xref: /linux/arch/x86/kvm/vmx/nested.c (revision 0de1020f7bbb3e1c9cd5b6f3eb4bdd661b1ff735)
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 	kvm_clear_available_registers(vcpu, 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 	kvm_reset_dirty_registers(vcpu);
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 					 bool from_hardware)
416 {
417 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
418 	struct vcpu_vmx *vmx = to_vmx(vcpu);
419 	unsigned long exit_qualification;
420 	u32 vm_exit_reason;
421 
422 	if (vmx->nested.pml_full) {
423 		vm_exit_reason = EXIT_REASON_PML_FULL;
424 		vmx->nested.pml_full = false;
425 
426 		/*
427 		 * It should be impossible to trigger a nested PML Full VM-Exit
428 		 * for anything other than an EPT Violation from L2.  KVM *can*
429 		 * trigger nEPT page fault injection in response to an EPT
430 		 * Misconfig, e.g. if the MMIO SPTE was stale and L1's EPT
431 		 * tables also changed, but KVM should not treat EPT Misconfig
432 		 * VM-Exits as writes.
433 		 */
434 		WARN_ON_ONCE(vmx->vt.exit_reason.basic != EXIT_REASON_EPT_VIOLATION);
435 
436 		/*
437 		 * PML Full and EPT Violation VM-Exits both use bit 12 to report
438 		 * "NMI unblocking due to IRET", i.e. the bit can be propagated
439 		 * as-is from the original EXIT_QUALIFICATION.
440 		 */
441 		exit_qualification = vmx_get_exit_qual(vcpu) & INTR_INFO_UNBLOCK_NMI;
442 	} else {
443 		if (fault->error_code & PFERR_RSVD_MASK) {
444 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
445 			exit_qualification = 0;
446 		} else {
447 			u64 mask = EPT_VIOLATION_GVA_IS_VALID |
448 				   EPT_VIOLATION_GVA_TRANSLATED;
449 
450 			if (vmx->nested.msrs.ept_caps & VMX_EPT_ADVANCED_VMEXIT_INFO_BIT)
451 				mask |= EPT_VIOLATION_GVA_USER |
452 					EPT_VIOLATION_GVA_WRITABLE |
453 					EPT_VIOLATION_GVA_NX;
454 
455 			exit_qualification = fault->exit_qualification & ~mask;
456 
457 			/*
458 			 * Use the EXIT_QUALIFICATION from the VMCS if and only
459 			 * if the hardware VM-Exit from L2 was an EPT Violation.
460 			 * If the fault is synthesized, then EXIT_QUALIFICATION
461 			 * is stale and/or holds entirely different data.  And
462 			 * conversely, KVM _must_ rely on EXIT_QUALIFICATION if
463 			 * the fault came from hardware, because KVM only sees
464 			 * and walks the faulting GPA.
465 			 */
466 			if (from_hardware)
467 				exit_qualification |= vmx_get_exit_qual(vcpu) & mask;
468 			else
469 				exit_qualification |= fault->exit_qualification & mask;
470 
471 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
472 		}
473 
474 		/*
475 		 * Although the caller (kvm_inject_emulated_page_fault) would
476 		 * have already synced the faulting address in the shadow EPT
477 		 * tables for the current EPTP12, we also need to sync it for
478 		 * any other cached EPTP02s based on the same EP4TA, since the
479 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
480 		 */
481 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
482 					   fault->address);
483 	}
484 
485 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
486 	vmcs12->guest_physical_address = fault->address;
487 }
488 
489 static inline bool nested_ept_mbec_enabled(struct kvm_vcpu *vcpu)
490 {
491 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
492 
493 	return nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC);
494 }
495 
496 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
497 {
498 	struct vcpu_vmx *vmx = to_vmx(vcpu);
499 	bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
500 	int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
501 
502 	kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
503 				nested_ept_ad_enabled(vcpu),
504 				nested_ept_mbec_enabled(vcpu),
505 				nested_ept_get_eptp(vcpu));
506 }
507 
508 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
509 {
510 	WARN_ON(mmu_is_nested(vcpu));
511 
512 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
513 	nested_ept_new_eptp(vcpu);
514 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
515 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
516 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
517 
518 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
519 }
520 
521 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
522 {
523 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
524 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
525 }
526 
527 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
528 					    u16 error_code)
529 {
530 	bool inequality, bit;
531 
532 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
533 	inequality =
534 		(error_code & vmcs12->page_fault_error_code_mask) !=
535 		 vmcs12->page_fault_error_code_match;
536 	return inequality ^ bit;
537 }
538 
539 static bool nested_vmx_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
540 					   u32 error_code)
541 {
542 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
543 
544 	/*
545 	 * Drop bits 31:16 of the error code when performing the #PF mask+match
546 	 * check.  All VMCS fields involved are 32 bits, but Intel CPUs never
547 	 * set bits 31:16 and VMX disallows setting bits 31:16 in the injected
548 	 * error code.  Including the to-be-dropped bits in the check might
549 	 * result in an "impossible" or missed exit from L1's perspective.
550 	 */
551 	if (vector == PF_VECTOR)
552 		return nested_vmx_is_page_fault_vmexit(vmcs12, (u16)error_code);
553 
554 	return (vmcs12->exception_bitmap & (1u << vector));
555 }
556 
557 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
558 					       struct vmcs12 *vmcs12)
559 {
560 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
561 		return 0;
562 
563 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
564 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
565 		return -EINVAL;
566 
567 	return 0;
568 }
569 
570 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
571 						struct vmcs12 *vmcs12)
572 {
573 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
574 		return 0;
575 
576 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
577 		return -EINVAL;
578 
579 	return 0;
580 }
581 
582 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
583 						struct vmcs12 *vmcs12)
584 {
585 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
586 		return 0;
587 
588 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
589 		return -EINVAL;
590 
591 	if (CC(!nested_cpu_has_vid(vmcs12) && vmcs12->tpr_threshold >> 4))
592 		return -EINVAL;
593 
594 	return 0;
595 }
596 
597 /*
598  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
599  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
600  * only the "disable intercept" case needs to be handled.
601  */
602 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
603 							unsigned long *msr_bitmap_l0,
604 							u32 msr, int type)
605 {
606 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
607 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
608 
609 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
610 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
611 }
612 
613 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
614 {
615 	int msr;
616 
617 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
618 		unsigned word = msr / BITS_PER_LONG;
619 
620 		msr_bitmap[word] = ~0;
621 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
622 	}
623 }
624 
625 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
626 static inline									\
627 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
628 					 unsigned long *msr_bitmap_l1,		\
629 					 unsigned long *msr_bitmap_l0, u32 msr)	\
630 {										\
631 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
632 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
633 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
634 	else									\
635 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
636 }
637 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
638 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
639 
640 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
641 						    unsigned long *msr_bitmap_l1,
642 						    unsigned long *msr_bitmap_l0,
643 						    u32 msr, int types)
644 {
645 	if (types & MSR_TYPE_R)
646 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
647 						  msr_bitmap_l0, msr);
648 	if (types & MSR_TYPE_W)
649 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
650 						   msr_bitmap_l0, msr);
651 }
652 
653 #define nested_vmx_merge_msr_bitmaps(msr, type)	\
654 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1,	\
655 					 msr_bitmap_l0, msr, type)
656 
657 #define nested_vmx_merge_msr_bitmaps_read(msr) \
658 	nested_vmx_merge_msr_bitmaps(msr, MSR_TYPE_R)
659 
660 #define nested_vmx_merge_msr_bitmaps_write(msr) \
661 	nested_vmx_merge_msr_bitmaps(msr, MSR_TYPE_W)
662 
663 #define nested_vmx_merge_msr_bitmaps_rw(msr) \
664 	nested_vmx_merge_msr_bitmaps(msr, MSR_TYPE_RW)
665 
666 static void nested_vmx_merge_pmu_msr_bitmaps(struct kvm_vcpu *vcpu,
667 					     unsigned long *msr_bitmap_l1,
668 					     unsigned long *msr_bitmap_l0)
669 {
670 	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
671 	struct vcpu_vmx *vmx = to_vmx(vcpu);
672 	int i;
673 
674 	/*
675 	 * Skip the merges if the vCPU doesn't have a mediated PMU MSR, i.e. if
676 	 * none of the MSRs can possibly be passed through to L1.
