xref: /linux/arch/x86/kvm/vmx/nested.c (revision 4215ee0d7bb5358882375c84d3cd0488bb5813b2)
1 // SPDX-License-Identifier: GPL-2.0
2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3 
4 #include <linux/objtool.h>
5 #include <linux/percpu.h>
6 
7 #include <asm/debugreg.h>
8 #include <asm/mmu_context.h>
9 #include <asm/msr.h>
10 
11 #include "x86.h"
12 #include "cpuid.h"
13 #include "hyperv.h"
14 #include "mmu.h"
15 #include "nested.h"
16 #include "pmu.h"
17 #include "posted_intr.h"
18 #include "sgx.h"
19 #include "trace.h"
20 #include "vmx.h"
21 #include "smm.h"
22 #include "x86_ops.h"
23 
24 static bool __read_mostly enable_shadow_vmcs = 1;
25 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
26 
27 static bool __ro_after_init warn_on_missed_cc;
28 module_param(warn_on_missed_cc, bool, 0444);
29 
30 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
31 
32 /*
33  * Hyper-V requires all of these, so mark them as supported even though
34  * they are just treated the same as all-context.
35  */
36 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
37 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
38 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
39 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
40 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
41 
42 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
43 
44 enum {
45 	VMX_VMREAD_BITMAP,
46 	VMX_VMWRITE_BITMAP,
47 	VMX_BITMAP_NR
48 };
49 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
50 
51 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
52 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
53 
54 struct shadow_vmcs_field {
55 	u16	encoding;
56 	u16	offset;
57 };
58 static struct shadow_vmcs_field shadow_read_only_fields[] = {
59 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
60 #include "vmcs_shadow_fields.h"
61 };
62 static int max_shadow_read_only_fields =
63 	ARRAY_SIZE(shadow_read_only_fields);
64 
65 static struct shadow_vmcs_field shadow_read_write_fields[] = {
66 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
67 #include "vmcs_shadow_fields.h"
68 };
69 static int max_shadow_read_write_fields =
70 	ARRAY_SIZE(shadow_read_write_fields);
71 
72 static void init_vmcs_shadow_fields(void)
73 {
74 	int i, j;
75 
76 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
77 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
78 
79 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
80 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
81 		u16 field = entry.encoding;
82 
83 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
84 		    (i + 1 == max_shadow_read_only_fields ||
85 		     shadow_read_only_fields[i + 1].encoding != field + 1))
86 			pr_err("Missing field from shadow_read_only_field %x\n",
87 			       field + 1);
88 
89 		if (get_vmcs12_field_offset(field) < 0)
90 			continue;
91 
92 		clear_bit(field, vmx_vmread_bitmap);
93 		if (field & 1)
94 #ifdef CONFIG_X86_64
95 			continue;
96 #else
97 			entry.offset += sizeof(u32);
98 #endif
99 		shadow_read_only_fields[j++] = entry;
100 	}
101 	max_shadow_read_only_fields = j;
102 
103 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
104 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
105 		u16 field = entry.encoding;
106 
107 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
108 		    (i + 1 == max_shadow_read_write_fields ||
109 		     shadow_read_write_fields[i + 1].encoding != field + 1))
110 			pr_err("Missing field from shadow_read_write_field %x\n",
111 			       field + 1);
112 
113 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
114 			  field <= GUEST_TR_AR_BYTES,
115 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
116 
117 		if (get_vmcs12_field_offset(field) < 0)
118 			continue;
119 
120 		/*
121 		 * KVM emulates PML and the VMX preemption timer irrespective
122 		 * of hardware support, but shadowing their related VMCS fields
123 		 * requires hardware support as the CPU will reject VMWRITEs to
124 		 * fields that don't exist.
125 		 */
126 		switch (field) {
127 		case GUEST_PML_INDEX:
128 			if (!cpu_has_vmx_pml())
129 				continue;
130 			break;
131 		case VMX_PREEMPTION_TIMER_VALUE:
132 			if (!cpu_has_vmx_preemption_timer())
133 				continue;
134 			break;
135 		default:
136 			break;
137 		}
138 
139 		clear_bit(field, vmx_vmwrite_bitmap);
140 		clear_bit(field, vmx_vmread_bitmap);
141 		if (field & 1)
142 #ifdef CONFIG_X86_64
143 			continue;
144 #else
145 			entry.offset += sizeof(u32);
146 #endif
147 		shadow_read_write_fields[j++] = entry;
148 	}
149 	max_shadow_read_write_fields = j;
150 }
151 
152 /*
153  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
154  * set the success or error code of an emulated VMX instruction (as specified
155  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
156  * instruction.
157  */
158 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
159 {
160 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
161 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
162 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
163 	return kvm_skip_emulated_instruction(vcpu);
164 }
165 
166 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
167 {
168 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
169 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
170 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
171 			| X86_EFLAGS_CF);
172 	return kvm_skip_emulated_instruction(vcpu);
173 }
174 
175 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
176 				u32 vm_instruction_error)
177 {
178 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
179 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
180 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
181 			| X86_EFLAGS_ZF);
182 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
183 	/*
184 	 * We don't need to force sync to shadow VMCS because
185 	 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
186 	 * fields and thus must be synced.
187 	 */
188 	if (nested_vmx_is_evmptr12_set(to_vmx(vcpu)))
189 		to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
190 
191 	return kvm_skip_emulated_instruction(vcpu);
192 }
193 
194 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
195 {
196 	struct vcpu_vmx *vmx = to_vmx(vcpu);
197 
198 	/*
199 	 * failValid writes the error number to the current VMCS, which
200 	 * can't be done if there isn't a current VMCS.
201 	 */
202 	if (vmx->nested.current_vmptr == INVALID_GPA &&
203 	    !nested_vmx_is_evmptr12_valid(vmx))
204 		return nested_vmx_failInvalid(vcpu);
205 
206 	return nested_vmx_failValid(vcpu, vm_instruction_error);
207 }
208 
209 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
210 {
211 	/* TODO: not to reset guest simply here. */
212 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
213 	pr_debug_ratelimited("nested vmx abort, indicator %d\n", indicator);
214 }
215 
216 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
217 {
218 	return fixed_bits_valid(control, low, high);
219 }
220 
221 static inline u64 vmx_control_msr(u32 low, u32 high)
222 {
223 	return low | ((u64)high << 32);
224 }
225 
226 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
227 {
228 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
229 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
230 	vmx->nested.need_vmcs12_to_shadow_sync = false;
231 }
232 
233 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
234 {
235 #ifdef CONFIG_KVM_HYPERV
236 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
237 	struct vcpu_vmx *vmx = to_vmx(vcpu);
238 
239 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map);
240 	vmx->nested.hv_evmcs = NULL;
241 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
242 
243 	if (hv_vcpu) {
244 		hv_vcpu->nested.pa_page_gpa = INVALID_GPA;
245 		hv_vcpu->nested.vm_id = 0;
246 		hv_vcpu->nested.vp_id = 0;
247 	}
248 #endif
249 }
250 
251 static bool nested_evmcs_handle_vmclear(struct kvm_vcpu *vcpu, gpa_t vmptr)
252 {
253 #ifdef CONFIG_KVM_HYPERV
254 	struct vcpu_vmx *vmx = to_vmx(vcpu);
255 	/*
256 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
257 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
258 	 * way to distinguish it from VMCS12) and we must not corrupt it by
259 	 * writing to the non-existent 'launch_state' field. The area doesn't
260 	 * have to be the currently active EVMCS on the calling CPU and there's
261 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
262 	 * state. It is possible that the area will stay mapped as
263 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
264 	 */
265 	if (!guest_cpu_cap_has_evmcs(vcpu) ||
266 	    !evmptr_is_valid(nested_get_evmptr(vcpu)))
267 		return false;
268 
269 	if (nested_vmx_evmcs(vmx) && vmptr == vmx->nested.hv_evmcs_vmptr)
270 		nested_release_evmcs(vcpu);
271 
272 	return true;
273 #else
274 	return false;
275 #endif
276 }
277 
278 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
279 				     struct loaded_vmcs *prev)
280 {
281 	struct vmcs_host_state *dest, *src;
282 
283 	if (unlikely(!vmx->vt.guest_state_loaded))
284 		return;
285 
286 	src = &prev->host_state;
287 	dest = &vmx->loaded_vmcs->host_state;
288 
289 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
290 	dest->ldt_sel = src->ldt_sel;
291 #ifdef CONFIG_X86_64
292 	dest->ds_sel = src->ds_sel;
293 	dest->es_sel = src->es_sel;
294 #endif
295 }
296 
297 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
298 {
299 	struct vcpu_vmx *vmx = to_vmx(vcpu);
300 	struct loaded_vmcs *prev;
301 	int cpu;
302 
303 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
304 		return;
305 
306 	cpu = get_cpu();
307 	prev = vmx->loaded_vmcs;
308 	vmx->loaded_vmcs = vmcs;
309 	vmx_vcpu_load_vmcs(vcpu, cpu);
310 	vmx_sync_vmcs_host_state(vmx, prev);
311 	put_cpu();
312 
313 	vcpu->arch.regs_avail = ~VMX_REGS_LAZY_LOAD_SET;
314 
315 	/*
316 	 * All lazily updated registers will be reloaded from VMCS12 on both
317 	 * vmentry and vmexit.
318 	 */
319 	vcpu->arch.regs_dirty = 0;
320 }
321 
322 static void nested_put_vmcs12_pages(struct kvm_vcpu *vcpu)
323 {
324 	struct vcpu_vmx *vmx = to_vmx(vcpu);
325 
326 	kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map);
327 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map);
328 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map);
329 	vmx->nested.pi_desc = NULL;
330 }
331 
332 /*
333  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
334  * just stops using VMX.
335  */
336 static void free_nested(struct kvm_vcpu *vcpu)
337 {
338 	struct vcpu_vmx *vmx = to_vmx(vcpu);
339 
340 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
341 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
342 
343 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
344 		return;
345 
346 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
347 
348 	vmx->nested.vmxon = false;
349 	vmx->nested.smm.vmxon = false;
350 	vmx->nested.vmxon_ptr = INVALID_GPA;
351 	free_vpid(vmx->nested.vpid02);
352 	vmx->nested.posted_intr_nv = -1;
353 	vmx->nested.current_vmptr = INVALID_GPA;
354 	if (enable_shadow_vmcs) {
355 		vmx_disable_shadow_vmcs(vmx);
356 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
357 		free_vmcs(vmx->vmcs01.shadow_vmcs);
358 		vmx->vmcs01.shadow_vmcs = NULL;
359 	}
360 	kfree(vmx->nested.cached_vmcs12);
361 	vmx->nested.cached_vmcs12 = NULL;
362 	kfree(vmx->nested.cached_shadow_vmcs12);
363 	vmx->nested.cached_shadow_vmcs12 = NULL;
364 
365 	nested_put_vmcs12_pages(vcpu);
366 
367 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
368 
369 	nested_release_evmcs(vcpu);
370 
371 	free_loaded_vmcs(&vmx->nested.vmcs02);
372 }
373 
374 /*
375  * Ensure that the current vmcs of the logical processor is the
376  * vmcs01 of the vcpu before calling free_nested().
377  */
378 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
379 {
380 	vcpu_load(vcpu);
381 	vmx_leave_nested(vcpu);
382 	vcpu_put(vcpu);
383 }
384 
385 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
386 
387 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
388 {
389 	return VALID_PAGE(root_hpa) &&
390 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
391 }
392 
393 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
394 				       gpa_t addr)
395 {
396 	unsigned long roots = 0;
397 	uint i;
398 	struct kvm_mmu_root_info *cached_root;
399 
400 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
401 
402 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
403 		cached_root = &vcpu->arch.mmu->prev_roots[i];
404 
405 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
406 					    eptp))
407 			roots |= KVM_MMU_ROOT_PREVIOUS(i);
408 	}
409 	if (roots)
410 		kvm_mmu_invalidate_addr(vcpu, vcpu->arch.mmu, addr, roots);
411 }
412 
413 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
414 		struct x86_exception *fault)
415 {
416 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
417 	struct vcpu_vmx *vmx = to_vmx(vcpu);
418 	unsigned long exit_qualification;
419 	u32 vm_exit_reason;
420 
421 	if (vmx->nested.pml_full) {
422 		vm_exit_reason = EXIT_REASON_PML_FULL;
423 		vmx->nested.pml_full = false;
424 
425 		/*
426 		 * It should be impossible to trigger a nested PML Full VM-Exit
427 		 * for anything other than an EPT Violation from L2.  KVM *can*
428 		 * trigger nEPT page fault injection in response to an EPT
429 		 * Misconfig, e.g. if the MMIO SPTE was stale and L1's EPT
430 		 * tables also changed, but KVM should not treat EPT Misconfig
431 		 * VM-Exits as writes.
432 		 */
433 		WARN_ON_ONCE(vmx->vt.exit_reason.basic != EXIT_REASON_EPT_VIOLATION);
434 
435 		/*
436 		 * PML Full and EPT Violation VM-Exits both use bit 12 to report
437 		 * "NMI unblocking due to IRET", i.e. the bit can be propagated
438 		 * as-is from the original EXIT_QUALIFICATION.
439 		 */
440 		exit_qualification = vmx_get_exit_qual(vcpu) & INTR_INFO_UNBLOCK_NMI;
441 	} else {
442 		if (fault->error_code & PFERR_RSVD_MASK) {
443 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
444 			exit_qualification = 0;
445 		} else {
446 			exit_qualification = fault->exit_qualification;
447 			exit_qualification |= vmx_get_exit_qual(vcpu) &
448 					      (EPT_VIOLATION_GVA_IS_VALID |
449 					       EPT_VIOLATION_GVA_TRANSLATED);
450 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
451 		}
452 
453 		/*
454 		 * Although the caller (kvm_inject_emulated_page_fault) would
455 		 * have already synced the faulting address in the shadow EPT
456 		 * tables for the current EPTP12, we also need to sync it for
457 		 * any other cached EPTP02s based on the same EP4TA, since the
458 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
459 		 */
460 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
461 					   fault->address);
462 	}
463 
464 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
465 	vmcs12->guest_physical_address = fault->address;
466 }
467 
468 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
469 {
470 	struct vcpu_vmx *vmx = to_vmx(vcpu);
471 	bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
472 	int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
473 
474 	kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
475 				nested_ept_ad_enabled(vcpu),
476 				nested_ept_get_eptp(vcpu));
477 }
478 
479 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
480 {
481 	WARN_ON(mmu_is_nested(vcpu));
482 
483 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
484 	nested_ept_new_eptp(vcpu);
485 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
486 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
487 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
488 
489 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
490 }
491 
492 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
493 {
494 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
495 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
496 }
497 
498 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
499 					    u16 error_code)
500 {
501 	bool inequality, bit;
502 
503 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
504 	inequality =
505 		(error_code & vmcs12->page_fault_error_code_mask) !=
506 		 vmcs12->page_fault_error_code_match;
507 	return inequality ^ bit;
508 }
509 
510 static bool nested_vmx_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
511 					   u32 error_code)
512 {
513 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
514 
515 	/*
516 	 * Drop bits 31:16 of the error code when performing the #PF mask+match
517 	 * check.  All VMCS fields involved are 32 bits, but Intel CPUs never
518 	 * set bits 31:16 and VMX disallows setting bits 31:16 in the injected
519 	 * error code.  Including the to-be-dropped bits in the check might
520 	 * result in an "impossible" or missed exit from L1's perspective.
521 	 */
522 	if (vector == PF_VECTOR)
523 		return nested_vmx_is_page_fault_vmexit(vmcs12, (u16)error_code);
524 
525 	return (vmcs12->exception_bitmap & (1u << vector));
526 }
527 
528 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
529 					       struct vmcs12 *vmcs12)
530 {
531 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
532 		return 0;
533 
534 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
535 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
536 		return -EINVAL;
537 
538 	return 0;
539 }
540 
541 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
542 						struct vmcs12 *vmcs12)
543 {
544 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
545 		return 0;
546 
547 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
548 		return -EINVAL;
549 
550 	return 0;
551 }
552 
553 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
554 						struct vmcs12 *vmcs12)
555 {
556 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
557 		return 0;
558 
559 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
560 		return -EINVAL;
561 
562 	if (CC(!nested_cpu_has_vid(vmcs12) && vmcs12->tpr_threshold >> 4))
563 		return -EINVAL;
564 
565 	return 0;
566 }
567 
568 /*
569  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
570  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
571  * only the "disable intercept" case needs to be handled.
572  */
573 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
574 							unsigned long *msr_bitmap_l0,
575 							u32 msr, int type)
576 {
577 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
578 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
579 
580 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
581 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
582 }
583 
584 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
585 {
586 	int msr;
587 
588 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
589 		unsigned word = msr / BITS_PER_LONG;
590 
591 		msr_bitmap[word] = ~0;
592 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
593 	}
594 }
595 
596 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
597 static inline									\
598 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
599 					 unsigned long *msr_bitmap_l1,		\
600 					 unsigned long *msr_bitmap_l0, u32 msr)	\
601 {										\
602 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
603 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
604 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
605 	else									\
606 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
607 }
608 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
609 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
610 
611 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
612 						    unsigned long *msr_bitmap_l1,
613 						    unsigned long *msr_bitmap_l0,
614 						    u32 msr, int types)
615 {
616 	if (types & MSR_TYPE_R)
617 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
618 						  msr_bitmap_l0, msr);
619 	if (types & MSR_TYPE_W)
620 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
621 						   msr_bitmap_l0, msr);
622 }
623 
624 /*
625  * Merge L0's and L1's MSR bitmap, return false to indicate that
626  * we do not use the hardware.
627  */
628 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
629 						 struct vmcs12 *vmcs12)
630 {
631 	struct vcpu_vmx *vmx = to_vmx(vcpu);
632 	int msr;
633 	unsigned long *msr_bitmap_l1;
634 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
635 	struct kvm_host_map map;
636 
637 	/* Nothing to do if the MSR bitmap is not in use.  */
638 	if (!cpu_has_vmx_msr_bitmap() ||
639 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
640 		return false;
641 
642 	/*
643 	 * MSR bitmap update can be skipped when:
644 	 * - MSR bitmap for L1 hasn't changed.
645 	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
646 	 *   before.
647 	 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
648 	 *   and tells KVM (L0) there were no changes in MSR bitmap for L2.
649 	 */
650 	if (!vmx->nested.force_msr_bitmap_recalc) {
651 		struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
652 
653 		if (evmcs && evmcs->hv_enlightenments_control.msr_bitmap &&
654 		    evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
655 			return true;
656 	}
657 
658 	if (kvm_vcpu_map_readonly(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), &map))
659 		return false;
660 
661 	msr_bitmap_l1 = (unsigned long *)map.hva;
662 
663 	/*
664 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
665 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
666 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
667 	 */
668 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
669 
670 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
671 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
672 			/*
673 			 * L0 need not intercept reads for MSRs between 0x800
674 			 * and 0x8ff, it just lets the processor take the value
675 			 * from the virtual-APIC page; take those 256 bits
676 			 * directly from the L1 bitmap.
677 			 */
678 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
679 				unsigned word = msr / BITS_PER_LONG;
680 
681 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
682 			}
683 		}
684 
685 		nested_vmx_disable_intercept_for_x2apic_msr(
686 			msr_bitmap_l1, msr_bitmap_l0,
687 			X2APIC_MSR(APIC_TASKPRI),
688 			MSR_TYPE_R | MSR_TYPE_W);
689 
690 		if (nested_cpu_has_vid(vmcs12)) {
691 			nested_vmx_disable_intercept_for_x2apic_msr(
692 				msr_bitmap_l1, msr_bitmap_l0,
693 				X2APIC_MSR(APIC_EOI),
694 				MSR_TYPE_W);
695 			nested_vmx_disable_intercept_for_x2apic_msr(
696 				msr_bitmap_l1, msr_bitmap_l0,
697 				X2APIC_MSR(APIC_SELF_IPI),
698 				MSR_TYPE_W);
699 		}
700 	}
701 
702 	/*
703 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
704 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
705 	 */
706 #ifdef CONFIG_X86_64
707 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
708 					 MSR_FS_BASE, MSR_TYPE_RW);
709 
710 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
711 					 MSR_GS_BASE, MSR_TYPE_RW);
712 
713 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
714 					 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
715 #endif
716 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
717 					 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
718 
719 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
720 					 MSR_IA32_PRED_CMD, MSR_TYPE_W);
721 
722 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
723 					 MSR_IA32_FLUSH_CMD, MSR_TYPE_W);
724 
725 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
726 					 MSR_IA32_APERF, MSR_TYPE_R);
727 
728 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
729 					 MSR_IA32_MPERF, MSR_TYPE_R);
730 
731 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
732 					 MSR_IA32_U_CET, MSR_TYPE_RW);
733 
734 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
735 					 MSR_IA32_S_CET, MSR_TYPE_RW);
736 
737 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
738 					 MSR_IA32_PL0_SSP, MSR_TYPE_RW);
739 
740 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
741 					 MSR_IA32_PL1_SSP, MSR_TYPE_RW);
742 
743 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
744 					 MSR_IA32_PL2_SSP, MSR_TYPE_RW);
745 
746 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
747 					 MSR_IA32_PL3_SSP, MSR_TYPE_RW);
748 
749 	kvm_vcpu_unmap(vcpu, &map);
750 
751 	vmx->nested.force_msr_bitmap_recalc = false;
752 
753 	return true;
754 }
755 
756 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
757 				       struct vmcs12 *vmcs12)
758 {
759 	struct vcpu_vmx *vmx = to_vmx(vcpu);
760 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
761 
762 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
763 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
764 		return;
765 
766 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
767 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
768 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
769 		return;
770 
771 	kvm_read_guest_cached(vcpu->kvm, ghc, get_shadow_vmcs12(vcpu),
772 			      VMCS12_SIZE);
773 }
774 
775 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
776 					      struct vmcs12 *vmcs12)
777 {
778 	struct vcpu_vmx *vmx = to_vmx(vcpu);
779 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
780 
781 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
782 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
783 		return;
784 
785 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
786 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
787 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
788 		return;
789 
790 	kvm_write_guest_cached(vcpu->kvm, ghc, get_shadow_vmcs12(vcpu),
791 			       VMCS12_SIZE);
792 }
793 
794 /*
795  * In nested virtualization, check if L1 has set
796  * VM_EXIT_ACK_INTR_ON_EXIT
797  */
798 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
799 {
800 	return get_vmcs12(vcpu)->vm_exit_controls &
801 		VM_EXIT_ACK_INTR_ON_EXIT;
802 }
803 
804 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
805 					  struct vmcs12 *vmcs12)
806 {
807 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
808 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
809 		return -EINVAL;
810 	else
811 		return 0;
812 }
813 
814 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
815 					   struct vmcs12 *vmcs12)
816 {
817 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
818 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
819 	    !nested_cpu_has_vid(vmcs12) &&
820 	    !nested_cpu_has_posted_intr(vmcs12))
821 		return 0;
822 
823 	/*
824 	 * If virtualize x2apic mode is enabled,
825 	 * virtualize apic access must be disabled.
826 	 */
827 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
828 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
829 		return -EINVAL;
830 
831 	/*
832 	 * If virtual interrupt delivery is enabled,
833 	 * we must exit on external interrupts.
834 	 */
835 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
836 		return -EINVAL;
837 
838 	/*
839 	 * bits 15:8 should be zero in posted_intr_nv,
840 	 * the descriptor address has been already checked
841 	 * in nested_get_vmcs12_pages.
842 	 *
843 	 * bits 5:0 of posted_intr_desc_addr should be zero.
844 	 */
845 	if (nested_cpu_has_posted_intr(vmcs12) &&
846 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
847 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
848 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
849 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
850 		return -EINVAL;
851 
852 	/* tpr shadow is needed by all apicv features. */
853 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
854 		return -EINVAL;
855 
856 	return 0;
857 }
858 
859 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
860 {
861 	struct vcpu_vmx *vmx = to_vmx(vcpu);
862 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
863 				       vmx->nested.msrs.misc_high);
864 
865 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
866 }
867 
868 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
869 				       u32 count, u64 addr)
870 {
871 	if (count == 0)
872 		return 0;
873 
874 	/*
875 	 * Exceeding the limit results in architecturally _undefined_ behavior,
876 	 * i.e. KVM is allowed to do literally anything in response to a bad
877 	 * limit.  Immediately generate a consistency check so that code that
878 	 * consumes the count doesn't need to worry about extreme edge cases.
879 	 */
880 	if (count > nested_vmx_max_atomic_switch_msrs(vcpu))
881 		return -EINVAL;
882 
883 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
884 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
885 		return -EINVAL;
886 
887 	return 0;
888 }
889 
890 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
891 						     struct vmcs12 *vmcs12)
892 {
893 	if (CC(nested_vmx_check_msr_switch(vcpu,
894 					   vmcs12->vm_exit_msr_load_count,
895 					   vmcs12->vm_exit_msr_load_addr)) ||
896 	    CC(nested_vmx_check_msr_switch(vcpu,
897 					   vmcs12->vm_exit_msr_store_count,
898 					   vmcs12->vm_exit_msr_store_addr)))
899 		return -EINVAL;
900 
901 	return 0;
902 }
903 
904 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
905                                                       struct vmcs12 *vmcs12)
906 {
907 	if (CC(nested_vmx_check_msr_switch(vcpu,
908 					   vmcs12->vm_entry_msr_load_count,
909 					   vmcs12->vm_entry_msr_load_addr)))
910                 return -EINVAL;
911 
912 	return 0;
913 }
914 
915 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
916 					 struct vmcs12 *vmcs12)
917 {
918 	if (!nested_cpu_has_pml(vmcs12))
919 		return 0;
920 
921 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
922 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
923 		return -EINVAL;
924 
925 	return 0;
926 }
927 
928 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
929 							struct vmcs12 *vmcs12)
930 {
931 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
932 	       !nested_cpu_has_ept(vmcs12)))
933 		return -EINVAL;
934 	return 0;
935 }
936 
937 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
938 							 struct vmcs12 *vmcs12)
939 {
940 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
941 	       !nested_cpu_has_ept(vmcs12)))
942 		return -EINVAL;
943 	return 0;
944 }
945 
946 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
947 						 struct vmcs12 *vmcs12)
948 {
949 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
950 		return 0;
951 
952 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
953 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
954 		return -EINVAL;
955 
956 	return 0;
957 }
958 
959 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
960 				       struct vmx_msr_entry *e)
961 {
962 	/* x2APIC MSR accesses are not allowed */
963 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
964 		return -EINVAL;
965 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
966 	    CC(e->index == MSR_IA32_UCODE_REV))
967 		return -EINVAL;
968 	if (CC(e->reserved != 0))
969 		return -EINVAL;
970 	return 0;
971 }
972 
973 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
974 				     struct vmx_msr_entry *e)
975 {
976 	if (CC(e->index == MSR_FS_BASE) ||
977 	    CC(e->index == MSR_GS_BASE) ||
978 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
979 	    nested_vmx_msr_check_common(vcpu, e))
980 		return -EINVAL;
981 	return 0;
982 }
983 
984 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
985 				      struct vmx_msr_entry *e)
986 {
987 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
988 	    nested_vmx_msr_check_common(vcpu, e))
989 		return -EINVAL;
990 	return 0;
991 }
992 
993 /*
994  * Load guest's/host's msr at nested entry/exit.
995  * return 0 for success, entry index for failure.
996  *
997  * One of the failure modes for MSR load/store is when a list exceeds the
998  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
999  * as possible, process all valid entries before failing rather than precheck
1000  * for a capacity violation.
1001  */
1002 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1003 {
1004 	u32 i;
1005 	struct vmx_msr_entry e;
1006 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1007 
1008 	for (i = 0; i < count; i++) {
1009 		if (WARN_ON_ONCE(i >= max_msr_list_size))
1010 			goto fail;
1011 
1012 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
1013 					&e, sizeof(e))) {
1014 			pr_debug_ratelimited(
1015 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
1016 				__func__, i, gpa + i * sizeof(e));
1017 			goto fail;
1018 		}
1019 		if (nested_vmx_load_msr_check(vcpu, &e)) {
1020 			pr_debug_ratelimited(
1021 				"%s check failed (%u, 0x%x, 0x%x)\n",
1022 				__func__, i, e.index, e.reserved);
1023 			goto fail;
1024 		}
1025 		if (kvm_emulate_msr_write(vcpu, e.index, e.value)) {
1026 			pr_debug_ratelimited(
1027 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1028 				__func__, i, e.index, e.value);
1029 			goto fail;
1030 		}
1031 	}
1032 	return 0;
1033 fail:
1034 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
1035 	return i + 1;
1036 }
1037 
1038 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
1039 					    u32 msr_index,
1040 					    u64 *data)
1041 {
1042 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1043 
1044 	/*
1045 	 * If the L0 hypervisor stored a more accurate value for the TSC that
1046 	 * does not include the time taken for emulation of the L2->L1
1047 	 * VM-exit in L0, use the more accurate value.
