions that are not debug traps before MTF.
3886 	 *
3887 	 * Note that only a pending nested run can block a pending exception.
3888 	 * Otherwise an injected NMI/interrupt should either be
3889 	 * lost or delivered to the nested hypervisor in the IDT_VECTORING_INFO,
3890 	 * while delivering the pending exception.
3891 	 */
3892 
3893 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3894 		if (vmx->nested.nested_run_pending)
3895 			return -EBUSY;
3896 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3897 			goto no_vmexit;
3898 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3899 		return 0;
3900 	}
3901 
3902 	if (mtf_pending) {
3903 		if (block_nested_events)
3904 			return -EBUSY;
3905 		nested_vmx_update_pending_dbg(vcpu);
3906 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3907 		return 0;
3908 	}
3909 
3910 	if (vcpu->arch.exception.pending) {
3911 		if (vmx->nested.nested_run_pending)
3912 			return -EBUSY;
3913 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3914 			goto no_vmexit;
3915 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3916 		return 0;
3917 	}
3918 
3919 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3920 		if (block_nested_events)
3921 			return -EBUSY;
3922 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3923 		return 0;
3924 	}
3925 
3926 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3927 		if (block_nested_events)
3928 			return -EBUSY;
3929 		goto no_vmexit;
3930 	}
3931 
3932 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3933 		if (block_nested_events)
3934 			return -EBUSY;
3935 		if (!nested_exit_on_nmi(vcpu))
3936 			goto no_vmexit;
3937 
3938 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3939 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3940 				  INTR_INFO_VALID_MASK, 0);
3941 		/*
3942 		 * The NMI-triggered VM exit counts as injection:
3943 		 * clear this one and block further NMIs.
3944 		 */
3945 		vcpu->arch.nmi_pending = 0;
3946 		vmx_set_nmi_mask(vcpu, true);
3947 		return 0;
3948 	}
3949 
3950 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3951 		if (block_nested_events)
3952 			return -EBUSY;
3953 		if (!nested_exit_on_intr(vcpu))
3954 			goto no_vmexit;
3955 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3956 		return 0;
3957 	}
3958 
3959 no_vmexit:
3960 	return vmx_complete_nested_posted_interrupt(vcpu);
3961 }
3962 
3963 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3964 {
3965 	ktime_t remaining =
3966 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3967 	u64 value;
3968 
3969 	if (ktime_to_ns(remaining) <= 0)
3970 		return 0;
3971 
3972 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3973 	do_div(value, 1000000);
3974 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3975 }
3976 
3977 static bool is_vmcs12_ext_field(unsigned long field)
3978 {
3979 	switch (field) {
3980 	case GUEST_ES_SELECTOR:
3981 	case GUEST_CS_SELECTOR:
3982 	case GUEST_SS_SELECTOR:
3983 	case GUEST_DS_SELECTOR:
3984 	case GUEST_FS_SELECTOR:
3985 	case GUEST_GS_SELECTOR:
3986 	case GUEST_LDTR_SELECTOR:
3987 	case GUEST_TR_SELECTOR:
3988 	case GUEST_ES_LIMIT:
3989 	case GUEST_CS_LIMIT:
3990 	case GUEST_SS_LIMIT:
3991 	case GUEST_DS_LIMIT:
3992 	case GUEST_FS_LIMIT:
3993 	case GUEST_GS_LIMIT:
3994 	case GUEST_LDTR_LIMIT:
3995 	case GUEST_TR_LIMIT:
3996 	case GUEST_GDTR_LIMIT:
3997 	case GUEST_IDTR_LIMIT:
3998 	case GUEST_ES_AR_BYTES:
3999 	case GUEST_DS_AR_BYTES:
4000 	case GUEST_FS_AR_BYTES:
4001 	case GUEST_GS_AR_BYTES:
4002 	case GUEST_LDTR_AR_BYTES:
4003 	case GUEST_TR_AR_BYTES:
4004 	case GUEST_ES_BASE:
4005 	case GUEST_CS_BASE:
4006 	case GUEST_SS_BASE:
4007 	case GUEST_DS_BASE:
4008 	case GUEST_FS_BASE:
4009 	case GUEST_GS_BASE:
4010 	case GUEST_LDTR_BASE:
4011 	case GUEST_TR_BASE:
4012 	case GUEST_GDTR_BASE:
4013 	case GUEST_IDTR_BASE:
4014 	case GUEST_PENDING_DBG_EXCEPTIONS:
4015 	case GUEST_BNDCFGS:
4016 		return true;
4017 	default:
4018 		break;
4019 	}
4020 
4021 	return false;
4022 }
4023 
4024 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4025 				       struct vmcs12 *vmcs12)
4026 {
4027 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4028 
4029 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4030 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4031 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4032 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4033 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4034 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4035 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4036 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4037 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4038 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4039 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4040 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4041 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4042 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4043 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4044 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4045 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4046 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4047 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4048 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4049 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4050 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4051 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4052 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4053 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4054 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4055 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4056 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4057 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4058 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4059 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4060 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4061 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4062 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4063 	vmcs12->guest_pending_dbg_exceptions =
4064 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4065 	if (kvm_mpx_supported())
4066 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
4067 
4068 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4069 }
4070 
4071 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4072 				       struct vmcs12 *vmcs12)
4073 {
4074 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4075 	int cpu;
4076 
4077 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4078 		return;
4079 
4080 
4081 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4082 
4083 	cpu = get_cpu();
4084 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4085 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4086 
4087 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4088 
4089 	vmx->loaded_vmcs = &vmx->vmcs01;
4090 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4091 	put_cpu();
4092 }
4093 
4094 /*
4095  * Update the guest state fields of vmcs12 to reflect changes that
4096  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4097  * VM-entry controls is also updated, since this is really a guest
4098  * state bit.)
4099  */
4100 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4101 {
4102 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4103 
4104 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4105 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4106 
4107 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4108 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4109 
4110 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4111 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4112 
4113 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4114 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4115 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4116 
4117 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4118 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4119 
4120 	vmcs12->guest_interruptibility_info =
4121 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4122 
4123 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4124 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4125 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4126 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4127 	else
4128 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4129 
4130 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4131 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4132 	    !vmx->nested.nested_run_pending)
4133 		vmcs12->vmx_preemption_timer_value =
4134 			vmx_get_preemption_timer_value(vcpu);
4135 
4136 	/*
4137 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4138 	 * own CR3 without exiting. If it has changed it, we must keep it.
4139 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4140 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4141 	 *
4142 	 * Additionally, restore L2's PDPTR to vmcs12.
4143 	 */
4144 	if (enable_ept) {
4145 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4146 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4147 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4148 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4149 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4150 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4151 		}
4152 	}
4153 
4154 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4155 
4156 	if (nested_cpu_has_vid(vmcs12))
4157 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4158 
4159 	vmcs12->vm_entry_controls =
4160 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4161 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4162 
4163 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4164 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4165 
4166 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4167 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4168 }
4169 
4170 /*
4171  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4172  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4173  * and this function updates it to reflect the changes to the guest state while
4174  * L2 was running (and perhaps made some exits which were handled directly by L0
4175  * without going back to L1), and to reflect the exit reason.
4176  * Note that we do not have to copy here all VMCS fields, just those that
4177  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4178  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4179  * which already writes to vmcs12 directly.
4180  */
4181 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4182 			   u32 vm_exit_reason, u32 exit_intr_info,
4183 			   unsigned long exit_qualification)
4184 {
4185 	/* update exit information fields: */
4186 	vmcs12->vm_exit_reason = vm_exit_reason;
4187 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4188 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4189 	vmcs12->exit_qualification = exit_qualification;
4190 	vmcs12->vm_exit_intr_info = exit_intr_info;
4191 
4192 	vmcs12->idt_vectoring_info_field = 0;
4193 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4194 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4195 
4196 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4197 		vmcs12->launch_state = 1;
4198 
4199 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4200 		 * instead of reading the real value. */
4201 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4202 
4203 		/*
4204 		 * Transfer the event that L0 or L1 may wanted to inject into
4205 		 * L2 to IDT_VECTORING_INFO_FIELD.
4206 		 */
4207 		vmcs12_save_pending_event(vcpu, vmcs12);
4208 
4209 		/*
4210 		 * According to spec, there's no need to store the guest's
4211 		 * MSRs if the exit is due to a VM-entry failure that occurs
4212 		 * during or after loading the guest state. Since this exit
4213 		 * does not fall in that category, we need to save the MSRs.
4214 		 */
4215 		if (nested_vmx_store_msr(vcpu,
4216 					 vmcs12->vm_exit_msr_store_addr,
4217 					 vmcs12->vm_exit_msr_store_count))
4218 			nested_vmx_abort(vcpu,
4219 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4220 	}
4221 
4222 	/*
4223 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4224 	 * preserved above and would only end up incorrectly in L1.
4225 	 */
4226 	vcpu->arch.nmi_injected = false;
4227 	kvm_clear_exception_queue(vcpu);
4228 	kvm_clear_interrupt_queue(vcpu);
4229 }
4230 
4231 /*
4232  * A part of what we need to when the nested L2 guest exits and we want to
4233  * run its L1 parent, is to reset L1's guest state to the host state specified
4234  * in vmcs12.
4235  * This function is to be called not only on normal nested exit, but also on
4236  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4237  * Failures During or After Loading Guest State").
4238  * This function should be called when the active VMCS is L1's (vmcs01).
4239  */
4240 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4241 				   struct vmcs12 *vmcs12)
4242 {
4243 	enum vm_entry_failure_code ignored;
4244 	struct kvm_segment seg;
4245 
4246 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4247 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4248 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4249 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4250 	else
4251 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4252 	vmx_set_efer(vcpu, vcpu->arch.efer);
4253 
4254 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4255 	kvm_rip_write(vcpu, vmcs12->host_rip);
4256 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4257 	vmx_set_interrupt_shadow(vcpu, 0);
4258 
4259 	/*
4260 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4261 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4262 	 *
4263 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4264 	 * (KVM doesn't change it);
4265 	 */
4266 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4267 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4268 
4269 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4270 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4271 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4272 
4273 	nested_ept_uninit_mmu_context(vcpu);
4274 
4275 	/*
4276 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4277 	 * couldn't have changed.