677 	 */
678 	if (!kvm_vcpu_has_mediated_pmu(vcpu))
679 		return;
680 
681 	for (i = 0; i < pmu->nr_arch_gp_counters; i++) {
682 		nested_vmx_merge_msr_bitmaps_rw(MSR_IA32_PERFCTR0 + i);
683 		nested_vmx_merge_msr_bitmaps_rw(MSR_IA32_PMC0 + i);
684 	}
685 
686 	for (i = 0; i < pmu->nr_arch_fixed_counters; i++)
687 		nested_vmx_merge_msr_bitmaps_rw(MSR_CORE_PERF_FIXED_CTR0 + i);
688 
689 	nested_vmx_merge_msr_bitmaps_rw(MSR_CORE_PERF_GLOBAL_CTRL);
690 	nested_vmx_merge_msr_bitmaps_read(MSR_CORE_PERF_GLOBAL_STATUS);
691 	nested_vmx_merge_msr_bitmaps_write(MSR_CORE_PERF_GLOBAL_OVF_CTRL);
692 }
693 
694 /*
695  * Merge L0's and L1's MSR bitmap, return false to indicate that
696  * we do not use the hardware.
697  */
698 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
699 						 struct vmcs12 *vmcs12)
700 {
701 	struct vcpu_vmx *vmx = to_vmx(vcpu);
702 	int msr;
703 	unsigned long *msr_bitmap_l1;
704 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
705 	struct kvm_host_map map;
706 
707 	/* Nothing to do if the MSR bitmap is not in use.  */
708 	if (!cpu_has_vmx_msr_bitmap() ||
709 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
710 		return false;
711 
712 	/*
713 	 * MSR bitmap update can be skipped when:
714 	 * - MSR bitmap for L1 hasn't changed.
715 	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
716 	 *   before.
717 	 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
718 	 *   and tells KVM (L0) there were no changes in MSR bitmap for L2.
719 	 */
720 	if (!vmx->nested.force_msr_bitmap_recalc) {
721 		struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
722 
723 		if (evmcs && evmcs->hv_enlightenments_control.msr_bitmap &&
724 		    evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
725 			return true;
726 	}
727 
728 	if (kvm_vcpu_map_readonly(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), &map))
729 		return false;
730 
731 	msr_bitmap_l1 = (unsigned long *)map.hva;
732 
733 	/*
734 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
735 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
736 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
737 	 */
738 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
739 
740 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
741 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
742 			/*
743 			 * L0 need not intercept reads for MSRs between 0x800
744 			 * and 0x8ff, it just lets the processor take the value
745 			 * from the virtual-APIC page; take those 256 bits
746 			 * directly from the L1 bitmap.
747 			 */
748 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
749 				unsigned word = msr / BITS_PER_LONG;
750 
751 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
752 			}
753 		}
754 
755 		nested_vmx_disable_intercept_for_x2apic_msr(
756 			msr_bitmap_l1, msr_bitmap_l0,
757 			X2APIC_MSR(APIC_TASKPRI),
758 			MSR_TYPE_R | MSR_TYPE_W);
759 
760 		if (nested_cpu_has_vid(vmcs12)) {
761 			nested_vmx_disable_intercept_for_x2apic_msr(
762 				msr_bitmap_l1, msr_bitmap_l0,
763 				X2APIC_MSR(APIC_EOI),
764 				MSR_TYPE_W);
765 			nested_vmx_disable_intercept_for_x2apic_msr(
766 				msr_bitmap_l1, msr_bitmap_l0,
767 				X2APIC_MSR(APIC_SELF_IPI),
768 				MSR_TYPE_W);
769 		}
770 	}
771 
772 	/*
773 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
774 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
775 	 */
776 #ifdef CONFIG_X86_64
777 	nested_vmx_merge_msr_bitmaps_rw(MSR_FS_BASE);
778 	nested_vmx_merge_msr_bitmaps_rw(MSR_GS_BASE);
779 	nested_vmx_merge_msr_bitmaps_rw(MSR_KERNEL_GS_BASE);
780 #endif
781 	nested_vmx_merge_msr_bitmaps_rw(MSR_IA32_SPEC_CTRL);
782 	nested_vmx_merge_msr_bitmaps_write(MSR_IA32_PRED_CMD);
783 	nested_vmx_merge_msr_bitmaps_write(MSR_IA32_FLUSH_CMD);
784 
785 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
786 					 MSR_IA32_APERF, MSR_TYPE_R);
787 
788 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
789 					 MSR_IA32_MPERF, MSR_TYPE_R);
790 
791 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
792 					 MSR_IA32_U_CET, MSR_TYPE_RW);
793 
794 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
795 					 MSR_IA32_S_CET, MSR_TYPE_RW);
796 
797 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
798 					 MSR_IA32_PL0_SSP, MSR_TYPE_RW);
799 
800 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
801 					 MSR_IA32_PL1_SSP, MSR_TYPE_RW);
802 
803 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
804 					 MSR_IA32_PL2_SSP, MSR_TYPE_RW);
805 
806 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
807 					 MSR_IA32_PL3_SSP, MSR_TYPE_RW);
808 
809 	nested_vmx_merge_pmu_msr_bitmaps(vcpu, msr_bitmap_l1, msr_bitmap_l0);
810 
811 	kvm_vcpu_unmap(vcpu, &map);
812 
813 	vmx->nested.force_msr_bitmap_recalc = false;
814 
815 	return true;
816 }
817 
818 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
819 				       struct vmcs12 *vmcs12)
820 {
821 	struct vcpu_vmx *vmx = to_vmx(vcpu);
822 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
823 
824 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
825 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
826 		return;
827 
828 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
829 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
830 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
831 		return;
832 
833 	kvm_read_guest_cached(vcpu->kvm, ghc, get_shadow_vmcs12(vcpu),
834 			      VMCS12_SIZE);
835 }
836 
837 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
838 					      struct vmcs12 *vmcs12)
839 {
840 	struct vcpu_vmx *vmx = to_vmx(vcpu);
841 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
842 
843 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
844 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
845 		return;
846 
847 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
848 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
849 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
850 		return;
851 
852 	kvm_write_guest_cached(vcpu->kvm, ghc, get_shadow_vmcs12(vcpu),
853 			       VMCS12_SIZE);
854 }
855 
856 /*
857  * In nested virtualization, check if L1 has set
858  * VM_EXIT_ACK_INTR_ON_EXIT
859  */
860 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
861 {
862 	return get_vmcs12(vcpu)->vm_exit_controls &
863 		VM_EXIT_ACK_INTR_ON_EXIT;
864 }
865 
866 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
867 					  struct vmcs12 *vmcs12)
868 {
869 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
870 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
871 		return -EINVAL;
872 	else
873 		return 0;
874 }
875 
876 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
877 					   struct vmcs12 *vmcs12)
878 {
879 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
880 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
881 	    !nested_cpu_has_vid(vmcs12) &&
882 	    !nested_cpu_has_posted_intr(vmcs12))
883 		return 0;
884 
885 	/*
886 	 * If virtualize x2apic mode is enabled,
887 	 * virtualize apic access must be disabled.