1048 	 */
1049 	if (msr_index == MSR_IA32_TSC) {
1050 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
1051 						    MSR_IA32_TSC);
1052 
1053 		if (i >= 0) {
1054 			u64 val = vmx->msr_autostore.guest.val[i].value;
1055 
1056 			*data = kvm_read_l1_tsc(vcpu, val);
1057 			return true;
1058 		}
1059 	}
1060 
1061 	if (kvm_emulate_msr_read(vcpu, msr_index, data)) {
1062 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
1063 			msr_index);
1064 		return false;
1065 	}
1066 	return true;
1067 }
1068 
1069 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
1070 				     struct vmx_msr_entry *e)
1071 {
1072 	if (kvm_vcpu_read_guest(vcpu,
1073 				gpa + i * sizeof(*e),
1074 				e, 2 * sizeof(u32))) {
1075 		pr_debug_ratelimited(
1076 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
1077 			__func__, i, gpa + i * sizeof(*e));
1078 		return false;
1079 	}
1080 	if (nested_vmx_store_msr_check(vcpu, e)) {
1081 		pr_debug_ratelimited(
1082 			"%s check failed (%u, 0x%x, 0x%x)\n",
1083 			__func__, i, e->index, e->reserved);
1084 		return false;
1085 	}
1086 	return true;
1087 }
1088 
1089 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1090 {
1091 	u64 data;
1092 	u32 i;
1093 	struct vmx_msr_entry e;
1094 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1095 
1096 	for (i = 0; i < count; i++) {
1097 		if (WARN_ON_ONCE(i >= max_msr_list_size))
1098 			return -EINVAL;
1099 
1100 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1101 			return -EINVAL;
1102 
1103 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1104 			return -EINVAL;
1105 
1106 		if (kvm_vcpu_write_guest(vcpu,
1107 					 gpa + i * sizeof(e) +
1108 					     offsetof(struct vmx_msr_entry, value),
1109 					 &data, sizeof(data))) {
1110 			pr_debug_ratelimited(
1111 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1112 				__func__, i, e.index, data);
1113 			return -EINVAL;
1114 		}
1115 	}
1116 	return 0;
1117 }
1118 
1119 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1120 {
1121 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1122 	u32 count = vmcs12->vm_exit_msr_store_count;
1123 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1124 	struct vmx_msr_entry e;
1125 	u32 i;
1126 
1127 	for (i = 0; i < count; i++) {
1128 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1129 			return false;
1130 
1131 		if (e.index == msr_index)
1132 			return true;
1133 	}
1134 	return false;
1135 }
1136 
1137 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1138 					   u32 msr_index)
1139 {
1140 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1141 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1142 	bool in_vmcs12_store_list;
1143 	int msr_autostore_slot;
1144 	bool in_autostore_list;
1145 	int last;
1146 
1147 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1148 	in_autostore_list = msr_autostore_slot >= 0;
1149 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1150 
1151 	if (in_vmcs12_store_list && !in_autostore_list) {
1152 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1153 			/*
1154 			 * Emulated VMEntry does not fail here.  Instead a less
1155 			 * accurate value will be returned by
1156 			 * nested_vmx_get_vmexit_msr_value() by reading KVM's
1157 			 * internal MSR state instead of reading the value from
1158 			 * the vmcs02 VMExit MSR-store area.
1159 			 */
1160 			pr_warn_ratelimited(
1161 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1162 				msr_index);
1163 			return;
1164 		}
1165 		last = autostore->nr++;
1166 		autostore->val[last].index = msr_index;
1167 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1168 		last = --autostore->nr;
1169 		autostore->val[msr_autostore_slot] = autostore->val[last];
1170 	}
1171 }
1172 
1173 /*
1174  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1175  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1176  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1177  * @entry_failure_code.
1178  */
1179 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1180 			       bool nested_ept, bool reload_pdptrs,
1181 			       enum vm_entry_failure_code *entry_failure_code)
1182 {
1183 	if (CC(!kvm_vcpu_is_legal_cr3(vcpu, cr3))) {
1184 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1185 		return -EINVAL;
1186 	}
1187 
1188 	/*
1189 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1190 	 * must not be dereferenced.
1191 	 */
1192 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1193 	    CC(!load_pdptrs(vcpu, cr3))) {
1194 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1195 		return -EINVAL;
1196 	}
1197 
1198 	vcpu->arch.cr3 = cr3;
1199 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1200 
1201 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1202 	kvm_init_mmu(vcpu);
1203 
1204 	if (!nested_ept)
1205 		kvm_mmu_new_pgd(vcpu, cr3);
1206 
1207 	return 0;
1208 }
1209 
1210 /*
1211  * Returns if KVM is able to config CPU to tag TLB entries
1212  * populated by L2 differently than TLB entries populated
1213  * by L1.
1214  *
1215  * If L0 uses EPT, L1 and L2 run with different EPTP because
1216  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1217  * are tagged with different EPTP.
1218  *
1219  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1220  * with different VPID (L1 entries are tagged with vmx->vpid
1221  * while L2 entries are tagged with vmx->nested.vpid02).
1222  */
1223 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1224 {
1225 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1226 
1227 	return enable_ept ||
1228 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1229 }
1230 
1231 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1232 					    struct vmcs12 *vmcs12,
1233 					    bool is_vmenter)
1234 {
1235 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1236 
1237 	/* Handle pending Hyper-V TLB flush requests */
1238 	kvm_hv_nested_transtion_tlb_flush(vcpu, enable_ept);
1239 
1240 	/*
1241 	 * If VPID is disabled, then guest TLB accesses use VPID=0, i.e. the
1242 	 * same VPID as the host, and so architecturally, linear and combined
1243 	 * mappings for VPID=0 must be flushed at VM-Enter and VM-Exit.  KVM
1244 	 * emulates L2 sharing L1's VPID=0 by using vpid01 while running L2,
1245 	 * and so KVM must also emulate TLB flush of VPID=0, i.e. vpid01.  This
1246 	 * is required if VPID is disabled in KVM, as a TLB flush (there are no
1247 	 * VPIDs) still occurs from L1's perspective, and KVM may need to
1248 	 * synchronize the MMU in response to the guest TLB flush.
1249 	 *
1250 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1251 	 * EPT is a special snowflake, as guest-physical mappings aren't
1252 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1253 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1254 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1255 	 * those mappings.
1256 	 */
1257 	if (!nested_cpu_has_vpid(vmcs12)) {
1258 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1259 		return;
1260 	}
1261 
1262 	/* L2 should never have a VPID if VPID is disabled. */
1263 	WARN_ON(!enable_vpid);
1264 
1265 	/*
1266 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1267 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1268 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1269 	 * that the new vpid12 has never been used and thus represents a new
1270 	 * guest ASID that cannot have entries in the TLB.
1271 	 */
1272 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1273 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1274 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1275 		return;
1276 	}
1277 
1278 	/*
1279 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1280 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1281 	 * KVM was unable to allocate a VPID for L2, flush the current context
1282 	 * as the effective ASID is common to both L1 and L2.
1283 	 */
1284 	if (!nested_has_guest_tlb_tag(vcpu))
1285 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1286 }
1287 
1288 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1289 {
1290 	superset &= mask;
1291 	subset &= mask;
1292 
1293 	return (superset | subset) == superset;
1294 }
1295 
1296 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1297 {
1298 	const u64 feature_bits = VMX_BASIC_DUAL_MONITOR_TREATMENT |
1299 				 VMX_BASIC_INOUT |
1300 				 VMX_BASIC_TRUE_CTLS |
1301 				 VMX_BASIC_NO_HW_ERROR_CODE_CC;
1302 
1303 	const u64 reserved_bits = GENMASK_ULL(63, 57) |
1304 				  GENMASK_ULL(47, 45) |
1305 				  BIT_ULL(31);
1306 
1307 	u64 vmx_basic = vmcs_config.nested.basic;
1308 
1309 	BUILD_BUG_ON(feature_bits & reserved_bits);
1310 
1311 	/*
1312 	 * Except for 32BIT_PHYS_ADDR_ONLY, which is an anti-feature bit (has
1313 	 * inverted polarity), the incoming value must not set feature bits or
1314 	 * reserved bits that aren't allowed/supported by KVM.  Fields, i.e.
1315 	 * multi-bit values, are explicitly checked below.
1316 	 */
1317 	if (!is_bitwise_subset(vmx_basic, data, feature_bits | reserved_bits))
1318 		return -EINVAL;
1319 
1320 	/*
1321 	 * KVM does not emulate a version of VMX that constrains physical
1322 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1323 	 */
1324 	if (data & VMX_BASIC_32BIT_PHYS_ADDR_ONLY)
1325 		return -EINVAL;
1326 
1327 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1328 	    vmx_basic_vmcs_revision_id(data))
1329 		return -EINVAL;
1330 
1331 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1332 		return -EINVAL;
1333 
1334 	vmx->nested.msrs.basic = data;
1335 	return 0;
1336 }
1337 
1338 static void vmx_get_control_msr(struct nested_vmx_msrs *msrs, u32 msr_index,
1339 				u32 **low, u32 **high)
1340 {
1341 	switch (msr_index) {
1342 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1343 		*low = &msrs->pinbased_ctls_low;
1344 		*high = &msrs->pinbased_ctls_high;
1345 		break;
1346 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1347 		*low = &msrs->procbased_ctls_low;
1348 		*high = &msrs->procbased_ctls_high;
1349 		break;
1350 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1351 		*low = &msrs->exit_ctls_low;
1352 		*high = &msrs->exit_ctls_high;
1353 		break;
1354 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1355 		*low = &msrs->entry_ctls_low;
1356 		*high = &msrs->entry_ctls_high;
1357 		break;
1358 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1359 		*low = &msrs->secondary_ctls_low;
1360 		*high = &msrs->secondary_ctls_high;
1361 		break;
1362 	default:
1363 		BUG();
1364 	}
1365 }
1366 
1367 static int
1368 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1369 {
1370 	u32 *lowp, *highp;
1371 	u64 supported;
1372 
1373 	vmx_get_control_msr(&vmcs_config.nested, msr_index, &lowp, &highp);
1374 
1375 	supported = vmx_control_msr(*lowp, *highp);
1376 
1377 	/* Check must-be-1 bits are still 1. */
1378 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1379 		return -EINVAL;
1380 
1381 	/* Check must-be-0 bits are still 0. */
1382 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1383 		return -EINVAL;
1384 
1385 	vmx_get_control_msr(&vmx->nested.msrs, msr_index, &lowp, &highp);
1386 	*lowp = data;
1387 	*highp = data >> 32;
1388 	return 0;
1389 }
1390 
1391 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1392 {
1393 	const u64 feature_bits = VMX_MISC_SAVE_EFER_LMA |
1394 				 VMX_MISC_ACTIVITY_HLT |
1395 				 VMX_MISC_ACTIVITY_SHUTDOWN |
1396 				 VMX_MISC_ACTIVITY_WAIT_SIPI |
1397 				 VMX_MISC_INTEL_PT |
1398 				 VMX_MISC_RDMSR_IN_SMM |
1399 				 VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
1400 				 VMX_MISC_VMXOFF_BLOCK_SMI |
1401 				 VMX_MISC_ZERO_LEN_INS;
1402 
1403 	const u64 reserved_bits = BIT_ULL(31) | GENMASK_ULL(13, 9);
1404 
1405 	u64 vmx_misc = vmx_control_msr(vmcs_config.nested.misc_low,
1406 				       vmcs_config.nested.misc_high);
1407 
1408 	BUILD_BUG_ON(feature_bits & reserved_bits);
1409 
1410 	/*
1411 	 * The incoming value must not set feature bits or reserved bits that
1412 	 * aren't allowed/supported by KVM.  Fields, i.e. multi-bit values, are
1413 	 * explicitly checked below.
1414 	 */
1415 	if (!is_bitwise_subset(vmx_misc, data, feature_bits | reserved_bits))
1416 		return -EINVAL;
1417 
1418 	if ((vmx->nested.msrs.pinbased_ctls_high &
1419 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1420 	    vmx_misc_preemption_timer_rate(data) !=
1421 	    vmx_misc_preemption_timer_rate(vmx_misc))
1422 		return -EINVAL;
1423 
1424 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1425 		return -EINVAL;
1426 
1427 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1428 		return -EINVAL;
1429 
1430 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1431 		return -EINVAL;
1432 
1433 	vmx->nested.msrs.misc_low = data;
1434 	vmx->nested.msrs.misc_high = data >> 32;
1435 
1436 	return 0;
1437 }
1438 
1439 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1440 {
1441 	u64 vmx_ept_vpid_cap = vmx_control_msr(vmcs_config.nested.ept_caps,
1442 					       vmcs_config.nested.vpid_caps);
1443 
1444 	/* Every bit is either reserved or a feature bit. */
1445 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1446 		return -EINVAL;
1447 
1448 	vmx->nested.msrs.ept_caps = data;
1449 	vmx->nested.msrs.vpid_caps = data >> 32;
1450 	return 0;
1451 }
1452 
1453 static u64 *vmx_get_fixed0_msr(struct nested_vmx_msrs *msrs, u32 msr_index)
1454 {
1455 	switch (msr_index) {
1456 	case MSR_IA32_VMX_CR0_FIXED0:
1457 		return &msrs->cr0_fixed0;
1458 	case MSR_IA32_VMX_CR4_FIXED0:
1459 		return &msrs->cr4_fixed0;
1460 	default:
1461 		BUG();
1462 	}
1463 }
1464 
1465 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1466 {
1467 	const u64 *msr = vmx_get_fixed0_msr(&vmcs_config.nested, msr_index);
1468 
1469 	/*
1470 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1471 	 * must be 1 in the restored value.
1472 	 */
1473 	if (!is_bitwise_subset(data, *msr, -1ULL))
1474 		return -EINVAL;
1475 
1476 	*vmx_get_fixed0_msr(&vmx->nested.msrs, msr_index) = data;
1477 	return 0;
1478 }
1479 
1480 /*
1481  * Called when userspace is restoring VMX MSRs.
1482  *
1483  * Returns 0 on success, non-0 otherwise.
1484  */
1485 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1486 {
1487 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1488 
1489 	/*
1490 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1491 	 * is in VMX operation.
1492 	 */
1493 	if (vmx->nested.vmxon)
1494 		return -EBUSY;
1495 
1496 	switch (msr_index) {
1497 	case MSR_IA32_VMX_BASIC:
1498 		return vmx_restore_vmx_basic(vmx, data);
1499 	case MSR_IA32_VMX_PINBASED_CTLS:
1500 	case MSR_IA32_VMX_PROCBASED_CTLS:
1501 	case MSR_IA32_VMX_EXIT_CTLS:
1502 	case MSR_IA32_VMX_ENTRY_CTLS:
1503 		/*
1504 		 * The "non-true" VMX capability MSRs are generated from the
1505 		 * "true" MSRs, so we do not support restoring them directly.
1506 		 *
1507 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1508 		 * should restore the "true" MSRs with the must-be-1 bits
1509 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1510 		 * DEFAULT SETTINGS".
1511 		 */
1512 		return -EINVAL;
1513 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1514 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1515 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1516 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1517 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1518 		return vmx_restore_control_msr(vmx, msr_index, data);
1519 	case MSR_IA32_VMX_MISC:
1520 		return vmx_restore_vmx_misc(vmx, data);
1521 	case MSR_IA32_VMX_CR0_FIXED0:
1522 	case MSR_IA32_VMX_CR4_FIXED0:
1523 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1524 	case MSR_IA32_VMX_CR0_FIXED1:
1525 	case MSR_IA32_VMX_CR4_FIXED1:
1526 		/*
1527 		 * These MSRs are generated based on the vCPU's CPUID, so we
1528 		 * do not support restoring them directly.
1529 		 */
1530 		return -EINVAL;
1531 	case MSR_IA32_VMX_EPT_VPID_CAP:
1532 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1533 	case MSR_IA32_VMX_VMCS_ENUM:
1534 		vmx->nested.msrs.vmcs_enum = data;
1535 		return 0;
1536 	case MSR_IA32_VMX_VMFUNC:
1537 		if (data & ~vmcs_config.nested.vmfunc_controls)
1538 			return -EINVAL;
1539 		vmx->nested.msrs.vmfunc_controls = data;
1540 		return 0;
1541 	default:
1542 		/*
1543 		 * The rest of the VMX capability MSRs do not support restore.
1544 		 */
1545 		return -EINVAL;
1546 	}
1547 }
1548 
1549 /* Returns 0 on success, non-0 otherwise. */
1550 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1551 {
1552 	switch (msr_index) {
1553 	case MSR_IA32_VMX_BASIC:
1554 		*pdata = msrs->basic;
1555 		break;
1556 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1557 	case MSR_IA32_VMX_PINBASED_CTLS:
1558 		*pdata = vmx_control_msr(
1559 			msrs->pinbased_ctls_low,
1560 			msrs->pinbased_ctls_high);
1561 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1562 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1563 		break;
1564 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1565 	case MSR_IA32_VMX_PROCBASED_CTLS:
1566 		*pdata = vmx_control_msr(
1567 			msrs->procbased_ctls_low,
1568 			msrs->procbased_ctls_high);
1569 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1570 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1571 		break;
1572 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1573 	case MSR_IA32_VMX_EXIT_CTLS:
1574 		*pdata = vmx_control_msr(
1575 			msrs->exit_ctls_low,
1576 			msrs->exit_ctls_high);
1577 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1578 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1579 		break;
1580 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1581 	case MSR_IA32_VMX_ENTRY_CTLS:
1582 		*pdata = vmx_control_msr(
1583 			msrs->entry_ctls_low,
1584 			msrs->entry_ctls_high);
1585 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1586 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1587 		break;
1588 	case MSR_IA32_VMX_MISC:
1589 		*pdata = vmx_control_msr(
1590 			msrs->misc_low,
1591 			msrs->misc_high);
1592 		break;
1593 	case MSR_IA32_VMX_CR0_FIXED0:
1594 		*pdata = msrs->cr0_fixed0;
1595 		break;
1596 	case MSR_IA32_VMX_CR0_FIXED1:
1597 		*pdata = msrs->cr0_fixed1;
1598 		break;
1599 	case MSR_IA32_VMX_CR4_FIXED0:
1600 		*pdata = msrs->cr4_fixed0;
1601 		break;
1602 	case MSR_IA32_VMX_CR4_FIXED1:
1603 		*pdata = msrs->cr4_fixed1;
1604 		break;
1605 	case MSR_IA32_VMX_VMCS_ENUM:
1606 		*pdata = msrs->vmcs_enum;
1607 		break;
1608 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1609 		*pdata = vmx_control_msr(
1610 			msrs->secondary_ctls_low,
1611 			msrs->secondary_ctls_high);
1612 		break;
1613 	case MSR_IA32_VMX_EPT_VPID_CAP:
1614 		*pdata = msrs->ept_caps |
1615 			((u64)msrs->vpid_caps << 32);
1616 		break;
1617 	case MSR_IA32_VMX_VMFUNC:
1618 		*pdata = msrs->vmfunc_controls;
1619 		break;
1620 	default:
1621 		return 1;
1622 	}
1623 
1624 	return 0;
1625 }
1626 
1627 /*
1628  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1629  * been modified by the L1 guest.  Note, "writable" in this context means
1630  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1631  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1632  * VM-exit information fields (which are actually writable if the vCPU is
1633  * configured to support "VMWRITE to any supported field in the VMCS").
1634  */
1635 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1636 {
1637 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1638 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1639 	struct shadow_vmcs_field field;
1640 	unsigned long val;
1641 	int i;
1642 
1643 	if (WARN_ON(!shadow_vmcs))
1644 		return;
1645 
1646 	preempt_disable();
1647 
1648 	vmcs_load(shadow_vmcs);
1649 
1650 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1651 		field = shadow_read_write_fields[i];
1652 		val = __vmcs_readl(field.encoding);
1653 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1654 	}
1655 
1656 	vmcs_clear(shadow_vmcs);
1657 	vmcs_load(vmx->loaded_vmcs->vmcs);
1658 
1659 	preempt_enable();
1660 }
1661 
1662 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1663 {
1664 	const struct shadow_vmcs_field *fields[] = {
1665 		shadow_read_write_fields,
1666 		shadow_read_only_fields
1667 	};
1668 	const int max_fields[] = {
1669 		max_shadow_read_write_fields,
1670 		max_shadow_read_only_fields
1671 	};
1672 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1673 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1674 	struct shadow_vmcs_field field;
1675 	unsigned long val;
1676 	int i, q;
1677 
1678 	if (WARN_ON(!shadow_vmcs))
1679 		return;
1680 
1681 	vmcs_load(shadow_vmcs);
1682 
1683 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1684 		for (i = 0; i < max_fields[q]; i++) {
1685 			field = fields[q][i];
1686 			val = vmcs12_read_any(vmcs12, field.encoding,
1687 					      field.offset);
1688 			__vmcs_writel(field.encoding, val);
1689 		}
1690 	}
1691 
1692 	vmcs_clear(shadow_vmcs);
1693 	vmcs_load(vmx->loaded_vmcs->vmcs);
1694 }
1695 
1696 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1697 {
1698 #ifdef CONFIG_KVM_HYPERV
1699 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1700 	struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
1701 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(&vmx->vcpu);
1702 
1703 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1704 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1705 	vmcs12->guest_rip = evmcs->guest_rip;
1706 
1707 	if (unlikely(!(hv_clean_fields &
1708 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_ENLIGHTENMENTSCONTROL))) {
1709 		hv_vcpu->nested.pa_page_gpa = evmcs->partition_assist_page;
1710 		hv_vcpu->nested.vm_id = evmcs->hv_vm_id;
1711 		hv_vcpu->nested.vp_id = evmcs->hv_vp_id;
1712 	}
1713 
1714 	if (unlikely(!(hv_clean_fields &
1715 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1716 		vmcs12->guest_rsp = evmcs->guest_rsp;
1717 		vmcs12->guest_rflags = evmcs->guest_rflags;
1718 		vmcs12->guest_interruptibility_info =
1719 			evmcs->guest_interruptibility_info;
1720 		/*
1721 		 * Not present in struct vmcs12:
1722 		 * vmcs12->guest_ssp = evmcs->guest_ssp;
1723 		 */
1724 	}
1725 
1726 	if (unlikely(!(hv_clean_fields &
1727 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1728 		vmcs12->cpu_based_vm_exec_control =
1729 			evmcs->cpu_based_vm_exec_control;
1730 	}
1731 
1732 	if (unlikely(!(hv_clean_fields &
1733 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1734 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1735 	}
1736 
1737 	if (unlikely(!(hv_clean_fields &
1738 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1739 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1740 	}
1741 
1742 	if (unlikely(!(hv_clean_fields &
1743 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1744 		vmcs12->vm_entry_intr_info_field =
1745 			evmcs->vm_entry_intr_info_field;
1746 		vmcs12->vm_entry_exception_error_code =
1747 			evmcs->vm_entry_exception_error_code;
1748 		vmcs12->vm_entry_instruction_len =
1749 			evmcs->vm_entry_instruction_len;
1750 	}
1751 
1752 	if (unlikely(!(hv_clean_fields &
1753 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1754 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1755 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1756 		vmcs12->host_cr0 = evmcs->host_cr0;
1757 		vmcs12->host_cr3 = evmcs->host_cr3;
1758 		vmcs12->host_cr4 = evmcs->host_cr4;
1759 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1760 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1761 		vmcs12->host_rip = evmcs->host_rip;
1762 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1763 		vmcs12->host_es_selector = evmcs->host_es_selector;
1764 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1765 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1766 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1767 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1768 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1769 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1770 		vmcs12->host_ia32_perf_global_ctrl = evmcs->host_ia32_perf_global_ctrl;
1771 		/*
1772 		 * Not present in struct vmcs12:
1773 		 * vmcs12->host_ia32_s_cet = evmcs->host_ia32_s_cet;
1774 		 * vmcs12->host_ssp = evmcs->host_ssp;
1775 		 * vmcs12->host_ia32_int_ssp_table_addr = evmcs->host_ia32_int_ssp_table_addr;
1776 		 */
1777 	}
1778 
1779 	if (unlikely(!(hv_clean_fields &
1780 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1781 		vmcs12->pin_based_vm_exec_control =
1782 			evmcs->pin_based_vm_exec_control;
1783 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1784 		vmcs12->secondary_vm_exec_control =
1785 			evmcs->secondary_vm_exec_control;
1786 	}
1787 
1788 	if (unlikely(!(hv_clean_fields &
1789 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1790 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1791 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1792 	}
1793 
1794 	if (unlikely(!(hv_clean_fields &
1795 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1796 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1797 	}
1798 
1799 	if (unlikely(!(hv_clean_fields &
1800 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1801 		vmcs12->guest_es_base = evmcs->guest_es_base;
1802 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1803 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1804 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1805 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1806 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1807 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1808 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1809 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1810 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1811 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1812 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1813 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1814 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1815 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1816 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1817 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1818 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1819 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1820 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1821 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1822 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1823 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1824 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1825 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1826 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1827 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1828 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1829 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1830 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1831 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1832 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1833 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1834 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1835 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1836 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1837 	}
1838 
1839 	if (unlikely(!(hv_clean_fields &
1840 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1841 		vmcs12->tsc_offset = evmcs->tsc_offset;
1842 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1843 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1844 		vmcs12->encls_exiting_bitmap = evmcs->encls_exiting_bitmap;
1845 		vmcs12->tsc_multiplier = evmcs->tsc_multiplier;
1846 	}
1847 
1848 	if (unlikely(!(hv_clean_fields &
1849 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1850 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1851 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1852 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1853 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1854 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1855 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1856 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1857 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1858 	}
1859 
1860 	if (unlikely(!(hv_clean_fields &
1861 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1862 		vmcs12->host_fs_base = evmcs->host_fs_base;
1863 		vmcs12->host_gs_base = evmcs->host_gs_base;
1864 		vmcs12->host_tr_base = evmcs->host_tr_base;
1865 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1866 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1867 		vmcs12->host_rsp = evmcs->host_rsp;
1868 	}
1869 
1870 	if (unlikely(!(hv_clean_fields &
1871 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1872 		vmcs12->ept_pointer = evmcs->ept_pointer;
1873 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1874 	}
1875 
1876 	if (unlikely(!(hv_clean_fields &
1877 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1878 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1879 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1880 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1881 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1882 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1883 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1884 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1885 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1886 		vmcs12->guest_pending_dbg_exceptions =
1887 			evmcs->guest_pending_dbg_exceptions;
1888 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1889 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1890 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1891 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1892 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1893 		vmcs12->guest_ia32_perf_global_ctrl = evmcs->guest_ia32_perf_global_ctrl;
1894 		/*
1895 		 * Not present in struct vmcs12:
1896 		 * vmcs12->guest_ia32_s_cet = evmcs->guest_ia32_s_cet;
1897 		 * vmcs12->guest_ia32_lbr_ctl = evmcs->guest_ia32_lbr_ctl;
1898 		 * vmcs12->guest_ia32_int_ssp_table_addr = evmcs->guest_ia32_int_ssp_table_addr;
1899 		 */
1900 	}
1901 
1902 	/*
1903 	 * Not used?