4278 	 */
4279 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4280 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4281 
4282 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4283 
4284 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4285 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4286 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4287 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4288 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4289 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4290 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4291 
4292 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4293 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4294 		vmcs_write64(GUEST_BNDCFGS, 0);
4295 
4296 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4297 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4298 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4299 	}
4300 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4301 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4302 					 vmcs12->host_ia32_perf_global_ctrl));
4303 
4304 	/* Set L1 segment info according to Intel SDM
4305 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4306 	seg = (struct kvm_segment) {
4307 		.base = 0,
4308 		.limit = 0xFFFFFFFF,
4309 		.selector = vmcs12->host_cs_selector,
4310 		.type = 11,
4311 		.present = 1,
4312 		.s = 1,
4313 		.g = 1
4314 	};
4315 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4316 		seg.l = 1;
4317 	else
4318 		seg.db = 1;
4319 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4320 	seg = (struct kvm_segment) {
4321 		.base = 0,
4322 		.limit = 0xFFFFFFFF,
4323 		.type = 3,
4324 		.present = 1,
4325 		.s = 1,
4326 		.db = 1,
4327 		.g = 1
4328 	};
4329 	seg.selector = vmcs12->host_ds_selector;
4330 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4331 	seg.selector = vmcs12->host_es_selector;
4332 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4333 	seg.selector = vmcs12->host_ss_selector;
4334 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4335 	seg.selector = vmcs12->host_fs_selector;
4336 	seg.base = vmcs12->host_fs_base;
4337 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4338 	seg.selector = vmcs12->host_gs_selector;
4339 	seg.base = vmcs12->host_gs_base;
4340 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4341 	seg = (struct kvm_segment) {
4342 		.base = vmcs12->host_tr_base,
4343 		.limit = 0x67,
4344 		.selector = vmcs12->host_tr_selector,
4345 		.type = 11,
4346 		.present = 1
4347 	};
4348 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4349 
4350 	memset(&seg, 0, sizeof(seg));
4351 	seg.unusable = 1;
4352 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4353 
4354 	kvm_set_dr(vcpu, 7, 0x400);
4355 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4356 
4357 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4358 				vmcs12->vm_exit_msr_load_count))
4359 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4360 
4361 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4362 }
4363 
4364 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4365 {
4366 	struct vmx_uret_msr *efer_msr;
4367 	unsigned int i;
4368 
4369 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4370 		return vmcs_read64(GUEST_IA32_EFER);
4371 
4372 	if (cpu_has_load_ia32_efer())
4373 		return host_efer;
4374 
4375 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4376 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4377 			return vmx->msr_autoload.guest.val[i].value;
4378 	}
4379 
4380 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4381 	if (efer_msr)
4382 		return efer_msr->data;
4383 
4384 	return host_efer;
4385 }
4386 
4387 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4388 {
4389 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4390 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4391 	struct vmx_msr_entry g, h;
4392 	gpa_t gpa;
4393 	u32 i, j;
4394 
4395 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4396 
4397 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4398 		/*
4399 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4400 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4401 		 * and vcpu->arch.dr7 is not squirreled away before the
4402 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4403 		 */
4404 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4405 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4406 		else
4407 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4408 	}
4409 
4410 	/*
4411 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4412 	 * handle a variety of side effects to KVM's software model.
4413 	 */
4414 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4415 
4416 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4417 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4418 
4419 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4420 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4421 
4422 	nested_ept_uninit_mmu_context(vcpu);
4423 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4424 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4425 
4426 	/*
4427 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4428 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4429 	 * VMFail, like everything else we just need to ensure our
4430 	 * software model is up-to-date.
4431 	 */
4432 	if (enable_ept && is_pae_paging(vcpu))
4433 		ept_save_pdptrs(vcpu);
4434 
4435 	kvm_mmu_reset_context(vcpu);
4436 
4437 	/*
4438 	 * This nasty bit of open coding is a compromise between blindly
4439 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4440 	 * of VMFail), leaving the nested VM's MSRs in the software model
4441 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4442 	 * expensive since the lists are unbound by hardware).  For each
4443 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4444 	 * list, reload it from the exit load list if it exists and differs
4445 	 * from the guest value.  The intent is to stuff host state as
4446 	 * silently as possible, not to fully process the exit load list.
4447 	 */
4448 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4449 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4450 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4451 			pr_debug_ratelimited(
4452 				"%s read MSR index failed (%u, 0x%08llx)\n",
4453 				__func__, i, gpa);
4454 			goto vmabort;
4455 		}
4456 
4457 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4458 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4459 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4460 				pr_debug_ratelimited(
4461 					"%s read MSR failed (%u, 0x%08llx)\n",
4462 					__func__, j, gpa);
4463 				goto vmabort;
4464 			}
4465 			if (h.index != g.index)
4466 				continue;
4467 			if (h.value == g.value)
4468 				break;
4469 
4470 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4471 				pr_debug_ratelimited(
4472 					"%s check failed (%u, 0x%x, 0x%x)\n",
4473 					__func__, j, h.index, h.reserved);
4474 				goto vmabort;
4475 			}
4476 
4477 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4478 				pr_debug_ratelimited(
4479 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4480 					__func__, j, h.index, h.value);
4481 				goto vmabort;
4482 			}
4483 		}
4484 	}
4485 
4486 	return;
4487 
4488 vmabort:
4489 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4490 }
4491 
4492 /*
4493  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4494  * and modify vmcs12 to make it see what it would expect to see there if
4495  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4496  */
4497 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4498 		       u32 exit_intr_info, unsigned long exit_qualification)
4499 {
4500 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4501 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4502 
4503 	/* trying to cancel vmlaunch/vmresume is a bug */
4504 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4505 
4506 	/* Similarly, triple faults in L2 should never escape. */
4507 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
4508 
4509 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4510 		/*
4511 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4512 		 * Enlightened VMCS after migration and we still need to
4513 		 * do that when something is forcing L2->L1 exit prior to
4514 		 * the first L2 run.
4515 		 */
4516 		(void)nested_get_evmcs_page(vcpu);
4517 	}
4518 
4519 	/* Service the TLB flush request for L2 before switching to L1. */
4520 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
4521 		kvm_vcpu_flush_tlb_current(vcpu);
4522 
4523 	/*
4524 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4525 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4526 	 * up-to-date before switching to L1.
4527 	 */
4528 	if (enable_ept && is_pae_paging(vcpu))
4529 		vmx_ept_load_pdptrs(vcpu);
4530 
4531 	leave_guest_mode(vcpu);
4532 
4533 	if (nested_cpu_has_preemption_timer(vmcs12))
4534 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4535 
4536 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4537 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4538 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4539 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4540 	}
4541 
4542 	if (likely(!vmx->fail)) {
4543 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4544 
4545 		if (vm_exit_reason != -1)
4546 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4547 				       exit_intr_info, exit_qualification);
4548 
4549 		/*
4550 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4551 		 * also be used to capture vmcs12 cache as part of
4552 		 * capturing nVMX state for snapshot (migration).
4553 		 *
4554 		 * Otherwise, this flush will dirty guest memory at a
4555 		 * point it is already assumed by user-space to be
4556 		 * immutable.
4557 		 */
4558 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4559 	} else {
4560 		/*
4561 		 * The only expected VM-instruction error is "VM entry with
4562 		 * invalid control field(s)." Anything else indicates a
4563 		 * problem with L0.  And we should never get here with a
4564 		 * VMFail of any type if early consistency checks are enabled.
4565 		 */
4566 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4567 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4568 		WARN_ON_ONCE(nested_early_check);
4569 	}
4570 
4571 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4572 
4573 	/* Update any VMCS fields that might have changed while L2 ran */
4574 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4575 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4576 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4577 	if (kvm_has_tsc_control)
4578 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4579 
4580 	if (vmx->nested.l1_tpr_threshold != -1)
4581 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4582 
4583 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4584 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4585 		vmx_set_virtual_apic_mode(vcpu);
4586 	}
4587 
4588 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4589 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4590 		vmx_update_cpu_dirty_logging(vcpu);
4591 	}
4592 
4593 	/* Unpin physical memory we referred to in vmcs02 */
4594 	if (vmx->nested.apic_access_page) {
4595 		kvm_release_page_clean(vmx->nested.apic_access_page);
4596 		vmx->nested.apic_access_page = NULL;
4597 	}
4598 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4599 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4600 	vmx->nested.pi_desc = NULL;
4601 
4602 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4603 		vmx->nested.reload_vmcs01_apic_access_page = false;
4604 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4605 	}
4606 
4607 	if ((vm_exit_reason != -1) &&
4608 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4609 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4610 
4611 	/* in case we halted in L2 */
4612 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4613 
4614 	if (likely(!vmx->fail)) {
4615 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4616 		    nested_exit_intr_ack_set(vcpu)) {
4617 			int irq = kvm_cpu_get_interrupt(vcpu);
4618 			WARN_ON(irq < 0);
4619 			vmcs12->vm_exit_intr_info = irq |
4620 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4621 		}
4622 
4623 		if (vm_exit_reason != -1)
4624 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4625 						       vmcs12->exit_qualification,
4626 						       vmcs12->idt_vectoring_info_field,
4627 						       vmcs12->vm_exit_intr_info,
4628 						       vmcs12->vm_exit_intr_error_code,
4629 						       KVM_ISA_VMX);
4630 
4631 		load_vmcs12_host_state(vcpu, vmcs12);
4632 
4633 		return;
4634 	}
4635 
4636 	/*
4637 	 * After an early L2 VM-entry failure, we're now back
4638 	 * in L1 which thinks it just finished a VMLAUNCH or
4639 	 * VMRESUME instruction, so we need to set the failure
4640 	 * flag and the VM-instruction error field of the VMCS
4641 	 * accordingly, and skip the emulated instruction.
4642 	 */
4643 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4644 
4645 	/*
4646 	 * Restore L1's host state to KVM's software model.  We're here
4647 	 * because a consistency check was caught by hardware, which
4648 	 * means some amount of guest state has been propagated to KVM's
4649 	 * model and needs to be unwound to the host's state.