888 	 */
889 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
890 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
891 		return -EINVAL;
892 
893 	/*
894 	 * If virtual interrupt delivery is enabled,
895 	 * we must exit on external interrupts.
896 	 */
897 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
898 		return -EINVAL;
899 
900 	/*
901 	 * bits 15:8 should be zero in posted_intr_nv,
902 	 * the descriptor address has been already checked
903 	 * in nested_get_vmcs12_pages.
904 	 *
905 	 * bits 5:0 of posted_intr_desc_addr should be zero.
906 	 */
907 	if (nested_cpu_has_posted_intr(vmcs12) &&
908 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
909 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
910 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
911 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
912 		return -EINVAL;
913 
914 	/* tpr shadow is needed by all apicv features. */
915 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
916 		return -EINVAL;
917 
918 	return 0;
919 }
920 
921 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
922 {
923 	struct vcpu_vmx *vmx = to_vmx(vcpu);
924 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
925 				       vmx->nested.msrs.misc_high);
926 
927 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
928 }
929 
930 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
931 				       u32 count, u64 addr)
932 {
933 	if (count == 0)
934 		return 0;
935 
936 	/*
937 	 * Exceeding the limit results in architecturally _undefined_ behavior,
938 	 * i.e. KVM is allowed to do literally anything in response to a bad
939 	 * limit.  Immediately generate a consistency check so that code that
940 	 * consumes the count doesn't need to worry about extreme edge cases.
941 	 */
942 	if (count > nested_vmx_max_atomic_switch_msrs(vcpu))
943 		return -EINVAL;
944 
945 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
946 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
947 		return -EINVAL;
948 
949 	return 0;
950 }
951 
952 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
953 						     struct vmcs12 *vmcs12)
954 {
955 	if (CC(nested_vmx_check_msr_switch(vcpu,
956 					   vmcs12->vm_exit_msr_load_count,
957 					   vmcs12->vm_exit_msr_load_addr)) ||
958 	    CC(nested_vmx_check_msr_switch(vcpu,
959 					   vmcs12->vm_exit_msr_store_count,
960 					   vmcs12->vm_exit_msr_store_addr)))
961 		return -EINVAL;
962 
963 	return 0;
964 }
965 
966 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
967                                                       struct vmcs12 *vmcs12)
968 {
969 	if (CC(nested_vmx_check_msr_switch(vcpu,
970 					   vmcs12->vm_entry_msr_load_count,
971 					   vmcs12->vm_entry_msr_load_addr)))
972                 return -EINVAL;
973 
974 	return 0;
975 }
976 
977 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
978 					 struct vmcs12 *vmcs12)
979 {
980 	if (!nested_cpu_has_pml(vmcs12))
981 		return 0;
982 
983 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
984 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
985 		return -EINVAL;
986 
987 	return 0;
988 }
989 
990 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
991 							struct vmcs12 *vmcs12)
992 {
993 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
994 	       !nested_cpu_has_ept(vmcs12)))
995 		return -EINVAL;
996 	return 0;
997 }
998 
999 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
1000 							 struct vmcs12 *vmcs12)
1001 {
1002 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
1003 	       !nested_cpu_has_ept(vmcs12)))
1004 		return -EINVAL;
1005 	return 0;
1006 }
1007 
1008 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
1009 						 struct vmcs12 *vmcs12)
1010 {
1011 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
1012 		return 0;
1013 
1014 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
1015 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
1016 		return -EINVAL;
1017 
1018 	return 0;
1019 }
1020 
1021 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
1022 				       struct vmx_msr_entry *e)
1023 {
1024 	/* x2APIC MSR accesses are not allowed */
1025 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
1026 		return -EINVAL;
1027 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
1028 	    CC(e->index == MSR_IA32_UCODE_REV))
1029 		return -EINVAL;
1030 	if (CC(e->reserved != 0))
1031 		return -EINVAL;
1032 	return 0;
1033 }
1034 
1035 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
1036 				     struct vmx_msr_entry *e)
1037 {
1038 	if (CC(e->index == MSR_FS_BASE) ||
1039 	    CC(e->index == MSR_GS_BASE) ||
1040 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
1041 	    nested_vmx_msr_check_common(vcpu, e))
1042 		return -EINVAL;
1043 	return 0;
1044 }
1045 
1046 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
1047 				      struct vmx_msr_entry *e)
1048 {
1049 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
1050 	    nested_vmx_msr_check_common(vcpu, e))
1051 		return -EINVAL;
1052 	return 0;
1053 }
1054 
1055 /*
1056  * Load guest's/host's msr at nested entry/exit.
1057  * return 0 for success, entry index for failure.
1058  *
1059  * One of the failure modes for MSR load/store is when a list exceeds the
1060  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
1061  * as possible, process all valid entries before failing rather than precheck
1062  * for a capacity violation.
1063  */
1064 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1065 {
1066 	u32 i;
1067 	struct vmx_msr_entry e;
1068 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1069 
1070 	for (i = 0; i < count; i++) {
1071 		if (WARN_ON_ONCE(i >= max_msr_list_size))
1072 			goto fail;
1073 
1074 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
1075 					&e, sizeof(e))) {
1076 			pr_debug_ratelimited(
1077 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
1078 				__func__, i, gpa + i * sizeof(e));
1079 			goto fail;
1080 		}
1081 		if (nested_vmx_load_msr_check(vcpu, &e)) {
1082 			pr_debug_ratelimited(
1083 				"%s check failed (%u, 0x%x, 0x%x)\n",
1084 				__func__, i, e.index, e.reserved);
1085 			goto fail;
1086 		}
1087 		if (kvm_emulate_msr_write(vcpu, e.index, e.value)) {
1088 			pr_debug_ratelimited(
1089 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1090 				__func__, i, e.index, e.value);
1091 			goto fail;
1092 		}
1093 	}
1094 	return 0;
1095 fail:
1096 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
1097 	return i + 1;
1098 }
1099 
1100 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
1101 					    u32 msr_index,
1102 					    u64 *data)
1103 {
1104 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1105 
1106 	/*
1107 	 * If the L0 hypervisor stored a more accurate value for the TSC that
1108 	 * does not include the time taken for emulation of the L2->L1
1109 	 * VM-exit in L0, use the more accurate value.