1904 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1905 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1906 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1907 	 * vmcs12->page_fault_error_code_mask =
1908 	 *		evmcs->page_fault_error_code_mask;
1909 	 * vmcs12->page_fault_error_code_match =
1910 	 *		evmcs->page_fault_error_code_match;
1911 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1912 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1913 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1914 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1915 	 */
1916 
1917 	/*
1918 	 * Read only fields:
1919 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1920 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1921 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1922 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1923 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1924 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1925 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1926 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1927 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1928 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1929 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1930 	 *
1931 	 * Not present in struct vmcs12:
1932 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1933 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1934 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1935 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1936 	 */
1937 
1938 	return;
1939 #else /* CONFIG_KVM_HYPERV */
1940 	KVM_BUG_ON(1, vmx->vcpu.kvm);
1941 #endif /* CONFIG_KVM_HYPERV */
1942 }
1943 
1944 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1945 {
1946 #ifdef CONFIG_KVM_HYPERV
1947 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1948 	struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
1949 
1950 	/*
1951 	 * Should not be changed by KVM:
1952 	 *
1953 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1954 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1955 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1956 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1957 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1958 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1959 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1960 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1961 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1962 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1963 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1964 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1965 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1966 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1967 	 * evmcs->host_rip = vmcs12->host_rip;
1968 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1969 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1970 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1971 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1972 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1973 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1974 	 * evmcs->host_rsp = vmcs12->host_rsp;
1975 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1976 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1977 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1978 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1979 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1980 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1981 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1982 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1983 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1984 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1985 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1986 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1987 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1988 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1989 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1990 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1991 	 * evmcs->page_fault_error_code_mask =
1992 	 *		vmcs12->page_fault_error_code_mask;
1993 	 * evmcs->page_fault_error_code_match =
1994 	 *		vmcs12->page_fault_error_code_match;
1995 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1996 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1997 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1998 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1999 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
2000 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
2001 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
2002 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
2003 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
2004 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
2005 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
2006 	 * evmcs->guest_ia32_perf_global_ctrl = vmcs12->guest_ia32_perf_global_ctrl;
2007 	 * evmcs->host_ia32_perf_global_ctrl = vmcs12->host_ia32_perf_global_ctrl;
2008 	 * evmcs->encls_exiting_bitmap = vmcs12->encls_exiting_bitmap;
2009 	 * evmcs->tsc_multiplier = vmcs12->tsc_multiplier;
2010 	 *
2011 	 * Not present in struct vmcs12:
2012 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
2013 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
2014 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
2015 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
2016 	 * evmcs->host_ia32_s_cet = vmcs12->host_ia32_s_cet;
2017 	 * evmcs->host_ssp = vmcs12->host_ssp;
2018 	 * evmcs->host_ia32_int_ssp_table_addr = vmcs12->host_ia32_int_ssp_table_addr;
2019 	 * evmcs->guest_ia32_s_cet = vmcs12->guest_ia32_s_cet;
2020 	 * evmcs->guest_ia32_lbr_ctl = vmcs12->guest_ia32_lbr_ctl;
2021 	 * evmcs->guest_ia32_int_ssp_table_addr = vmcs12->guest_ia32_int_ssp_table_addr;
2022 	 * evmcs->guest_ssp = vmcs12->guest_ssp;
2023 	 */
2024 
2025 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
2026 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
2027 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
2028 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
2029 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
2030 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
2031 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
2032 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
2033 
2034 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
2035 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
2036 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
2037 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
2038 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
2039 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
2040 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
2041 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
2042 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
2043 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
2044 
2045 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
2046 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
2047 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
2048 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
2049 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
2050 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
2051 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
2052 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
2053 
2054 	evmcs->guest_es_base = vmcs12->guest_es_base;
2055 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
2056 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
2057 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
2058 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
2059 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
2060 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
2061 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
2062 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
2063 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
2064 
2065 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
2066 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
2067 
2068 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
2069 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
2070 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
2071 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
2072 
2073 	evmcs->guest_pending_dbg_exceptions =
2074 		vmcs12->guest_pending_dbg_exceptions;
2075 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
2076 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
2077 
2078 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
2079 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
2080 
2081 	evmcs->guest_cr0 = vmcs12->guest_cr0;
2082 	evmcs->guest_cr3 = vmcs12->guest_cr3;
2083 	evmcs->guest_cr4 = vmcs12->guest_cr4;
2084 	evmcs->guest_dr7 = vmcs12->guest_dr7;
2085 
2086 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
2087 
2088 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
2089 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
2090 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
2091 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
2092 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
2093 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
2094 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
2095 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
2096 
2097 	evmcs->exit_qualification = vmcs12->exit_qualification;
2098 
2099 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
2100 	evmcs->guest_rsp = vmcs12->guest_rsp;
2101 	evmcs->guest_rflags = vmcs12->guest_rflags;
2102 
2103 	evmcs->guest_interruptibility_info =
2104 		vmcs12->guest_interruptibility_info;
2105 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
2106 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
2107 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
2108 	evmcs->vm_entry_exception_error_code =
2109 		vmcs12->vm_entry_exception_error_code;
2110 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
2111 
2112 	evmcs->guest_rip = vmcs12->guest_rip;
2113 
2114 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
2115 
2116 	return;
2117 #else /* CONFIG_KVM_HYPERV */
2118 	KVM_BUG_ON(1, vmx->vcpu.kvm);
2119 #endif /* CONFIG_KVM_HYPERV */
2120 }
2121 
2122 /*
2123  * This is an equivalent of the nested hypervisor executing the vmptrld
2124  * instruction.
2125  */
2126 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
2127 	struct kvm_vcpu *vcpu, bool from_launch)
2128 {
2129 #ifdef CONFIG_KVM_HYPERV
2130 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2131 	bool evmcs_gpa_changed = false;
2132 	u64 evmcs_gpa;
2133 
2134 	if (likely(!guest_cpu_cap_has_evmcs(vcpu)))
2135 		return EVMPTRLD_DISABLED;
2136 
2137 	evmcs_gpa = nested_get_evmptr(vcpu);
2138 	if (!evmptr_is_valid(evmcs_gpa)) {
2139 		nested_release_evmcs(vcpu);
2140 		return EVMPTRLD_DISABLED;
2141 	}
2142 
2143 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
2144 		vmx->nested.current_vmptr = INVALID_GPA;
2145 
2146 		nested_release_evmcs(vcpu);
2147 
2148 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2149 				 &vmx->nested.hv_evmcs_map))
2150 			return EVMPTRLD_ERROR;
2151 
2152 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2153 
2154 		/*
2155 		 * Currently, KVM only supports eVMCS version 1
2156 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2157 		 * value to first u32 field of eVMCS which should specify eVMCS
2158 		 * VersionNumber.
2159 		 *
2160 		 * Guest should be aware of supported eVMCS versions by host by
2161 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2162 		 * expected to set this CPUID leaf according to the value
2163 		 * returned in vmcs_version from nested_enable_evmcs().
2164 		 *
2165 		 * However, it turns out that Microsoft Hyper-V fails to comply
2166 		 * to their own invented interface: When Hyper-V use eVMCS, it
2167 		 * just sets first u32 field of eVMCS to revision_id specified
2168 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2169 		 * which is one of the supported versions specified in
2170 		 * CPUID.0x4000000A.EAX[0:15].
2171 		 *
2172 		 * To overcome Hyper-V bug, we accept here either a supported
2173 		 * eVMCS version or VMCS12 revision_id as valid values for first
2174 		 * u32 field of eVMCS.
2175 		 */
2176 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2177 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2178 			nested_release_evmcs(vcpu);
2179 			return EVMPTRLD_VMFAIL;
2180 		}
2181 
2182 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2183 
2184 		evmcs_gpa_changed = true;
2185 		/*
2186 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2187 		 * reloaded from guest's memory (read only fields, fields not
2188 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2189 		 * are no leftovers.
2190 		 */
2191 		if (from_launch) {
2192 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2193 			memset(vmcs12, 0, sizeof(*vmcs12));
2194 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2195 		}
2196 
2197 	}
2198 
2199 	/*
2200 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2201 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2202 	 */
2203 	if (from_launch || evmcs_gpa_changed) {
2204 		vmx->nested.hv_evmcs->hv_clean_fields &=
2205 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2206 
2207 		vmx->nested.force_msr_bitmap_recalc = true;
2208 	}
2209 
2210 	return EVMPTRLD_SUCCEEDED;
2211 #else
2212 	return EVMPTRLD_DISABLED;
2213 #endif
2214 }
2215 
2216 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2217 {
2218 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2219 
2220 	if (nested_vmx_is_evmptr12_valid(vmx))
2221 		copy_vmcs12_to_enlightened(vmx);
2222 	else
2223 		copy_vmcs12_to_shadow(vmx);
2224 
2225 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2226 }
2227 
2228 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2229 {
2230 	struct vcpu_vmx *vmx =
2231 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2232 
2233 	vmx->nested.preemption_timer_expired = true;
2234 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2235 	kvm_vcpu_kick(&vmx->vcpu);
2236 
2237 	return HRTIMER_NORESTART;
2238 }
2239 
2240 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2241 {
2242 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2243 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2244 
2245 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2246 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2247 
2248 	if (!vmx->nested.has_preemption_timer_deadline) {
2249 		vmx->nested.preemption_timer_deadline =
2250 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2251 		vmx->nested.has_preemption_timer_deadline = true;
2252 	}
2253 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2254 }
2255 
2256 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2257 					u64 preemption_timeout)
2258 {
2259 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2260 
2261 	/*
2262 	 * A timer value of zero is architecturally guaranteed to cause
2263 	 * a VMExit prior to executing any instructions in the guest.
2264 	 */
2265 	if (preemption_timeout == 0) {
2266 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2267 		return;
2268 	}
2269 
2270 	if (vcpu->arch.virtual_tsc_khz == 0)
2271 		return;
2272 
2273 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2274 	preemption_timeout *= 1000000;
2275 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2276 	hrtimer_start(&vmx->nested.preemption_timer,
2277 		      ktime_add_ns(ktime_get(), preemption_timeout),
2278 		      HRTIMER_MODE_ABS_PINNED);
2279 }
2280 
2281 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2282 {
2283 	if (vmx->nested.nested_run_pending &&
2284 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2285 		return vmcs12->guest_ia32_efer;
2286 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2287 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2288 	else
2289 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2290 }
2291 
2292 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2293 {
2294 	struct kvm *kvm = vmx->vcpu.kvm;
2295 
2296 	/*
2297 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2298 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2299 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2300 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2301 	 */
2302 	if (vmx->nested.vmcs02_initialized)
2303 		return;
2304 	vmx->nested.vmcs02_initialized = true;
2305 
2306 	if (vmx->ve_info)
2307 		vmcs_write64(VE_INFORMATION_ADDRESS, __pa(vmx->ve_info));
2308 
2309 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2310 	if (cpu_has_vmx_vmfunc())
2311 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2312 
2313 	if (cpu_has_vmx_posted_intr())
2314 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2315 
2316 	if (cpu_has_vmx_msr_bitmap())
2317 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2318 
2319 	/*
2320 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2321 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2322 	 * deal with filtering out L2 GPAs from the buffer.
2323 	 */
2324 	if (enable_pml) {
2325 		vmcs_write64(PML_ADDRESS, 0);
2326 		vmcs_write16(GUEST_PML_INDEX, -1);
2327 	}
2328 
2329 	if (cpu_has_vmx_encls_vmexit())
2330 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2331 
2332 	if (kvm_notify_vmexit_enabled(kvm))
2333 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
2334 
2335 	/*
2336 	 * Set the MSR load/store lists to match L0's settings.  Only the
2337 	 * addresses are constant (for vmcs02), the counts can change based
2338 	 * on L2's behavior, e.g. switching to/from long mode.
2339 	 */
2340 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2341 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2342 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2343 
2344 	vmx_set_constant_host_state(vmx);
2345 }
2346 
2347 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2348 				      struct vmcs12 *vmcs12)
2349 {
2350 	prepare_vmcs02_constant_state(vmx);
2351 
2352 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2353 
2354 	/*
2355 	 * If VPID is disabled, then guest TLB accesses use VPID=0, i.e. the
2356 	 * same VPID as the host.  Emulate this behavior by using vpid01 for L2
2357 	 * if VPID is disabled in vmcs12.  Note, if VPID is disabled, VM-Enter
2358 	 * and VM-Exit are architecturally required to flush VPID=0, but *only*
2359 	 * VPID=0.  I.e. using vpid02 would be ok (so long as KVM emulates the
2360 	 * required flushes), but doing so would cause KVM to over-flush.  E.g.
2361 	 * if L1 runs L2 X with VPID12=1, then runs L2 Y with VPID12 disabled,
2362 	 * and then runs L2 X again, then KVM can and should retain TLB entries
2363 	 * for VPID12=1.
2364 	 */
2365 	if (enable_vpid) {
2366 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2367 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2368 		else
2369 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2370 	}
2371 }
2372 
2373 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2374 				 struct vmcs12 *vmcs12)
2375 {
2376 	u32 exec_control;
2377 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2378 
2379 	if (vmx->nested.dirty_vmcs12 || nested_vmx_is_evmptr12_valid(vmx))
2380 		prepare_vmcs02_early_rare(vmx, vmcs12);
2381 
2382 	/*
2383 	 * PIN CONTROLS
2384 	 */
2385 	exec_control = __pin_controls_get(vmcs01);
2386 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2387 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2388 
2389 	/* Posted interrupts setting is only taken from vmcs12.  */
2390 	vmx->nested.pi_pending = false;
2391 	if (nested_cpu_has_posted_intr(vmcs12)) {
2392 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2393 	} else {
2394 		vmx->nested.posted_intr_nv = -1;
2395 		exec_control &= ~PIN_BASED_POSTED_INTR;
2396 	}
2397 	pin_controls_set(vmx, exec_control);
2398 
2399 	/*
2400 	 * EXEC CONTROLS
2401 	 */
2402 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2403 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2404 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2405 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2406 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2407 
2408 	vmx->nested.l1_tpr_threshold = -1;
2409 	if (exec_control & CPU_BASED_TPR_SHADOW)
2410 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2411 #ifdef CONFIG_X86_64
2412 	else
2413 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2414 				CPU_BASED_CR8_STORE_EXITING;
2415 #endif
2416 
2417 	/*
2418 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2419 	 * for I/O port accesses.
2420 	 */
2421 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2422 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2423 
2424 	/*
2425 	 * This bit will be computed in nested_get_vmcs12_pages, because
2426 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2427 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2428 	 * only set/clear this bit.
2429 	 */
2430 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2431 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2432 
2433 	exec_controls_set(vmx, exec_control);
2434 
2435 	/*
2436 	 * SECONDARY EXEC CONTROLS
2437 	 */
2438 	if (cpu_has_secondary_exec_ctrls()) {
2439 		exec_control = __secondary_exec_controls_get(vmcs01);
2440 
2441 		/* Take the following fields only from vmcs12 */
2442 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2443 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2444 				  SECONDARY_EXEC_ENABLE_INVPCID |
2445 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2446 				  SECONDARY_EXEC_ENABLE_XSAVES |
2447 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2448 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2449 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2450 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2451 				  SECONDARY_EXEC_DESC);
2452 
2453 		if (nested_cpu_has(vmcs12,
2454 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2455 			exec_control |= vmcs12->secondary_vm_exec_control;
2456 
2457 		/* PML is emulated and never enabled in hardware for L2. */
2458 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2459 
2460 		/* VMCS shadowing for L2 is emulated for now */
2461 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2462 
2463 		/*
2464 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2465 		 * will not have to rewrite the controls just for this bit.
2466 		 */
2467 		if (vmx_umip_emulated() && (vmcs12->guest_cr4 & X86_CR4_UMIP))
2468 			exec_control |= SECONDARY_EXEC_DESC;
2469 
2470 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2471 			vmcs_write16(GUEST_INTR_STATUS,
2472 				vmcs12->guest_intr_status);
2473 
2474 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2475 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2476 
2477 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2478 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2479 
2480 		secondary_exec_controls_set(vmx, exec_control);
2481 	}
2482 
2483 	/*
2484 	 * ENTRY CONTROLS
2485 	 *
2486 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2487 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2488 	 * on the related bits (if supported by the CPU) in the hope that
2489 	 * we can avoid VMWrites during vmx_set_efer().
2490 	 *
2491 	 * Similarly, take vmcs01's PERF_GLOBAL_CTRL in the hope that if KVM is
2492 	 * loading PERF_GLOBAL_CTRL via the VMCS for L1, then KVM will want to
2493 	 * do the same for L2.
2494 	 */
2495 	exec_control = __vm_entry_controls_get(vmcs01);
2496 	exec_control |= (vmcs12->vm_entry_controls &
2497 			 ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
2498 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2499 	if (cpu_has_load_ia32_efer()) {
2500 		if (guest_efer & EFER_LMA)
2501 			exec_control |= VM_ENTRY_IA32E_MODE;
2502 		if (guest_efer != kvm_host.efer)
2503 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2504 	}
2505 	vm_entry_controls_set(vmx, exec_control);
2506 
2507 	/*
2508 	 * EXIT CONTROLS
2509 	 *
2510 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2511 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2512 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2513 	 */
2514 	exec_control = __vm_exit_controls_get(vmcs01);
2515 	if (cpu_has_load_ia32_efer() && guest_efer != kvm_host.efer)
2516 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2517 	else
2518 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2519 	vm_exit_controls_set(vmx, exec_control);
2520 
2521 	/*
2522 	 * Interrupt/Exception Fields
2523 	 */
2524 	if (vmx->nested.nested_run_pending) {
2525 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2526 			     vmcs12->vm_entry_intr_info_field);
2527 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2528 			     vmcs12->vm_entry_exception_error_code);
2529 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2530 			     vmcs12->vm_entry_instruction_len);
2531 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2532 			     vmcs12->guest_interruptibility_info);
2533 		vmx->loaded_vmcs->nmi_known_unmasked =
2534 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2535 	} else {
2536 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2537 	}
2538 }
2539 
2540 static void vmcs_read_cet_state(struct kvm_vcpu *vcpu, u64 *s_cet,
2541 				u64 *ssp, u64 *ssp_tbl)
2542 {
2543 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_IBT) ||
2544 	    guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK))
2545 		*s_cet = vmcs_readl(GUEST_S_CET);
2546 
2547 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK)) {
2548 		*ssp = vmcs_readl(GUEST_SSP);
2549 		*ssp_tbl = vmcs_readl(GUEST_INTR_SSP_TABLE);
2550 	}
2551 }
2552 
2553 static void vmcs_write_cet_state(struct kvm_vcpu *vcpu, u64 s_cet,
2554 				 u64 ssp, u64 ssp_tbl)
2555 {
2556 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_IBT) ||
2557 	    guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK))
2558 		vmcs_writel(GUEST_S_CET, s_cet);
2559 
2560 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK)) {
2561 		vmcs_writel(GUEST_SSP, ssp);
2562 		vmcs_writel(GUEST_INTR_SSP_TABLE, ssp_tbl);
2563 	}
2564 }
2565 
2566 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2567 {
2568 	struct hv_enlightened_vmcs *hv_evmcs = nested_vmx_evmcs(vmx);
2569 
2570 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2571 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2572 
2573 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2574 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2575 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2576 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2577 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2578 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2579 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2580 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2581 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2582 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2583 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2584 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2585 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2586 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2587 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2588 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2589 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2590 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2591 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2592 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2593 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2594 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2595 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2596 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2597 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2598 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2599 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2600 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2601 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2602 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2603 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2604 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2605 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2606 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2607 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2608 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2609 
2610 		vmx_segment_cache_clear(vmx);
2611 	}
2612 
2613 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2614 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2615 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2616 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2617 			    vmcs12->guest_pending_dbg_exceptions);
2618 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2619 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2620 
2621 		/*
2622 		 * L1 may access the L2's PDPTR, so save them to construct
2623 		 * vmcs12
2624 		 */
2625 		if (enable_ept) {
2626 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2627 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2628 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2629 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2630 		}
2631 
2632 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2633 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2634 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2635 	}
2636 
2637 	if (nested_cpu_has_xsaves(vmcs12))
2638 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2639 
2640 	/*
2641 	 * Whether page-faults are trapped is determined by a combination of
2642 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2643 	 * doesn't care about page faults then we should set all of these to
2644 	 * L1's desires. However, if L0 does care about (some) page faults, it
2645 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2646 	 * simply ask to exit on each and every L2 page fault. This is done by
2647 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2648 	 * Note that below we don't need special code to set EB.PF beyond the
2649 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2650 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2651 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2652 	 */
2653 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2654 		/*
2655 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2656 		 * go ahead and use it?
2657 		 */
2658 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2659 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2660 	} else {
2661 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2662 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2663 	}
2664 
2665 	if (cpu_has_vmx_apicv()) {
2666 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2667 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2668 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2669 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2670 	}
2671 
2672 	/*
2673 	 * Make sure the msr_autostore list is up to date before we set the
2674 	 * count in the vmcs02.
2675 	 */
2676 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2677 
2678 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2679 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2680 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2681 
2682 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)
2683 		vmcs_write_cet_state(&vmx->vcpu, vmcs12->guest_s_cet,
2684 				     vmcs12->guest_ssp, vmcs12->guest_ssp_tbl);
2685 
2686 	set_cr4_guest_host_mask(vmx);
2687 }
2688 
2689 /*
2690  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2691  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2692  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2693  * guest in a way that will both be appropriate to L1's requests, and our
2694  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2695  * function also has additional necessary side-effects, like setting various
2696  * vcpu->arch fields.
2697  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2698  * is assigned to entry_failure_code on failure.
2699  */
2700 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2701 			  bool from_vmentry,
2702 			  enum vm_entry_failure_code *entry_failure_code)
2703 {
2704 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2705 	struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
2706 	bool load_guest_pdptrs_vmcs12 = false;
2707 
2708 	if (vmx->nested.dirty_vmcs12 || nested_vmx_is_evmptr12_valid(vmx)) {
2709 		prepare_vmcs02_rare(vmx, vmcs12);
2710 		vmx->nested.dirty_vmcs12 = false;
2711 
2712 		load_guest_pdptrs_vmcs12 = !nested_vmx_is_evmptr12_valid(vmx) ||
2713 			!(evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2714 	}
2715 
2716 	if (vmx->nested.nested_run_pending &&
2717 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2718 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2719 		vmx_guest_debugctl_write(vcpu, vmcs12->guest_ia32_debugctl &
2720 					       vmx_get_supported_debugctl(vcpu, false));
2721 	} else {
2722 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2723 		vmx_guest_debugctl_write(vcpu, vmx->nested.pre_vmenter_debugctl);
2724 	}
2725 
2726 	if (!vmx->nested.nested_run_pending ||
2727 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE))
2728 		vmcs_write_cet_state(vcpu, vmx->nested.pre_vmenter_s_cet,
2729 				     vmx->nested.pre_vmenter_ssp,
2730 				     vmx->nested.pre_vmenter_ssp_tbl);
2731 
2732 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2733 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2734 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.pre_vmenter_bndcfgs);
2735 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2736 
2737 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2738 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2739 	 * trap. Note that CR0.TS also needs updating - we do this later.
2740 	 */
2741 	vmx_update_exception_bitmap(vcpu);
2742 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2743 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2744 
2745 	if (vmx->nested.nested_run_pending &&
2746 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2747 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2748 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2749 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2750 		vmcs_write64(GUEST_IA32_PAT, vcpu->arch.pat);
2751 	}
2752 
2753 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2754 			vcpu->arch.l1_tsc_offset,
2755 			vmx_get_l2_tsc_offset(vcpu),
2756 			vmx_get_l2_tsc_multiplier(vcpu));
2757 
2758 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2759 			vcpu->arch.l1_tsc_scaling_ratio,
2760 			vmx_get_l2_tsc_multiplier(vcpu));
2761 
2762 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2763 	if (kvm_caps.has_tsc_control)
2764 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2765 
2766 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2767 
2768 	if (nested_cpu_has_ept(vmcs12))
2769 		nested_ept_init_mmu_context(vcpu);
2770 
2771 	/*
2772 	 * Override the CR0/CR4 read shadows after setting the effective guest
2773 	 * CR0/CR4.  The common helpers also set the shadows, but they don't
2774 	 * account for vmcs12's cr0/4_guest_host_mask.
2775 	 */
2776 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2777 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2778 
2779 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2780 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2781 
2782 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2783 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2784 	vmx_set_efer(vcpu, vcpu->arch.efer);
2785 
2786 	/*
2787 	 * Guest state is invalid and unrestricted guest is disabled,
2788 	 * which means L1 attempted VMEntry to L2 with invalid state.
2789 	 * Fail the VMEntry.
2790 	 *
2791 	 * However when force loading the guest state (SMM exit or
2792 	 * loading nested state after migration, it is possible to
2793 	 * have invalid guest state now, which will be later fixed by
2794 	 * restoring L2 register state
2795 	 */
2796 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2797 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2798 		return -EINVAL;
2799 	}
2800 
2801 	/* Shadow page tables on either EPT or shadow page tables. */
2802 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2803 				from_vmentry, entry_failure_code))
2804 		return -EINVAL;
2805 
2806 	/*
2807 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2808 	 * on nested VM-Exit, which can occur without actually running L2 and
2809 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2810 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2811 	 * transition to HLT instead of running L2.
2812 	 */
2813 	if (enable_ept)
2814 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2815 
2816 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2817 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2818 	    is_pae_paging(vcpu)) {
2819 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2820 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2821 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2822 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2823 	}
2824 
2825 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2826 	    kvm_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)) &&
2827 	    WARN_ON_ONCE(__kvm_emulate_msr_write(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2828 						 vmcs12->guest_ia32_perf_global_ctrl))) {
2829 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2830 		return -EINVAL;
2831 	}
2832 
2833 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2834 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2835 
2836 	/*
2837 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2838 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2839 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2840 	 * here.
2841 	 */
2842 	if (nested_vmx_is_evmptr12_valid(vmx))
2843 		evmcs->hv_clean_fields |= HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2844 
2845 	return 0;
2846 }
2847 
2848 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2849 {
2850 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2851 	       nested_cpu_has_virtual_nmis(vmcs12)))
2852 		return -EINVAL;
2853 
2854 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2855 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2856 		return -EINVAL;
2857 
2858 	return 0;
2859 }
2860 
2861 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2862 {
2863 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2864 
2865 	/* Check for memory type validity */
2866 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2867 	case VMX_EPTP_MT_UC:
2868 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2869 			return false;
2870 		break;
2871 	case VMX_EPTP_MT_WB:
2872 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2873 			return false;
2874 		break;
2875 	default:
2876 		return false;
2877 	}
2878 
2879 	/* Page-walk levels validity. */
2880 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2881 	case VMX_EPTP_PWL_5:
2882 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2883 			return false;
2884 		break;
2885 	case VMX_EPTP_PWL_4:
2886 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2887 			return false;
2888 		break;
2889 	default:
2890 		return false;
2891 	}
2892 
2893 	/* Reserved bits should not be set */
2894 	if (CC(!kvm_vcpu_is_legal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2895 		return false;
2896 
2897 	/* AD, if set, should be supported */
2898 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2899 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2900 			return false;
2901 	}
2902 
2903 	return true;
2904 }
2905 
2906 /*
2907  * Checks related to VM-Execution Control Fields
2908  */
2909 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2910                                               struct vmcs12 *vmcs12)
2911 {
2912 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2913 
2914 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2915 				   vmx->nested.msrs.pinbased_ctls_low,
2916 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2917 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2918 				   vmx->nested.msrs.procbased_ctls_low,
2919 				   vmx->nested.msrs.procbased_ctls_high)))
2920 		return -EINVAL;
2921 
2922 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2923 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2924 				   vmx->nested.msrs.secondary_ctls_low,
2925 				   vmx->nested.msrs.secondary_ctls_high)))
2926 		return -EINVAL;
2927 
2928 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2929 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2930 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2931 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2932 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2933 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2934 	    nested_vmx_check_nmi_controls(vmcs12) ||
2935 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2936 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2937 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2938 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2939 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2940 		return -EINVAL;
2941 
2942 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2943 	    nested_cpu_has_save_preemption_timer(vmcs12))
2944 		return -EINVAL;
2945 
2946 	if (nested_cpu_has_ept(vmcs12) &&
2947 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2948 		return -EINVAL;
2949 
2950 	if (nested_cpu_has_vmfunc(vmcs12)) {
2951 		if (CC(vmcs12->vm_function_control &
2952 		       ~vmx->nested.msrs.vmfunc_controls))
2953 			return -EINVAL;
2954 
2955 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2956 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2957 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2958 				return -EINVAL;
2959 		}
2960 	}
2961 
2962 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING) &&
2963 	    CC(!vmcs12->tsc_multiplier))
2964 		return -EINVAL;
2965 
2966 	return 0;
2967 }
2968 
2969 /*
2970  * Checks related to VM-Exit Control Fields
2971  */
2972 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2973                                          struct vmcs12 *vmcs12)
2974 {
2975 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2976 
2977 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2978 				    vmx->nested.msrs.exit_ctls_low,
2979 				    vmx->nested.msrs.exit_ctls_high)) ||
2980 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2981 		return -EINVAL;
2982 
2983 	return 0;
2984 }
2985 
2986 /*
2987  * Checks related to VM-Entry Control Fields
2988  */
2989 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2990 					  struct vmcs12 *vmcs12)
2991 {
2992 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2993 
2994 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2995 				    vmx->nested.msrs.entry_ctls_low,
2996 				    vmx->nested.msrs.entry_ctls_high)))
2997 		return -EINVAL;
2998 
2999 	/*
3000 	 * From the Intel SDM, volume 3:
3001 	 * Fields relevant to VM-entry event injection must be set properly.
3002 	 * These fields are the VM-entry interruption-information field, the
3003 	 * VM-entry exception error code, and the VM-entry instruction length.
3004 	 */
3005 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
3006 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
3007 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
3008 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
3009 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
3010 		bool urg = nested_cpu_has2(vmcs12,
3011 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
3012 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
3013 
3014 		/* VM-entry interruption-info field: interruption type */
3015 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
3016 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
3017 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
3018 			return -EINVAL;
3019 
3020 		/* VM-entry interruption-info field: vector */
3021 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
3022 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
3023 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
3024 			return -EINVAL;
3025 
3026 		/*
3027 		 * Cannot deliver error code in real mode or if the interrupt
3028 		 * type is not hardware exception. For other cases, do the
3029 		 * consistency check only if the vCPU doesn't enumerate
3030 		 * VMX_BASIC_NO_HW_ERROR_CODE_CC.