4650 	 */
4651 	nested_vmx_restore_host_state(vcpu);
4652 
4653 	vmx->fail = 0;
4654 }
4655 
4656 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4657 {
4658 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4659 }
4660 
4661 /*
4662  * Decode the memory-address operand of a vmx instruction, as recorded on an
4663  * exit caused by such an instruction (run by a guest hypervisor).
4664  * On success, returns 0. When the operand is invalid, returns 1 and throws
4665  * #UD, #GP, or #SS.
4666  */
4667 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4668 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4669 {
4670 	gva_t off;
4671 	bool exn;
4672 	struct kvm_segment s;
4673 
4674 	/*
4675 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4676 	 * Execution", on an exit, vmx_instruction_info holds most of the
4677 	 * addressing components of the operand. Only the displacement part
4678 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4679 	 * For how an actual address is calculated from all these components,
4680 	 * refer to Vol. 1, "Operand Addressing".
4681 	 */
4682 	int  scaling = vmx_instruction_info & 3;
4683 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4684 	bool is_reg = vmx_instruction_info & (1u << 10);
4685 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4686 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4687 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4688 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4689 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4690 
4691 	if (is_reg) {
4692 		kvm_queue_exception(vcpu, UD_VECTOR);
4693 		return 1;
4694 	}
4695 
4696 	/* Addr = segment_base + offset */
4697 	/* offset = base + [index * scale] + displacement */
4698 	off = exit_qualification; /* holds the displacement */
4699 	if (addr_size == 1)
4700 		off = (gva_t)sign_extend64(off, 31);
4701 	else if (addr_size == 0)
4702 		off = (gva_t)sign_extend64(off, 15);
4703 	if (base_is_valid)
4704 		off += kvm_register_read(vcpu, base_reg);
4705 	if (index_is_valid)
4706 		off += kvm_register_read(vcpu, index_reg) << scaling;
4707 	vmx_get_segment(vcpu, &s, seg_reg);
4708 
4709 	/*
4710 	 * The effective address, i.e. @off, of a memory operand is truncated
4711 	 * based on the address size of the instruction.  Note that this is
4712 	 * the *effective address*, i.e. the address prior to accounting for
4713 	 * the segment's base.
4714 	 */
4715 	if (addr_size == 1) /* 32 bit */
4716 		off &= 0xffffffff;
4717 	else if (addr_size == 0) /* 16 bit */
4718 		off &= 0xffff;
4719 
4720 	/* Checks for #GP/#SS exceptions. */
4721 	exn = false;
4722 	if (is_long_mode(vcpu)) {
4723 		/*
4724 		 * The virtual/linear address is never truncated in 64-bit
4725 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4726 		 * address when using FS/GS with a non-zero base.
4727 		 */
4728 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4729 			*ret = s.base + off;
4730 		else
4731 			*ret = off;
4732 
4733 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4734 		 * non-canonical form. This is the only check on the memory
4735 		 * destination for long mode!
4736 		 */
4737 		exn = is_noncanonical_address(*ret, vcpu);
4738 	} else {
4739 		/*
4740 		 * When not in long mode, the virtual/linear address is
4741 		 * unconditionally truncated to 32 bits regardless of the
4742 		 * address size.
4743 		 */
4744 		*ret = (s.base + off) & 0xffffffff;
4745 
4746 		/* Protected mode: apply checks for segment validity in the
4747 		 * following order:
4748 		 * - segment type check (#GP(0) may be thrown)
4749 		 * - usability check (#GP(0)/#SS(0))
4750 		 * - limit check (#GP(0)/#SS(0))
4751 		 */
4752 		if (wr)
4753 			/* #GP(0) if the destination operand is located in a
4754 			 * read-only data segment or any code segment.
4755 			 */
4756 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4757 		else
4758 			/* #GP(0) if the source operand is located in an
4759 			 * execute-only code segment
4760 			 */
4761 			exn = ((s.type & 0xa) == 8);
4762 		if (exn) {
4763 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4764 			return 1;
4765 		}
4766 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4767 		 */
4768 		exn = (s.unusable != 0);
4769 
4770 		/*
4771 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4772 		 * outside the segment limit.  All CPUs that support VMX ignore
4773 		 * limit checks for flat segments, i.e. segments with base==0,
4774 		 * limit==0xffffffff and of type expand-up data or code.
4775 		 */
4776 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4777 		     ((s.type & 8) || !(s.type & 4))))
4778 			exn = exn || ((u64)off + len - 1 > s.limit);
4779 	}
4780 	if (exn) {
4781 		kvm_queue_exception_e(vcpu,
4782 				      seg_reg == VCPU_SREG_SS ?
4783 						SS_VECTOR : GP_VECTOR,
4784 				      0);
4785 		return 1;
4786 	}
4787 
4788 	return 0;
4789 }
4790 
4791 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4792 {
4793 	struct vcpu_vmx *vmx;
4794 
4795 	if (!nested_vmx_allowed(vcpu))
4796 		return;
4797 
4798 	vmx = to_vmx(vcpu);
4799 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4800 		vmx->nested.msrs.entry_ctls_high |=
4801 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4802 		vmx->nested.msrs.exit_ctls_high |=
4803 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4804 	} else {
4805 		vmx->nested.msrs.entry_ctls_high &=
4806 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4807 		vmx->nested.msrs.exit_ctls_high &=
4808 				~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4809 	}
4810 }
4811 
4812 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4813 				int *ret)
4814 {
4815 	gva_t gva;
4816 	struct x86_exception e;
4817 	int r;
4818 
4819 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4820 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4821 				sizeof(*vmpointer), &gva)) {
4822 		*ret = 1;
4823 		return -EINVAL;
4824 	}
4825 
4826 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4827 	if (r != X86EMUL_CONTINUE) {
4828 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4829 		return -EINVAL;
4830 	}
4831 
4832 	return 0;
4833 }
4834 
4835 /*
4836  * Allocate a shadow VMCS and associate it with the currently loaded
4837  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4838  * VMCS is also VMCLEARed, so that it is ready for use.
4839  */
4840 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4841 {
4842 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4843 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4844 
4845 	/*
4846 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4847 	 * executes VMXON and free it when L1 executes VMXOFF.
4848 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4849 	 * here when vmcs01 already have an allocated shadow vmcs.
4850 	 */
4851 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4852 
4853 	if (!loaded_vmcs->shadow_vmcs) {
4854 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4855 		if (loaded_vmcs->shadow_vmcs)
4856 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4857 	}
4858 	return loaded_vmcs->shadow_vmcs;
4859 }
4860 
4861 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4862 {
4863 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4864 	int r;
4865 
4866 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4867 	if (r < 0)
4868 		goto out_vmcs02;
4869 
4870 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4871 	if (!vmx->nested.cached_vmcs12)
4872 		goto out_cached_vmcs12;
4873 
4874 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4875 	if (!vmx->nested.cached_shadow_vmcs12)
4876 		goto out_cached_shadow_vmcs12;
4877 
4878 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4879 		goto out_shadow_vmcs;
4880 
4881 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4882 		     HRTIMER_MODE_ABS_PINNED);
4883 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4884 
4885 	vmx->nested.vpid02 = allocate_vpid();
4886 
4887 	vmx->nested.vmcs02_initialized = false;
4888 	vmx->nested.vmxon = true;
4889 
4890 	if (vmx_pt_mode_is_host_guest()) {
4891 		vmx->pt_desc.guest.ctl = 0;
4892 		pt_update_intercept_for_msr(vcpu);
4893 	}
4894 
4895 	return 0;
4896 
4897 out_shadow_vmcs:
4898 	kfree(vmx->nested.cached_shadow_vmcs12);
4899 
4900 out_cached_shadow_vmcs12:
4901 	kfree(vmx->nested.cached_vmcs12);
4902 
4903 out_cached_vmcs12:
4904 	free_loaded_vmcs(&vmx->nested.vmcs02);
4905 
4906 out_vmcs02:
4907 	return -ENOMEM;
4908 }
4909 
4910 /* Emulate the VMXON instruction. */
4911 static int handle_vmon(struct kvm_vcpu *vcpu)
4912 {
4913 	int ret;
4914 	gpa_t vmptr;
4915 	uint32_t revision;
4916 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4917 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4918 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4919 
4920 	/*
4921 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4922 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4923 	 * 1 (see vmx_is_valid_cr4() for when we allow the guest to set this).
4924 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4925 	 * have already been checked by hardware, prior to the VM-exit for
4926 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4927 	 * that bit set to 1 in non-root mode.