1110 	 */
1111 	if (msr_index == MSR_IA32_TSC && vmx->nested.tsc_autostore_slot >= 0) {
1112 		int slot = vmx->nested.tsc_autostore_slot;
1113 		u64 host_tsc = vmx->msr_autostore.val[slot].value;
1114 
1115 		*data = kvm_read_l1_tsc(vcpu, host_tsc);
1116 		return true;
1117 	}
1118 
1119 	if (kvm_emulate_msr_read(vcpu, msr_index, data)) {
1120 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
1121 			msr_index);
1122 		return false;
1123 	}
1124 	return true;
1125 }
1126 
1127 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
1128 				     struct vmx_msr_entry *e)
1129 {
1130 	if (kvm_vcpu_read_guest(vcpu,
1131 				gpa + i * sizeof(*e),
1132 				e, 2 * sizeof(u32))) {
1133 		pr_debug_ratelimited(
1134 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
1135 			__func__, i, gpa + i * sizeof(*e));
1136 		return false;
1137 	}
1138 	if (nested_vmx_store_msr_check(vcpu, e)) {
1139 		pr_debug_ratelimited(
1140 			"%s check failed (%u, 0x%x, 0x%x)\n",
1141 			__func__, i, e->index, e->reserved);
1142 		return false;
1143 	}
1144 	return true;
1145 }
1146 
1147 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1148 {
1149 	u64 data;
1150 	u32 i;
1151 	struct vmx_msr_entry e;
1152 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1153 
1154 	for (i = 0; i < count; i++) {
1155 		if (WARN_ON_ONCE(i >= max_msr_list_size))
1156 			return -EINVAL;
1157 
1158 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1159 			return -EINVAL;
1160 
1161 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1162 			return -EINVAL;
1163 
1164 		if (kvm_vcpu_write_guest(vcpu,
1165 					 gpa + i * sizeof(e) +
1166 					     offsetof(struct vmx_msr_entry, value),
1167 					 &data, sizeof(data))) {
1168 			pr_debug_ratelimited(
1169 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1170 				__func__, i, e.index, data);
1171 			return -EINVAL;
1172 		}
1173 	}
1174 	return 0;
1175 }
1176 
1177 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1178 {
1179 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1180 	u32 count = vmcs12->vm_exit_msr_store_count;
1181 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1182 	struct vmx_msr_entry e;
1183 	u32 i;
1184 
1185 	for (i = 0; i < count; i++) {
1186 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1187 			return false;
1188 
1189 		if (e.index == msr_index)
1190 			return true;
1191 	}
1192 	return false;
1193 }
1194 
1195 /*
1196  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1197  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1198  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1199  * @entry_failure_code.
1200  */
1201 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1202 			       bool nested_ept, bool reload_pdptrs,
1203 			       enum vm_entry_failure_code *entry_failure_code)
1204 {
1205 	if (CC(!kvm_vcpu_is_legal_cr3(vcpu, cr3))) {
1206 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1207 		return -EINVAL;
1208 	}
1209 
1210 	/*
1211 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1212 	 * must not be dereferenced.
1213 	 */
1214 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1215 	    CC(!load_pdptrs(vcpu, cr3))) {
1216 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1217 		return -EINVAL;
1218 	}
1219 
1220 	vcpu->arch.cr3 = cr3;
1221 	kvm_register_mark_dirty(vcpu, VCPU_REG_CR3);
1222 
1223 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1224 	kvm_init_mmu(vcpu);
1225 
1226 	if (!nested_ept)
1227 		kvm_mmu_new_pgd(vcpu, cr3);
1228 
1229 	return 0;
1230 }
1231 
1232 /*
1233  * Returns if KVM is able to config CPU to tag TLB entries
1234  * populated by L2 differently than TLB entries populated
1235  * by L1.
1236  *
1237  * If L0 uses EPT, L1 and L2 run with different EPTP because
1238  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1239  * are tagged with different EPTP.
1240  *
1241  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1242  * with different VPID (L1 entries are tagged with vmx->vpid
1243  * while L2 entries are tagged with vmx->nested.vpid02).
1244  */
1245 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1246 {
1247 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1248 
1249 	return enable_ept ||
1250 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1251 }
1252 
1253 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1254 					    struct vmcs12 *vmcs12,
1255 					    bool is_vmenter)
1256 {
1257 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1258 
1259 	/* Handle pending Hyper-V TLB flush requests */
1260 	kvm_hv_nested_transtion_tlb_flush(vcpu, enable_ept);
1261 
1262 	/*
1263 	 * If VPID is disabled, then guest TLB accesses use VPID=0, i.e. the
1264 	 * same VPID as the host, and so architecturally, linear and combined
1265 	 * mappings for VPID=0 must be flushed at VM-Enter and VM-Exit.  KVM
1266 	 * emulates L2 sharing L1's VPID=0 by using vpid01 while running L2,
1267 	 * and so KVM must also emulate TLB flush of VPID=0, i.e. vpid01.  This
1268 	 * is required if VPID is disabled in KVM, as a TLB flush (there are no
1269 	 * VPIDs) still occurs from L1's perspective, and KVM may need to
1270 	 * synchronize the MMU in response to the guest TLB flush.
1271 	 *
1272 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1273 	 * EPT is a special snowflake, as guest-physical mappings aren't
1274 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1275 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1276 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1277 	 * those mappings.
1278 	 */
1279 	if (!nested_cpu_has_vpid(vmcs12)) {
1280 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1281 		return;
1282 	}
1283 
1284 	/* L2 should never have a VPID if VPID is disabled. */
1285 	WARN_ON(!enable_vpid);
1286 
1287 	/*
1288 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1289 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1290 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1291 	 * that the new vpid12 has never been used and thus represents a new
1292 	 * guest ASID that cannot have entries in the TLB.
1293 	 */
1294 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1295 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1296 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1297 		return;
1298 	}
1299 
1300 	/*
1301 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1302 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1303 	 * KVM was unable to allocate a VPID for L2, flush the current context
1304 	 * as the effective ASID is common to both L1 and L2.
1305 	 */
1306 	if (!nested_has_guest_tlb_tag(vcpu))
1307 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1308 }
1309 
1310 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1311 {
1312 	superset &= mask;
1313 	subset &= mask;
1314 
1315 	return (superset | subset) == superset;
1316 }
1317 
1318 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1319 {
1320 	const u64 feature_bits = VMX_BASIC_DUAL_MONITOR_TREATMENT |
1321 				 VMX_BASIC_INOUT |
1322 				 VMX_BASIC_TRUE_CTLS |
1323 				 VMX_BASIC_NO_HW_ERROR_CODE_CC;
1324 
1325 	const u64 reserved_bits = GENMASK_ULL(63, 57) |
1326 				  GENMASK_ULL(47, 45) |
1327 				  BIT_ULL(31);
1328 
1329 	u64 vmx_basic = vmcs_config.nested.basic;
1330 
1331 	BUILD_BUG_ON(feature_bits & reserved_bits);
1332 
1333 	/*
1334 	 * Except for 32BIT_PHYS_ADDR_ONLY, which is an anti-feature bit (has
1335 	 * inverted polarity), the incoming value must not set feature bits or
1336 	 * reserved bits that aren't allowed/supported by KVM.  Fields, i.e.
1337 	 * multi-bit values, are explicitly checked below.