3031 		 */
3032 		if (!prot_mode || intr_type != INTR_TYPE_HARD_EXCEPTION) {
3033 			if (CC(has_error_code))
3034 				return -EINVAL;
3035 		} else if (!nested_cpu_has_no_hw_errcode_cc(vcpu)) {
3036 			if (CC(has_error_code != x86_exception_has_error_code(vector)))
3037 				return -EINVAL;
3038 		}
3039 
3040 		/* VM-entry exception error code */
3041 		if (CC(has_error_code &&
3042 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
3043 			return -EINVAL;
3044 
3045 		/* VM-entry interruption-info field: reserved bits */
3046 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
3047 			return -EINVAL;
3048 
3049 		/* VM-entry instruction length */
3050 		switch (intr_type) {
3051 		case INTR_TYPE_SOFT_EXCEPTION:
3052 		case INTR_TYPE_SOFT_INTR:
3053 		case INTR_TYPE_PRIV_SW_EXCEPTION:
3054 			if (CC(vmcs12->vm_entry_instruction_len > X86_MAX_INSTRUCTION_LENGTH) ||
3055 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
3056 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
3057 				return -EINVAL;
3058 		}
3059 	}
3060 
3061 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
3062 		return -EINVAL;
3063 
3064 	return 0;
3065 }
3066 
3067 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
3068 				     struct vmcs12 *vmcs12)
3069 {
3070 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
3071 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
3072 	    nested_check_vm_entry_controls(vcpu, vmcs12))
3073 		return -EINVAL;
3074 
3075 #ifdef CONFIG_KVM_HYPERV
3076 	if (guest_cpu_cap_has_evmcs(vcpu))
3077 		return nested_evmcs_check_controls(vmcs12);
3078 #endif
3079 
3080 	return 0;
3081 }
3082 
3083 static int nested_vmx_check_controls_late(struct kvm_vcpu *vcpu,
3084 					  struct vmcs12 *vmcs12)
3085 {
3086 	void *vapic = to_vmx(vcpu)->nested.virtual_apic_map.hva;
3087 	u32 vtpr = vapic ? (*(u32 *)(vapic + APIC_TASKPRI)) >> 4 : 0;
3088 
3089 	/*
3090 	 * Don't bother with the consistency checks if KVM isn't configured to
3091 	 * WARN on missed consistency checks, as KVM needs to rely on hardware
3092 	 * to fully detect an illegal vTPR vs. TRP Threshold combination due to
3093 	 * the vTPR being writable by L1 at all times (it's an in-memory value,
3094 	 * not a VMCS field).  I.e. even if the check passes now, it might fail
3095 	 * at the actual VM-Enter.
3096 	 *
3097 	 * Keying off the module param also allows treating an invalid vAPIC
3098 	 * mapping as a consistency check failure without increasing the risk
3099 	 * of breaking a "real" VM.
3100 	 */
3101 	if (!warn_on_missed_cc)
3102 		return 0;
3103 
3104 	if ((exec_controls_get(to_vmx(vcpu)) & CPU_BASED_TPR_SHADOW) &&
3105 	    nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW) &&
3106 	    !nested_cpu_has_vid(vmcs12) &&
3107 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
3108 	    (CC(!vapic) ||
3109 	     CC((vmcs12->tpr_threshold & GENMASK(3, 0)) > (vtpr & GENMASK(3, 0)))))
3110 		return -EINVAL;
3111 
3112 	return 0;
3113 }
3114 
3115 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
3116 				       struct vmcs12 *vmcs12)
3117 {
3118 #ifdef CONFIG_X86_64
3119 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
3120 		!!(vcpu->arch.efer & EFER_LMA)))
3121 		return -EINVAL;
3122 #endif
3123 	return 0;
3124 }
3125 
3126 static bool is_l1_noncanonical_address_on_vmexit(u64 la, struct vmcs12 *vmcs12)
3127 {
3128 	/*
3129 	 * Check that the given linear address is canonical after a VM exit
3130 	 * from L2, based on HOST_CR4.LA57 value that will be loaded for L1.
3131 	 */
3132 	u8 l1_address_bits_on_exit = (vmcs12->host_cr4 & X86_CR4_LA57) ? 57 : 48;
3133 
3134 	return !__is_canonical_address(la, l1_address_bits_on_exit);
3135 }
3136 
3137 static int nested_vmx_check_cet_state_common(struct kvm_vcpu *vcpu, u64 s_cet,
3138 					     u64 ssp, u64 ssp_tbl)
3139 {
3140 	if (CC(!kvm_is_valid_u_s_cet(vcpu, s_cet)) || CC(!IS_ALIGNED(ssp, 4)) ||
3141 	    CC(is_noncanonical_msr_address(ssp_tbl, vcpu)))
3142 		return -EINVAL;
3143 
3144 	return 0;
3145 }
3146 
3147 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
3148 				       struct vmcs12 *vmcs12)
3149 {
3150 	bool ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
3151 
3152 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
3153 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
3154 	    CC(!kvm_vcpu_is_legal_cr3(vcpu, vmcs12->host_cr3)))
3155 		return -EINVAL;
3156 
3157 	if (CC(vmcs12->host_cr4 & X86_CR4_CET && !(vmcs12->host_cr0 & X86_CR0_WP)))
3158 		return -EINVAL;
3159 
3160 	if (CC(is_noncanonical_msr_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
3161 	    CC(is_noncanonical_msr_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
3162 		return -EINVAL;
3163 
3164 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
3165 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
3166 		return -EINVAL;
3167 
3168 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3169 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3170 					   vmcs12->host_ia32_perf_global_ctrl)))
3171 		return -EINVAL;
3172 
3173 	if (ia32e) {
3174 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
3175 			return -EINVAL;
3176 	} else {
3177 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
3178 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
3179 		    CC((vmcs12->host_rip) >> 32))
3180 			return -EINVAL;
3181 	}
3182 
3183 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3184 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3185 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3186 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3187 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3188 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3189 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3190 	    CC(vmcs12->host_cs_selector == 0) ||
3191 	    CC(vmcs12->host_tr_selector == 0) ||
3192 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
3193 		return -EINVAL;
3194 
3195 	if (CC(is_noncanonical_base_address(vmcs12->host_fs_base, vcpu)) ||
3196 	    CC(is_noncanonical_base_address(vmcs12->host_gs_base, vcpu)) ||
3197 	    CC(is_noncanonical_base_address(vmcs12->host_gdtr_base, vcpu)) ||
3198 	    CC(is_noncanonical_base_address(vmcs12->host_idtr_base, vcpu)) ||
3199 	    CC(is_noncanonical_base_address(vmcs12->host_tr_base, vcpu)) ||
3200 	    CC(is_l1_noncanonical_address_on_vmexit(vmcs12->host_rip, vmcs12)))
3201 		return -EINVAL;
3202 
3203 	/*
3204 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
3205 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
3206 	 * the values of the LMA and LME bits in the field must each be that of
3207 	 * the host address-space size VM-exit control.
3208 	 */
3209 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
3210 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
3211 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
3212 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
3213 			return -EINVAL;
3214 	}
3215 
3216 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_CET_STATE) {
3217 		if (nested_vmx_check_cet_state_common(vcpu, vmcs12->host_s_cet,
3218 						      vmcs12->host_ssp,
3219 						      vmcs12->host_ssp_tbl))
3220 			return -EINVAL;
3221 
3222 		/*
3223 		 * IA32_S_CET and SSP must be canonical if the host will
3224 		 * enter 64-bit mode after VM-exit; otherwise, higher
3225 		 * 32-bits must be all 0s.
3226 		 */
3227 		if (ia32e) {
3228 			if (CC(is_noncanonical_msr_address(vmcs12->host_s_cet, vcpu)) ||
3229 			    CC(is_noncanonical_msr_address(vmcs12->host_ssp, vcpu)))
3230 				return -EINVAL;
3231 		} else {
3232 			if (CC(vmcs12->host_s_cet >> 32) || CC(vmcs12->host_ssp >> 32))
3233 				return -EINVAL;
3234 		}
3235 	}
3236 
3237 	return 0;
3238 }
3239 
3240 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
3241 					  struct vmcs12 *vmcs12)
3242 {
3243 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3244 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
3245 	struct vmcs_hdr hdr;
3246 
3247 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
3248 		return 0;
3249 
3250 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
3251 		return -EINVAL;
3252 
3253 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
3254 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
3255 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
3256                 return -EINVAL;
3257 
3258 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
3259 					    offsetof(struct vmcs12, hdr),
3260 					    sizeof(hdr))))
3261 		return -EINVAL;
3262 
3263 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
3264 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
3265 		return -EINVAL;
3266 
3267 	return 0;
3268 }
3269 
3270 /*
3271  * Checks related to Guest Non-register State
3272  */
3273 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
3274 {
3275 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
3276 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
3277 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
3278 		return -EINVAL;
3279 
3280 	return 0;
3281 }
3282 
3283 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
3284 					struct vmcs12 *vmcs12,
3285 					enum vm_entry_failure_code *entry_failure_code)
3286 {
3287 	bool ia32e = !!(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE);
3288 
3289 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
3290 
3291 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3292 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3293 		return -EINVAL;
3294 
3295 	if (CC(vmcs12->guest_cr4 & X86_CR4_CET && !(vmcs12->guest_cr0 & X86_CR0_WP)))
3296 		return -EINVAL;
3297 
3298 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
3299 	    (CC(!kvm_dr7_valid(vmcs12->guest_dr7)) ||
3300 	     CC(!vmx_is_valid_debugctl(vcpu, vmcs12->guest_ia32_debugctl, false))))
3301 		return -EINVAL;
3302 
3303 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3304 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3305 		return -EINVAL;
3306 
3307 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3308 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3309 		return -EINVAL;
3310 	}
3311 
3312 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3313 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3314 					   vmcs12->guest_ia32_perf_global_ctrl)))
3315 		return -EINVAL;
3316 
3317 	if (CC((vmcs12->guest_cr0 & (X86_CR0_PG | X86_CR0_PE)) == X86_CR0_PG))
3318 		return -EINVAL;
3319 
3320 	if (CC(ia32e && !(vmcs12->guest_cr4 & X86_CR4_PAE)) ||
3321 	    CC(ia32e && !(vmcs12->guest_cr0 & X86_CR0_PG)))
3322 		return -EINVAL;
3323 
3324 	/*
3325 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3326 	 * are performed on the field for the IA32_EFER MSR:
3327 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3328 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3329 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3330 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3331 	 *   CR0.PG) is 1.
3332 	 */
3333 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3334 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3335 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3336 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3337 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3338 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3339 			return -EINVAL;
3340 	}
3341 
3342 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3343 	    (CC(is_noncanonical_msr_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3344 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3345 		return -EINVAL;
3346 
3347 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE) {
3348 		if (nested_vmx_check_cet_state_common(vcpu, vmcs12->guest_s_cet,
3349 						      vmcs12->guest_ssp,
3350 						      vmcs12->guest_ssp_tbl))
3351 			return -EINVAL;
3352 
3353 		/*
3354 		 * Guest SSP must have 63:N bits identical, rather than
3355 		 * be canonical (i.e., 63:N-1 bits identical), where N is
3356 		 * the CPU's maximum linear-address width. Similar to
3357 		 * is_noncanonical_msr_address(), use the host's
3358 		 * linear-address width.
3359 		 */
3360 		if (CC(!__is_canonical_address(vmcs12->guest_ssp, max_host_virt_addr_bits() + 1)))
3361 			return -EINVAL;
3362 	}
3363 
3364 	if (nested_check_guest_non_reg_state(vmcs12))
3365 		return -EINVAL;
3366 
3367 	return 0;
3368 }
3369 
3370 #ifdef CONFIG_KVM_HYPERV
3371 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3372 {
3373 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3374 
3375 	/*
3376 	 * hv_evmcs may end up being not mapped after migration (when
3377 	 * L2 was running), map it here to make sure vmcs12 changes are
3378 	 * properly reflected.
3379 	 */
3380 	if (guest_cpu_cap_has_evmcs(vcpu) &&
3381 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3382 		enum nested_evmptrld_status evmptrld_status =
3383 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3384 
3385 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3386 		    evmptrld_status == EVMPTRLD_ERROR)
3387 			return false;
3388 
3389 		/*
3390 		 * Post migration VMCS12 always provides the most actual
3391 		 * information, copy it to eVMCS upon entry.
3392 		 */
3393 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3394 	}
3395 
3396 	return true;
3397 }
3398 #endif
3399 
3400 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3401 {
3402 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3403 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3404 	struct kvm_host_map *map;
3405 
3406 	if (!vcpu->arch.pdptrs_from_userspace &&
3407 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3408 		/*
3409 		 * Reload the guest's PDPTRs since after a migration
3410 		 * the guest CR3 might be restored prior to setting the nested
3411 		 * state which can lead to a load of wrong PDPTRs.
3412 		 */
3413 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3414 			return false;
3415 	}
3416 
3417 
3418 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3419 		map = &vmx->nested.apic_access_page_map;
3420 
3421 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->apic_access_addr), map)) {
3422 			vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(map->pfn));
3423 		} else {
3424 			pr_debug_ratelimited("%s: no backing for APIC-access address in vmcs12\n",
3425 					     __func__);
3426 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3427 			vcpu->run->internal.suberror =
3428 				KVM_INTERNAL_ERROR_EMULATION;
3429 			vcpu->run->internal.ndata = 0;
3430 			return false;
3431 		}
3432 	}
3433 
3434 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3435 		map = &vmx->nested.virtual_apic_map;
3436 
3437 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3438 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3439 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3440 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3441 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3442 			/*
3443 			 * The processor will never use the TPR shadow, simply
3444 			 * clear the bit from the execution control.  Such a
3445 			 * configuration is useless, but it happens in tests.
3446 			 * For any other configuration, failing the vm entry is
3447 			 * _not_ what the processor does but it's basically the
3448 			 * only possibility we have.
3449 			 */
3450 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3451 		} else {
3452 			/*
3453 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3454 			 * force VM-Entry to fail.
3455 			 */
3456 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3457 		}
3458 	}
3459 
3460 	if (nested_cpu_has_posted_intr(vmcs12)) {
3461 		map = &vmx->nested.pi_desc_map;
3462 
3463 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3464 			vmx->nested.pi_desc =
3465 				(struct pi_desc *)(((void *)map->hva) +
3466 				offset_in_page(vmcs12->posted_intr_desc_addr));
3467 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3468 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3469 		} else {
3470 			/*
3471 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3472 			 * access the contents of the VMCS12 posted interrupt
3473 			 * descriptor. (Note that KVM may do this when it
3474 			 * should not, per the architectural specification.)
3475 			 */
3476 			vmx->nested.pi_desc = NULL;
3477 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3478 		}
3479 	}
3480 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3481 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3482 	else
3483 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3484 
3485 	return true;
3486 }
3487 
3488 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3489 {
3490 #ifdef CONFIG_KVM_HYPERV
3491 	/*
3492 	 * Note: nested_get_evmcs_page() also updates 'vp_assist_page' copy
3493 	 * in 'struct kvm_vcpu_hv' in case eVMCS is in use, this is mandatory
3494 	 * to make nested_evmcs_l2_tlb_flush_enabled() work correctly post
3495 	 * migration.
3496 	 */
3497 	if (!nested_get_evmcs_page(vcpu)) {
3498 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3499 				     __func__);
3500 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3501 		vcpu->run->internal.suberror =
3502 			KVM_INTERNAL_ERROR_EMULATION;
3503 		vcpu->run->internal.ndata = 0;
3504 
3505 		return false;
3506 	}
3507 #endif
3508 
3509 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3510 		return false;
3511 
3512 	return true;
3513 }
3514 
3515 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3516 {
3517 	struct vmcs12 *vmcs12;
3518 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3519 	gpa_t dst;
3520 
3521 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3522 		return 0;
3523 
3524 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3525 		return 1;
3526 
3527 	/*
3528 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3529 	 * set is already checked as part of A/D emulation.
3530 	 */
3531 	vmcs12 = get_vmcs12(vcpu);
3532 	if (!nested_cpu_has_pml(vmcs12))
3533 		return 0;
3534 
3535 	if (vmcs12->guest_pml_index >= PML_LOG_NR_ENTRIES) {
3536 		vmx->nested.pml_full = true;
3537 		return 1;
3538 	}
3539 
3540 	gpa &= ~0xFFFull;
3541 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3542 
3543 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3544 				 offset_in_page(dst), sizeof(gpa)))
3545 		return 0;
3546 
3547 	vmcs12->guest_pml_index--;
3548 
3549 	return 0;
3550 }
3551 
3552 /*
3553  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3554  * for running VMX instructions (except VMXON, whose prerequisites are
3555  * slightly different). It also specifies what exception to inject otherwise.
3556  * Note that many of these exceptions have priority over VM exits, so they
3557  * don't have to be checked again here.
3558  */
3559 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3560 {
3561 	if (!to_vmx(vcpu)->nested.vmxon) {
3562 		kvm_queue_exception(vcpu, UD_VECTOR);
3563 		return 0;
3564 	}
3565 
3566 	if (vmx_get_cpl(vcpu)) {
3567 		kvm_inject_gp(vcpu, 0);
3568 		return 0;
3569 	}
3570 
3571 	return 1;
3572 }
3573 
3574 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3575 				   struct vmcs12 *vmcs12);
3576 
3577 /*
3578  * If from_vmentry is false, this is being called from state restore (either RSM
3579  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3580  *
3581  * Returns:
3582  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3583  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3584  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3585  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3586  */
3587 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3588 							bool from_vmentry)
3589 {
3590 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3591 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3592 	enum vm_entry_failure_code entry_failure_code;
3593 	union vmx_exit_reason exit_reason = {
3594 		.basic = EXIT_REASON_INVALID_STATE,
3595 		.failed_vmentry = 1,
3596 	};
3597 	u32 failed_index;
3598 
3599 	trace_kvm_nested_vmenter(kvm_rip_read(vcpu),
3600 				 vmx->nested.current_vmptr,
3601 				 vmcs12->guest_rip,
3602 				 vmcs12->guest_intr_status,
3603 				 vmcs12->vm_entry_intr_info_field,
3604 				 vmcs12->secondary_vm_exec_control & SECONDARY_EXEC_ENABLE_EPT,
3605 				 vmcs12->ept_pointer,
3606 				 vmcs12->guest_cr3,
3607 				 KVM_ISA_VMX);
3608 
3609 	kvm_service_local_tlb_flush_requests(vcpu);
3610 
3611 	if (!vmx->nested.nested_run_pending ||
3612 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3613 		vmx->nested.pre_vmenter_debugctl = vmx_guest_debugctl_read();
3614 	if (kvm_mpx_supported() &&
3615 	    (!vmx->nested.nested_run_pending ||
3616 	     !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
3617 		vmx->nested.pre_vmenter_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3618 
3619 	if (!vmx->nested.nested_run_pending ||
3620 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE))
3621 		vmcs_read_cet_state(vcpu, &vmx->nested.pre_vmenter_s_cet,
3622 				    &vmx->nested.pre_vmenter_ssp,
3623 				    &vmx->nested.pre_vmenter_ssp_tbl);
3624 
3625 	/*
3626 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled.  In the
3627 	 * event of a "late" VM-Fail, i.e. a VM-Fail detected by hardware but
3628 	 * not KVM, KVM must unwind its software model to the pre-VM-Entry host
3629 	 * state.  When EPT is disabled, GUEST_CR3 holds KVM's shadow CR3, not
3630 	 * L1's "real" CR3, which causes nested_vmx_restore_host_state() to
3631 	 * corrupt vcpu->arch.cr3.  Stuffing vmcs01.GUEST_CR3 results in the
3632 	 * unwind naturally setting arch.cr3 to the correct value.  Smashing
3633 	 * vmcs01.GUEST_CR3 is safe because nested VM-Exits, and the unwind,
3634 	 * reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is guaranteed to be
3635 	 * overwritten with a shadow CR3 prior to re-entering L1.
3636 	 */
3637 	if (!enable_ept)
3638 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3639 
3640 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3641 
3642 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3643 
3644 	if (from_vmentry) {
3645 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3646 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3647 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3648 		}
3649 
3650 		if (nested_vmx_check_controls_late(vcpu, vmcs12)) {
3651 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3652 			return NVMX_VMENTRY_VMFAIL;
3653 		}
3654 
3655 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3656 						 &entry_failure_code)) {
3657 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3658 			vmcs12->exit_qualification = entry_failure_code;
3659 			goto vmentry_fail_vmexit;
3660 		}
3661 	}
3662 
3663 	enter_guest_mode(vcpu);
3664 
3665 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3666 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3667 		vmcs12->exit_qualification = entry_failure_code;
3668 		goto vmentry_fail_vmexit_guest_mode;
3669 	}
3670 
3671 	if (from_vmentry) {
3672 		failed_index = nested_vmx_load_msr(vcpu,
3673 						   vmcs12->vm_entry_msr_load_addr,
3674 						   vmcs12->vm_entry_msr_load_count);
3675 		if (failed_index) {
3676 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3677 			vmcs12->exit_qualification = failed_index;
3678 			goto vmentry_fail_vmexit_guest_mode;
3679 		}
3680 	} else {
3681 		/*
3682 		 * The MMU is not initialized to point at the right entities yet and
3683 		 * "get pages" would need to read data from the guest (i.e. we will
3684 		 * need to perform gpa to hpa translation). Request a call
3685 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3686 		 * have already been set at vmentry time and should not be reset.
3687 		 */
3688 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3689 	}
3690 
3691 	/*
3692 	 * Re-evaluate pending events if L1 had a pending IRQ/NMI/INIT/SIPI
3693 	 * when it executed VMLAUNCH/VMRESUME, as entering non-root mode can
3694 	 * effectively unblock various events, e.g. INIT/SIPI cause VM-Exit
3695 	 * unconditionally.  Take care to pull data from vmcs01 as appropriate,
3696 	 * e.g. when checking for interrupt windows, as vmcs02 is now loaded.
3697 	 */
3698 	if ((__exec_controls_get(&vmx->vmcs01) & (CPU_BASED_INTR_WINDOW_EXITING |
3699 						  CPU_BASED_NMI_WINDOW_EXITING)) ||
3700 	    kvm_apic_has_pending_init_or_sipi(vcpu) ||
3701 	    kvm_apic_has_interrupt(vcpu))
3702 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3703 
3704 	/*
3705 	 * Do not start the preemption timer hrtimer until after we know
3706 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3707 	 * the timer.
3708 	 */
3709 	vmx->nested.preemption_timer_expired = false;
3710 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3711 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3712 		vmx_start_preemption_timer(vcpu, timer_value);
3713 	}
3714 
3715 	/*
3716 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3717 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3718 	 * returned as far as L1 is concerned. It will only return (and set
3719 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3720 	 */
3721 	return NVMX_VMENTRY_SUCCESS;
3722 
3723 	/*
3724 	 * A failed consistency check that leads to a VMExit during L1's
3725 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3726 	 * 26.7 "VM-entry failures during or after loading guest state".
3727 	 */
3728 vmentry_fail_vmexit_guest_mode:
3729 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3730 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3731 	leave_guest_mode(vcpu);
3732 
3733 vmentry_fail_vmexit:
3734 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3735 
3736 	if (!from_vmentry)
3737 		return NVMX_VMENTRY_VMEXIT;
3738 
3739 	load_vmcs12_host_state(vcpu, vmcs12);
3740 	vmcs12->vm_exit_reason = exit_reason.full;
3741 	if (enable_shadow_vmcs || nested_vmx_is_evmptr12_valid(vmx))
3742 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3743 	return NVMX_VMENTRY_VMEXIT;
3744 }
3745 
3746 /*
3747  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3748  * for running an L2 nested guest.
3749  */
3750 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3751 {
3752 	struct vmcs12 *vmcs12;
3753 	enum nvmx_vmentry_status status;
3754 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3755 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3756 	enum nested_evmptrld_status evmptrld_status;
3757 
3758 	if (!nested_vmx_check_permission(vcpu))
3759 		return 1;
3760 
3761 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3762 	if (evmptrld_status == EVMPTRLD_ERROR) {
3763 		kvm_queue_exception(vcpu, UD_VECTOR);
3764 		return 1;
3765 	}
3766 
3767 	kvm_pmu_branch_retired(vcpu);
3768 
3769 	if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3770 		return nested_vmx_failInvalid(vcpu);
3771 
3772 	if (CC(!nested_vmx_is_evmptr12_valid(vmx) &&
3773 	       vmx->nested.current_vmptr == INVALID_GPA))
3774 		return nested_vmx_failInvalid(vcpu);
3775 
3776 	vmcs12 = get_vmcs12(vcpu);
3777 
3778 	/*
3779 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3780 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3781 	 * rather than RFLAGS.ZF, and no error number is stored to the
3782 	 * VM-instruction error field.
3783 	 */
3784 	if (CC(vmcs12->hdr.shadow_vmcs))
3785 		return nested_vmx_failInvalid(vcpu);
3786 
3787 	if (nested_vmx_is_evmptr12_valid(vmx)) {
3788 		struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
3789 
3790 		copy_enlightened_to_vmcs12(vmx, evmcs->hv_clean_fields);
3791 		/* Enlightened VMCS doesn't have launch state */
3792 		vmcs12->launch_state = !launch;
3793 	} else if (enable_shadow_vmcs) {
3794 		copy_shadow_to_vmcs12(vmx);
3795 	}
3796 
3797 	/*
3798 	 * The nested entry process starts with enforcing various prerequisites
3799 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3800 	 * they fail: As the SDM explains, some conditions should cause the
3801 	 * instruction to fail, while others will cause the instruction to seem
3802 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3803 	 * To speed up the normal (success) code path, we should avoid checking
3804 	 * for misconfigurations which will anyway be caught by the processor
3805 	 * when using the merged vmcs02.
3806 	 */
3807 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3808 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3809 
3810 	if (CC(vmcs12->launch_state == launch))
3811 		return nested_vmx_fail(vcpu,
3812 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3813 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3814 
3815 	if (nested_vmx_check_controls(vcpu, vmcs12))
3816 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3817 
3818 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3819 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3820 
3821 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3822 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3823 
3824 	/*
3825 	 * We're finally done with prerequisite checking, and can start with
3826 	 * the nested entry.
3827 	 */
3828 	vmx->nested.nested_run_pending = 1;
3829 	vmx->nested.has_preemption_timer_deadline = false;
3830 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3831 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3832 		goto vmentry_failed;
3833 
3834 	/* Hide L1D cache contents from the nested guest.  */
3835 	kvm_request_l1tf_flush_l1d();
3836 
3837 	/*
3838 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3839 	 * also be used as part of restoring nVMX state for
3840 	 * snapshot restore (migration).
3841 	 *
3842 	 * In this flow, it is assumed that vmcs12 cache was
3843 	 * transferred as part of captured nVMX state and should
3844 	 * therefore not be read from guest memory (which may not
3845 	 * exist on destination host yet).
3846 	 */
3847 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3848 
3849 	switch (vmcs12->guest_activity_state) {
3850 	case GUEST_ACTIVITY_HLT:
3851 		/*
3852 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3853 		 * awakened by event injection or by an NMI-window VM-exit or
3854 		 * by an interrupt-window VM-exit, halt the vcpu.
3855 		 */
3856 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3857 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3858 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3859 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3860 			vmx->nested.nested_run_pending = 0;
3861 			return kvm_emulate_halt_noskip(vcpu);
3862 		}
3863 		break;
3864 	case GUEST_ACTIVITY_WAIT_SIPI:
3865 		vmx->nested.nested_run_pending = 0;
3866 		kvm_set_mp_state(vcpu, KVM_MP_STATE_INIT_RECEIVED);
3867 		break;
3868 	default:
3869 		break;
3870 	}
3871 
3872 	return 1;
3873 
3874 vmentry_failed:
3875 	vmx->nested.nested_run_pending = 0;
3876 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3877 		return 0;
3878 	if (status == NVMX_VMENTRY_VMEXIT)
3879 		return 1;
3880 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3881 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3882 }
3883 
3884 /*
3885  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3886  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3887  * This function returns the new value we should put in vmcs12.guest_cr0.
3888  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3889  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3890  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3891  *     didn't trap the bit, because if L1 did, so would L0).
3892  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3893  *     been modified by L2, and L1 knows it. So just leave the old value of
3894  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3895  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3896  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3897  *     changed these bits, and therefore they need to be updated, but L0
3898  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3899  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3900  */
3901 static inline unsigned long
3902 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3903 {
3904 	return
3905 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3906 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3907 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3908 			vcpu->arch.cr0_guest_owned_bits));
3909 }
3910 
3911 static inline unsigned long
3912 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3913 {
3914 	return
3915 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3916 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3917 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3918 			vcpu->arch.cr4_guest_owned_bits));
3919 }
3920 
3921 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3922 				      struct vmcs12 *vmcs12,
3923 				      u32 vm_exit_reason, u32 exit_intr_info)
3924 {
3925 	u32 idt_vectoring;
3926 	unsigned int nr;
3927 
3928 	/*
3929 	 * Per the SDM, VM-Exits due to double and triple faults are never
3930 	 * considered to occur during event delivery, even if the double/triple
3931 	 * fault is the result of an escalating vectoring issue.