4928 	 */
4929 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4930 		kvm_queue_exception(vcpu, UD_VECTOR);
4931 		return 1;
4932 	}
4933 
4934 	/* CPL=0 must be checked manually. */
4935 	if (vmx_get_cpl(vcpu)) {
4936 		kvm_inject_gp(vcpu, 0);
4937 		return 1;
4938 	}
4939 
4940 	if (vmx->nested.vmxon)
4941 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4942 
4943 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4944 			!= VMXON_NEEDED_FEATURES) {
4945 		kvm_inject_gp(vcpu, 0);
4946 		return 1;
4947 	}
4948 
4949 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4950 		return ret;
4951 
4952 	/*
4953 	 * SDM 3: 24.11.5
4954 	 * The first 4 bytes of VMXON region contain the supported
4955 	 * VMCS revision identifier
4956 	 *
4957 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4958 	 * which replaces physical address width with 32
4959 	 */
4960 	if (!page_address_valid(vcpu, vmptr))
4961 		return nested_vmx_failInvalid(vcpu);
4962 
4963 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4964 	    revision != VMCS12_REVISION)
4965 		return nested_vmx_failInvalid(vcpu);
4966 
4967 	vmx->nested.vmxon_ptr = vmptr;
4968 	ret = enter_vmx_operation(vcpu);
4969 	if (ret)
4970 		return ret;
4971 
4972 	return nested_vmx_succeed(vcpu);
4973 }
4974 
4975 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4976 {
4977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4978 
4979 	if (vmx->nested.current_vmptr == INVALID_GPA)
4980 		return;
4981 
4982 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4983 
4984 	if (enable_shadow_vmcs) {
4985 		/* copy to memory all shadowed fields in case
4986 		   they were modified */
4987 		copy_shadow_to_vmcs12(vmx);
4988 		vmx_disable_shadow_vmcs(vmx);
4989 	}
4990 	vmx->nested.posted_intr_nv = -1;
4991 
4992 	/* Flush VMCS12 to guest memory */
4993 	kvm_vcpu_write_guest_page(vcpu,
4994 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4995 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4996 
4997 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4998 
4999 	vmx->nested.current_vmptr = INVALID_GPA;
5000 }
5001 
5002 /* Emulate the VMXOFF instruction */
5003 static int handle_vmoff(struct kvm_vcpu *vcpu)
5004 {
5005 	if (!nested_vmx_check_permission(vcpu))
5006 		return 1;
5007 
5008 	free_nested(vcpu);
5009 
5010 	/* Process a latched INIT during time CPU was in VMX operation */
5011 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5012 
5013 	return nested_vmx_succeed(vcpu);
5014 }
5015 
5016 /* Emulate the VMCLEAR instruction */
5017 static int handle_vmclear(struct kvm_vcpu *vcpu)
5018 {
5019 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5020 	u32 zero = 0;
5021 	gpa_t vmptr;
5022 	u64 evmcs_gpa;
5023 	int r;
5024 
5025 	if (!nested_vmx_check_permission(vcpu))
5026 		return 1;
5027 
5028 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5029 		return r;
5030 
5031 	if (!page_address_valid(vcpu, vmptr))
5032 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5033 
5034 	if (vmptr == vmx->nested.vmxon_ptr)
5035 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5036 
5037 	/*
5038 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
5039 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
5040 	 * way to distinguish it from VMCS12) and we must not corrupt it by
5041 	 * writing to the non-existent 'launch_state' field. The area doesn't
5042 	 * have to be the currently active EVMCS on the calling CPU and there's
5043 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5044 	 * state. It is possible that the area will stay mapped as
5045 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5046 	 */
5047 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
5048 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
5049 		if (vmptr == vmx->nested.current_vmptr)
5050 			nested_release_vmcs12(vcpu);
5051 
5052 		kvm_vcpu_write_guest(vcpu,
5053 				     vmptr + offsetof(struct vmcs12,
5054 						      launch_state),
5055 				     &zero, sizeof(zero));
5056 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5057 		nested_release_evmcs(vcpu);
5058 	}
5059 
5060 	return nested_vmx_succeed(vcpu);
5061 }
5062 
5063 /* Emulate the VMLAUNCH instruction */
5064 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5065 {
5066 	return nested_vmx_run(vcpu, true);
5067 }
5068 
5069 /* Emulate the VMRESUME instruction */
5070 static int handle_vmresume(struct kvm_vcpu *vcpu)
5071 {
5072 
5073 	return nested_vmx_run(vcpu, false);
5074 }
5075 
5076 static int handle_vmread(struct kvm_vcpu *vcpu)
5077 {
5078 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5079 						    : get_vmcs12(vcpu);
5080 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5081 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5082 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5083 	struct x86_exception e;
5084 	unsigned long field;
5085 	u64 value;
5086 	gva_t gva = 0;
5087 	short offset;
5088 	int len, r;
5089 
5090 	if (!nested_vmx_check_permission(vcpu))
5091 		return 1;
5092 
5093 	/*
5094 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5095 	 * any VMREAD sets the ALU flags for VMfailInvalid.
5096 	 */
5097 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5098 	    (is_guest_mode(vcpu) &&
5099 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5100 		return nested_vmx_failInvalid(vcpu);
5101 
5102 	/* Decode instruction info and find the field to read */
5103 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5104 
5105 	offset = vmcs_field_to_offset(field);
5106 	if (offset < 0)
5107 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5108 
5109 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5110 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5111 
5112 	/* Read the field, zero-extended to a u64 value */
5113 	value = vmcs12_read_any(vmcs12, field, offset);
5114 
5115 	/*
5116 	 * Now copy part of this value to register or memory, as requested.
5117 	 * Note that the number of bits actually copied is 32 or 64 depending
5118 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5119 	 */
5120 	if (instr_info & BIT(10)) {
5121 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5122 	} else {
5123 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5124 		if (get_vmx_mem_address(vcpu, exit_qualification,
5125 					instr_info, true, len, &gva))
5126 			return 1;
5127 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5128 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5129 		if (r != X86EMUL_CONTINUE)
5130 			return kvm_handle_memory_failure(vcpu, r, &e);
5131 	}
5132 
5133 	return nested_vmx_succeed(vcpu);
5134 }
5135 
5136 static bool is_shadow_field_rw(unsigned long field)
5137 {
5138 	switch (field) {
5139 #define SHADOW_FIELD_RW(x, y) case x:
5140 #include "vmcs_shadow_fields.h"
5141 		return true;
5142 	default:
5143 		break;
5144 	}
5145 	return false;
5146 }
5147 
5148 static bool is_shadow_field_ro(unsigned long field)
5149 {
5150 	switch (field) {
5151 #define SHADOW_FIELD_RO(x, y) case x:
5152 #include "vmcs_shadow_fields.h"
5153 		return true;
5154 	default:
5155 		break;
5156 	}
5157 	return false;
5158 }
5159 
5160 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5161 {
5162 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5163 						    : get_vmcs12(vcpu);
5164 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5165 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5166 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5167 	struct x86_exception e;
5168 	unsigned long field;
5169 	short offset;
5170 	gva_t gva;
5171 	int len, r;
5172 
5173 	/*
5174 	 * The value to write might be 32 or 64 bits, depending on L1's long
5175 	 * mode, and eventually we need to write that into a field of several
5176 	 * possible lengths. The code below first zero-extends the value to 64
5177 	 * bit (value), and then copies only the appropriate number of
5178 	 * bits into the vmcs12 field.
5179 	 */
5180 	u64 value = 0;
5181 
5182 	if (!nested_vmx_check_permission(vcpu))
5183 		return 1;
5184 
5185 	/*
5186 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5187 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5188 	 */
5189 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5190 	    (is_guest_mode(vcpu) &&
5191 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5192 		return nested_vmx_failInvalid(vcpu);
5193 
5194 	if (instr_info & BIT(10))
5195 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5196 	else {
5197 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5198 		if (get_vmx_mem_address(vcpu, exit_qualification,
5199 					instr_info, false, len, &gva))
5200 			return 1;
5201 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5202 		if (r != X86EMUL_CONTINUE)
5203 			return kvm_handle_memory_failure(vcpu, r, &e);
5204 	}
5205 
5206 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5207 
5208 	offset = vmcs_field_to_offset(field);
5209 	if (offset < 0)
5210 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5211 
5212 	/*
5213 	 * If the vCPU supports "VMWRITE to any supported field in the
5214 	 * VMCS," then the "read-only" fields are actually read/write.
5215 	 */
5216 	if (vmcs_field_readonly(field) &&
5217 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5218 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5219 
5220 	/*
5221 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5222 	 * vmcs12, else we may crush a field or consume a stale value.
5223 	 */
5224 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5225 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5226 
5227 	/*
5228 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5229 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5230 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5231 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5232 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5233 	 * the stripped down value, L2 sees the full value as stored by KVM).
5234 	 */
5235 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5236 		value &= 0x1f0ff;
5237 
5238 	vmcs12_write_any(vmcs12, field, offset, value);
5239 
5240 	/*
5241 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5242 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5243 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5244 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5245 	 */
5246 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5247 		/*
5248 		 * L1 can read these fields without exiting, ensure the
5249 		 * shadow VMCS is up-to-date.
5250 		 */
5251 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5252 			preempt_disable();
5253 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5254 
5255 			__vmcs_writel(field, value);
5256 
5257 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5258 			vmcs_load(vmx->loaded_vmcs->vmcs);
5259 			preempt_enable();
5260 		}
5261 		vmx->nested.dirty_vmcs12 = true;
5262 	}
5263 
5264 	return nested_vmx_succeed(vcpu);
5265 }
5266 
5267 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5268 {
5269 	vmx->nested.current_vmptr = vmptr;
5270 	if (enable_shadow_vmcs) {
5271 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5272 		vmcs_write64(VMCS_LINK_POINTER,
5273 			     __pa(vmx->vmcs01.shadow_vmcs));
5274 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5275 	}
5276 	vmx->nested.dirty_vmcs12 = true;
5277 }
5278 
5279 /* Emulate the VMPTRLD instruction */
5280 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5281 {
5282 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5283 	gpa_t vmptr;
5284 	int r;
5285 
5286 	if (!nested_vmx_check_permission(vcpu))
5287 		return 1;
5288 
5289 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5290 		return r;
5291 
5292 	if (!page_address_valid(vcpu, vmptr))
5293 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5294 
5295 	if (vmptr == vmx->nested.vmxon_ptr)
5296 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5297 
5298 	/* Forbid normal VMPTRLD if Enlightened version was used */
5299 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5300 		return 1;
5301 
5302 	if (vmx->nested.current_vmptr != vmptr) {
5303 		struct kvm_host_map map;
5304 		struct vmcs12 *new_vmcs12;
5305 
5306 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
5307 			/*
5308 			 * Reads from an unbacked page return all 1s,
5309 			 * which means that the 32 bits located at the
5310 			 * given physical address won't match the required
5311 			 * VMCS12_REVISION identifier.
5312 			 */
5313 			return nested_vmx_fail(vcpu,
5314 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5315 		}
5316 
5317 		new_vmcs12 = map.hva;
5318 
5319 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5320 		    (new_vmcs12->hdr.shadow_vmcs &&
5321 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5322 			kvm_vcpu_unmap(vcpu, &map, false);
5323 			return nested_vmx_fail(vcpu,
5324 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5325 		}
5326 
5327 		nested_release_vmcs12(vcpu);
5328 
5329 		/*
5330 		 * Load VMCS12 from guest memory since it is not already
5331 		 * cached.