1338 	 */
1339 	if (!is_bitwise_subset(vmx_basic, data, feature_bits | reserved_bits))
1340 		return -EINVAL;
1341 
1342 	/*
1343 	 * KVM does not emulate a version of VMX that constrains physical
1344 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1345 	 */
1346 	if (data & VMX_BASIC_32BIT_PHYS_ADDR_ONLY)
1347 		return -EINVAL;
1348 
1349 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1350 	    vmx_basic_vmcs_revision_id(data))
1351 		return -EINVAL;
1352 
1353 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1354 		return -EINVAL;
1355 
1356 	vmx->nested.msrs.basic = data;
1357 	return 0;
1358 }
1359 
1360 static void vmx_get_control_msr(struct nested_vmx_msrs *msrs, u32 msr_index,
1361 				u32 **low, u32 **high)
1362 {
1363 	switch (msr_index) {
1364 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1365 		*low = &msrs->pinbased_ctls_low;
1366 		*high = &msrs->pinbased_ctls_high;
1367 		break;
1368 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1369 		*low = &msrs->procbased_ctls_low;
1370 		*high = &msrs->procbased_ctls_high;
1371 		break;
1372 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1373 		*low = &msrs->exit_ctls_low;
1374 		*high = &msrs->exit_ctls_high;
1375 		break;
1376 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1377 		*low = &msrs->entry_ctls_low;
1378 		*high = &msrs->entry_ctls_high;
1379 		break;
1380 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1381 		*low = &msrs->secondary_ctls_low;
1382 		*high = &msrs->secondary_ctls_high;
1383 		break;
1384 	default:
1385 		BUG();
1386 	}
1387 }
1388 
1389 static int
1390 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1391 {
1392 	u32 *lowp, *highp;
1393 	u64 supported;
1394 
1395 	vmx_get_control_msr(&vmcs_config.nested, msr_index, &lowp, &highp);
1396 
1397 	supported = vmx_control_msr(*lowp, *highp);
1398 
1399 	/* Check must-be-1 bits are still 1. */
1400 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1401 		return -EINVAL;
1402 
1403 	/* Check must-be-0 bits are still 0. */
1404 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1405 		return -EINVAL;
1406 
1407 	vmx_get_control_msr(&vmx->nested.msrs, msr_index, &lowp, &highp);
1408 	*lowp = data;
1409 	*highp = data >> 32;
1410 	return 0;
1411 }
1412 
1413 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1414 {
1415 	const u64 feature_bits = VMX_MISC_SAVE_EFER_LMA |
1416 				 VMX_MISC_ACTIVITY_HLT |
1417 				 VMX_MISC_ACTIVITY_SHUTDOWN |
1418 				 VMX_MISC_ACTIVITY_WAIT_SIPI |
1419 				 VMX_MISC_INTEL_PT |
1420 				 VMX_MISC_RDMSR_IN_SMM |
1421 				 VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
1422 				 VMX_MISC_VMXOFF_BLOCK_SMI |
1423 				 VMX_MISC_ZERO_LEN_INS;
1424 
1425 	const u64 reserved_bits = BIT_ULL(31) | GENMASK_ULL(13, 9);
1426 
1427 	u64 vmx_misc = vmx_control_msr(vmcs_config.nested.misc_low,
1428 				       vmcs_config.nested.misc_high);
1429 
1430 	BUILD_BUG_ON(feature_bits & reserved_bits);
1431 
1432 	/*
1433 	 * The incoming value must not set feature bits or reserved bits that
1434 	 * aren't allowed/supported by KVM.  Fields, i.e. multi-bit values, are
1435 	 * explicitly checked below.
1436 	 */
1437 	if (!is_bitwise_subset(vmx_misc, data, feature_bits | reserved_bits))
1438 		return -EINVAL;
1439 
1440 	if ((vmx->nested.msrs.pinbased_ctls_high &
1441 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1442 	    vmx_misc_preemption_timer_rate(data) !=
1443 	    vmx_misc_preemption_timer_rate(vmx_misc))
1444 		return -EINVAL;
1445 
1446 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1447 		return -EINVAL;
1448 
1449 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1450 		return -EINVAL;
1451 
1452 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1453 		return -EINVAL;
1454 
1455 	vmx->nested.msrs.misc_low = data;
1456 	vmx->nested.msrs.misc_high = data >> 32;
1457 
1458 	return 0;
1459 }
1460 
1461 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1462 {
1463 	u64 vmx_ept_vpid_cap = vmx_control_msr(vmcs_config.nested.ept_caps,
1464 					       vmcs_config.nested.vpid_caps);
1465 
1466 	/* Every bit is either reserved or a feature bit. */
1467 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1468 		return -EINVAL;
1469 
1470 	vmx->nested.msrs.ept_caps = data;
1471 	vmx->nested.msrs.vpid_caps = data >> 32;
1472 	return 0;
1473 }
1474 
1475 static u64 *vmx_get_fixed0_msr(struct nested_vmx_msrs *msrs, u32 msr_index)
1476 {
1477 	switch (msr_index) {
1478 	case MSR_IA32_VMX_CR0_FIXED0:
1479 		return &msrs->cr0_fixed0;
1480 	case MSR_IA32_VMX_CR4_FIXED0:
1481 		return &msrs->cr4_fixed0;
1482 	default:
1483 		BUG();
1484 	}
1485 }
1486 
1487 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1488 {
1489 	const u64 *msr = vmx_get_fixed0_msr(&vmcs_config.nested, msr_index);
1490 
1491 	/*
1492 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1493 	 * must be 1 in the restored value.
1494 	 */
1495 	if (!is_bitwise_subset(data, *msr, -1ULL))
1496 		return -EINVAL;
1497 
1498 	*vmx_get_fixed0_msr(&vmx->nested.msrs, msr_index) = data;
1499 	return 0;
1500 }
1501 
1502 /*
1503  * Called when userspace is restoring VMX MSRs.
1504  *
1505  * Returns 0 on success, non-0 otherwise.
1506  */
1507 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1508 {
1509 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1510 
1511 	/*
1512 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1513 	 * is in VMX operation.
1514 	 */
1515 	if (vmx->nested.vmxon)
1516 		return -EBUSY;
1517 
1518 	switch (msr_index) {
1519 	case MSR_IA32_VMX_BASIC:
1520 		return vmx_restore_vmx_basic(vmx, data);
1521 	case MSR_IA32_VMX_PINBASED_CTLS:
1522 	case MSR_IA32_VMX_PROCBASED_CTLS:
1523 	case MSR_IA32_VMX_EXIT_CTLS:
1524 	case MSR_IA32_VMX_ENTRY_CTLS:
1525 		/*
1526 		 * The "non-true" VMX capability MSRs are generated from the
1527 		 * "true" MSRs, so we do not support restoring them directly.
1528 		 *
1529 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1530 		 * should restore the "true" MSRs with the must-be-1 bits
1531 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1532 		 * DEFAULT SETTINGS".