3932 	 *
3933 	 * Note, the SDM qualifies the double fault behavior with "The original
3934 	 * event results in a double-fault exception".  It's unclear why the
3935 	 * qualification exists since exits due to double fault can occur only
3936 	 * while vectoring a different exception (injected events are never
3937 	 * subject to interception), i.e. there's _always_ an original event.
3938 	 *
3939 	 * The SDM also uses NMI as a confusing example for the "original event
3940 	 * causes the VM exit directly" clause.  NMI isn't special in any way,
3941 	 * the same rule applies to all events that cause an exit directly.
3942 	 * NMI is an odd choice for the example because NMIs can only occur on
3943 	 * instruction boundaries, i.e. they _can't_ occur during vectoring.
3944 	 */
3945 	if ((u16)vm_exit_reason == EXIT_REASON_TRIPLE_FAULT ||
3946 	    ((u16)vm_exit_reason == EXIT_REASON_EXCEPTION_NMI &&
3947 	     is_double_fault(exit_intr_info))) {
3948 		vmcs12->idt_vectoring_info_field = 0;
3949 	} else if (vcpu->arch.exception.injected) {
3950 		nr = vcpu->arch.exception.vector;
3951 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3952 
3953 		if (kvm_exception_is_soft(nr)) {
3954 			vmcs12->vm_exit_instruction_len =
3955 				vcpu->arch.event_exit_inst_len;
3956 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3957 		} else
3958 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3959 
3960 		if (vcpu->arch.exception.has_error_code) {
3961 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3962 			vmcs12->idt_vectoring_error_code =
3963 				vcpu->arch.exception.error_code;
3964 		}
3965 
3966 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3967 	} else if (vcpu->arch.nmi_injected) {
3968 		vmcs12->idt_vectoring_info_field =
3969 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3970 	} else if (vcpu->arch.interrupt.injected) {
3971 		nr = vcpu->arch.interrupt.nr;
3972 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3973 
3974 		if (vcpu->arch.interrupt.soft) {
3975 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3976 			vmcs12->vm_entry_instruction_len =
3977 				vcpu->arch.event_exit_inst_len;
3978 		} else
3979 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3980 
3981 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3982 	} else {
3983 		vmcs12->idt_vectoring_info_field = 0;
3984 	}
3985 }
3986 
3987 
3988 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3989 {
3990 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3991 	gfn_t gfn;
3992 
3993 	/*
3994 	 * Don't need to mark the APIC access page dirty; it is never
3995 	 * written to by the CPU during APIC virtualization.
3996 	 */
3997 
3998 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3999 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
4000 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
4001 	}
4002 
4003 	if (nested_cpu_has_posted_intr(vmcs12)) {
4004 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
4005 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
4006 	}
4007 }
4008 
4009 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4010 {
4011 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4012 	int max_irr;
4013 	void *vapic_page;
4014 	u16 status;
4015 
4016 	if (!vmx->nested.pi_pending)
4017 		return 0;
4018 
4019 	if (!vmx->nested.pi_desc)
4020 		goto mmio_needed;
4021 
4022 	vmx->nested.pi_pending = false;
4023 
4024 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4025 		return 0;
4026 
4027 	max_irr = pi_find_highest_vector(vmx->nested.pi_desc);
4028 	if (max_irr > 0) {
4029 		vapic_page = vmx->nested.virtual_apic_map.hva;
4030 		if (!vapic_page)
4031 			goto mmio_needed;
4032 
4033 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
4034 			vapic_page, &max_irr);
4035 		status = vmcs_read16(GUEST_INTR_STATUS);
4036 		if ((u8)max_irr > ((u8)status & 0xff)) {
4037 			status &= ~0xff;
4038 			status |= (u8)max_irr;
4039 			vmcs_write16(GUEST_INTR_STATUS, status);
4040 		}
4041 	}
4042 
4043 	nested_mark_vmcs12_pages_dirty(vcpu);
4044 	return 0;
4045 
4046 mmio_needed:
4047 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
4048 	return -ENXIO;
4049 }
4050 
4051 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu)
4052 {
4053 	struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
4054 	u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
4055 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4056 	unsigned long exit_qual;
4057 
4058 	if (ex->has_payload) {
4059 		exit_qual = ex->payload;
4060 	} else if (ex->vector == PF_VECTOR) {
4061 		exit_qual = vcpu->arch.cr2;
4062 	} else if (ex->vector == DB_VECTOR) {
4063 		exit_qual = vcpu->arch.dr6;
4064 		exit_qual &= ~DR6_BT;
4065 		exit_qual ^= DR6_ACTIVE_LOW;
4066 	} else {
4067 		exit_qual = 0;
4068 	}
4069 
4070 	/*
4071 	 * Unlike AMD's Paged Real Mode, which reports an error code on #PF
4072 	 * VM-Exits even if the CPU is in Real Mode, Intel VMX never sets the
4073 	 * "has error code" flags on VM-Exit if the CPU is in Real Mode.
4074 	 */
4075 	if (ex->has_error_code && is_protmode(vcpu)) {
4076 		/*
4077 		 * Intel CPUs do not generate error codes with bits 31:16 set,
4078 		 * and more importantly VMX disallows setting bits 31:16 in the
4079 		 * injected error code for VM-Entry.  Drop the bits to mimic
4080 		 * hardware and avoid inducing failure on nested VM-Entry if L1
4081 		 * chooses to inject the exception back to L2.  AMD CPUs _do_
4082 		 * generate "full" 32-bit error codes, so KVM allows userspace
4083 		 * to inject exception error codes with bits 31:16 set.
4084 		 */
4085 		vmcs12->vm_exit_intr_error_code = (u16)ex->error_code;
4086 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
4087 	}
4088 
4089 	if (kvm_exception_is_soft(ex->vector))
4090 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
4091 	else
4092 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
4093 
4094 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
4095 	    vmx_get_nmi_mask(vcpu))
4096 		intr_info |= INTR_INFO_UNBLOCK_NMI;
4097 
4098 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
4099 }
4100 
4101 /*
4102  * Returns true if a debug trap is (likely) pending delivery.  Infer the class
4103  * of a #DB (trap-like vs. fault-like) from the exception payload (to-be-DR6).
4104  * Using the payload is flawed because code breakpoints (fault-like) and data
4105  * breakpoints (trap-like) set the same bits in DR6 (breakpoint detected), i.e.
4106  * this will return false positives if a to-be-injected code breakpoint #DB is
4107  * pending (from KVM's perspective, but not "pending" across an instruction
4108  * boundary).  ICEBP, a.k.a. INT1, is also not reflected here even though it
4109  * too is trap-like.
4110  *
4111  * KVM "works" despite these flaws as ICEBP isn't currently supported by the
4112  * emulator, Monitor Trap Flag is not marked pending on intercepted #DBs (the
4113  * #DB has already happened), and MTF isn't marked pending on code breakpoints
4114  * from the emulator (because such #DBs are fault-like and thus don't trigger
4115  * actions that fire on instruction retire).
4116  */
4117 static unsigned long vmx_get_pending_dbg_trap(struct kvm_queued_exception *ex)
4118 {
4119 	if (!ex->pending || ex->vector != DB_VECTOR)
4120 		return 0;
4121 
4122 	/* General Detect #DBs are always fault-like. */
4123 	return ex->payload & ~DR6_BD;
4124 }
4125 
4126 /*
4127  * Returns true if there's a pending #DB exception that is lower priority than
4128  * a pending Monitor Trap Flag VM-Exit.  TSS T-flag #DBs are not emulated by
4129  * KVM, but could theoretically be injected by userspace.  Note, this code is
4130  * imperfect, see above.
4131  */
4132 static bool vmx_is_low_priority_db_trap(struct kvm_queued_exception *ex)
4133 {
4134 	return vmx_get_pending_dbg_trap(ex) & ~DR6_BT;
4135 }
4136 
4137 /*
4138  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
4139  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
4140  * represents these debug traps with a payload that is said to be compatible
4141  * with the 'pending debug exceptions' field, write the payload to the VMCS
4142  * field if a VM-exit is delivered before the debug trap.
4143  */
4144 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
4145 {
4146 	unsigned long pending_dbg;
4147 
4148 	pending_dbg = vmx_get_pending_dbg_trap(&vcpu->arch.exception);
4149 	if (pending_dbg)
4150 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, pending_dbg);
4151 }
4152 
4153 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
4154 {
4155 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
4156 	       to_vmx(vcpu)->nested.preemption_timer_expired;
4157 }
4158 
4159 static bool vmx_has_nested_events(struct kvm_vcpu *vcpu, bool for_injection)
4160 {
4161 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4162 	void *vapic = vmx->nested.virtual_apic_map.hva;
4163 	int max_irr, vppr;
4164 
4165 	if (nested_vmx_preemption_timer_pending(vcpu) ||
4166 	    vmx->nested.mtf_pending)
4167 		return true;
4168 
4169 	/*
4170 	 * Virtual Interrupt Delivery doesn't require manual injection.  Either
4171 	 * the interrupt is already in GUEST_RVI and will be recognized by CPU
4172 	 * at VM-Entry, or there is a KVM_REQ_EVENT pending and KVM will move
4173 	 * the interrupt from the PIR to RVI prior to entering the guest.
4174 	 */
4175 	if (for_injection)
4176 		return false;
4177 
4178 	if (!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
4179 	    __vmx_interrupt_blocked(vcpu))
4180 		return false;
4181 
4182 	if (!vapic)
4183 		return false;
4184 
4185 	vppr = *((u32 *)(vapic + APIC_PROCPRI));
4186 
4187 	max_irr = vmx_get_rvi();
4188 	if ((max_irr & 0xf0) > (vppr & 0xf0))
4189 		return true;
4190 
4191 	if (vmx->nested.pi_pending && vmx->nested.pi_desc &&
4192 	    pi_test_on(vmx->nested.pi_desc)) {
4193 		max_irr = pi_find_highest_vector(vmx->nested.pi_desc);
4194 		if (max_irr > 0 && (max_irr & 0xf0) > (vppr & 0xf0))
4195 			return true;
4196 	}
4197 
4198 	return false;
4199 }
4200 
4201 /*
4202  * Per the Intel SDM's table "Priority Among Concurrent Events", with minor
4203  * edits to fill in missing examples, e.g. #DB due to split-lock accesses,
4204  * and less minor edits to splice in the priority of VMX Non-Root specific
4205  * events, e.g. MTF and NMI/INTR-window exiting.
4206  *
4207  * 1 Hardware Reset and Machine Checks
4208  *	- RESET
4209  *	- Machine Check
4210  *
4211  * 2 Trap on Task Switch
4212  *	- T flag in TSS is set (on task switch)
4213  *
4214  * 3 External Hardware Interventions
4215  *	- FLUSH
4216  *	- STOPCLK
4217  *	- SMI
4218  *	- INIT
4219  *
4220  * 3.5 Monitor Trap Flag (MTF) VM-exit[1]
4221  *
4222  * 4 Traps on Previous Instruction
4223  *	- Breakpoints
4224  *	- Trap-class Debug Exceptions (#DB due to TF flag set, data/I-O
4225  *	  breakpoint, or #DB due to a split-lock access)
4226  *
4227  * 4.3	VMX-preemption timer expired VM-exit
4228  *
4229  * 4.6	NMI-window exiting VM-exit[2]
4230  *
4231  * 5 Nonmaskable Interrupts (NMI)
4232  *
4233  * 5.5 Interrupt-window exiting VM-exit and Virtual-interrupt delivery
4234  *
4235  * 6 Maskable Hardware Interrupts
4236  *
4237  * 7 Code Breakpoint Fault
4238  *
4239  * 8 Faults from Fetching Next Instruction
4240  *	- Code-Segment Limit Violation
4241  *	- Code Page Fault
4242  *	- Control protection exception (missing ENDBRANCH at target of indirect
4243  *					call or jump)
4244  *
4245  * 9 Faults from Decoding Next Instruction
4246  *	- Instruction length > 15 bytes
4247  *	- Invalid Opcode
4248  *	- Coprocessor Not Available
4249  *
4250  *10 Faults on Executing Instruction
4251  *	- Overflow
4252  *	- Bound error
4253  *	- Invalid TSS
4254  *	- Segment Not Present
4255  *	- Stack fault
4256  *	- General Protection
4257  *	- Data Page Fault
4258  *	- Alignment Check
4259  *	- x86 FPU Floating-point exception
4260  *	- SIMD floating-point exception
4261  *	- Virtualization exception
4262  *	- Control protection exception
4263  *
4264  * [1] Per the "Monitor Trap Flag" section: System-management interrupts (SMIs),
4265  *     INIT signals, and higher priority events take priority over MTF VM exits.
4266  *     MTF VM exits take priority over debug-trap exceptions and lower priority
4267  *     events.
4268  *
4269  * [2] Debug-trap exceptions and higher priority events take priority over VM exits
4270  *     caused by the VMX-preemption timer.  VM exits caused by the VMX-preemption
4271  *     timer take priority over VM exits caused by the "NMI-window exiting"
4272  *     VM-execution control and lower priority events.
4273  *
4274  * [3] Debug-trap exceptions and higher priority events take priority over VM exits
4275  *     caused by "NMI-window exiting".  VM exits caused by this control take
4276  *     priority over non-maskable interrupts (NMIs) and lower priority events.
4277  *
4278  * [4] Virtual-interrupt delivery has the same priority as that of VM exits due to
4279  *     the 1-setting of the "interrupt-window exiting" VM-execution control.  Thus,
4280  *     non-maskable interrupts (NMIs) and higher priority events take priority over
4281  *     delivery of a virtual interrupt; delivery of a virtual interrupt takes
4282  *     priority over external interrupts and lower priority events.
4283  */
4284 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
4285 {
4286 	struct kvm_lapic *apic = vcpu->arch.apic;
4287 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4288 	/*
4289 	 * Only a pending nested run blocks a pending exception.  If there is a
4290 	 * previously injected event, the pending exception occurred while said
4291 	 * event was being delivered and thus needs to be handled.
4292 	 */
4293 	bool block_nested_exceptions = vmx->nested.nested_run_pending;
4294 	/*
4295 	 * Events that don't require injection, i.e. that are virtualized by
4296 	 * hardware, aren't blocked by a pending VM-Enter as KVM doesn't need
4297 	 * to regain control in order to deliver the event, and hardware will
4298 	 * handle event ordering, e.g. with respect to injected exceptions.
4299 	 *
4300 	 * But, new events (not exceptions) are only recognized at instruction
4301 	 * boundaries.  If an event needs reinjection, then KVM is handling a
4302 	 * VM-Exit that occurred _during_ instruction execution; new events,
4303 	 * irrespective of whether or not they're injected, are blocked until
4304 	 * the instruction completes.
4305 	 */
4306 	bool block_non_injected_events = kvm_event_needs_reinjection(vcpu);
4307 	/*
4308 	 * Inject events are blocked by nested VM-Enter, as KVM is responsible
4309 	 * for managing priority between concurrent events, i.e. KVM needs to
4310 	 * wait until after VM-Enter completes to deliver injected events.
4311 	 */
4312 	bool block_nested_events = block_nested_exceptions ||
4313 				   block_non_injected_events;
4314 
4315 	if (lapic_in_kernel(vcpu) &&
4316 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
4317 		if (block_nested_events)
4318 			return -EBUSY;
4319 		nested_vmx_update_pending_dbg(vcpu);
4320 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
4321 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
4322 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
4323 
4324 		/* MTF is discarded if the vCPU is in WFS. */
4325 		vmx->nested.mtf_pending = false;
4326 		return 0;
4327 	}
4328 
4329 	if (lapic_in_kernel(vcpu) &&
4330 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
4331 		if (block_nested_events)
4332 			return -EBUSY;
4333 
4334 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
4335 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
4336 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
4337 						apic->sipi_vector & 0xFFUL);
4338 			return 0;
4339 		}
4340 		/* Fallthrough, the SIPI is completely ignored. */
4341 	}
4342 
4343 	/*
4344 	 * Process exceptions that are higher priority than Monitor Trap Flag:
4345 	 * fault-like exceptions, TSS T flag #DB (not emulated by KVM, but
4346 	 * could theoretically come in from userspace), and ICEBP (INT1).
4347 	 *
4348 	 * TODO: SMIs have higher priority than MTF and trap-like #DBs (except
4349 	 * for TSS T flag #DBs).  KVM also doesn't save/restore pending MTF
4350 	 * across SMI/RSM as it should; that needs to be addressed in order to
4351 	 * prioritize SMI over MTF and trap-like #DBs.
4352 	 */
4353 	if (vcpu->arch.exception_vmexit.pending &&
4354 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception_vmexit)) {
4355 		if (block_nested_exceptions)
4356 			return -EBUSY;
4357 
4358 		nested_vmx_inject_exception_vmexit(vcpu);
4359 		return 0;
4360 	}
4361 
4362 	if (vcpu->arch.exception.pending &&
4363 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception)) {
4364 		if (block_nested_exceptions)
4365 			return -EBUSY;
4366 		goto no_vmexit;
4367 	}
4368 
4369 	if (vmx->nested.mtf_pending) {
4370 		if (block_nested_events)
4371 			return -EBUSY;
4372 		nested_vmx_update_pending_dbg(vcpu);
4373 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
4374 		return 0;
4375 	}
4376 
4377 	if (vcpu->arch.exception_vmexit.pending) {
4378 		if (block_nested_exceptions)
4379 			return -EBUSY;
4380 
4381 		nested_vmx_inject_exception_vmexit(vcpu);
4382 		return 0;
4383 	}
4384 
4385 	if (vcpu->arch.exception.pending) {
4386 		if (block_nested_exceptions)
4387 			return -EBUSY;
4388 		goto no_vmexit;
4389 	}
4390 
4391 	if (nested_vmx_preemption_timer_pending(vcpu)) {
4392 		if (block_nested_events)
4393 			return -EBUSY;
4394 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
4395 		return 0;
4396 	}
4397 
4398 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
4399 		if (block_nested_events)
4400 			return -EBUSY;
4401 		goto no_vmexit;
4402 	}
4403 
4404 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
4405 		if (block_nested_events)
4406 			return -EBUSY;
4407 		if (!nested_exit_on_nmi(vcpu))
4408 			goto no_vmexit;
4409 
4410 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
4411 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
4412 				  INTR_INFO_VALID_MASK, 0);
4413 		/*
4414 		 * The NMI-triggered VM exit counts as injection:
4415 		 * clear this one and block further NMIs.
4416 		 */
4417 		vcpu->arch.nmi_pending = 0;
4418 		vmx_set_nmi_mask(vcpu, true);
4419 		return 0;
4420 	}
4421 
4422 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
4423 		int irq;
4424 
4425 		if (!nested_exit_on_intr(vcpu)) {
4426 			if (block_nested_events)
4427 				return -EBUSY;
4428 
4429 			goto no_vmexit;
4430 		}
4431 
4432 		if (!nested_exit_intr_ack_set(vcpu)) {
4433 			if (block_nested_events)
4434 				return -EBUSY;
4435 
4436 			nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
4437 			return 0;
4438 		}
4439 
4440 		irq = kvm_cpu_get_extint(vcpu);
4441 		if (irq != -1) {
4442 			if (block_nested_events)
4443 				return -EBUSY;
4444 
4445 			nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT,
4446 					  INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR | irq, 0);
4447 			return 0;
4448 		}
4449 
4450 		irq = kvm_apic_has_interrupt(vcpu);
4451 		if (WARN_ON_ONCE(irq < 0))
4452 			goto no_vmexit;
4453 
4454 		/*
4455 		 * If the IRQ is L2's PI notification vector, process posted
4456 		 * interrupts for L2 instead of injecting VM-Exit, as the
4457 		 * detection/morphing architecturally occurs when the IRQ is
4458 		 * delivered to the CPU.  Note, only interrupts that are routed
4459 		 * through the local APIC trigger posted interrupt processing,
4460 		 * and enabling posted interrupts requires ACK-on-exit.
4461 		 */
4462 		if (irq == vmx->nested.posted_intr_nv) {
4463 			/*
4464 			 * Nested posted interrupts are delivered via RVI, i.e.
4465 			 * aren't injected by KVM, and so can be queued even if
4466 			 * manual event injection is disallowed.
4467 			 */
4468 			if (block_non_injected_events)
4469 				return -EBUSY;
4470 
4471 			vmx->nested.pi_pending = true;
4472 			kvm_apic_clear_irr(vcpu, irq);
4473 			goto no_vmexit;
4474 		}
4475 
4476 		if (block_nested_events)
4477 			return -EBUSY;
4478 
4479 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT,
4480 				  INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR | irq, 0);
4481 
4482 		/*
4483 		 * ACK the interrupt _after_ emulating VM-Exit, as the IRQ must
4484 		 * be marked as in-service in vmcs01.GUEST_INTERRUPT_STATUS.SVI
4485 		 * if APICv is active.
4486 		 */
4487 		kvm_apic_ack_interrupt(vcpu, irq);
4488 		return 0;
4489 	}
4490 
4491 no_vmexit:
4492 	return vmx_complete_nested_posted_interrupt(vcpu);
4493 }
4494 
4495 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
4496 {
4497 	ktime_t remaining =
4498 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
4499 	u64 value;
4500 
4501 	if (ktime_to_ns(remaining) <= 0)
4502 		return 0;
4503 
4504 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
4505 	do_div(value, 1000000);
4506 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
4507 }
4508 
4509 static bool is_vmcs12_ext_field(unsigned long field)
4510 {
4511 	switch (field) {
4512 	case GUEST_ES_SELECTOR:
4513 	case GUEST_CS_SELECTOR:
4514 	case GUEST_SS_SELECTOR:
4515 	case GUEST_DS_SELECTOR:
4516 	case GUEST_FS_SELECTOR:
4517 	case GUEST_GS_SELECTOR:
4518 	case GUEST_LDTR_SELECTOR:
4519 	case GUEST_TR_SELECTOR:
4520 	case GUEST_ES_LIMIT:
4521 	case GUEST_CS_LIMIT:
4522 	case GUEST_SS_LIMIT:
4523 	case GUEST_DS_LIMIT:
4524 	case GUEST_FS_LIMIT:
4525 	case GUEST_GS_LIMIT:
4526 	case GUEST_LDTR_LIMIT:
4527 	case GUEST_TR_LIMIT:
4528 	case GUEST_GDTR_LIMIT:
4529 	case GUEST_IDTR_LIMIT:
4530 	case GUEST_ES_AR_BYTES:
4531 	case GUEST_DS_AR_BYTES:
4532 	case GUEST_FS_AR_BYTES:
4533 	case GUEST_GS_AR_BYTES:
4534 	case GUEST_LDTR_AR_BYTES:
4535 	case GUEST_TR_AR_BYTES:
4536 	case GUEST_ES_BASE:
4537 	case GUEST_CS_BASE:
4538 	case GUEST_SS_BASE:
4539 	case GUEST_DS_BASE:
4540 	case GUEST_FS_BASE:
4541 	case GUEST_GS_BASE:
4542 	case GUEST_LDTR_BASE:
4543 	case GUEST_TR_BASE:
4544 	case GUEST_GDTR_BASE:
4545 	case GUEST_IDTR_BASE:
4546 	case GUEST_PENDING_DBG_EXCEPTIONS:
4547 	case GUEST_BNDCFGS:
4548 		return true;
4549 	default:
4550 		break;
4551 	}
4552 
4553 	return false;
4554 }
4555 
4556 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4557 				       struct vmcs12 *vmcs12)
4558 {
4559 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4560 
4561 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4562 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4563 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4564 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4565 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4566 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4567 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4568 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4569 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4570 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4571 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4572 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4573 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4574 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4575 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4576 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4577 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4578 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4579 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4580 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4581 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4582 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4583 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4584 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4585 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4586 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4587 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4588 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4589 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4590 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4591 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4592 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4593 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4594 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4595 	vmcs12->guest_pending_dbg_exceptions =
4596 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4597 
4598 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4599 }
4600 
4601 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4602 				       struct vmcs12 *vmcs12)
4603 {
4604 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4605 	int cpu;
4606 
4607 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4608 		return;
4609 
4610 
4611 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4612 
4613 	cpu = get_cpu();
4614 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4615 	vmx_vcpu_load_vmcs(vcpu, cpu);
4616 
4617 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4618 
4619 	vmx->loaded_vmcs = &vmx->vmcs01;
4620 	vmx_vcpu_load_vmcs(vcpu, cpu);
4621 	put_cpu();
4622 }
4623 
4624 /*
4625  * Update the guest state fields of vmcs12 to reflect changes that
4626  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4627  * VM-entry controls is also updated, since this is really a guest
4628  * state bit.)
4629  */
4630 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4631 {
4632 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4633 
4634 	if (nested_vmx_is_evmptr12_valid(vmx))
4635 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4636 
4637 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4638 		!nested_vmx_is_evmptr12_valid(vmx);
4639 
4640 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4641 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4642 
4643 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4644 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4645 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4646 
4647 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4648 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4649 
4650 	vmcs12->guest_interruptibility_info =
4651 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4652 
4653 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4654 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4655 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4656 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4657 	else
4658 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4659 
4660 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4661 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4662 	    !vmx->nested.nested_run_pending)
4663 		vmcs12->vmx_preemption_timer_value =
4664 			vmx_get_preemption_timer_value(vcpu);
4665 
4666 	/*
4667 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4668 	 * own CR3 without exiting. If it has changed it, we must keep it.
4669 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4670 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4671 	 *
4672 	 * Additionally, restore L2's PDPTR to vmcs12.
4673 	 */
4674 	if (enable_ept) {
4675 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4676 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4677 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4678 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4679 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4680 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4681 		}
4682 	}
4683 
4684 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4685 
4686 	if (nested_cpu_has_vid(vmcs12))
4687 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4688 
4689 	vmcs12->vm_entry_controls =
4690 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4691 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4692 
4693 	/*
4694 	 * Note!  Save DR7, but intentionally don't grab DEBUGCTL from vmcs02.
4695 	 * Writes to DEBUGCTL that aren't intercepted by L1 are immediately
4696 	 * propagated to vmcs12 (see vmx_set_msr()), as the value loaded into
4697 	 * vmcs02 doesn't strictly track vmcs12.
4698 	 */
4699 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4700 		vmcs12->guest_dr7 = vcpu->arch.dr7;
4701 
4702 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4703 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4704 
4705 	vmcs_read_cet_state(&vmx->vcpu, &vmcs12->guest_s_cet,
4706 			    &vmcs12->guest_ssp,
4707 			    &vmcs12->guest_ssp_tbl);
4708 }
4709 
4710 /*
4711  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4712  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4713  * and this function updates it to reflect the changes to the guest state while
4714  * L2 was running (and perhaps made some exits which were handled directly by L0
4715  * without going back to L1), and to reflect the exit reason.
4716  * Note that we do not have to copy here all VMCS fields, just those that
4717  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4718  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4719  * which already writes to vmcs12 directly.
4720  */
4721 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4722 			   u32 vm_exit_reason, u32 exit_intr_info,
4723 			   unsigned long exit_qualification, u32 exit_insn_len)
4724 {
4725 	/* update exit information fields: */
4726 	vmcs12->vm_exit_reason = vm_exit_reason;
4727 	if (vmx_get_exit_reason(vcpu).enclave_mode)
4728 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4729 	vmcs12->exit_qualification = exit_qualification;
4730 
4731 	/*
4732 	 * On VM-Exit due to a failed VM-Entry, the VMCS isn't marked launched
4733 	 * and only EXIT_REASON and EXIT_QUALIFICATION are updated, all other
4734 	 * exit info fields are unmodified.
4735 	 */
4736 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4737 		vmcs12->launch_state = 1;
4738 
4739 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4740 		 * instead of reading the real value. */
4741 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4742 
4743 		/*
4744 		 * Transfer the event that L0 or L1 may wanted to inject into
4745 		 * L2 to IDT_VECTORING_INFO_FIELD.
4746 		 */
4747 		vmcs12_save_pending_event(vcpu, vmcs12,
4748 					  vm_exit_reason, exit_intr_info);
4749 
4750 		vmcs12->vm_exit_intr_info = exit_intr_info;
4751 		vmcs12->vm_exit_instruction_len = exit_insn_len;
4752 		vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4753 
4754 		/*
4755 		 * According to spec, there's no need to store the guest's
4756 		 * MSRs if the exit is due to a VM-entry failure that occurs
4757 		 * during or after loading the guest state. Since this exit
4758 		 * does not fall in that category, we need to save the MSRs.
4759 		 */
4760 		if (nested_vmx_store_msr(vcpu,
4761 					 vmcs12->vm_exit_msr_store_addr,
4762 					 vmcs12->vm_exit_msr_store_count))
4763 			nested_vmx_abort(vcpu,
4764 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4765 	}
4766 }
4767 
4768 /*
4769  * A part of what we need to when the nested L2 guest exits and we want to
4770  * run its L1 parent, is to reset L1's guest state to the host state specified
4771  * in vmcs12.