5332 		 */
5333 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
5334 		kvm_vcpu_unmap(vcpu, &map, false);
5335 
5336 		set_current_vmptr(vmx, vmptr);
5337 	}
5338 
5339 	return nested_vmx_succeed(vcpu);
5340 }
5341 
5342 /* Emulate the VMPTRST instruction */
5343 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5344 {
5345 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5346 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5347 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5348 	struct x86_exception e;
5349 	gva_t gva;
5350 	int r;
5351 
5352 	if (!nested_vmx_check_permission(vcpu))
5353 		return 1;
5354 
5355 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5356 		return 1;
5357 
5358 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5359 				true, sizeof(gpa_t), &gva))
5360 		return 1;
5361 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5362 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5363 					sizeof(gpa_t), &e);
5364 	if (r != X86EMUL_CONTINUE)
5365 		return kvm_handle_memory_failure(vcpu, r, &e);
5366 
5367 	return nested_vmx_succeed(vcpu);
5368 }
5369 
5370 /* Emulate the INVEPT instruction */
5371 static int handle_invept(struct kvm_vcpu *vcpu)
5372 {
5373 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5374 	u32 vmx_instruction_info, types;
5375 	unsigned long type, roots_to_free;
5376 	struct kvm_mmu *mmu;
5377 	gva_t gva;
5378 	struct x86_exception e;
5379 	struct {
5380 		u64 eptp, gpa;
5381 	} operand;
5382 	int i, r;
5383 
5384 	if (!(vmx->nested.msrs.secondary_ctls_high &
5385 	      SECONDARY_EXEC_ENABLE_EPT) ||
5386 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5387 		kvm_queue_exception(vcpu, UD_VECTOR);
5388 		return 1;
5389 	}
5390 
5391 	if (!nested_vmx_check_permission(vcpu))
5392 		return 1;
5393 
5394 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5395 	type = kvm_register_read(vcpu, (vmx_instruction_info >> 28) & 0xf);
5396 
5397 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5398 
5399 	if (type >= 32 || !(types & (1 << type)))
5400 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5401 
5402 	/* According to the Intel VMX instruction reference, the memory
5403 	 * operand is read even if it isn't needed (e.g., for type==global)
5404 	 */
5405 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5406 			vmx_instruction_info, false, sizeof(operand), &gva))
5407 		return 1;
5408 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5409 	if (r != X86EMUL_CONTINUE)
5410 		return kvm_handle_memory_failure(vcpu, r, &e);
5411 
5412 	/*
5413 	 * Nested EPT roots are always held through guest_mmu,
5414 	 * not root_mmu.
5415 	 */
5416 	mmu = &vcpu->arch.guest_mmu;
5417 
5418 	switch (type) {
5419 	case VMX_EPT_EXTENT_CONTEXT:
5420 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5421 			return nested_vmx_fail(vcpu,
5422 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5423 
5424 		roots_to_free = 0;
5425 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
5426 					    operand.eptp))
5427 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5428 
5429 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5430 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5431 						    mmu->prev_roots[i].pgd,
5432 						    operand.eptp))
5433 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5434 		}
5435 		break;
5436 	case VMX_EPT_EXTENT_GLOBAL:
5437 		roots_to_free = KVM_MMU_ROOTS_ALL;
5438 		break;
5439 	default:
5440 		BUG();
5441 		break;
5442 	}
5443 
5444 	if (roots_to_free)
5445 		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
5446 
5447 	return nested_vmx_succeed(vcpu);
5448 }
5449 
5450 static int handle_invvpid(struct kvm_vcpu *vcpu)
5451 {
5452 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5453 	u32 vmx_instruction_info;
5454 	unsigned long type, types;
5455 	gva_t gva;
5456 	struct x86_exception e;
5457 	struct {
5458 		u64 vpid;
5459 		u64 gla;
5460 	} operand;
5461 	u16 vpid02;
5462 	int r;
5463 
5464 	if (!(vmx->nested.msrs.secondary_ctls_high &
5465 	      SECONDARY_EXEC_ENABLE_VPID) ||
5466 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5467 		kvm_queue_exception(vcpu, UD_VECTOR);
5468 		return 1;
5469 	}
5470 
5471 	if (!nested_vmx_check_permission(vcpu))
5472 		return 1;
5473 
5474 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5475 	type = kvm_register_read(vcpu, (vmx_instruction_info >> 28) & 0xf);
5476 
5477 	types = (vmx->nested.msrs.vpid_caps &
5478 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5479 
5480 	if (type >= 32 || !(types & (1 << type)))
5481 		return nested_vmx_fail(vcpu,
5482 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5483 
5484 	/* according to the intel vmx instruction reference, the memory
5485 	 * operand is read even if it isn't needed (e.g., for type==global)
5486 	 */
5487 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5488 			vmx_instruction_info, false, sizeof(operand), &gva))
5489 		return 1;
5490 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5491 	if (r != X86EMUL_CONTINUE)
5492 		return kvm_handle_memory_failure(vcpu, r, &e);
5493 
5494 	if (operand.vpid >> 16)
5495 		return nested_vmx_fail(vcpu,
5496 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5497 
5498 	vpid02 = nested_get_vpid02(vcpu);
5499 	switch (type) {
5500 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5501 		if (!operand.vpid ||
5502 		    is_noncanonical_address(operand.gla, vcpu))
5503 			return nested_vmx_fail(vcpu,
5504 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5505 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5506 		break;
5507 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5508 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5509 		if (!operand.vpid)
5510 			return nested_vmx_fail(vcpu,
5511 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5512 		vpid_sync_context(vpid02);
5513 		break;
5514 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5515 		vpid_sync_context(vpid02);
5516 		break;
5517 	default:
5518 		WARN_ON_ONCE(1);
5519 		return kvm_skip_emulated_instruction(vcpu);
5520 	}
5521 
5522 	/*
5523 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5524 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5525 	 * roots as VPIDs are not tracked in the MMU role.
5526 	 *
5527 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5528 	 * an MMU when EPT is disabled.
5529 	 *
5530 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5531 	 */
5532 	if (!enable_ept)
5533 		kvm_mmu_free_guest_mode_roots(vcpu, &vcpu->arch.root_mmu);
5534 
5535 	return nested_vmx_succeed(vcpu);
5536 }
5537 
5538 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5539 				     struct vmcs12 *vmcs12)
5540 {
5541 	u32 index = kvm_rcx_read(vcpu);
5542 	u64 new_eptp;
5543 
5544 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5545 		return 1;
5546 	if (index >= VMFUNC_EPTP_ENTRIES)
5547 		return 1;
5548 
5549 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5550 				     &new_eptp, index * 8, 8))
5551 		return 1;
5552 
5553 	/*
5554 	 * If the (L2) guest does a vmfunc to the currently
5555 	 * active ept pointer, we don't have to do anything else
5556 	 */
5557 	if (vmcs12->ept_pointer != new_eptp) {
5558 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5559 			return 1;
5560 
5561 		vmcs12->ept_pointer = new_eptp;
5562 		nested_ept_new_eptp(vcpu);
5563 
5564 		if (!nested_cpu_has_vpid(vmcs12))
5565 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5566 	}
5567 
5568 	return 0;
5569 }
5570 
5571 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5572 {
5573 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5574 	struct vmcs12 *vmcs12;
5575 	u32 function = kvm_rax_read(vcpu);
5576 
5577 	/*
5578 	 * VMFUNC is only supported for nested guests, but we always enable the
5579 	 * secondary control for simplicity; for non-nested mode, fake that we
5580 	 * didn't by injecting #UD.
5581 	 */
5582 	if (!is_guest_mode(vcpu)) {
5583 		kvm_queue_exception(vcpu, UD_VECTOR);
5584 		return 1;
5585 	}
5586 
5587 	vmcs12 = get_vmcs12(vcpu);
5588 
5589 	/*
5590 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5591 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5592 	 */
5593 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5594 		kvm_queue_exception(vcpu, UD_VECTOR);
5595 		return 1;
5596 	}
5597 
5598 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5599 		goto fail;
5600 
5601 	switch (function) {
5602 	case 0:
5603 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5604 			goto fail;
5605 		break;
5606 	default:
5607 		goto fail;
5608 	}
5609 	return kvm_skip_emulated_instruction(vcpu);
5610 
5611 fail:
5612 	/*
5613 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5614 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5615 	 * EXIT_REASON_VMFUNC as the exit reason.
5616 	 */
5617 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5618 			  vmx_get_intr_info(vcpu),
5619 			  vmx_get_exit_qual(vcpu));
5620 	return 1;
5621 }
5622 
5623 /*
5624  * Return true if an IO instruction with the specified port and size should cause
5625  * a VM-exit into L1.
5626  */
5627 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5628 				 int size)
5629 {
5630 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5631 	gpa_t bitmap, last_bitmap;
5632 	u8 b;
5633 
5634 	last_bitmap = INVALID_GPA;
5635 	b = -1;
5636 
5637 	while (size > 0) {
5638 		if (port < 0x8000)
5639 			bitmap = vmcs12->io_bitmap_a;
5640 		else if (port < 0x10000)
5641 			bitmap = vmcs12->io_bitmap_b;
5642 		else
5643 			return true;
5644 		bitmap += (port & 0x7fff) / 8;
5645 
5646 		if (last_bitmap != bitmap)
5647 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5648 				return true;
5649 		if (b & (1 << (port & 7)))
5650 			return true;
5651 
5652 		port++;
5653 		size--;
5654 		last_bitmap = bitmap;
5655 	}
5656 
5657 	return false;
5658 }
5659 
5660 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5661 				       struct vmcs12 *vmcs12)
5662 {
5663 	unsigned long exit_qualification;
5664 	unsigned short port;
5665 	int size;
5666 
5667 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5668 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5669 
5670 	exit_qualification = vmx_get_exit_qual(vcpu);
5671 
5672 	port = exit_qualification >> 16;
5673 	size = (exit_qualification & 7) + 1;
5674 
5675 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5676 }
5677 
5678 /*
5679  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5680  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5681  * disinterest in the current event (read or write a specific MSR) by using an
5682  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5683  */
5684 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5685 					struct vmcs12 *vmcs12,
5686 					union vmx_exit_reason exit_reason)
5687 {
5688 	u32 msr_index = kvm_rcx_read(vcpu);
5689 	gpa_t bitmap;
5690 
5691 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5692 		return true;
5693 
5694 	/*
5695 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5696 	 * for the four combinations of read/write and low/high MSR numbers.
5697 	 * First we need to figure out which of the four to use:
5698 	 */
5699 	bitmap = vmcs12->msr_bitmap;
5700 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5701 		bitmap += 2048;
5702 	if (msr_index >= 0xc0000000) {
5703 		msr_index -= 0xc0000000;
5704 		bitmap += 1024;
5705 	}
5706 
5707 	/* Then read the msr_index'th bit from this bitmap: */
5708 	if (msr_index < 1024*8) {
5709 		unsigned char b;
5710 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5711 			return true;
5712 		return 1 & (b >> (msr_index & 7));
5713 	} else
5714 		return true; /* let L1 handle the wrong parameter */
5715 }
5716 
5717 /*
5718  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5719  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5720  * intercept (via guest_host_mask etc.) the current event.