1533 		 */
1534 		return -EINVAL;
1535 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1536 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1537 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1538 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1539 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1540 		return vmx_restore_control_msr(vmx, msr_index, data);
1541 	case MSR_IA32_VMX_MISC:
1542 		return vmx_restore_vmx_misc(vmx, data);
1543 	case MSR_IA32_VMX_CR0_FIXED0:
1544 	case MSR_IA32_VMX_CR4_FIXED0:
1545 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1546 	case MSR_IA32_VMX_CR0_FIXED1:
1547 	case MSR_IA32_VMX_CR4_FIXED1:
1548 		/*
1549 		 * These MSRs are generated based on the vCPU's CPUID, so we
1550 		 * do not support restoring them directly.
1551 		 */
1552 		return -EINVAL;
1553 	case MSR_IA32_VMX_EPT_VPID_CAP:
1554 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1555 	case MSR_IA32_VMX_VMCS_ENUM:
1556 		vmx->nested.msrs.vmcs_enum = data;
1557 		return 0;
1558 	case MSR_IA32_VMX_VMFUNC:
1559 		if (data & ~vmcs_config.nested.vmfunc_controls)
1560 			return -EINVAL;
1561 		vmx->nested.msrs.vmfunc_controls = data;
1562 		return 0;
1563 	default:
1564 		/*
1565 		 * The rest of the VMX capability MSRs do not support restore.
1566 		 */
1567 		return -EINVAL;
1568 	}
1569 }
1570 
1571 /* Returns 0 on success, non-0 otherwise. */
1572 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1573 {
1574 	switch (msr_index) {
1575 	case MSR_IA32_VMX_BASIC:
1576 		*pdata = msrs->basic;
1577 		break;
1578 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1579 	case MSR_IA32_VMX_PINBASED_CTLS:
1580 		*pdata = vmx_control_msr(
1581 			msrs->pinbased_ctls_low,
1582 			msrs->pinbased_ctls_high);
1583 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1584 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1585 		break;
1586 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1587 	case MSR_IA32_VMX_PROCBASED_CTLS:
1588 		*pdata = vmx_control_msr(
1589 			msrs->procbased_ctls_low,
1590 			msrs->procbased_ctls_high);
1591 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1592 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1593 		break;
1594 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1595 	case MSR_IA32_VMX_EXIT_CTLS:
1596 		*pdata = vmx_control_msr(
1597 			msrs->exit_ctls_low,
1598 			msrs->exit_ctls_high);
1599 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1600 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1601 		break;
1602 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1603 	case MSR_IA32_VMX_ENTRY_CTLS:
1604 		*pdata = vmx_control_msr(
1605 			msrs->entry_ctls_low,
1606 			msrs->entry_ctls_high);
1607 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1608 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1609 		break;
1610 	case MSR_IA32_VMX_MISC:
1611 		*pdata = vmx_control_msr(
1612 			msrs->misc_low,
1613 			msrs->misc_high);
1614 		break;
1615 	case MSR_IA32_VMX_CR0_FIXED0:
1616 		*pdata = msrs->cr0_fixed0;
1617 		break;
1618 	case MSR_IA32_VMX_CR0_FIXED1:
1619 		*pdata = msrs->cr0_fixed1;
1620 		break;
1621 	case MSR_IA32_VMX_CR4_FIXED0:
1622 		*pdata = msrs->cr4_fixed0;
1623 		break;
1624 	case MSR_IA32_VMX_CR4_FIXED1:
1625 		*pdata = msrs->cr4_fixed1;
1626 		break;
1627 	case MSR_IA32_VMX_VMCS_ENUM:
1628 		*pdata = msrs->vmcs_enum;
1629 		break;
1630 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1631 		*pdata = vmx_control_msr(
1632 			msrs->secondary_ctls_low,
1633 			msrs->secondary_ctls_high);
1634 		break;
1635 	case MSR_IA32_VMX_EPT_VPID_CAP:
1636 		*pdata = msrs->ept_caps |
1637 			((u64)msrs->vpid_caps << 32);
1638 		break;
1639 	case MSR_IA32_VMX_VMFUNC:
1640 		*pdata = msrs->vmfunc_controls;
1641 		break;
1642 	default:
1643 		return 1;
1644 	}
1645 
1646 	return 0;
1647 }
1648 
1649 /*
1650  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1651  * been modified by the L1 guest.  Note, "writable" in this context means
1652  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1653  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1654  * VM-exit information fields (which are actually writable if the vCPU is
1655  * configured to support "VMWRITE to any supported field in the VMCS").
1656  */
1657 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1658 {
1659 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1660 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1661 	struct shadow_vmcs_field field;
1662 	unsigned long val;
1663 	int i;
1664 
1665 	if (WARN_ON(!shadow_vmcs))
1666 		return;
1667 
1668 	preempt_disable();
1669 
1670 	vmcs_load(shadow_vmcs);
1671 
1672 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1673 		field = shadow_read_write_fields[i];
1674 		val = __vmcs_readl(field.encoding);
1675 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1676 	}
1677 
1678 	vmcs_clear(shadow_vmcs);
1679 	vmcs_load(vmx->loaded_vmcs->vmcs);
1680 
1681 	preempt_enable();
1682 }
1683 
1684 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1685 {
1686 	const struct shadow_vmcs_field *fields[] = {
1687 		shadow_read_write_fields,
1688 		shadow_read_only_fields
1689 	};
1690 	const int max_fields[] = {
1691 		max_shadow_read_write_fields,
1692 		max_shadow_read_only_fields
1693 	};
1694 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1695 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1696 	struct shadow_vmcs_field field;
1697 	unsigned long val;
1698 	int i, q;
1699 
1700 	if (WARN_ON(!shadow_vmcs))
1701 		return;
1702 
1703 	vmcs_load(shadow_vmcs);
1704 
1705 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1706 		for (i = 0; i < max_fields[q]; i++) {
1707 			field = fields[q][i];
1708 			val = vmcs12_read_any(vmcs12, field.encoding,
1709 					      field.offset);
1710 			__vmcs_writel(field.encoding, val);
1711 		}
1712 	}
1713 
1714 	vmcs_clear(shadow_vmcs);
1715 	vmcs_load(vmx->loaded_vmcs->vmcs);
1716 }
1717 
1718 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1719 {
1720 #ifdef CONFIG_KVM_HYPERV
1721 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1722 	struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
1723 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(&vmx->vcpu);
1724 
1725 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1726 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1727 	vmcs12->guest_rip = evmcs->guest_rip;
1728 
1729 	if (unlikely(!(hv_clean_fields &
1730 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_ENLIGHTENMENTSCONTROL))) {
1731 		hv_vcpu->nested.pa_page_gpa = evmcs->partition_assist_page;
1732 		hv_vcpu->nested.vm_id = evmcs->hv_vm_id;
1733 		hv_vcpu->nested.vp_id = evmcs->hv_vp_id;
1734 	}
1735 
1736 	if (unlikely(!(hv_clean_fields &
1737 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1738 		vmcs12->guest_rsp = evmcs->guest_rsp;
1739 		vmcs12->guest_rflags = evmcs->guest_rflags;
1740 		vmcs12->guest_interruptibility_info =
1741 			evmcs->guest_interruptibility_info;
1742 		/*
1743 		 * Not present in struct vmcs12:
1744 		 * vmcs12->guest_ssp = evmcs->guest_ssp;
1745 		 */
1746 	}
1747 
1748 	if (unlikely(!(hv_clean_fields &
1749 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1750 		vmcs12->cpu_based_vm_exec_control =
1751 			evmcs->cpu_based_vm_exec_control;
1752 	}
1753 
1754 	if (unlikely(!(hv_clean_fields &
1755 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1756 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1757 	}
1758 
1759 	if (unlikely(!(hv_clean_fields &
1760 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1761 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1762 	}
1763 
1764 	if (unlikely(!(hv_clean_fields &
1765 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1766 		vmcs12->vm_entry_intr_info_field =
1767 			evmcs->vm_entry_intr_info_field;
1768 		vmcs12->vm_entry_exception_error_code =
1769 			evmcs->vm_entry_exception_error_code;
1770 		vmcs12->vm_entry_instruction_len =
1771 			evmcs->vm_entry_instruction_len;
1772 	}
1773 
1774 	if (unlikely(!(hv_clean_fields &
1775 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1776 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1777 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1778 		vmcs12->host_cr0 = evmcs->host_cr0;
1779 		vmcs12->host_cr3 = evmcs->host_cr3;
1780 		vmcs12->host_cr4 = evmcs->host_cr4;
1781 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1782 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1783 		vmcs12->host_rip = evmcs->host_rip;
1784 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1785 		vmcs12->host_es_selector = evmcs->host_es_selector;
1786 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1787 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1788 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1789 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1790 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1791 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1792 		vmcs12->host_ia32_perf_global_ctrl = evmcs->host_ia32_perf_global_ctrl;
1793 		/*
1794 		 * Not present in struct vmcs12:
1795 		 * vmcs12->host_ia32_s_cet = evmcs->host_ia32_s_cet;
1796 		 * vmcs12->host_ssp = evmcs->host_ssp;
1797 		 * vmcs12->host_ia32_int_ssp_table_addr = evmcs->host_ia32_int_ssp_table_addr;
1798 		 */
1799 	}
1800 
1801 	if (unlikely(!(hv_clean_fields &
1802 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1803 		vmcs12->pin_based_vm_exec_control =
1804 			evmcs->pin_based_vm_exec_control;
1805 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1806 		vmcs12->secondary_vm_exec_control =
1807 			evmcs->secondary_vm_exec_control;
1808 	}
1809 
1810 	if (unlikely(!(hv_clean_fields &
1811 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1812 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1813 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1814 	}
1815 
1816 	if (unlikely(!(hv_clean_fields &
1817 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1818 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1819 	}
1820 
1821 	if (unlikely(!(hv_clean_fields &