4772  * This function is to be called not only on normal nested exit, but also on
4773  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4774  * Failures During or After Loading Guest State").
4775  * This function should be called when the active VMCS is L1's (vmcs01).
4776  */
4777 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4778 				   struct vmcs12 *vmcs12)
4779 {
4780 	enum vm_entry_failure_code ignored;
4781 	struct kvm_segment seg;
4782 
4783 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4784 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4785 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4786 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4787 	else
4788 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4789 	vmx_set_efer(vcpu, vcpu->arch.efer);
4790 
4791 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4792 	kvm_rip_write(vcpu, vmcs12->host_rip);
4793 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4794 	vmx_set_interrupt_shadow(vcpu, 0);
4795 
4796 	/*
4797 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4798 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4799 	 *
4800 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4801 	 * (KVM doesn't change it);
4802 	 */
4803 	vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4804 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4805 
4806 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4807 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4808 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4809 
4810 	nested_ept_uninit_mmu_context(vcpu);
4811 
4812 	/*
4813 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4814 	 * couldn't have changed.
4815 	 */
4816 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4817 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4818 
4819 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4820 
4821 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4822 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4823 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4824 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4825 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4826 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4827 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4828 
4829 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4830 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4831 		vmcs_write64(GUEST_BNDCFGS, 0);
4832 
4833 	/*
4834 	 * Load CET state from host state if VM_EXIT_LOAD_CET_STATE is set.
4835 	 * otherwise CET state should be retained across VM-exit, i.e.,
4836 	 * guest values should be propagated from vmcs12 to vmcs01.
4837 	 */
4838 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_CET_STATE)
4839 		vmcs_write_cet_state(vcpu, vmcs12->host_s_cet, vmcs12->host_ssp,
4840 				     vmcs12->host_ssp_tbl);
4841 	else
4842 		vmcs_write_cet_state(vcpu, vmcs12->guest_s_cet, vmcs12->guest_ssp,
4843 				     vmcs12->guest_ssp_tbl);
4844 
4845 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4846 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4847 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4848 	}
4849 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
4850 	    kvm_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)))
4851 		WARN_ON_ONCE(__kvm_emulate_msr_write(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4852 						     vmcs12->host_ia32_perf_global_ctrl));
4853 
4854 	/* Set L1 segment info according to Intel SDM
4855 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4856 	seg = (struct kvm_segment) {
4857 		.base = 0,
4858 		.limit = 0xFFFFFFFF,
4859 		.selector = vmcs12->host_cs_selector,
4860 		.type = 11,
4861 		.present = 1,
4862 		.s = 1,
4863 		.g = 1
4864 	};
4865 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4866 		seg.l = 1;
4867 	else
4868 		seg.db = 1;
4869 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4870 	seg = (struct kvm_segment) {
4871 		.base = 0,
4872 		.limit = 0xFFFFFFFF,
4873 		.type = 3,
4874 		.present = 1,
4875 		.s = 1,
4876 		.db = 1,
4877 		.g = 1
4878 	};
4879 	seg.selector = vmcs12->host_ds_selector;
4880 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4881 	seg.selector = vmcs12->host_es_selector;
4882 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4883 	seg.selector = vmcs12->host_ss_selector;
4884 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4885 	seg.selector = vmcs12->host_fs_selector;
4886 	seg.base = vmcs12->host_fs_base;
4887 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4888 	seg.selector = vmcs12->host_gs_selector;
4889 	seg.base = vmcs12->host_gs_base;
4890 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4891 	seg = (struct kvm_segment) {
4892 		.base = vmcs12->host_tr_base,
4893 		.limit = 0x67,
4894 		.selector = vmcs12->host_tr_selector,
4895 		.type = 11,
4896 		.present = 1
4897 	};
4898 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4899 
4900 	memset(&seg, 0, sizeof(seg));
4901 	seg.unusable = 1;
4902 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4903 
4904 	kvm_set_dr(vcpu, 7, 0x400);
4905 	vmx_guest_debugctl_write(vcpu, 0);
4906 
4907 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4908 				vmcs12->vm_exit_msr_load_count))
4909 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4910 
4911 	to_vt(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4912 }
4913 
4914 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4915 {
4916 	struct vmx_uret_msr *efer_msr;
4917 	unsigned int i;
4918 
4919 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4920 		return vmcs_read64(GUEST_IA32_EFER);
4921 
4922 	if (cpu_has_load_ia32_efer())
4923 		return kvm_host.efer;
4924 
4925 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4926 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4927 			return vmx->msr_autoload.guest.val[i].value;
4928 	}
4929 
4930 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4931 	if (efer_msr)
4932 		return efer_msr->data;
4933 
4934 	return kvm_host.efer;
4935 }
4936 
4937 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4938 {
4939 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4940 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4941 	struct vmx_msr_entry g, h;
4942 	gpa_t gpa;
4943 	u32 i, j;
4944 
4945 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4946 
4947 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4948 		/*
4949 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4950 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4951 		 * and vcpu->arch.dr7 is not squirreled away before the
4952 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4953 		 */
4954 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4955 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4956 		else
4957 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4958 	}
4959 
4960 	/* Reload DEBUGCTL to ensure vmcs01 has a fresh FREEZE_IN_SMM value. */
4961 	vmx_reload_guest_debugctl(vcpu);
4962 
4963 	/*
4964 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4965 	 * handle a variety of side effects to KVM's software model.
4966 	 */
4967 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4968 
4969 	vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4970 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4971 
4972 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4973 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4974 
4975 	nested_ept_uninit_mmu_context(vcpu);
4976 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4977 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4978 
4979 	/*
4980 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4981 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4982 	 * VMFail, like everything else we just need to ensure our
4983 	 * software model is up-to-date.
4984 	 */
4985 	if (enable_ept && is_pae_paging(vcpu))
4986 		ept_save_pdptrs(vcpu);
4987 
4988 	kvm_mmu_reset_context(vcpu);
4989 
4990 	/*
4991 	 * This nasty bit of open coding is a compromise between blindly
4992 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4993 	 * of VMFail), leaving the nested VM's MSRs in the software model
4994 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4995 	 * expensive since the lists are unbound by hardware).  For each
4996 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4997 	 * list, reload it from the exit load list if it exists and differs
4998 	 * from the guest value.  The intent is to stuff host state as
4999 	 * silently as possible, not to fully process the exit load list.
5000 	 */
5001 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
5002 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
5003 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
5004 			pr_debug_ratelimited(
5005 				"%s read MSR index failed (%u, 0x%08llx)\n",
5006 				__func__, i, gpa);
5007 			goto vmabort;
5008 		}
5009 
5010 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
5011 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
5012 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
5013 				pr_debug_ratelimited(
5014 					"%s read MSR failed (%u, 0x%08llx)\n",
5015 					__func__, j, gpa);
5016 				goto vmabort;
5017 			}
5018 			if (h.index != g.index)
5019 				continue;
5020 			if (h.value == g.value)
5021 				break;
5022 
5023 			if (nested_vmx_load_msr_check(vcpu, &h)) {
5024 				pr_debug_ratelimited(
5025 					"%s check failed (%u, 0x%x, 0x%x)\n",
5026 					__func__, j, h.index, h.reserved);
5027 				goto vmabort;
5028 			}
5029 
5030 			if (kvm_emulate_msr_write(vcpu, h.index, h.value)) {
5031 				pr_debug_ratelimited(
5032 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
5033 					__func__, j, h.index, h.value);
5034 				goto vmabort;
5035 			}
5036 		}
5037 	}
5038 
5039 	return;
5040 
5041 vmabort:
5042 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
5043 }
5044 
5045 /*
5046  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
5047  * and modify vmcs12 to make it see what it would expect to see there if
5048  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
5049  */
5050 void __nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
5051 			 u32 exit_intr_info, unsigned long exit_qualification,
5052 			 u32 exit_insn_len)
5053 {
5054 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5055 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5056 
5057 	/* Pending MTF traps are discarded on VM-Exit. */
5058 	vmx->nested.mtf_pending = false;
5059 
5060 	/* trying to cancel vmlaunch/vmresume is a bug */
5061 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
5062 
5063 #ifdef CONFIG_KVM_HYPERV
5064 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
5065 		/*
5066 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
5067 		 * Enlightened VMCS after migration and we still need to
5068 		 * do that when something is forcing L2->L1 exit prior to
5069 		 * the first L2 run.
5070 		 */
5071 		(void)nested_get_evmcs_page(vcpu);
5072 	}
5073 #endif
5074 
5075 	/* Service pending TLB flush requests for L2 before switching to L1. */
5076 	kvm_service_local_tlb_flush_requests(vcpu);
5077 
5078 	/*
5079 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
5080 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
5081 	 * up-to-date before switching to L1.
5082 	 */
5083 	if (enable_ept && is_pae_paging(vcpu))
5084 		vmx_ept_load_pdptrs(vcpu);
5085 
5086 	leave_guest_mode(vcpu);
5087 
5088 	if (nested_cpu_has_preemption_timer(vmcs12))
5089 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
5090 
5091 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
5092 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
5093 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
5094 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
5095 	}
5096 
5097 	if (likely(!vmx->fail)) {
5098 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
5099 
5100 		if (vm_exit_reason != -1)
5101 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
5102 				       exit_intr_info, exit_qualification,
5103 				       exit_insn_len);
5104 
5105 		/*
5106 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
5107 		 * also be used to capture vmcs12 cache as part of
5108 		 * capturing nVMX state for snapshot (migration).
5109 		 *
5110 		 * Otherwise, this flush will dirty guest memory at a
5111 		 * point it is already assumed by user-space to be
5112 		 * immutable.
5113 		 */
5114 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
5115 	} else {
5116 		/*
5117 		 * The only expected VM-instruction error is "VM entry with
5118 		 * invalid control field(s)." Anything else indicates a
5119 		 * problem with L0.
5120 		 */
5121 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
5122 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
5123 
5124 		/* VM-Fail at VM-Entry means KVM missed a consistency check. */
5125 		WARN_ON_ONCE(warn_on_missed_cc);
5126 	}
5127 
5128 	/*
5129 	 * Drop events/exceptions that were queued for re-injection to L2
5130 	 * (picked up via vmx_complete_interrupts()), as well as exceptions
5131 	 * that were pending for L2.  Note, this must NOT be hoisted above
5132 	 * prepare_vmcs12(), events/exceptions queued for re-injection need to
5133 	 * be captured in vmcs12 (see vmcs12_save_pending_event()).
5134 	 */
5135 	vcpu->arch.nmi_injected = false;
5136 	kvm_clear_exception_queue(vcpu);
5137 	kvm_clear_interrupt_queue(vcpu);
5138 
5139 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
5140 
5141 	kvm_nested_vmexit_handle_ibrs(vcpu);
5142 
5143 	/* Update any VMCS fields that might have changed while L2 ran */
5144 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
5145 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
5146 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
5147 	if (kvm_caps.has_tsc_control)
5148 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
5149 
5150 	if (vmx->nested.l1_tpr_threshold != -1)
5151 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
5152 
5153 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
5154 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
5155 		vmx_set_virtual_apic_mode(vcpu);
5156 	}
5157 
5158 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
5159 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
5160 		vmx_update_cpu_dirty_logging(vcpu);
5161 	}
5162 
5163 	nested_put_vmcs12_pages(vcpu);
5164 
5165 	if (vmx->nested.reload_vmcs01_apic_access_page) {
5166 		vmx->nested.reload_vmcs01_apic_access_page = false;
5167 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5168 	}
5169 
5170 	if (vmx->nested.update_vmcs01_apicv_status) {
5171 		vmx->nested.update_vmcs01_apicv_status = false;
5172 		vmx_refresh_apicv_exec_ctrl(vcpu);
5173 	}
5174 
5175 	if (vmx->nested.update_vmcs01_hwapic_isr) {
5176 		vmx->nested.update_vmcs01_hwapic_isr = false;
5177 		kvm_apic_update_hwapic_isr(vcpu);
5178 	}
5179 
5180 	if ((vm_exit_reason != -1) &&
5181 	    (enable_shadow_vmcs || nested_vmx_is_evmptr12_valid(vmx)))
5182 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5183 
5184 	/* in case we halted in L2 */
5185 	kvm_set_mp_state(vcpu, KVM_MP_STATE_RUNNABLE);
5186 
5187 	if (likely(!vmx->fail)) {
5188 		if (vm_exit_reason != -1)
5189 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
5190 						       vmcs12->exit_qualification,
5191 						       vmcs12->idt_vectoring_info_field,
5192 						       vmcs12->vm_exit_intr_info,
5193 						       vmcs12->vm_exit_intr_error_code,
5194 						       KVM_ISA_VMX);
5195 
5196 		load_vmcs12_host_state(vcpu, vmcs12);
5197 
5198 		/*
5199 		 * Process events if an injectable IRQ or NMI is pending, even
5200 		 * if the event is blocked (RFLAGS.IF is cleared on VM-Exit).
5201 		 * If an event became pending while L2 was active, KVM needs to
5202 		 * either inject the event or request an IRQ/NMI window.  SMIs
5203 		 * don't need to be processed as SMM is mutually exclusive with
5204 		 * non-root mode.  INIT/SIPI don't need to be checked as INIT
5205 		 * is blocked post-VMXON, and SIPIs are ignored.
5206 		 */
5207 		if (kvm_cpu_has_injectable_intr(vcpu) || vcpu->arch.nmi_pending)
5208 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5209 		return;
5210 	}
5211 
5212 	/*
5213 	 * After an early L2 VM-entry failure, we're now back
5214 	 * in L1 which thinks it just finished a VMLAUNCH or
5215 	 * VMRESUME instruction, so we need to set the failure
5216 	 * flag and the VM-instruction error field of the VMCS
5217 	 * accordingly, and skip the emulated instruction.
5218 	 */
5219 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
5220 
5221 	/*
5222 	 * Restore L1's host state to KVM's software model.  We're here
5223 	 * because a consistency check was caught by hardware, which
5224 	 * means some amount of guest state has been propagated to KVM's
5225 	 * model and needs to be unwound to the host's state.
5226 	 */
5227 	nested_vmx_restore_host_state(vcpu);
5228 
5229 	vmx->fail = 0;
5230 }
5231 
5232 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
5233 {
5234 	kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5235 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
5236 }
5237 
5238 /*
5239  * Decode the memory-address operand of a vmx instruction, as recorded on an
5240  * exit caused by such an instruction (run by a guest hypervisor).
5241  * On success, returns 0. When the operand is invalid, returns 1 and throws
5242  * #UD, #GP, or #SS.
5243  */
5244 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
5245 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
5246 {
5247 	gva_t off;
5248 	bool exn;
5249 	struct kvm_segment s;
5250 
5251 	/*
5252 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5253 	 * Execution", on an exit, vmx_instruction_info holds most of the
5254 	 * addressing components of the operand. Only the displacement part
5255 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5256 	 * For how an actual address is calculated from all these components,
5257 	 * refer to Vol. 1, "Operand Addressing".
5258 	 */
5259 	int  scaling = vmx_instruction_info & 3;
5260 	int  addr_size = (vmx_instruction_info >> 7) & 7;
5261 	bool is_reg = vmx_instruction_info & (1u << 10);
5262 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
5263 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5264 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5265 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5266 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5267 
5268 	if (is_reg) {
5269 		kvm_queue_exception(vcpu, UD_VECTOR);
5270 		return 1;
5271 	}
5272 
5273 	/* Addr = segment_base + offset */
5274 	/* offset = base + [index * scale] + displacement */
5275 	off = exit_qualification; /* holds the displacement */
5276 	if (addr_size == 1)
5277 		off = (gva_t)sign_extend64(off, 31);
5278 	else if (addr_size == 0)
5279 		off = (gva_t)sign_extend64(off, 15);
5280 	if (base_is_valid)
5281 		off += kvm_register_read(vcpu, base_reg);
5282 	if (index_is_valid)
5283 		off += kvm_register_read(vcpu, index_reg) << scaling;
5284 	vmx_get_segment(vcpu, &s, seg_reg);
5285 
5286 	/*
5287 	 * The effective address, i.e. @off, of a memory operand is truncated
5288 	 * based on the address size of the instruction.  Note that this is
5289 	 * the *effective address*, i.e. the address prior to accounting for
5290 	 * the segment's base.
5291 	 */
5292 	if (addr_size == 1) /* 32 bit */
5293 		off &= 0xffffffff;
5294 	else if (addr_size == 0) /* 16 bit */
5295 		off &= 0xffff;
5296 
5297 	/* Checks for #GP/#SS exceptions. */
5298 	exn = false;
5299 	if (is_long_mode(vcpu)) {
5300 		/*
5301 		 * The virtual/linear address is never truncated in 64-bit
5302 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
5303 		 * address when using FS/GS with a non-zero base.
5304 		 */
5305 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
5306 			*ret = s.base + off;
5307 		else
5308 			*ret = off;
5309 
5310 		*ret = vmx_get_untagged_addr(vcpu, *ret, 0);
5311 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
5312 		 * non-canonical form. This is the only check on the memory
5313 		 * destination for long mode!
5314 		 */
5315 		exn = is_noncanonical_address(*ret, vcpu, 0);
5316 	} else {
5317 		/*
5318 		 * When not in long mode, the virtual/linear address is
5319 		 * unconditionally truncated to 32 bits regardless of the
5320 		 * address size.
5321 		 */
5322 		*ret = (s.base + off) & 0xffffffff;
5323 
5324 		/* Protected mode: apply checks for segment validity in the
5325 		 * following order:
5326 		 * - segment type check (#GP(0) may be thrown)
5327 		 * - usability check (#GP(0)/#SS(0))
5328 		 * - limit check (#GP(0)/#SS(0))
5329 		 */
5330 		if (wr)
5331 			/* #GP(0) if the destination operand is located in a
5332 			 * read-only data segment or any code segment.
5333 			 */
5334 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
5335 		else
5336 			/* #GP(0) if the source operand is located in an
5337 			 * execute-only code segment
5338 			 */
5339 			exn = ((s.type & 0xa) == 8);
5340 		if (exn) {
5341 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5342 			return 1;
5343 		}
5344 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
5345 		 */
5346 		exn = (s.unusable != 0);
5347 
5348 		/*
5349 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
5350 		 * outside the segment limit.  All CPUs that support VMX ignore
5351 		 * limit checks for flat segments, i.e. segments with base==0,
5352 		 * limit==0xffffffff and of type expand-up data or code.
5353 		 */
5354 		if (!(s.base == 0 && s.limit == 0xffffffff &&
5355 		     ((s.type & 8) || !(s.type & 4))))
5356 			exn = exn || ((u64)off + len - 1 > s.limit);
5357 	}
5358 	if (exn) {
5359 		kvm_queue_exception_e(vcpu,
5360 				      seg_reg == VCPU_SREG_SS ?
5361 						SS_VECTOR : GP_VECTOR,
5362 				      0);
5363 		return 1;
5364 	}
5365 
5366 	return 0;
5367 }
5368 
5369 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
5370 				int *ret)
5371 {
5372 	gva_t gva;
5373 	struct x86_exception e;
5374 	int r;
5375 
5376 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5377 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
5378 				sizeof(*vmpointer), &gva)) {
5379 		*ret = 1;
5380 		return -EINVAL;
5381 	}
5382 
5383 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
5384 	if (r != X86EMUL_CONTINUE) {
5385 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
5386 		return -EINVAL;
5387 	}
5388 
5389 	return 0;
5390 }
5391 
5392 /*
5393  * Allocate a shadow VMCS and associate it with the currently loaded
5394  * VMCS, unless such a shadow VMCS already exists. The newly allocated
5395  * VMCS is also VMCLEARed, so that it is ready for use.
5396  */
5397 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
5398 {
5399 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5400 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
5401 
5402 	/*
5403 	 * KVM allocates a shadow VMCS only when L1 executes VMXON and frees it
5404 	 * when L1 executes VMXOFF or the vCPU is forced out of nested
5405 	 * operation.  VMXON faults if the CPU is already post-VMXON, so it
5406 	 * should be impossible to already have an allocated shadow VMCS.  KVM
5407 	 * doesn't support virtualization of VMCS shadowing, so vmcs01 should
5408 	 * always be the loaded VMCS.
5409 	 */
5410 	if (WARN_ON(loaded_vmcs != &vmx->vmcs01 || loaded_vmcs->shadow_vmcs))
5411 		return loaded_vmcs->shadow_vmcs;
5412 
5413 	loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
5414 	if (loaded_vmcs->shadow_vmcs)
5415 		vmcs_clear(loaded_vmcs->shadow_vmcs);
5416 
5417 	return loaded_vmcs->shadow_vmcs;
5418 }
5419 
5420 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
5421 {
5422 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5423 	int r;
5424 
5425 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
5426 	if (r < 0)
5427 		goto out_vmcs02;
5428 
5429 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5430 	if (!vmx->nested.cached_vmcs12)
5431 		goto out_cached_vmcs12;
5432 
5433 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
5434 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5435 	if (!vmx->nested.cached_shadow_vmcs12)
5436 		goto out_cached_shadow_vmcs12;
5437 
5438 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
5439 		goto out_shadow_vmcs;
5440 
5441 	hrtimer_setup(&vmx->nested.preemption_timer, vmx_preemption_timer_fn, CLOCK_MONOTONIC,
5442 		      HRTIMER_MODE_ABS_PINNED);
5443 
5444 	vmx->nested.vpid02 = allocate_vpid();
5445 
5446 	vmx->nested.vmcs02_initialized = false;
5447 	vmx->nested.vmxon = true;
5448 
5449 	if (vmx_pt_mode_is_host_guest()) {
5450 		vmx->pt_desc.guest.ctl = 0;
5451 		pt_update_intercept_for_msr(vcpu);
5452 	}
5453 
5454 	return 0;
5455 
5456 out_shadow_vmcs:
5457 	kfree(vmx->nested.cached_shadow_vmcs12);
5458 
5459 out_cached_shadow_vmcs12:
5460 	kfree(vmx->nested.cached_vmcs12);
5461 
5462 out_cached_vmcs12:
5463 	free_loaded_vmcs(&vmx->nested.vmcs02);
5464 
5465 out_vmcs02:
5466 	return -ENOMEM;
5467 }
5468 
5469 /* Emulate the VMXON instruction. */
5470 static int handle_vmxon(struct kvm_vcpu *vcpu)
5471 {
5472 	int ret;
5473 	gpa_t vmptr;
5474 	uint32_t revision;
5475 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5476 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
5477 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
5478 
5479 	/*
5480 	 * Manually check CR4.VMXE checks, KVM must force CR4.VMXE=1 to enter
5481 	 * the guest and so cannot rely on hardware to perform the check,
5482 	 * which has higher priority than VM-Exit (see Intel SDM's pseudocode
5483 	 * for VMXON).
5484 	 *
5485 	 * Rely on hardware for the other pre-VM-Exit checks, CR0.PE=1, !VM86
5486 	 * and !COMPATIBILITY modes.  For an unrestricted guest, KVM doesn't
5487 	 * force any of the relevant guest state.  For a restricted guest, KVM
5488 	 * does force CR0.PE=1, but only to also force VM86 in order to emulate
5489 	 * Real Mode, and so there's no need to check CR0.PE manually.
5490 	 */
5491 	if (!kvm_is_cr4_bit_set(vcpu, X86_CR4_VMXE)) {
5492 		kvm_queue_exception(vcpu, UD_VECTOR);
5493 		return 1;
5494 	}
5495 
5496 	/*
5497 	 * The CPL is checked for "not in VMX operation" and for "in VMX root",
5498 	 * and has higher priority than the VM-Fail due to being post-VMXON,
5499 	 * i.e. VMXON #GPs outside of VMX non-root if CPL!=0.  In VMX non-root,
5500 	 * VMXON causes VM-Exit and KVM unconditionally forwards VMXON VM-Exits
5501 	 * from L2 to L1, i.e. there's no need to check for the vCPU being in
5502 	 * VMX non-root.
5503 	 *
5504 	 * Forwarding the VM-Exit unconditionally, i.e. without performing the
5505 	 * #UD checks (see above), is functionally ok because KVM doesn't allow
5506 	 * L1 to run L2 without CR4.VMXE=0, and because KVM never modifies L2's
5507 	 * CR0 or CR4, i.e. it's L2's responsibility to emulate #UDs that are
5508 	 * missed by hardware due to shadowing CR0 and/or CR4.
5509 	 */
5510 	if (vmx_get_cpl(vcpu)) {
5511 		kvm_inject_gp(vcpu, 0);
5512 		return 1;
5513 	}
5514 
5515 	if (vmx->nested.vmxon)
5516 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5517 
5518 	/*
5519 	 * Invalid CR0/CR4 generates #GP.  These checks are performed if and
5520 	 * only if the vCPU isn't already in VMX operation, i.e. effectively
5521 	 * have lower priority than the VM-Fail above.
5522 	 */
5523 	if (!nested_host_cr0_valid(vcpu, kvm_read_cr0(vcpu)) ||
5524 	    !nested_host_cr4_valid(vcpu, kvm_read_cr4(vcpu))) {
5525 		kvm_inject_gp(vcpu, 0);
5526 		return 1;
5527 	}
5528 
5529 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5530 			!= VMXON_NEEDED_FEATURES) {
5531 		kvm_inject_gp(vcpu, 0);
5532 		return 1;
5533 	}
5534 
5535 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
5536 		return ret;
5537 
5538 	/*
5539 	 * SDM 3: 24.11.5
5540 	 * The first 4 bytes of VMXON region contain the supported
5541 	 * VMCS revision identifier
5542 	 *
5543 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
5544 	 * which replaces physical address width with 32
5545 	 */
5546 	if (!page_address_valid(vcpu, vmptr))
5547 		return nested_vmx_failInvalid(vcpu);
5548 
5549 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
5550 	    revision != VMCS12_REVISION)
5551 		return nested_vmx_failInvalid(vcpu);
5552 
5553 	vmx->nested.vmxon_ptr = vmptr;
5554 	ret = enter_vmx_operation(vcpu);
5555 	if (ret)
5556 		return ret;
5557 
5558 	return nested_vmx_succeed(vcpu);
5559 }
5560 
5561 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
5562 {
5563 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5564 
5565 	if (vmx->nested.current_vmptr == INVALID_GPA)
5566 		return;
5567 
5568 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
5569 
5570 	if (enable_shadow_vmcs) {
5571 		/* copy to memory all shadowed fields in case
5572 		   they were modified */
5573 		copy_shadow_to_vmcs12(vmx);
5574 		vmx_disable_shadow_vmcs(vmx);
5575 	}
5576 	vmx->nested.posted_intr_nv = -1;
5577 
5578 	/* Flush VMCS12 to guest memory */
5579 	kvm_vcpu_write_guest_page(vcpu,
5580 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
5581 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5582 
5583 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5584 
5585 	vmx->nested.current_vmptr = INVALID_GPA;
5586 }
5587 
5588 /* Emulate the VMXOFF instruction */
5589 static int handle_vmxoff(struct kvm_vcpu *vcpu)
5590 {
5591 	if (!nested_vmx_check_permission(vcpu))
5592 		return 1;
5593 
5594 	free_nested(vcpu);
5595 
5596 	if (kvm_apic_has_pending_init_or_sipi(vcpu))
5597 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5598 
5599 	return nested_vmx_succeed(vcpu);
5600 }
5601 
5602 /* Emulate the VMCLEAR instruction */
5603 static int handle_vmclear(struct kvm_vcpu *vcpu)
5604 {
5605 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5606 	u32 zero = 0;
5607 	gpa_t vmptr;
5608 	int r;
5609 
5610 	if (!nested_vmx_check_permission(vcpu))
5611 		return 1;
5612 
5613 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5614 		return r;
5615 
5616 	if (!page_address_valid(vcpu, vmptr))
5617 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5618 
5619 	if (vmptr == vmx->nested.vmxon_ptr)
5620 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5621 
5622 	if (likely(!nested_evmcs_handle_vmclear(vcpu, vmptr))) {
5623 		if (vmptr == vmx->nested.current_vmptr)
5624 			nested_release_vmcs12(vcpu);
5625 
5626 		/*
5627 		 * Silently ignore memory errors on VMCLEAR, Intel's pseudocode
5628 		 * for VMCLEAR includes a "ensure that data for VMCS referenced
5629 		 * by the operand is in memory" clause that guards writes to
5630 		 * memory, i.e. doing nothing for I/O is architecturally valid.
5631 		 *
5632 		 * FIXME: Suppress failures if and only if no memslot is found,
5633 		 * i.e. exit to userspace if __copy_to_user() fails.