5721  */
5722 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5723 	struct vmcs12 *vmcs12)
5724 {
5725 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5726 	int cr = exit_qualification & 15;
5727 	int reg;
5728 	unsigned long val;
5729 
5730 	switch ((exit_qualification >> 4) & 3) {
5731 	case 0: /* mov to cr */
5732 		reg = (exit_qualification >> 8) & 15;
5733 		val = kvm_register_read(vcpu, reg);
5734 		switch (cr) {
5735 		case 0:
5736 			if (vmcs12->cr0_guest_host_mask &
5737 			    (val ^ vmcs12->cr0_read_shadow))
5738 				return true;
5739 			break;
5740 		case 3:
5741 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5742 				return true;
5743 			break;
5744 		case 4:
5745 			if (vmcs12->cr4_guest_host_mask &
5746 			    (vmcs12->cr4_read_shadow ^ val))
5747 				return true;
5748 			break;
5749 		case 8:
5750 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5751 				return true;
5752 			break;
5753 		}
5754 		break;
5755 	case 2: /* clts */
5756 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5757 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5758 			return true;
5759 		break;
5760 	case 1: /* mov from cr */
5761 		switch (cr) {
5762 		case 3:
5763 			if (vmcs12->cpu_based_vm_exec_control &
5764 			    CPU_BASED_CR3_STORE_EXITING)
5765 				return true;
5766 			break;
5767 		case 8:
5768 			if (vmcs12->cpu_based_vm_exec_control &
5769 			    CPU_BASED_CR8_STORE_EXITING)
5770 				return true;
5771 			break;
5772 		}
5773 		break;
5774 	case 3: /* lmsw */
5775 		/*
5776 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5777 		 * cr0. Other attempted changes are ignored, with no exit.
5778 		 */
5779 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5780 		if (vmcs12->cr0_guest_host_mask & 0xe &
5781 		    (val ^ vmcs12->cr0_read_shadow))
5782 			return true;
5783 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5784 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5785 		    (val & 0x1))
5786 			return true;
5787 		break;
5788 	}
5789 	return false;
5790 }
5791 
5792 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
5793 					  struct vmcs12 *vmcs12)
5794 {
5795 	u32 encls_leaf;
5796 
5797 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
5798 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
5799 		return false;
5800 
5801 	encls_leaf = kvm_rax_read(vcpu);
5802 	if (encls_leaf > 62)
5803 		encls_leaf = 63;
5804 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
5805 }
5806 
5807 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5808 	struct vmcs12 *vmcs12, gpa_t bitmap)
5809 {
5810 	u32 vmx_instruction_info;
5811 	unsigned long field;
5812 	u8 b;
5813 
5814 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5815 		return true;
5816 
5817 	/* Decode instruction info and find the field to access */
5818 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5819 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5820 
5821 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5822 	if (field >> 15)
5823 		return true;
5824 
5825 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5826 		return true;
5827 
5828 	return 1 & (b >> (field & 7));
5829 }
5830 
5831 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5832 {
5833 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5834 
5835 	if (nested_cpu_has_mtf(vmcs12))
5836 		return true;
5837 
5838 	/*
5839 	 * An MTF VM-exit may be injected into the guest by setting the
5840 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5841 	 * is the case regardless of the 'monitor trap flag' VM-execution
5842 	 * control.
5843 	 */
5844 	return entry_intr_info == (INTR_INFO_VALID_MASK
5845 				   | INTR_TYPE_OTHER_EVENT);
5846 }
5847 
5848 /*
5849  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5850  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5851  */
5852 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
5853 				     union vmx_exit_reason exit_reason)
5854 {
5855 	u32 intr_info;
5856 
5857 	switch ((u16)exit_reason.basic) {
5858 	case EXIT_REASON_EXCEPTION_NMI:
5859 		intr_info = vmx_get_intr_info(vcpu);
5860 		if (is_nmi(intr_info))
5861 			return true;
5862 		else if (is_page_fault(intr_info))
5863 			return vcpu->arch.apf.host_apf_flags ||
5864 			       vmx_need_pf_intercept(vcpu);
5865 		else if (is_debug(intr_info) &&
5866 			 vcpu->guest_debug &
5867 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5868 			return true;
5869 		else if (is_breakpoint(intr_info) &&
5870 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5871 			return true;
5872 		else if (is_alignment_check(intr_info) &&
5873 			 !vmx_guest_inject_ac(vcpu))
5874 			return true;
5875 		return false;
5876 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5877 		return true;
5878 	case EXIT_REASON_MCE_DURING_VMENTRY:
5879 		return true;
5880 	case EXIT_REASON_EPT_VIOLATION:
5881 		/*
5882 		 * L0 always deals with the EPT violation. If nested EPT is
5883 		 * used, and the nested mmu code discovers that the address is
5884 		 * missing in the guest EPT table (EPT12), the EPT violation
5885 		 * will be injected with nested_ept_inject_page_fault()
5886 		 */
5887 		return true;
5888 	case EXIT_REASON_EPT_MISCONFIG:
5889 		/*
5890 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5891 		 * table (shadow on EPT) or a merged EPT table that L0 built
5892 		 * (EPT on EPT). So any problems with the structure of the
5893 		 * table is L0's fault.
5894 		 */
5895 		return true;
5896 	case EXIT_REASON_PREEMPTION_TIMER:
5897 		return true;
5898 	case EXIT_REASON_PML_FULL:
5899 		/*
5900 		 * PML is emulated for an L1 VMM and should never be enabled in
5901 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
5902 		 */
5903 		return true;
5904 	case EXIT_REASON_VMFUNC:
5905 		/* VM functions are emulated through L2->L0 vmexits. */
5906 		return true;
5907 	case EXIT_REASON_BUS_LOCK:
5908 		/*
5909 		 * At present, bus lock VM exit is never exposed to L1.
5910 		 * Handle L2's bus locks in L0 directly.
5911 		 */
5912 		return true;
5913 	default:
5914 		break;
5915 	}
5916 	return false;
5917 }
5918 
5919 /*
5920  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5921  * is_guest_mode (L2).
5922  */
5923 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
5924 				     union vmx_exit_reason exit_reason)
5925 {
5926 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5927 	u32 intr_info;
5928 
5929 	switch ((u16)exit_reason.basic) {
5930 	case EXIT_REASON_EXCEPTION_NMI:
5931 		intr_info = vmx_get_intr_info(vcpu);
5932 		if (is_nmi(intr_info))
5933 			return true;
5934 		else if (is_page_fault(intr_info))
5935 			return true;
5936 		return vmcs12->exception_bitmap &
5937 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5938 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5939 		return nested_exit_on_intr(vcpu);
5940 	case EXIT_REASON_TRIPLE_FAULT:
5941 		return true;
5942 	case EXIT_REASON_INTERRUPT_WINDOW:
5943 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5944 	case EXIT_REASON_NMI_WINDOW:
5945 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5946 	case EXIT_REASON_TASK_SWITCH:
5947 		return true;
5948 	case EXIT_REASON_CPUID:
5949 		return true;
5950 	case EXIT_REASON_HLT:
5951 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5952 	case EXIT_REASON_INVD:
5953 		return true;
5954 	case EXIT_REASON_INVLPG:
5955 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5956 	case EXIT_REASON_RDPMC:
5957 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5958 	case EXIT_REASON_RDRAND:
5959 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5960 	case EXIT_REASON_RDSEED:
5961 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5962 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5963 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5964 	case EXIT_REASON_VMREAD:
5965 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5966 			vmcs12->vmread_bitmap);
5967 	case EXIT_REASON_VMWRITE:
5968 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5969 			vmcs12->vmwrite_bitmap);
5970 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5971 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5972 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5973 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5974 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5975 		/*
5976 		 * VMX instructions trap unconditionally. This allows L1 to
5977 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5978 		 */
5979 		return true;
5980 	case EXIT_REASON_CR_ACCESS:
5981 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5982 	case EXIT_REASON_DR_ACCESS:
5983 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5984 	case EXIT_REASON_IO_INSTRUCTION:
5985 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5986 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5987 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5988 	case EXIT_REASON_MSR_READ:
5989 	case EXIT_REASON_MSR_WRITE:
5990 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5991 	case EXIT_REASON_INVALID_STATE:
5992 		return true;
5993 	case EXIT_REASON_MWAIT_INSTRUCTION:
5994 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5995 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5996 		return nested_vmx_exit_handled_mtf(vmcs12);
5997 	case EXIT_REASON_MONITOR_INSTRUCTION:
5998 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5999 	case EXIT_REASON_PAUSE_INSTRUCTION:
6000 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6001 			nested_cpu_has2(vmcs12,
6002 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6003 	case EXIT_REASON_MCE_DURING_VMENTRY:
6004 		return true;
6005 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6006 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6007 	case EXIT_REASON_APIC_ACCESS:
6008 	case EXIT_REASON_APIC_WRITE:
6009 	case EXIT_REASON_EOI_INDUCED:
6010 		/*
6011 		 * The controls for "virtualize APIC accesses," "APIC-
6012 		 * register virtualization," and "virtual-interrupt
6013 		 * delivery" only come from vmcs12.
6014 		 */
6015 		return true;
6016 	case EXIT_REASON_INVPCID:
6017 		return
6018 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6019 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6020 	case EXIT_REASON_WBINVD:
6021 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6022 	case EXIT_REASON_XSETBV:
6023 		return true;
6024 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6025 		/*
6026 		 * This should never happen, since it is not possible to
6027 		 * set XSS to a non-zero value---neither in L1 nor in L2.
6028 		 * If if it were, XSS would have to be checked against
6029 		 * the XSS exit bitmap in vmcs12.
6030 		 */
6031 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
6032 	case EXIT_REASON_UMWAIT:
6033 	case EXIT_REASON_TPAUSE:
6034 		return nested_cpu_has2(vmcs12,
6035 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6036 	case EXIT_REASON_ENCLS:
6037 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6038 	default:
6039 		return true;
6040 	}
6041 }
6042 
6043 /*
6044  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6045  * reflected into L1.