1822 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1823 		vmcs12->guest_es_base = evmcs->guest_es_base;
1824 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1825 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1826 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1827 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1828 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1829 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1830 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1831 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1832 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1833 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1834 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1835 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1836 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1837 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1838 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1839 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1840 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1841 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1842 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1843 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1844 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1845 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1846 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1847 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1848 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1849 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1850 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1851 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1852 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1853 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1854 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1855 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1856 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1857 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1858 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1859 	}
1860 
1861 	if (unlikely(!(hv_clean_fields &
1862 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1863 		vmcs12->tsc_offset = evmcs->tsc_offset;
1864 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1865 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1866 		vmcs12->encls_exiting_bitmap = evmcs->encls_exiting_bitmap;
1867 		vmcs12->tsc_multiplier = evmcs->tsc_multiplier;
1868 	}
1869 
1870 	if (unlikely(!(hv_clean_fields &
1871 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1872 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1873 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1874 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1875 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1876 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1877 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1878 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1879 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1880 	}
1881 
1882 	if (unlikely(!(hv_clean_fields &
1883 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1884 		vmcs12->host_fs_base = evmcs->host_fs_base;
1885 		vmcs12->host_gs_base = evmcs->host_gs_base;
1886 		vmcs12->host_tr_base = evmcs->host_tr_base;
1887 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1888 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1889 		vmcs12->host_rsp = evmcs->host_rsp;
1890 	}
1891 
1892 	if (unlikely(!(hv_clean_fields &
1893 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1894 		vmcs12->ept_pointer = evmcs->ept_pointer;
1895 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1896 	}
1897 
1898 	if (unlikely(!(hv_clean_fields &
1899 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1900 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1901 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1902 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1903 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1904 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1905 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1906 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1907 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1908 		vmcs12->guest_pending_dbg_exceptions =
1909 			evmcs->guest_pending_dbg_exceptions;
1910 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1911 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1912 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1913 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1914 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1915 		vmcs12->guest_ia32_perf_global_ctrl = evmcs->guest_ia32_perf_global_ctrl;
1916 		/*
1917 		 * Not present in struct vmcs12:
1918 		 * vmcs12->guest_ia32_s_cet = evmcs->guest_ia32_s_cet;
1919 		 * vmcs12->guest_ia32_lbr_ctl = evmcs->guest_ia32_lbr_ctl;
1920 		 * vmcs12->guest_ia32_int_ssp_table_addr = evmcs->guest_ia32_int_ssp_table_addr;
1921 		 */
1922 	}
1923 
1924 	/*
1925 	 * Not used?
1926 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1927 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1928 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1929 	 * vmcs12->page_fault_error_code_mask =
1930 	 *		evmcs->page_fault_error_code_mask;
1931 	 * vmcs12->page_fault_error_code_match =
1932 	 *		evmcs->page_fault_error_code_match;
1933 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1934 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1935 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1936 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1937 	 */
1938 
1939 	/*
1940 	 * Read only fields:
1941 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1942 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1943 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1944 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1945 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1946 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1947 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1948 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1949 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1950 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1951 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1952 	 *
1953 	 * Not present in struct vmcs12:
1954 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1955 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1956 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1957 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1958 	 */
1959 
1960 	return;
1961 #else /* CONFIG_KVM_HYPERV */
1962 	KVM_BUG_ON(1, vmx->vcpu.kvm);
1963 #endif /* CONFIG_KVM_HYPERV */
1964 }
1965 
1966 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1967 {
1968 #ifdef CONFIG_KVM_HYPERV
1969 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1970 	struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
1971 
1972 	/*
1973 	 * Should not be changed by KVM:
1974 	 *
1975 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1976 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1977 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1978 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1979 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1980 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1981 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1982 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1983 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1984 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1985 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1986 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1987 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1988 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1989 	 * evmcs->host_rip = vmcs12->host_rip;
1990 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1991 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1992 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1993 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1994 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1995 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1996 	 * evmcs->host_rsp = vmcs12->host_rsp;
1997 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1998 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1999 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
2000 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
2001 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
2002 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
2003 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
2004 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
2005 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
2006 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
2007 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
2008 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
2009 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
2010 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
2011 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
2012 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
2013 	 * evmcs->page_fault_error_code_mask =
2014 	 *		vmcs12->page_fault_error_code_mask;
2015 	 * evmcs->page_fault_error_code_match =
2016 	 *		vmcs12->page_fault_error_code_match;
2017 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
2018 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
2019 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
2020 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
2021 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
2022 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
2023 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
2024 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
2025 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
2026 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
2027 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