5634 		 */
5635 		(void)kvm_vcpu_write_guest(vcpu,
5636 					   vmptr + offsetof(struct vmcs12,
5637 							    launch_state),
5638 					   &zero, sizeof(zero));
5639 	}
5640 
5641 	return nested_vmx_succeed(vcpu);
5642 }
5643 
5644 /* Emulate the VMLAUNCH instruction */
5645 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5646 {
5647 	return nested_vmx_run(vcpu, true);
5648 }
5649 
5650 /* Emulate the VMRESUME instruction */
5651 static int handle_vmresume(struct kvm_vcpu *vcpu)
5652 {
5653 
5654 	return nested_vmx_run(vcpu, false);
5655 }
5656 
5657 static int handle_vmread(struct kvm_vcpu *vcpu)
5658 {
5659 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5660 						    : get_vmcs12(vcpu);
5661 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5662 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5663 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5664 	struct x86_exception e;
5665 	unsigned long field;
5666 	u64 value;
5667 	gva_t gva = 0;
5668 	short offset;
5669 	int len, r;
5670 
5671 	if (!nested_vmx_check_permission(vcpu))
5672 		return 1;
5673 
5674 	/* Decode instruction info and find the field to read */
5675 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5676 
5677 	if (!nested_vmx_is_evmptr12_valid(vmx)) {
5678 		/*
5679 		 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5680 		 * any VMREAD sets the ALU flags for VMfailInvalid.
5681 		 */
5682 		if (vmx->nested.current_vmptr == INVALID_GPA ||
5683 		    (is_guest_mode(vcpu) &&
5684 		     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5685 			return nested_vmx_failInvalid(vcpu);
5686 
5687 		offset = get_vmcs12_field_offset(field);
5688 		if (offset < 0)
5689 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5690 
5691 		if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5692 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5693 
5694 		/* Read the field, zero-extended to a u64 value */
5695 		value = vmcs12_read_any(vmcs12, field, offset);
5696 	} else {
5697 		/*
5698 		 * Hyper-V TLFS (as of 6.0b) explicitly states, that while an
5699 		 * enlightened VMCS is active VMREAD/VMWRITE instructions are
5700 		 * unsupported. Unfortunately, certain versions of Windows 11
5701 		 * don't comply with this requirement which is not enforced in
5702 		 * genuine Hyper-V. Allow VMREAD from an enlightened VMCS as a
5703 		 * workaround, as misbehaving guests will panic on VM-Fail.
5704 		 * Note, enlightened VMCS is incompatible with shadow VMCS so
5705 		 * all VMREADs from L2 should go to L1.
5706 		 */
5707 		if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5708 			return nested_vmx_failInvalid(vcpu);
5709 
5710 		offset = evmcs_field_offset(field, NULL);
5711 		if (offset < 0)
5712 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5713 
5714 		/* Read the field, zero-extended to a u64 value */
5715 		value = evmcs_read_any(nested_vmx_evmcs(vmx), field, offset);
5716 	}
5717 
5718 	/*
5719 	 * Now copy part of this value to register or memory, as requested.
5720 	 * Note that the number of bits actually copied is 32 or 64 depending
5721 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5722 	 */
5723 	if (instr_info & BIT(10)) {
5724 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5725 	} else {
5726 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5727 		if (get_vmx_mem_address(vcpu, exit_qualification,
5728 					instr_info, true, len, &gva))
5729 			return 1;
5730 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5731 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5732 		if (r != X86EMUL_CONTINUE)
5733 			return kvm_handle_memory_failure(vcpu, r, &e);
5734 	}
5735 
5736 	return nested_vmx_succeed(vcpu);
5737 }
5738 
5739 static bool is_shadow_field_rw(unsigned long field)
5740 {
5741 	switch (field) {
5742 #define SHADOW_FIELD_RW(x, y) case x:
5743 #include "vmcs_shadow_fields.h"
5744 		return true;
5745 	default:
5746 		break;
5747 	}
5748 	return false;
5749 }
5750 
5751 static bool is_shadow_field_ro(unsigned long field)
5752 {
5753 	switch (field) {
5754 #define SHADOW_FIELD_RO(x, y) case x:
5755 #include "vmcs_shadow_fields.h"
5756 		return true;
5757 	default:
5758 		break;
5759 	}
5760 	return false;
5761 }
5762 
5763 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5764 {
5765 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5766 						    : get_vmcs12(vcpu);
5767 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5768 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5769 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5770 	struct x86_exception e;
5771 	unsigned long field;
5772 	short offset;
5773 	gva_t gva;
5774 	int len, r;
5775 
5776 	/*
5777 	 * The value to write might be 32 or 64 bits, depending on L1's long
5778 	 * mode, and eventually we need to write that into a field of several
5779 	 * possible lengths. The code below first zero-extends the value to 64
5780 	 * bit (value), and then copies only the appropriate number of
5781 	 * bits into the vmcs12 field.
5782 	 */
5783 	u64 value = 0;
5784 
5785 	if (!nested_vmx_check_permission(vcpu))
5786 		return 1;
5787 
5788 	/*
5789 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5790 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5791 	 */
5792 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5793 	    (is_guest_mode(vcpu) &&
5794 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5795 		return nested_vmx_failInvalid(vcpu);
5796 
5797 	if (instr_info & BIT(10))
5798 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5799 	else {
5800 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5801 		if (get_vmx_mem_address(vcpu, exit_qualification,
5802 					instr_info, false, len, &gva))
5803 			return 1;
5804 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5805 		if (r != X86EMUL_CONTINUE)
5806 			return kvm_handle_memory_failure(vcpu, r, &e);
5807 	}
5808 
5809 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5810 
5811 	offset = get_vmcs12_field_offset(field);
5812 	if (offset < 0)
5813 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5814 
5815 	/*
5816 	 * If the vCPU supports "VMWRITE to any supported field in the
5817 	 * VMCS," then the "read-only" fields are actually read/write.
5818 	 */
5819 	if (vmcs_field_readonly(field) &&
5820 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5821 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5822 
5823 	/*
5824 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5825 	 * vmcs12, else we may crush a field or consume a stale value.
5826 	 */
5827 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5828 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5829 
5830 	/*
5831 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5832 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5833 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5834 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5835 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5836 	 * the stripped down value, L2 sees the full value as stored by KVM).
5837 	 */
5838 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5839 		value &= 0x1f0ff;
5840 
5841 	vmcs12_write_any(vmcs12, field, offset, value);
5842 
5843 	/*
5844 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5845 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5846 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5847 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5848 	 */
5849 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5850 		/*
5851 		 * L1 can read these fields without exiting, ensure the
5852 		 * shadow VMCS is up-to-date.
5853 		 */
5854 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5855 			preempt_disable();
5856 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5857 
5858 			__vmcs_writel(field, value);
5859 
5860 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5861 			vmcs_load(vmx->loaded_vmcs->vmcs);
5862 			preempt_enable();
5863 		}
5864 		vmx->nested.dirty_vmcs12 = true;
5865 	}
5866 
5867 	return nested_vmx_succeed(vcpu);
5868 }
5869 
5870 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5871 {
5872 	vmx->nested.current_vmptr = vmptr;
5873 	if (enable_shadow_vmcs) {
5874 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5875 		vmcs_write64(VMCS_LINK_POINTER,
5876 			     __pa(vmx->vmcs01.shadow_vmcs));
5877 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5878 	}
5879 	vmx->nested.dirty_vmcs12 = true;
5880 	vmx->nested.force_msr_bitmap_recalc = true;
5881 }
5882 
5883 /* Emulate the VMPTRLD instruction */
5884 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5885 {
5886 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5887 	gpa_t vmptr;
5888 	int r;
5889 
5890 	if (!nested_vmx_check_permission(vcpu))
5891 		return 1;
5892 
5893 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5894 		return r;
5895 
5896 	if (!page_address_valid(vcpu, vmptr))
5897 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5898 
5899 	if (vmptr == vmx->nested.vmxon_ptr)
5900 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5901 
5902 	/* Forbid normal VMPTRLD if Enlightened version was used */
5903 	if (nested_vmx_is_evmptr12_valid(vmx))
5904 		return 1;
5905 
5906 	if (vmx->nested.current_vmptr != vmptr) {
5907 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5908 		struct vmcs_hdr hdr;
5909 
5910 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5911 			/*
5912 			 * Reads from an unbacked page return all 1s,
5913 			 * which means that the 32 bits located at the
5914 			 * given physical address won't match the required
5915 			 * VMCS12_REVISION identifier.
5916 			 */
5917 			return nested_vmx_fail(vcpu,
5918 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5919 		}
5920 
5921 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5922 						 offsetof(struct vmcs12, hdr),
5923 						 sizeof(hdr))) {
5924 			return nested_vmx_fail(vcpu,
5925 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5926 		}
5927 
5928 		if (hdr.revision_id != VMCS12_REVISION ||
5929 		    (hdr.shadow_vmcs &&
5930 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5931 			return nested_vmx_fail(vcpu,
5932 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5933 		}
5934 
5935 		nested_release_vmcs12(vcpu);
5936 
5937 		/*
5938 		 * Load VMCS12 from guest memory since it is not already
5939 		 * cached.
5940 		 */
5941 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5942 					  VMCS12_SIZE)) {
5943 			return nested_vmx_fail(vcpu,
5944 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5945 		}
5946 
5947 		set_current_vmptr(vmx, vmptr);
5948 	}
5949 
5950 	return nested_vmx_succeed(vcpu);
5951 }
5952 
5953 /* Emulate the VMPTRST instruction */
5954 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5955 {
5956 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5957 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5958 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5959 	struct x86_exception e;
5960 	gva_t gva;
5961 	int r;
5962 
5963 	if (!nested_vmx_check_permission(vcpu))
5964 		return 1;
5965 
5966 	if (unlikely(nested_vmx_is_evmptr12_valid(to_vmx(vcpu))))
5967 		return 1;
5968 
5969 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5970 				true, sizeof(gpa_t), &gva))
5971 		return 1;
5972 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5973 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5974 					sizeof(gpa_t), &e);
5975 	if (r != X86EMUL_CONTINUE)
5976 		return kvm_handle_memory_failure(vcpu, r, &e);
5977 
5978 	return nested_vmx_succeed(vcpu);
5979 }
5980 
5981 /* Emulate the INVEPT instruction */
5982 static int handle_invept(struct kvm_vcpu *vcpu)
5983 {
5984 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5985 	u32 vmx_instruction_info, types;
5986 	unsigned long type, roots_to_free;
5987 	struct kvm_mmu *mmu;
5988 	gva_t gva;
5989 	struct x86_exception e;
5990 	struct {
5991 		u64 eptp, gpa;
5992 	} operand;
5993 	int i, r, gpr_index;
5994 
5995 	if (!(vmx->nested.msrs.secondary_ctls_high &
5996 	      SECONDARY_EXEC_ENABLE_EPT) ||
5997 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5998 		kvm_queue_exception(vcpu, UD_VECTOR);
5999 		return 1;
6000 	}
6001 
6002 	if (!nested_vmx_check_permission(vcpu))
6003 		return 1;
6004 
6005 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6006 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
6007 	type = kvm_register_read(vcpu, gpr_index);
6008 
6009 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6010 
6011 	if (type >= 32 || !(types & (1 << type)))
6012 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6013 
6014 	/* According to the Intel VMX instruction reference, the memory
6015 	 * operand is read even if it isn't needed (e.g., for type==global)
6016 	 */
6017 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
6018 			vmx_instruction_info, false, sizeof(operand), &gva))
6019 		return 1;
6020 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
6021 	if (r != X86EMUL_CONTINUE)
6022 		return kvm_handle_memory_failure(vcpu, r, &e);
6023 
6024 	/*
6025 	 * Nested EPT roots are always held through guest_mmu,
6026 	 * not root_mmu.
6027 	 */
6028 	mmu = &vcpu->arch.guest_mmu;
6029 
6030 	switch (type) {
6031 	case VMX_EPT_EXTENT_CONTEXT:
6032 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
6033 			return nested_vmx_fail(vcpu,
6034 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6035 
6036 		roots_to_free = 0;
6037 		if (nested_ept_root_matches(mmu->root.hpa, mmu->root.pgd,
6038 					    operand.eptp))
6039 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
6040 
6041 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
6042 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
6043 						    mmu->prev_roots[i].pgd,
6044 						    operand.eptp))
6045 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
6046 		}
6047 		break;
6048 	case VMX_EPT_EXTENT_GLOBAL:
6049 		roots_to_free = KVM_MMU_ROOTS_ALL;
6050 		break;
6051 	default:
6052 		BUG();
6053 		break;
6054 	}
6055 
6056 	if (roots_to_free)
6057 		kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
6058 
6059 	return nested_vmx_succeed(vcpu);
6060 }
6061 
6062 static int handle_invvpid(struct kvm_vcpu *vcpu)
6063 {
6064 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6065 	u32 vmx_instruction_info;
6066 	unsigned long type, types;
6067 	gva_t gva;
6068 	struct x86_exception e;
6069 	struct {
6070 		u64 vpid;
6071 		u64 gla;
6072 	} operand;
6073 	u16 vpid02;
6074 	int r, gpr_index;
6075 
6076 	if (!(vmx->nested.msrs.secondary_ctls_high &
6077 	      SECONDARY_EXEC_ENABLE_VPID) ||
6078 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
6079 		kvm_queue_exception(vcpu, UD_VECTOR);
6080 		return 1;
6081 	}
6082 
6083 	if (!nested_vmx_check_permission(vcpu))
6084 		return 1;
6085 
6086 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6087 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
6088 	type = kvm_register_read(vcpu, gpr_index);
6089 
6090 	types = (vmx->nested.msrs.vpid_caps &
6091 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
6092 
6093 	if (type >= 32 || !(types & (1 << type)))
6094 		return nested_vmx_fail(vcpu,
6095 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6096 
6097 	/* according to the intel vmx instruction reference, the memory
6098 	 * operand is read even if it isn't needed (e.g., for type==global)
6099 	 */
6100 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
6101 			vmx_instruction_info, false, sizeof(operand), &gva))
6102 		return 1;
6103 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
6104 	if (r != X86EMUL_CONTINUE)
6105 		return kvm_handle_memory_failure(vcpu, r, &e);
6106 
6107 	if (operand.vpid >> 16)
6108 		return nested_vmx_fail(vcpu,
6109 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6110 
6111 	/*
6112 	 * Always flush the effective vpid02, i.e. never flush the current VPID
6113 	 * and never explicitly flush vpid01.  INVVPID targets a VPID, not a
6114 	 * VMCS, and so whether or not the current vmcs12 has VPID enabled is
6115 	 * irrelevant (and there may not be a loaded vmcs12).
6116 	 */
6117 	vpid02 = nested_get_vpid02(vcpu);
6118 	switch (type) {
6119 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
6120 		/*
6121 		 * LAM doesn't apply to addresses that are inputs to TLB
6122 		 * invalidation.
6123 		 */
6124 		if (!operand.vpid ||
6125 		    is_noncanonical_invlpg_address(operand.gla, vcpu))
6126 			return nested_vmx_fail(vcpu,
6127 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6128 		vpid_sync_vcpu_addr(vpid02, operand.gla);
6129 		break;
6130 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
6131 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
6132 		if (!operand.vpid)
6133 			return nested_vmx_fail(vcpu,
6134 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6135 		vpid_sync_context(vpid02);
6136 		break;
6137 	case VMX_VPID_EXTENT_ALL_CONTEXT:
6138 		vpid_sync_context(vpid02);
6139 		break;
6140 	default:
6141 		WARN_ON_ONCE(1);
6142 		return kvm_skip_emulated_instruction(vcpu);
6143 	}
6144 
6145 	/*
6146 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
6147 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
6148 	 * roots as VPIDs are not tracked in the MMU role.
6149 	 *
6150 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
6151 	 * an MMU when EPT is disabled.
6152 	 *
6153 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
6154 	 */
6155 	if (!enable_ept)
6156 		kvm_mmu_free_guest_mode_roots(vcpu->kvm, &vcpu->arch.root_mmu);
6157 
6158 	return nested_vmx_succeed(vcpu);
6159 }
6160 
6161 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
6162 				     struct vmcs12 *vmcs12)
6163 {
6164 	u32 index = kvm_rcx_read(vcpu);
6165 	u64 new_eptp;
6166 
6167 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
6168 		return 1;
6169 	if (index >= VMFUNC_EPTP_ENTRIES)
6170 		return 1;
6171 
6172 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
6173 				     &new_eptp, index * 8, 8))
6174 		return 1;
6175 
6176 	/*
6177 	 * If the (L2) guest does a vmfunc to the currently
6178 	 * active ept pointer, we don't have to do anything else
6179 	 */
6180 	if (vmcs12->ept_pointer != new_eptp) {
6181 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
6182 			return 1;
6183 
6184 		vmcs12->ept_pointer = new_eptp;
6185 		nested_ept_new_eptp(vcpu);
6186 
6187 		if (!nested_cpu_has_vpid(vmcs12))
6188 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
6189 	}
6190 
6191 	return 0;
6192 }
6193 
6194 static int handle_vmfunc(struct kvm_vcpu *vcpu)
6195 {
6196 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6197 	struct vmcs12 *vmcs12;
6198 	u32 function = kvm_rax_read(vcpu);
6199 
6200 	/*
6201 	 * VMFUNC should never execute cleanly while L1 is active; KVM supports
6202 	 * VMFUNC for nested VMs, but not for L1.
6203 	 */
6204 	if (WARN_ON_ONCE(!is_guest_mode(vcpu))) {
6205 		kvm_queue_exception(vcpu, UD_VECTOR);
6206 		return 1;
6207 	}
6208 
6209 	vmcs12 = get_vmcs12(vcpu);
6210 
6211 	/*
6212 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
6213 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
6214 	 */
6215 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
6216 		kvm_queue_exception(vcpu, UD_VECTOR);
6217 		return 1;
6218 	}
6219 
6220 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
6221 		goto fail;
6222 
6223 	switch (function) {
6224 	case 0:
6225 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
6226 			goto fail;
6227 		break;
6228 	default:
6229 		goto fail;
6230 	}
6231 	return kvm_skip_emulated_instruction(vcpu);
6232 
6233 fail:
6234 	/*
6235 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
6236 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
6237 	 * EXIT_REASON_VMFUNC as the exit reason.
6238 	 */
6239 	nested_vmx_vmexit(vcpu, vmx->vt.exit_reason.full,
6240 			  vmx_get_intr_info(vcpu),
6241 			  vmx_get_exit_qual(vcpu));
6242 	return 1;
6243 }
6244 
6245 /*
6246  * Return true if an IO instruction with the specified port and size should cause
6247  * a VM-exit into L1.
6248  */
6249 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
6250 				 int size)
6251 {
6252 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6253 	gpa_t bitmap, last_bitmap;
6254 	u8 b;
6255 
6256 	last_bitmap = INVALID_GPA;
6257 	b = -1;
6258 
6259 	while (size > 0) {
6260 		if (port < 0x8000)
6261 			bitmap = vmcs12->io_bitmap_a;
6262 		else if (port < 0x10000)
6263 			bitmap = vmcs12->io_bitmap_b;
6264 		else
6265 			return true;
6266 		bitmap += (port & 0x7fff) / 8;
6267 
6268 		if (last_bitmap != bitmap)
6269 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
6270 				return true;
6271 		if (b & (1 << (port & 7)))
6272 			return true;
6273 
6274 		port++;
6275 		size--;
6276 		last_bitmap = bitmap;
6277 	}
6278 
6279 	return false;
6280 }
6281 
6282 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6283 				       struct vmcs12 *vmcs12)
6284 {
6285 	unsigned long exit_qualification;
6286 	unsigned short port;
6287 	int size;
6288 
6289 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6290 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
6291 
6292 	exit_qualification = vmx_get_exit_qual(vcpu);
6293 
6294 	port = exit_qualification >> 16;
6295 	size = (exit_qualification & 7) + 1;
6296 
6297 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
6298 }
6299 
6300 /*
6301  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
6302  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6303  * disinterest in the current event (read or write a specific MSR) by using an
6304  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6305  */
6306 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6307 					struct vmcs12 *vmcs12,
6308 					union vmx_exit_reason exit_reason)
6309 {
6310 	u32 msr_index;
6311 	gpa_t bitmap;
6312 
6313 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6314 		return true;
6315 
6316 	if (exit_reason.basic == EXIT_REASON_MSR_READ_IMM ||
6317 	    exit_reason.basic == EXIT_REASON_MSR_WRITE_IMM)
6318 		msr_index = vmx_get_exit_qual(vcpu);
6319 	else
6320 		msr_index = kvm_rcx_read(vcpu);
6321 
6322 	/*
6323 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6324 	 * for the four combinations of read/write and low/high MSR numbers.
6325 	 * First we need to figure out which of the four to use:
6326 	 */
6327 	bitmap = vmcs12->msr_bitmap;
6328 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE ||
6329 	    exit_reason.basic == EXIT_REASON_MSR_WRITE_IMM)
6330 		bitmap += 2048;
6331 	if (msr_index >= 0xc0000000) {
6332 		msr_index -= 0xc0000000;
6333 		bitmap += 1024;
6334 	}
6335 
6336 	/* Then read the msr_index'th bit from this bitmap: */
6337 	if (msr_index < 1024*8) {
6338 		unsigned char b;
6339 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
6340 			return true;
6341 		return 1 & (b >> (msr_index & 7));
6342 	} else
6343 		return true; /* let L1 handle the wrong parameter */
6344 }
6345 
6346 /*
6347  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6348  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6349  * intercept (via guest_host_mask etc.) the current event.
6350  */
6351 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6352 	struct vmcs12 *vmcs12)
6353 {
6354 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
6355 	int cr = exit_qualification & 15;
6356 	int reg;
6357 	unsigned long val;
6358 
6359 	switch ((exit_qualification >> 4) & 3) {
6360 	case 0: /* mov to cr */
6361 		reg = (exit_qualification >> 8) & 15;
6362 		val = kvm_register_read(vcpu, reg);
6363 		switch (cr) {
6364 		case 0:
6365 			if (vmcs12->cr0_guest_host_mask &
6366 			    (val ^ vmcs12->cr0_read_shadow))
6367 				return true;
6368 			break;
6369 		case 3:
6370 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6371 				return true;
6372 			break;
6373 		case 4:
6374 			if (vmcs12->cr4_guest_host_mask &
6375 			    (vmcs12->cr4_read_shadow ^ val))
6376 				return true;
6377 			break;
6378 		case 8:
6379 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6380 				return true;
6381 			break;
6382 		}
6383 		break;
6384 	case 2: /* clts */
6385 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6386 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
6387 			return true;
6388 		break;
6389 	case 1: /* mov from cr */
6390 		switch (cr) {
6391 		case 3:
6392 			if (vmcs12->cpu_based_vm_exec_control &
6393 			    CPU_BASED_CR3_STORE_EXITING)
6394 				return true;
6395 			break;
6396 		case 8:
6397 			if (vmcs12->cpu_based_vm_exec_control &
6398 			    CPU_BASED_CR8_STORE_EXITING)
6399 				return true;
6400 			break;
6401 		}
6402 		break;
6403 	case 3: /* lmsw */
6404 		/*
6405 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6406 		 * cr0. Other attempted changes are ignored, with no exit.
6407 		 */
6408 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6409 		if (vmcs12->cr0_guest_host_mask & 0xe &
6410 		    (val ^ vmcs12->cr0_read_shadow))
6411 			return true;
6412 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6413 		    !(vmcs12->cr0_read_shadow & 0x1) &&
6414 		    (val & 0x1))
6415 			return true;
6416 		break;
6417 	}
6418 	return false;
6419 }
6420 
6421 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
6422 					  struct vmcs12 *vmcs12)
6423 {
6424 	u32 encls_leaf;
6425 
6426 	if (!guest_cpu_cap_has(vcpu, X86_FEATURE_SGX) ||
6427 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
6428 		return false;
6429 
6430 	encls_leaf = kvm_rax_read(vcpu);
6431 	if (encls_leaf > 62)
6432 		encls_leaf = 63;
6433 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
6434 }
6435 
6436 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
6437 	struct vmcs12 *vmcs12, gpa_t bitmap)
6438 {
6439 	u32 vmx_instruction_info;
6440 	unsigned long field;
6441 	u8 b;
6442 
6443 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
6444 		return true;
6445 
6446 	/* Decode instruction info and find the field to access */
6447 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6448 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6449 
6450 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
6451 	if (field >> 15)
6452 		return true;
6453 
6454 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
6455 		return true;
6456 
6457 	return 1 & (b >> (field & 7));
6458 }
6459 
6460 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
6461 {
6462 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
6463 
6464 	if (nested_cpu_has_mtf(vmcs12))
6465 		return true;
6466 
6467 	/*
6468 	 * An MTF VM-exit may be injected into the guest by setting the
6469 	 * interruption-type to 7 (other event) and the vector field to 0. Such
6470 	 * is the case regardless of the 'monitor trap flag' VM-execution
6471 	 * control.
6472 	 */
6473 	return entry_intr_info == (INTR_INFO_VALID_MASK
6474 				   | INTR_TYPE_OTHER_EVENT);
6475 }
6476 
6477 /*
6478  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
6479  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
6480  */
6481 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
6482 				     union vmx_exit_reason exit_reason)
6483 {
6484 	u32 intr_info;
6485 
6486 	switch ((u16)exit_reason.basic) {
6487 	case EXIT_REASON_EXCEPTION_NMI:
6488 		intr_info = vmx_get_intr_info(vcpu);
6489 		if (is_nmi(intr_info))
6490 			return true;
6491 		else if (is_page_fault(intr_info))
6492 			return vcpu->arch.apf.host_apf_flags ||
6493 			       vmx_need_pf_intercept(vcpu);
6494 		else if (is_debug(intr_info) &&
6495 			 vcpu->guest_debug &
6496 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6497 			return true;
6498 		else if (is_breakpoint(intr_info) &&
6499 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6500 			return true;
6501 		else if (is_alignment_check(intr_info) &&
6502 			 !vmx_guest_inject_ac(vcpu))
6503 			return true;
6504 		else if (is_ve_fault(intr_info))
6505 			return true;
6506 		return false;
6507 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6508 		return true;
6509 	case EXIT_REASON_MCE_DURING_VMENTRY:
6510 		return true;
6511 	case EXIT_REASON_EPT_VIOLATION:
6512 		/*
6513 		 * L0 always deals with the EPT violation. If nested EPT is
6514 		 * used, and the nested mmu code discovers that the address is
6515 		 * missing in the guest EPT table (EPT12), the EPT violation
6516 		 * will be injected with nested_ept_inject_page_fault()
6517 		 */
6518 		return true;
6519 	case EXIT_REASON_EPT_MISCONFIG:
6520 		/*
6521 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
6522 		 * table (shadow on EPT) or a merged EPT table that L0 built
6523 		 * (EPT on EPT). So any problems with the structure of the
6524 		 * table is L0's fault.
6525 		 */
6526 		return true;
6527 	case EXIT_REASON_PREEMPTION_TIMER:
6528 		return true;
6529 	case EXIT_REASON_PML_FULL:
6530 		/*
6531 		 * PML is emulated for an L1 VMM and should never be enabled in
6532 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
6533 		 */
6534 		return true;
6535 	case EXIT_REASON_VMFUNC:
6536 		/* VM functions are emulated through L2->L0 vmexits. */
6537 		return true;
6538 	case EXIT_REASON_BUS_LOCK:
6539 		/*
6540 		 * At present, bus lock VM exit is never exposed to L1.
6541 		 * Handle L2's bus locks in L0 directly.
6542 		 */
6543 		return true;
6544 #ifdef CONFIG_KVM_HYPERV
6545 	case EXIT_REASON_VMCALL:
6546 		/* Hyper-V L2 TLB flush hypercall is handled by L0 */
6547 		return guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
6548 			nested_evmcs_l2_tlb_flush_enabled(vcpu) &&
6549 			kvm_hv_is_tlb_flush_hcall(vcpu);
6550 #endif
6551 	default:
6552 		break;
6553 	}
6554 	return false;
6555 }
6556 
6557 /*
6558  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
6559  * is_guest_mode (L2).