6046  */
6047 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6048 {
6049 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6050 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6051 	unsigned long exit_qual;
6052 	u32 exit_intr_info;
6053 
6054 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6055 
6056 	/*
6057 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6058 	 * has already loaded L2's state.
6059 	 */
6060 	if (unlikely(vmx->fail)) {
6061 		trace_kvm_nested_vmenter_failed(
6062 			"hardware VM-instruction error: ",
6063 			vmcs_read32(VM_INSTRUCTION_ERROR));
6064 		exit_intr_info = 0;
6065 		exit_qual = 0;
6066 		goto reflect_vmexit;
6067 	}
6068 
6069 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6070 
6071 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6072 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6073 		return false;
6074 
6075 	/* If L1 doesn't want the exit, handle it in L0. */
6076 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6077 		return false;
6078 
6079 	/*
6080 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6081 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6082 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6083 	 * interrupts are never reflected to L1 so it's a non-issue.
6084 	 */
6085 	exit_intr_info = vmx_get_intr_info(vcpu);
6086 	if (is_exception_with_error_code(exit_intr_info)) {
6087 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6088 
6089 		vmcs12->vm_exit_intr_error_code =
6090 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6091 	}
6092 	exit_qual = vmx_get_exit_qual(vcpu);
6093 
6094 reflect_vmexit:
6095 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6096 	return true;
6097 }
6098 
6099 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6100 				struct kvm_nested_state __user *user_kvm_nested_state,
6101 				u32 user_data_size)
6102 {
6103 	struct vcpu_vmx *vmx;
6104 	struct vmcs12 *vmcs12;
6105 	struct kvm_nested_state kvm_state = {
6106 		.flags = 0,
6107 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6108 		.size = sizeof(kvm_state),
6109 		.hdr.vmx.flags = 0,
6110 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6111 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6112 		.hdr.vmx.preemption_timer_deadline = 0,
6113 	};
6114 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6115 		&user_kvm_nested_state->data.vmx[0];
6116 
6117 	if (!vcpu)
6118 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6119 
6120 	vmx = to_vmx(vcpu);
6121 	vmcs12 = get_vmcs12(vcpu);
6122 
6123 	if (nested_vmx_allowed(vcpu) &&
6124 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6125 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6126 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6127 
6128 		if (vmx_has_valid_vmcs12(vcpu)) {
6129 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6130 
6131 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6132 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6133 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6134 
6135 			if (is_guest_mode(vcpu) &&
6136 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6137 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6138 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6139 		}
6140 
6141 		if (vmx->nested.smm.vmxon)
6142 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6143 
6144 		if (vmx->nested.smm.guest_mode)
6145 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6146 
6147 		if (is_guest_mode(vcpu)) {
6148 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6149 
6150 			if (vmx->nested.nested_run_pending)
6151 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6152 
6153 			if (vmx->nested.mtf_pending)
6154 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6155 
6156 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6157 			    vmx->nested.has_preemption_timer_deadline) {
6158 				kvm_state.hdr.vmx.flags |=
6159 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6160 				kvm_state.hdr.vmx.preemption_timer_deadline =
6161 					vmx->nested.preemption_timer_deadline;
6162 			}
6163 		}
6164 	}
6165 
6166 	if (user_data_size < kvm_state.size)
6167 		goto out;
6168 
6169 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6170 		return -EFAULT;
6171 
6172 	if (!vmx_has_valid_vmcs12(vcpu))
6173 		goto out;
6174 
6175 	/*
6176 	 * When running L2, the authoritative vmcs12 state is in the
6177 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6178 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6179 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6180 	 * vmcs12 state is in the vmcs12 already.
6181 	 */
6182 	if (is_guest_mode(vcpu)) {
6183 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6184 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6185 	} else  {
6186 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6187 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6188 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6189 				/*
6190 				 * L1 hypervisor is not obliged to keep eVMCS
6191 				 * clean fields data always up-to-date while
6192 				 * not in guest mode, 'hv_clean_fields' is only
6193 				 * supposed to be actual upon vmentry so we need
6194 				 * to ignore it here and do full copy.
6195 				 */
6196 				copy_enlightened_to_vmcs12(vmx, 0);
6197 			else if (enable_shadow_vmcs)
6198 				copy_shadow_to_vmcs12(vmx);
6199 		}
6200 	}
6201 
6202 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6203 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6204 
6205 	/*
6206 	 * Copy over the full allocated size of vmcs12 rather than just the size
6207 	 * of the struct.
6208 	 */
6209 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6210 		return -EFAULT;
6211 
6212 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6213 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6214 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6215 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6216 			return -EFAULT;
6217 	}
6218 out:
6219 	return kvm_state.size;
6220 }
6221 
6222 /*
6223  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6224  */
6225 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6226 {
6227 	if (is_guest_mode(vcpu)) {
6228 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6229 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6230 	}
6231 	free_nested(vcpu);
6232 }
6233 
6234 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6235 				struct kvm_nested_state __user *user_kvm_nested_state,
6236 				struct kvm_nested_state *kvm_state)
6237 {
6238 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6239 	struct vmcs12 *vmcs12;
6240 	enum vm_entry_failure_code ignored;
6241 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6242 		&user_kvm_nested_state->data.vmx[0];
6243 	int ret;
6244 
6245 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6246 		return -EINVAL;
6247 
6248 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6249 		if (kvm_state->hdr.vmx.smm.flags)
6250 			return -EINVAL;
6251 
6252 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6253 			return -EINVAL;
6254 
6255 		/*
6256 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6257 		 * enable eVMCS capability on vCPU. However, since then
6258 		 * code was changed such that flag signals vmcs12 should
6259 		 * be copied into eVMCS in guest memory.
6260 		 *
6261 		 * To preserve backwards compatability, allow user
6262 		 * to set this flag even when there is no VMXON region.
6263 		 */
6264 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6265 			return -EINVAL;
6266 	} else {
6267 		if (!nested_vmx_allowed(vcpu))
6268 			return -EINVAL;
6269 
6270 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6271 			return -EINVAL;
6272 	}
6273 
6274 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6275 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6276 		return -EINVAL;
6277 
6278 	if (kvm_state->hdr.vmx.smm.flags &
6279 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6280 		return -EINVAL;
6281 
6282 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6283 		return -EINVAL;
6284 
6285 	/*
6286 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6287 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6288 	 * must be zero.
6289 	 */
6290 	if (is_smm(vcpu) ?
6291 		(kvm_state->flags &
6292 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6293 		: kvm_state->hdr.vmx.smm.flags)
6294 		return -EINVAL;
6295 
6296 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6297 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6298 		return -EINVAL;
6299 
6300 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6301 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6302 			return -EINVAL;
6303 
6304 	vmx_leave_nested(vcpu);
6305 
6306 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6307 		return 0;
6308 
6309 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6310 	ret = enter_vmx_operation(vcpu);
6311 	if (ret)
6312 		return ret;
6313 
6314 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6315 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6316 		/* See vmx_has_valid_vmcs12.  */
6317 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6318 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6319 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6320 			return -EINVAL;
6321 		else
6322 			return 0;
6323 	}
6324 
6325 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6326 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6327 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6328 			return -EINVAL;
6329 
6330 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6331 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6332 		/*
6333 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6334 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6335 		 * restored yet. EVMCS will be mapped from
6336 		 * nested_get_vmcs12_pages().
6337 		 */
6338 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6339 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6340 	} else {
6341 		return -EINVAL;
6342 	}
6343 
6344 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6345 		vmx->nested.smm.vmxon = true;
6346 		vmx->nested.vmxon = false;
6347 
6348 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6349 			vmx->nested.smm.guest_mode = true;
6350 	}
6351 
6352 	vmcs12 = get_vmcs12(vcpu);
6353 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6354 		return -EFAULT;
6355 
6356 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6357 		return -EINVAL;
6358 
6359 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6360 		return 0;
6361 
6362 	vmx->nested.nested_run_pending =
6363 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6364 
6365 	vmx->nested.mtf_pending =
6366 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6367 
6368 	ret = -EINVAL;
6369 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6370 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6371 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6372 
6373 		if (kvm_state->size <
6374 		    sizeof(*kvm_state) +
6375 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6376 			goto error_guest_mode;
6377 
6378 		if (copy_from_user(shadow_vmcs12,
6379 				   user_vmx_nested_state->shadow_vmcs12,
6380 				   sizeof(*shadow_vmcs12))) {
6381 			ret = -EFAULT;
6382 			goto error_guest_mode;
6383 		}
6384 
6385 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6386 		    !shadow_vmcs12->hdr.shadow_vmcs)
6387 			goto error_guest_mode;
6388 	}
6389 
6390 	vmx->nested.has_preemption_timer_deadline = false;
6391 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6392 		vmx->nested.has_preemption_timer_deadline = true;
6393 		vmx->nested.preemption_timer_deadline =
6394 			kvm_state->hdr.vmx.preemption_timer_deadline;
6395 	}
6396 
6397 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6398 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6399 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6400 		goto error_guest_mode;
6401 
6402 	vmx->nested.dirty_vmcs12 = true;
6403 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6404 	if (ret)
6405 		goto error_guest_mode;
6406 
6407 	return 0;
6408 
6409 error_guest_mode:
6410 	vmx->nested.nested_run_pending = 0;
6411 	return ret;
6412 }
6413 
6414 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6415 {
6416 	if (enable_shadow_vmcs) {
6417 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6418 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6419 	}
6420 }
6421 
6422 /*
6423  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6424  * that madness to get the encoding for comparison.
6425  */
6426 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6427 
6428 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6429 {
6430 	/*
6431 	 * Note these are the so called "index" of the VMCS field encoding, not
6432 	 * the index into vmcs12.
6433 	 */
6434 	unsigned int max_idx, idx;
6435 	int i;
6436 
6437 	/*
6438 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6439 	 * vmcs12, regardless of whether or not the associated feature is
6440 	 * exposed to L1.  Simply find the field with the highest index.