6560  */
6561 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
6562 				     union vmx_exit_reason exit_reason)
6563 {
6564 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6565 	u32 intr_info;
6566 
6567 	switch ((u16)exit_reason.basic) {
6568 	case EXIT_REASON_EXCEPTION_NMI:
6569 		intr_info = vmx_get_intr_info(vcpu);
6570 		if (is_nmi(intr_info))
6571 			return true;
6572 		else if (is_page_fault(intr_info))
6573 			return true;
6574 		return vmcs12->exception_bitmap &
6575 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
6576 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6577 		return nested_exit_on_intr(vcpu);
6578 	case EXIT_REASON_TRIPLE_FAULT:
6579 		return true;
6580 	case EXIT_REASON_INTERRUPT_WINDOW:
6581 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
6582 	case EXIT_REASON_NMI_WINDOW:
6583 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
6584 	case EXIT_REASON_TASK_SWITCH:
6585 		return true;
6586 	case EXIT_REASON_CPUID:
6587 		return true;
6588 	case EXIT_REASON_HLT:
6589 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6590 	case EXIT_REASON_INVD:
6591 		return true;
6592 	case EXIT_REASON_INVLPG:
6593 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6594 	case EXIT_REASON_RDPMC:
6595 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6596 	case EXIT_REASON_RDRAND:
6597 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
6598 	case EXIT_REASON_RDSEED:
6599 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
6600 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
6601 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6602 	case EXIT_REASON_VMREAD:
6603 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6604 			vmcs12->vmread_bitmap);
6605 	case EXIT_REASON_VMWRITE:
6606 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6607 			vmcs12->vmwrite_bitmap);
6608 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6609 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6610 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
6611 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6612 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
6613 		/*
6614 		 * VMX instructions trap unconditionally. This allows L1 to
6615 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6616 		 */
6617 		return true;
6618 	case EXIT_REASON_CR_ACCESS:
6619 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6620 	case EXIT_REASON_DR_ACCESS:
6621 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6622 	case EXIT_REASON_IO_INSTRUCTION:
6623 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
6624 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6625 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6626 	case EXIT_REASON_MSR_READ:
6627 	case EXIT_REASON_MSR_WRITE:
6628 	case EXIT_REASON_MSR_READ_IMM:
6629 	case EXIT_REASON_MSR_WRITE_IMM:
6630 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6631 	case EXIT_REASON_INVALID_STATE:
6632 		return true;
6633 	case EXIT_REASON_MWAIT_INSTRUCTION:
6634 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6635 	case EXIT_REASON_MONITOR_TRAP_FLAG:
6636 		return nested_vmx_exit_handled_mtf(vmcs12);
6637 	case EXIT_REASON_MONITOR_INSTRUCTION:
6638 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6639 	case EXIT_REASON_PAUSE_INSTRUCTION:
6640 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6641 			nested_cpu_has2(vmcs12,
6642 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6643 	case EXIT_REASON_MCE_DURING_VMENTRY:
6644 		return true;
6645 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6646 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6647 	case EXIT_REASON_APIC_ACCESS:
6648 	case EXIT_REASON_APIC_WRITE:
6649 	case EXIT_REASON_EOI_INDUCED:
6650 		/*
6651 		 * The controls for "virtualize APIC accesses," "APIC-
6652 		 * register virtualization," and "virtual-interrupt
6653 		 * delivery" only come from vmcs12.
6654 		 */
6655 		return true;
6656 	case EXIT_REASON_INVPCID:
6657 		return
6658 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6659 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6660 	case EXIT_REASON_WBINVD:
6661 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6662 	case EXIT_REASON_XSETBV:
6663 		return true;
6664 	case EXIT_REASON_XSAVES:
6665 	case EXIT_REASON_XRSTORS:
6666 		/*
6667 		 * Always forward XSAVES/XRSTORS to L1 as KVM doesn't utilize
6668 		 * XSS-bitmap, and always loads vmcs02 with vmcs12's XSS-bitmap
6669 		 * verbatim, i.e. any exit is due to L1's bitmap.  WARN if
6670 		 * XSAVES isn't enabled, as the CPU is supposed to inject #UD
6671 		 * in that case, before consulting the XSS-bitmap.
6672 		 */
6673 		WARN_ON_ONCE(!nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_XSAVES));
6674 		return true;
6675 	case EXIT_REASON_UMWAIT:
6676 	case EXIT_REASON_TPAUSE:
6677 		return nested_cpu_has2(vmcs12,
6678 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6679 	case EXIT_REASON_ENCLS:
6680 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6681 	case EXIT_REASON_NOTIFY:
6682 		/* Notify VM exit is not exposed to L1 */
6683 		return false;
6684 	case EXIT_REASON_SEAMCALL:
6685 	case EXIT_REASON_TDCALL:
6686 		/*
6687 		 * SEAMCALL and TDCALL unconditionally VM-Exit, but aren't
6688 		 * virtualized by KVM for L1 hypervisors, i.e. L1 should
6689 		 * never want or expect such an exit.
6690 		 */
6691 		return false;
6692 	default:
6693 		return true;
6694 	}
6695 }
6696 
6697 /*
6698  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6699  * reflected into L1.
6700  */
6701 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6702 {
6703 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6704 	union vmx_exit_reason exit_reason = vmx->vt.exit_reason;
6705 	unsigned long exit_qual;
6706 	u32 exit_intr_info;
6707 
6708 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6709 
6710 	/*
6711 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6712 	 * has already loaded L2's state.
6713 	 */
6714 	if (unlikely(vmx->fail)) {
6715 		trace_kvm_nested_vmenter_failed(
6716 			"hardware VM-instruction error: ",
6717 			vmcs_read32(VM_INSTRUCTION_ERROR));
6718 		exit_intr_info = 0;
6719 		exit_qual = 0;
6720 		goto reflect_vmexit;
6721 	}
6722 
6723 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6724 
6725 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6726 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6727 		return false;
6728 
6729 	/* If L1 doesn't want the exit, handle it in L0. */
6730 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6731 		return false;
6732 
6733 	/*
6734 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6735 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6736 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6737 	 * interrupts are never reflected to L1 so it's a non-issue.
6738 	 */
6739 	exit_intr_info = vmx_get_intr_info(vcpu);
6740 	if (is_exception_with_error_code(exit_intr_info)) {
6741 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6742 
6743 		vmcs12->vm_exit_intr_error_code =
6744 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6745 	}
6746 	exit_qual = vmx_get_exit_qual(vcpu);
6747 
6748 reflect_vmexit:
6749 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6750 	return true;
6751 }
6752 
6753 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6754 				struct kvm_nested_state __user *user_kvm_nested_state,
6755 				u32 user_data_size)
6756 {
6757 	struct vcpu_vmx *vmx;
6758 	struct vmcs12 *vmcs12;
6759 	struct kvm_nested_state kvm_state = {
6760 		.flags = 0,
6761 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6762 		.size = sizeof(kvm_state),
6763 		.hdr.vmx.flags = 0,
6764 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6765 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6766 		.hdr.vmx.preemption_timer_deadline = 0,
6767 	};
6768 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6769 		&user_kvm_nested_state->data.vmx[0];
6770 
6771 	if (!vcpu)
6772 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6773 
6774 	vmx = to_vmx(vcpu);
6775 	vmcs12 = get_vmcs12(vcpu);
6776 
6777 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_VMX) &&
6778 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6779 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6780 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6781 
6782 		if (vmx_has_valid_vmcs12(vcpu)) {
6783 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6784 
6785 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6786 			if (nested_vmx_is_evmptr12_set(vmx))
6787 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6788 
6789 			if (is_guest_mode(vcpu) &&
6790 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6791 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6792 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6793 		}
6794 
6795 		if (vmx->nested.smm.vmxon)
6796 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6797 
6798 		if (vmx->nested.smm.guest_mode)
6799 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6800 
6801 		if (is_guest_mode(vcpu)) {
6802 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6803 
6804 			if (vmx->nested.nested_run_pending)
6805 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6806 
6807 			if (vmx->nested.mtf_pending)
6808 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6809 
6810 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6811 			    vmx->nested.has_preemption_timer_deadline) {
6812 				kvm_state.hdr.vmx.flags |=
6813 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6814 				kvm_state.hdr.vmx.preemption_timer_deadline =
6815 					vmx->nested.preemption_timer_deadline;
6816 			}
6817 		}
6818 	}
6819 
6820 	if (user_data_size < kvm_state.size)
6821 		goto out;
6822 
6823 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6824 		return -EFAULT;
6825 
6826 	if (!vmx_has_valid_vmcs12(vcpu))
6827 		goto out;
6828 
6829 	/*
6830 	 * When running L2, the authoritative vmcs12 state is in the
6831 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6832 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6833 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6834 	 * vmcs12 state is in the vmcs12 already.
6835 	 */
6836 	if (is_guest_mode(vcpu)) {
6837 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6838 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6839 	} else  {
6840 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6841 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6842 			if (nested_vmx_is_evmptr12_valid(vmx))
6843 				/*
6844 				 * L1 hypervisor is not obliged to keep eVMCS
6845 				 * clean fields data always up-to-date while
6846 				 * not in guest mode, 'hv_clean_fields' is only
6847 				 * supposed to be actual upon vmentry so we need
6848 				 * to ignore it here and do full copy.
6849 				 */
6850 				copy_enlightened_to_vmcs12(vmx, 0);
6851 			else if (enable_shadow_vmcs)
6852 				copy_shadow_to_vmcs12(vmx);
6853 		}
6854 	}
6855 
6856 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6857 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6858 
6859 	/*
6860 	 * Copy over the full allocated size of vmcs12 rather than just the size
6861 	 * of the struct.
6862 	 */
6863 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6864 		return -EFAULT;
6865 
6866 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6867 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6868 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6869 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6870 			return -EFAULT;
6871 	}
6872 out:
6873 	return kvm_state.size;
6874 }
6875 
6876 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6877 {
6878 	if (is_guest_mode(vcpu)) {
6879 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6880 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6881 	}
6882 	free_nested(vcpu);
6883 }
6884 
6885 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6886 				struct kvm_nested_state __user *user_kvm_nested_state,
6887 				struct kvm_nested_state *kvm_state)
6888 {
6889 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6890 	struct vmcs12 *vmcs12;
6891 	enum vm_entry_failure_code ignored;
6892 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6893 		&user_kvm_nested_state->data.vmx[0];
6894 	int ret;
6895 
6896 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6897 		return -EINVAL;
6898 
6899 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6900 		if (kvm_state->hdr.vmx.smm.flags)
6901 			return -EINVAL;
6902 
6903 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6904 			return -EINVAL;
6905 
6906 		/*
6907 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6908 		 * enable eVMCS capability on vCPU. However, since then
6909 		 * code was changed such that flag signals vmcs12 should
6910 		 * be copied into eVMCS in guest memory.
6911 		 *
6912 		 * To preserve backwards compatibility, allow user
6913 		 * to set this flag even when there is no VMXON region.
6914 		 */
6915 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6916 			return -EINVAL;
6917 	} else {
6918 		if (!guest_cpu_cap_has(vcpu, X86_FEATURE_VMX))
6919 			return -EINVAL;
6920 
6921 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6922 			return -EINVAL;
6923 	}
6924 
6925 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6926 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6927 		return -EINVAL;
6928 
6929 	if (kvm_state->hdr.vmx.smm.flags &
6930 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6931 		return -EINVAL;
6932 
6933 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6934 		return -EINVAL;
6935 
6936 	/*
6937 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6938 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6939 	 * must be zero.
6940 	 */
6941 	if (is_smm(vcpu) ?
6942 		(kvm_state->flags &
6943 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6944 		: kvm_state->hdr.vmx.smm.flags)
6945 		return -EINVAL;
6946 
6947 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6948 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6949 		return -EINVAL;
6950 
6951 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6952 	    (!guest_cpu_cap_has(vcpu, X86_FEATURE_VMX) ||
6953 	     !vmx->nested.enlightened_vmcs_enabled))
6954 			return -EINVAL;
6955 
6956 	vmx_leave_nested(vcpu);
6957 
6958 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6959 		return 0;
6960 
6961 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6962 	ret = enter_vmx_operation(vcpu);
6963 	if (ret)
6964 		return ret;
6965 
6966 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6967 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6968 		/* See vmx_has_valid_vmcs12.  */
6969 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6970 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6971 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6972 			return -EINVAL;
6973 		else
6974 			return 0;
6975 	}
6976 
6977 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6978 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6979 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6980 			return -EINVAL;
6981 
6982 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6983 #ifdef CONFIG_KVM_HYPERV
6984 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6985 		/*
6986 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6987 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6988 		 * restored yet. EVMCS will be mapped from
6989 		 * nested_get_vmcs12_pages().
6990 		 */
6991 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6992 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6993 #endif
6994 	} else {
6995 		return -EINVAL;
6996 	}
6997 
6998 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6999 		vmx->nested.smm.vmxon = true;
7000 		vmx->nested.vmxon = false;
7001 
7002 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
7003 			vmx->nested.smm.guest_mode = true;
7004 	}
7005 
7006 	vmcs12 = get_vmcs12(vcpu);
7007 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
7008 		return -EFAULT;
7009 
7010 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
7011 		return -EINVAL;
7012 
7013 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
7014 		return 0;
7015 
7016 	vmx->nested.nested_run_pending =
7017 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
7018 
7019 	vmx->nested.mtf_pending =
7020 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
7021 
7022 	ret = -EINVAL;
7023 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
7024 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
7025 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
7026 
7027 		if (kvm_state->size <
7028 		    sizeof(*kvm_state) +
7029 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
7030 			goto error_guest_mode;
7031 
7032 		if (copy_from_user(shadow_vmcs12,
7033 				   user_vmx_nested_state->shadow_vmcs12,
7034 				   sizeof(*shadow_vmcs12))) {
7035 			ret = -EFAULT;
7036 			goto error_guest_mode;
7037 		}
7038 
7039 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
7040 		    !shadow_vmcs12->hdr.shadow_vmcs)
7041 			goto error_guest_mode;
7042 	}
7043 
7044 	vmx->nested.has_preemption_timer_deadline = false;
7045 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
7046 		vmx->nested.has_preemption_timer_deadline = true;
7047 		vmx->nested.preemption_timer_deadline =
7048 			kvm_state->hdr.vmx.preemption_timer_deadline;
7049 	}
7050 
7051 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
7052 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
7053 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
7054 		goto error_guest_mode;
7055 
7056 	vmx->nested.dirty_vmcs12 = true;
7057 	vmx->nested.force_msr_bitmap_recalc = true;
7058 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
7059 	if (ret)
7060 		goto error_guest_mode;
7061 
7062 	if (vmx->nested.mtf_pending)
7063 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7064 
7065 	return 0;
7066 
7067 error_guest_mode:
7068 	vmx->nested.nested_run_pending = 0;
7069 	return ret;
7070 }
7071 
7072 void nested_vmx_set_vmcs_shadowing_bitmap(void)
7073 {
7074 	if (enable_shadow_vmcs) {
7075 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
7076 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
7077 	}
7078 }
7079 
7080 static u64 nested_vmx_calc_vmcs_enum_msr(void)
7081 {
7082 	/*
7083 	 * Note these are the so called "index" of the VMCS field encoding, not
7084 	 * the index into vmcs12.
7085 	 */
7086 	unsigned int max_idx, idx;
7087 	int i;
7088 
7089 	/*
7090 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
7091 	 * vmcs12, regardless of whether or not the associated feature is
7092 	 * exposed to L1.  Simply find the field with the highest index.
7093 	 */
7094 	max_idx = 0;
7095 	for (i = 0; i < nr_vmcs12_fields; i++) {
7096 		/* The vmcs12 table is very, very sparsely populated. */
7097 		if (!vmcs12_field_offsets[i])
7098 			continue;
7099 
7100 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
7101 		if (idx > max_idx)
7102 			max_idx = idx;
7103 	}
7104 
7105 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
7106 }
7107 
7108 static void nested_vmx_setup_pinbased_ctls(struct vmcs_config *vmcs_conf,
7109 					   struct nested_vmx_msrs *msrs)
7110 {
7111 	msrs->pinbased_ctls_low =
7112 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
7113 
7114 	msrs->pinbased_ctls_high = vmcs_conf->pin_based_exec_ctrl;
7115 	msrs->pinbased_ctls_high &=
7116 		PIN_BASED_EXT_INTR_MASK |
7117 		PIN_BASED_NMI_EXITING |
7118 		PIN_BASED_VIRTUAL_NMIS |
7119 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
7120 	msrs->pinbased_ctls_high |=
7121 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
7122 		PIN_BASED_VMX_PREEMPTION_TIMER;
7123 }
7124 
7125 static void nested_vmx_setup_exit_ctls(struct vmcs_config *vmcs_conf,
7126 				       struct nested_vmx_msrs *msrs)
7127 {
7128 	msrs->exit_ctls_low =
7129 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
7130 
7131 	msrs->exit_ctls_high = vmcs_conf->vmexit_ctrl;
7132 	msrs->exit_ctls_high &=
7133 #ifdef CONFIG_X86_64
7134 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
7135 #endif
7136 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
7137 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_CET_STATE;
7138 	msrs->exit_ctls_high |=
7139 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
7140 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
7141 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT |
7142 		VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
7143 
7144 	if (!kvm_cpu_cap_has(X86_FEATURE_SHSTK) &&
7145 	    !kvm_cpu_cap_has(X86_FEATURE_IBT))
7146 		msrs->exit_ctls_high &= ~VM_EXIT_LOAD_CET_STATE;
7147 
7148 	/* We support free control of debug control saving. */
7149 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
7150 }
7151 
7152 static void nested_vmx_setup_entry_ctls(struct vmcs_config *vmcs_conf,
7153 					struct nested_vmx_msrs *msrs)
7154 {
7155 	msrs->entry_ctls_low =
7156 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
7157 
7158 	msrs->entry_ctls_high = vmcs_conf->vmentry_ctrl;
7159 	msrs->entry_ctls_high &=
7160 #ifdef CONFIG_X86_64
7161 		VM_ENTRY_IA32E_MODE |
7162 #endif
7163 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
7164 		VM_ENTRY_LOAD_CET_STATE;
7165 	msrs->entry_ctls_high |=
7166 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER |
7167 		 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
7168 
7169 	if (!kvm_cpu_cap_has(X86_FEATURE_SHSTK) &&
7170 	    !kvm_cpu_cap_has(X86_FEATURE_IBT))
7171 		msrs->entry_ctls_high &= ~VM_ENTRY_LOAD_CET_STATE;
7172 
7173 	/* We support free control of debug control loading. */
7174 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
7175 }
7176 
7177 static void nested_vmx_setup_cpubased_ctls(struct vmcs_config *vmcs_conf,
7178 					   struct nested_vmx_msrs *msrs)
7179 {
7180 	msrs->procbased_ctls_low =
7181 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
7182 
7183 	msrs->procbased_ctls_high = vmcs_conf->cpu_based_exec_ctrl;
7184 	msrs->procbased_ctls_high &=
7185 		CPU_BASED_INTR_WINDOW_EXITING |
7186 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
7187 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
7188 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
7189 		CPU_BASED_CR3_STORE_EXITING |
7190 #ifdef CONFIG_X86_64
7191 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
7192 #endif
7193 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
7194 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
7195 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
7196 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
7197 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
7198 	/*
7199 	 * We can allow some features even when not supported by the
7200 	 * hardware. For example, L1 can specify an MSR bitmap - and we
7201 	 * can use it to avoid exits to L1 - even when L0 runs L2
7202 	 * without MSR bitmaps.
7203 	 */
7204 	msrs->procbased_ctls_high |=
7205 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
7206 		CPU_BASED_USE_MSR_BITMAPS;
7207 
7208 	/* We support free control of CR3 access interception. */
7209 	msrs->procbased_ctls_low &=
7210 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
7211 }
7212 
7213 static void nested_vmx_setup_secondary_ctls(u32 ept_caps,
7214 					    struct vmcs_config *vmcs_conf,
7215 					    struct nested_vmx_msrs *msrs)
7216 {
7217 	msrs->secondary_ctls_low = 0;
7218 
7219 	msrs->secondary_ctls_high = vmcs_conf->cpu_based_2nd_exec_ctrl;
7220 	msrs->secondary_ctls_high &=
7221 		SECONDARY_EXEC_DESC |
7222 		SECONDARY_EXEC_ENABLE_RDTSCP |
7223 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7224 		SECONDARY_EXEC_WBINVD_EXITING |
7225 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
7226 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
7227 		SECONDARY_EXEC_RDRAND_EXITING |
7228 		SECONDARY_EXEC_ENABLE_INVPCID |
7229 		SECONDARY_EXEC_ENABLE_VMFUNC |
7230 		SECONDARY_EXEC_RDSEED_EXITING |
7231 		SECONDARY_EXEC_ENABLE_XSAVES |
7232 		SECONDARY_EXEC_TSC_SCALING |
7233 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
7234 
7235 	/*
7236 	 * We can emulate "VMCS shadowing," even if the hardware
7237 	 * doesn't support it.
7238 	 */
7239 	msrs->secondary_ctls_high |=
7240 		SECONDARY_EXEC_SHADOW_VMCS;
7241 
7242 	if (enable_ept) {
7243 		/* nested EPT: emulate EPT also to L1 */
7244 		msrs->secondary_ctls_high |=
7245 			SECONDARY_EXEC_ENABLE_EPT;
7246 		msrs->ept_caps =
7247 			VMX_EPT_PAGE_WALK_4_BIT |
7248 			VMX_EPT_PAGE_WALK_5_BIT |
7249 			VMX_EPTP_WB_BIT |
7250 			VMX_EPT_INVEPT_BIT |
7251 			VMX_EPT_EXECUTE_ONLY_BIT;
7252 
7253 		msrs->ept_caps &= ept_caps;
7254 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
7255 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
7256 			VMX_EPT_1GB_PAGE_BIT;
7257 		if (enable_ept_ad_bits) {
7258 			msrs->secondary_ctls_high |=
7259 				SECONDARY_EXEC_ENABLE_PML;
7260 			msrs->ept_caps |= VMX_EPT_AD_BIT;
7261 		}
7262 
7263 		/*
7264 		 * Advertise EPTP switching irrespective of hardware support,
7265 		 * KVM emulates it in software so long as VMFUNC is supported.
7266 		 */
7267 		if (cpu_has_vmx_vmfunc())
7268 			msrs->vmfunc_controls = VMX_VMFUNC_EPTP_SWITCHING;
7269 	}
7270 
7271 	/*
7272 	 * Old versions of KVM use the single-context version without
7273 	 * checking for support, so declare that it is supported even
7274 	 * though it is treated as global context.  The alternative is
7275 	 * not failing the single-context invvpid, and it is worse.
7276 	 */
7277 	if (enable_vpid) {
7278 		msrs->secondary_ctls_high |=
7279 			SECONDARY_EXEC_ENABLE_VPID;
7280 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
7281 			VMX_VPID_EXTENT_SUPPORTED_MASK;
7282 	}
7283 
7284 	if (enable_unrestricted_guest)
7285 		msrs->secondary_ctls_high |=
7286 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
7287 
7288 	if (flexpriority_enabled)
7289 		msrs->secondary_ctls_high |=
7290 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7291 
7292 	if (enable_sgx)
7293 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
7294 }
7295 
7296 static void nested_vmx_setup_misc_data(struct vmcs_config *vmcs_conf,
7297 				       struct nested_vmx_msrs *msrs)
7298 {
7299 	msrs->misc_low = (u32)vmcs_conf->misc & VMX_MISC_SAVE_EFER_LMA;
7300 	msrs->misc_low |=
7301 		VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
7302 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
7303 		VMX_MISC_ACTIVITY_HLT |
7304 		VMX_MISC_ACTIVITY_WAIT_SIPI;
7305 	msrs->misc_high = 0;
7306 }
7307 
7308 static void nested_vmx_setup_basic(struct nested_vmx_msrs *msrs)
7309 {
7310 	/*
7311 	 * This MSR reports some information about VMX support. We
7312 	 * should return information about the VMX we emulate for the
7313 	 * guest, and the VMCS structure we give it - not about the
7314 	 * VMX support of the underlying hardware.
7315 	 */
7316 	msrs->basic = vmx_basic_encode_vmcs_info(VMCS12_REVISION, VMCS12_SIZE,
7317 						 X86_MEMTYPE_WB);
7318 
7319 	msrs->basic |= VMX_BASIC_TRUE_CTLS;
7320 	if (cpu_has_vmx_basic_inout())
7321 		msrs->basic |= VMX_BASIC_INOUT;
7322 	if (cpu_has_vmx_basic_no_hw_errcode_cc())
7323 		msrs->basic |= VMX_BASIC_NO_HW_ERROR_CODE_CC;
7324 }
7325 
7326 static void nested_vmx_setup_cr_fixed(struct nested_vmx_msrs *msrs)
7327 {
7328 	/*
7329 	 * These MSRs specify bits which the guest must keep fixed on
7330 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
7331 	 * We picked the standard core2 setting.
7332 	 */
7333 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
7334 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
7335 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
7336 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
7337 
7338 	/* These MSRs specify bits which the guest must keep fixed off. */
7339 	rdmsrq(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
7340 	rdmsrq(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
7341 
7342 	if (vmx_umip_emulated())
7343 		msrs->cr4_fixed1 |= X86_CR4_UMIP;
7344 }
7345 
7346 /*
7347  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
7348  * returned for the various VMX controls MSRs when nested VMX is enabled.
7349  * The same values should also be used to verify that vmcs12 control fields are
7350  * valid during nested entry from L1 to L2.
7351  * Each of these control msrs has a low and high 32-bit half: A low bit is on
7352  * if the corresponding bit in the (32-bit) control field *must* be on, and a
7353  * bit in the high half is on if the corresponding bit in the control field
7354  * may be on. See also vmx_control_verify().
7355  */
7356 void nested_vmx_setup_ctls_msrs(struct vmcs_config *vmcs_conf, u32 ept_caps)
7357 {
7358 	struct nested_vmx_msrs *msrs = &vmcs_conf->nested;
7359 
7360 	/*
7361 	 * Note that as a general rule, the high half of the MSRs (bits in
7362 	 * the control fields which may be 1) should be initialized by the
7363 	 * intersection of the underlying hardware's MSR (i.e., features which
7364 	 * can be supported) and the list of features we want to expose -
7365 	 * because they are known to be properly supported in our code.
7366 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
7367 	 * be set to 0, meaning that L1 may turn off any of these bits. The
7368 	 * reason is that if one of these bits is necessary, it will appear
7369 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
7370 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
7371 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
7372 	 * These rules have exceptions below.
7373 	 */
7374 	nested_vmx_setup_pinbased_ctls(vmcs_conf, msrs);
7375 
7376 	nested_vmx_setup_exit_ctls(vmcs_conf, msrs);
7377 
7378 	nested_vmx_setup_entry_ctls(vmcs_conf, msrs);
7379 
7380 	nested_vmx_setup_cpubased_ctls(vmcs_conf, msrs);
7381 
7382 	nested_vmx_setup_secondary_ctls(ept_caps, vmcs_conf, msrs);
7383 
7384 	nested_vmx_setup_misc_data(vmcs_conf, msrs);
7385 
7386 	nested_vmx_setup_basic(msrs);
7387 
7388 	nested_vmx_setup_cr_fixed(msrs);
7389 
7390 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
7391 }
7392 
7393 void nested_vmx_hardware_unsetup(void)
7394 {
7395 	int i;
7396 
7397 	if (enable_shadow_vmcs) {
7398 		for (i = 0; i < VMX_BITMAP_NR; i++)
7399 			free_page((unsigned long)vmx_bitmap[i]);
7400 	}
7401 }
7402 
7403 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
7404 {
7405 	int i;
7406 
7407 	/*
7408 	 * Note!  The set of supported vmcs12 fields is consumed by both VMX
7409 	 * MSR and shadow VMCS setup.
7410 	 */
7411 	nested_vmx_setup_vmcs12_fields();
7412 
7413 	nested_vmx_setup_ctls_msrs(&vmcs_config, vmx_capability.ept);
7414 
7415 	if (!cpu_has_vmx_shadow_vmcs())
7416 		enable_shadow_vmcs = 0;
7417 	if (enable_shadow_vmcs) {
7418 		for (i = 0; i < VMX_BITMAP_NR; i++) {
7419 			/*
7420 			 * The vmx_bitmap is not tied to a VM and so should
7421 			 * not be charged to a memcg.
7422 			 */
7423 			vmx_bitmap[i] = (unsigned long *)
7424 				__get_free_page(GFP_KERNEL);
7425 			if (!vmx_bitmap[i]) {
7426 				nested_vmx_hardware_unsetup();
7427 				return -ENOMEM;
7428 			}
7429 		}
7430 
7431 		init_vmcs_shadow_fields();
7432 	}
7433 
7434 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
7435 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
7436 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
7437 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
7438 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
7439 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
7440 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
7441 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmxoff;
7442 	exit_handlers[EXIT_REASON_VMON]		= handle_vmxon;
7443 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
7444 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
7445 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
7446 
7447 	return 0;
7448 }
7449 
7450 struct kvm_x86_nested_ops vmx_nested_ops = {
7451 	.leave_nested = vmx_leave_nested,
7452 	.is_exception_vmexit = nested_vmx_is_exception_vmexit,
7453 	.check_events = vmx_check_nested_events,
7454 	.has_events = vmx_has_nested_events,
7455 	.triple_fault = nested_vmx_triple_fault,
7456 	.get_state = vmx_get_nested_state,
7457 	.set_state = vmx_set_nested_state,
7458 	.get_nested_state_pages = vmx_get_nested_state_pages,
7459 	.write_log_dirty = nested_vmx_write_pml_buffer,
7460 #ifdef CONFIG_KVM_HYPERV
7461 	.enable_evmcs = nested_enable_evmcs,
7462 	.get_evmcs_version = nested_get_evmcs_version,
7463 	.hv_inject_synthetic_vmexit_post_tlb_flush = vmx_hv_inject_synthetic_vmexit_post_tlb_flush,
7464 #endif
7465 };
7466