6441 	 */
6442 	max_idx = 0;
6443 	for (i = 0; i < nr_vmcs12_fields; i++) {
6444 		/* The vmcs12 table is very, very sparsely populated. */
6445 		if (!vmcs_field_to_offset_table[i])
6446 			continue;
6447 
6448 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6449 		if (idx > max_idx)
6450 			max_idx = idx;
6451 	}
6452 
6453 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6454 }
6455 
6456 /*
6457  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6458  * returned for the various VMX controls MSRs when nested VMX is enabled.
6459  * The same values should also be used to verify that vmcs12 control fields are
6460  * valid during nested entry from L1 to L2.
6461  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6462  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6463  * bit in the high half is on if the corresponding bit in the control field
6464  * may be on. See also vmx_control_verify().
6465  */
6466 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6467 {
6468 	/*
6469 	 * Note that as a general rule, the high half of the MSRs (bits in
6470 	 * the control fields which may be 1) should be initialized by the
6471 	 * intersection of the underlying hardware's MSR (i.e., features which
6472 	 * can be supported) and the list of features we want to expose -
6473 	 * because they are known to be properly supported in our code.
6474 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6475 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6476 	 * reason is that if one of these bits is necessary, it will appear
6477 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6478 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6479 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6480 	 * These rules have exceptions below.
6481 	 */
6482 
6483 	/* pin-based controls */
6484 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6485 		msrs->pinbased_ctls_low,
6486 		msrs->pinbased_ctls_high);
6487 	msrs->pinbased_ctls_low |=
6488 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6489 	msrs->pinbased_ctls_high &=
6490 		PIN_BASED_EXT_INTR_MASK |
6491 		PIN_BASED_NMI_EXITING |
6492 		PIN_BASED_VIRTUAL_NMIS |
6493 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6494 	msrs->pinbased_ctls_high |=
6495 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6496 		PIN_BASED_VMX_PREEMPTION_TIMER;
6497 
6498 	/* exit controls */
6499 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6500 		msrs->exit_ctls_low,
6501 		msrs->exit_ctls_high);
6502 	msrs->exit_ctls_low =
6503 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6504 
6505 	msrs->exit_ctls_high &=
6506 #ifdef CONFIG_X86_64
6507 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6508 #endif
6509 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6510 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6511 	msrs->exit_ctls_high |=
6512 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6513 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6514 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6515 
6516 	/* We support free control of debug control saving. */
6517 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6518 
6519 	/* entry controls */
6520 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6521 		msrs->entry_ctls_low,
6522 		msrs->entry_ctls_high);
6523 	msrs->entry_ctls_low =
6524 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6525 	msrs->entry_ctls_high &=
6526 #ifdef CONFIG_X86_64
6527 		VM_ENTRY_IA32E_MODE |
6528 #endif
6529 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6530 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6531 	msrs->entry_ctls_high |=
6532 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6533 
6534 	/* We support free control of debug control loading. */
6535 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6536 
6537 	/* cpu-based controls */
6538 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6539 		msrs->procbased_ctls_low,
6540 		msrs->procbased_ctls_high);
6541 	msrs->procbased_ctls_low =
6542 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6543 	msrs->procbased_ctls_high &=
6544 		CPU_BASED_INTR_WINDOW_EXITING |
6545 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6546 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6547 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6548 		CPU_BASED_CR3_STORE_EXITING |
6549 #ifdef CONFIG_X86_64
6550 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6551 #endif
6552 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6553 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6554 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6555 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6556 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6557 	/*
6558 	 * We can allow some features even when not supported by the
6559 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6560 	 * can use it to avoid exits to L1 - even when L0 runs L2
6561 	 * without MSR bitmaps.
6562 	 */
6563 	msrs->procbased_ctls_high |=
6564 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6565 		CPU_BASED_USE_MSR_BITMAPS;
6566 
6567 	/* We support free control of CR3 access interception. */
6568 	msrs->procbased_ctls_low &=
6569 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6570 
6571 	/*
6572 	 * secondary cpu-based controls.  Do not include those that
6573 	 * depend on CPUID bits, they are added later by
6574 	 * vmx_vcpu_after_set_cpuid.
6575 	 */
6576 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6577 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6578 		      msrs->secondary_ctls_low,
6579 		      msrs->secondary_ctls_high);
6580 
6581 	msrs->secondary_ctls_low = 0;
6582 	msrs->secondary_ctls_high &=
6583 		SECONDARY_EXEC_DESC |
6584 		SECONDARY_EXEC_ENABLE_RDTSCP |
6585 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6586 		SECONDARY_EXEC_WBINVD_EXITING |
6587 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6588 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6589 		SECONDARY_EXEC_RDRAND_EXITING |
6590 		SECONDARY_EXEC_ENABLE_INVPCID |
6591 		SECONDARY_EXEC_RDSEED_EXITING |
6592 		SECONDARY_EXEC_XSAVES |
6593 		SECONDARY_EXEC_TSC_SCALING;
6594 
6595 	/*
6596 	 * We can emulate "VMCS shadowing," even if the hardware
6597 	 * doesn't support it.
6598 	 */
6599 	msrs->secondary_ctls_high |=
6600 		SECONDARY_EXEC_SHADOW_VMCS;
6601 
6602 	if (enable_ept) {
6603 		/* nested EPT: emulate EPT also to L1 */
6604 		msrs->secondary_ctls_high |=
6605 			SECONDARY_EXEC_ENABLE_EPT;
6606 		msrs->ept_caps =
6607 			VMX_EPT_PAGE_WALK_4_BIT |
6608 			VMX_EPT_PAGE_WALK_5_BIT |
6609 			VMX_EPTP_WB_BIT |
6610 			VMX_EPT_INVEPT_BIT |
6611 			VMX_EPT_EXECUTE_ONLY_BIT;
6612 
6613 		msrs->ept_caps &= ept_caps;
6614 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6615 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6616 			VMX_EPT_1GB_PAGE_BIT;
6617 		if (enable_ept_ad_bits) {
6618 			msrs->secondary_ctls_high |=
6619 				SECONDARY_EXEC_ENABLE_PML;
6620 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6621 		}
6622 	}
6623 
6624 	if (cpu_has_vmx_vmfunc()) {
6625 		msrs->secondary_ctls_high |=
6626 			SECONDARY_EXEC_ENABLE_VMFUNC;
6627 		/*
6628 		 * Advertise EPTP switching unconditionally
6629 		 * since we emulate it
6630 		 */
6631 		if (enable_ept)
6632 			msrs->vmfunc_controls =
6633 				VMX_VMFUNC_EPTP_SWITCHING;
6634 	}
6635 
6636 	/*
6637 	 * Old versions of KVM use the single-context version without
6638 	 * checking for support, so declare that it is supported even
6639 	 * though it is treated as global context.  The alternative is
6640 	 * not failing the single-context invvpid, and it is worse.
6641 	 */
6642 	if (enable_vpid) {
6643 		msrs->secondary_ctls_high |=
6644 			SECONDARY_EXEC_ENABLE_VPID;
6645 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6646 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6647 	}
6648 
6649 	if (enable_unrestricted_guest)
6650 		msrs->secondary_ctls_high |=
6651 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6652 
6653 	if (flexpriority_enabled)
6654 		msrs->secondary_ctls_high |=
6655 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6656 
6657 	if (enable_sgx)
6658 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6659 
6660 	/* miscellaneous data */
6661 	rdmsr(MSR_IA32_VMX_MISC,
6662 		msrs->misc_low,
6663 		msrs->misc_high);
6664 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6665 	msrs->misc_low |=
6666 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6667 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6668 		VMX_MISC_ACTIVITY_HLT |
6669 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6670 	msrs->misc_high = 0;
6671 
6672 	/*
6673 	 * This MSR reports some information about VMX support. We
6674 	 * should return information about the VMX we emulate for the
6675 	 * guest, and the VMCS structure we give it - not about the
6676 	 * VMX support of the underlying hardware.
6677 	 */
6678 	msrs->basic =
6679 		VMCS12_REVISION |
6680 		VMX_BASIC_TRUE_CTLS |
6681 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6682 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6683 
6684 	if (cpu_has_vmx_basic_inout())
6685 		msrs->basic |= VMX_BASIC_INOUT;
6686 
6687 	/*
6688 	 * These MSRs specify bits which the guest must keep fixed on
6689 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6690 	 * We picked the standard core2 setting.
6691 	 */
6692 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6693 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6694 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6695 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6696 
6697 	/* These MSRs specify bits which the guest must keep fixed off. */
6698 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6699 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6700 
6701 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
6702 }
6703 
6704 void nested_vmx_hardware_unsetup(void)
6705 {
6706 	int i;
6707 
6708 	if (enable_shadow_vmcs) {
6709 		for (i = 0; i < VMX_BITMAP_NR; i++)
6710 			free_page((unsigned long)vmx_bitmap[i]);
6711 	}
6712 }
6713 
6714 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6715 {
6716 	int i;
6717 
6718 	if (!cpu_has_vmx_shadow_vmcs())
6719 		enable_shadow_vmcs = 0;
6720 	if (enable_shadow_vmcs) {
6721 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6722 			/*
6723 			 * The vmx_bitmap is not tied to a VM and so should
6724 			 * not be charged to a memcg.
6725 			 */
6726 			vmx_bitmap[i] = (unsigned long *)
6727 				__get_free_page(GFP_KERNEL);
6728 			if (!vmx_bitmap[i]) {
6729 				nested_vmx_hardware_unsetup();
6730 				return -ENOMEM;
6731 			}
6732 		}
6733 
6734 		init_vmcs_shadow_fields();
6735 	}
6736 
6737 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6738 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6739 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6740 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6741 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6742 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6743 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6744 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6745 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6746 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6747 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6748 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6749 
6750 	return 0;
6751 }
6752 
6753 struct kvm_x86_nested_ops vmx_nested_ops = {
6754 	.check_events = vmx_check_nested_events,
6755 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6756 	.triple_fault = nested_vmx_triple_fault,
6757 	.get_state = vmx_get_nested_state,
6758 	.set_state = vmx_set_nested_state,
6759 	.get_nested_state_pages = vmx_get_nested_state_pages,
6760 	.write_log_dirty = nested_vmx_write_pml_buffer,
6761 	.enable_evmcs = nested_enable_evmcs,
6762 	.get_evmcs_version = nested_get_evmcs_version,
6763 };
6764