2028 	 * evmcs->guest_ia32_perf_global_ctrl = vmcs12->guest_ia32_perf_global_ctrl;
2029 	 * evmcs->host_ia32_perf_global_ctrl = vmcs12->host_ia32_perf_global_ctrl;
2030 	 * evmcs->encls_exiting_bitmap = vmcs12->encls_exiting_bitmap;
2031 	 * evmcs->tsc_multiplier = vmcs12->tsc_multiplier;
2032 	 *
2033 	 * Not present in struct vmcs12:
2034 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
2035 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
2036 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
2037 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
2038 	 * evmcs->host_ia32_s_cet = vmcs12->host_ia32_s_cet;
2039 	 * evmcs->host_ssp = vmcs12->host_ssp;
2040 	 * evmcs->host_ia32_int_ssp_table_addr = vmcs12->host_ia32_int_ssp_table_addr;
2041 	 * evmcs->guest_ia32_s_cet = vmcs12->guest_ia32_s_cet;
2042 	 * evmcs->guest_ia32_lbr_ctl = vmcs12->guest_ia32_lbr_ctl;
2043 	 * evmcs->guest_ia32_int_ssp_table_addr = vmcs12->guest_ia32_int_ssp_table_addr;
2044 	 * evmcs->guest_ssp = vmcs12->guest_ssp;
2045 	 */
2046 
2047 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
2048 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
2049 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
2050 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
2051 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
2052 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
2053 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
2054 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
2055 
2056 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
2057 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
2058 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
2059 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
2060 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
2061 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
2062 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
2063 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
2064 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
2065 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
2066 
2067 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
2068 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
2069 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
2070 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
2071 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
2072 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
2073 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
2074 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
2075 
2076 	evmcs->guest_es_base = vmcs12->guest_es_base;
2077 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
2078 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
2079 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
2080 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
2081 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
2082 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
2083 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
2084 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
2085 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
2086 
2087 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
2088 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
2089 
2090 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
2091 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
2092 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
2093 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
2094 
2095 	evmcs->guest_pending_dbg_exceptions =
2096 		vmcs12->guest_pending_dbg_exceptions;
2097 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
2098 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
2099 
2100 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
2101 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
2102 
2103 	evmcs->guest_cr0 = vmcs12->guest_cr0;
2104 	evmcs->guest_cr3 = vmcs12->guest_cr3;
2105 	evmcs->guest_cr4 = vmcs12->guest_cr4;
2106 	evmcs->guest_dr7 = vmcs12->guest_dr7;
2107 
2108 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
2109 
2110 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
2111 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
2112 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
2113 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
2114 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
2115 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
2116 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
2117 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
2118 
2119 	evmcs->exit_qualification = vmcs12->exit_qualification;
2120 
2121 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
2122 	evmcs->guest_rsp = vmcs12->guest_rsp;
2123 	evmcs->guest_rflags = vmcs12->guest_rflags;
2124 
2125 	evmcs->guest_interruptibility_info =
2126 		vmcs12->guest_interruptibility_info;
2127 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
2128 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
2129 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
2130 	evmcs->vm_entry_exception_error_code =
2131 		vmcs12->vm_entry_exception_error_code;
2132 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
2133 
2134 	evmcs->guest_rip = vmcs12->guest_rip;
2135 
2136 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
2137 
2138 	return;
2139 #else /* CONFIG_KVM_HYPERV */
2140 	KVM_BUG_ON(1, vmx->vcpu.kvm);
2141 #endif /* CONFIG_KVM_HYPERV */
2142 }
2143 
2144 /*
2145  * This is an equivalent of the nested hypervisor executing the vmptrld
2146  * instruction.
2147  */
2148 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
2149 	struct kvm_vcpu *vcpu, bool from_launch)
2150 {
2151 #ifdef CONFIG_KVM_HYPERV
2152 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2153 	bool evmcs_gpa_changed = false;
2154 	u64 evmcs_gpa;
2155 
2156 	if (likely(!guest_cpu_cap_has_evmcs(vcpu)))
2157 		return EVMPTRLD_DISABLED;
2158 
2159 	evmcs_gpa = nested_get_evmptr(vcpu);
2160 	if (!evmptr_is_valid(evmcs_gpa)) {
2161 		nested_release_evmcs(vcpu);
2162 		return EVMPTRLD_DISABLED;
2163 	}
2164 
2165 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
2166 		vmx->nested.current_vmptr = INVALID_GPA;
2167 
2168 		nested_release_evmcs(vcpu);
2169 
2170 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2171 				 &vmx->nested.hv_evmcs_map))
2172 			return EVMPTRLD_ERROR;
2173 
2174 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2175 
2176 		/*
2177 		 * Currently, KVM only supports eVMCS version 1
2178 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2179 		 * value to first u32 field of eVMCS which should specify eVMCS
2180 		 * VersionNumber.
2181 		 *
2182 		 * Guest should be aware of supported eVMCS versions by host by
2183 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2184 		 * expected to set this CPUID leaf according to the value
2185 		 * returned in vmcs_version from nested_enable_evmcs().
2186 		 *
2187 		 * However, it turns out that Microsoft Hyper-V fails to comply
2188 		 * to their own invented interface: When Hyper-V use eVMCS, it
2189 		 * just sets first u32 field of eVMCS to revision_id specified
2190 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2191 		 * which is one of the supported versions specified in
2192 		 * CPUID.0x4000000A.EAX[0:15].
2193 		 *
2194 		 * To overcome Hyper-V bug, we accept here either a supported
2195 		 * eVMCS version or VMCS12 revision_id as valid values for first
2196 		 * u32 field of eVMCS.
2197 		 */
2198 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2199 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2200 			nested_release_evmcs(vcpu);
2201 			return EVMPTRLD_VMFAIL;
2202 		}
2203 
2204 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2205 
2206 		evmcs_gpa_changed = true;
2207 		/*
2208 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2209 		 * reloaded from guest's memory (read only fields, fields not
2210 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2211 		 * are no leftovers.
2212 		 */
2213 		if (from_launch) {
2214 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2215 			memset(vmcs12, 0, sizeof(*vmcs12));
2216 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2217 		}
2218 
2219 	}
2220 
2221 	/*
2222 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2223 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2224 	 */
2225 	if (from_launch || evmcs_gpa_changed) {
2226 		vmx->nested.hv_evmcs->hv_clean_fields &=
2227 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2228 
2229 		vmx->nested.force_msr_bitmap_recalc = true;
2230 	}
2231 
2232 	return EVMPTRLD_SUCCEEDED;
2233 #else
2234 	return EVMPTRLD_DISABLED;
2235 #endif
2236 }
2237 
2238 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2239 {
2240 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2241 
2242 	if (nested_vmx_is_evmptr12_valid(vmx))
2243 		copy_vmcs12_to_enlightened(vmx);
2244 	else
2245 		copy_vmcs12_to_shadow(vmx);
2246 
2247 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2248 }
2249 
2250 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2251 {
2252 	struct vcpu_vmx *vmx =
2253 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2254 
2255 	vmx->nested.preemption_timer_expired = true;
2256 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2257 	kvm_vcpu_kick(&vmx->vcpu);
2258 
2259 	return HRTIMER_NORESTART;
2260 }
2261 
2262 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2263 {
2264 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2265 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2266 
2267 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2268 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2269 
2270 	if (!vmx->nested.has_preemption_timer_deadline) {
2271 		vmx->nested.preemption_timer_deadline =
2272 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2273 		vmx->nested.has_preemption_timer_deadline = true;
2274 	}
2275 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2276 }
2277 
2278 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2279 					u64 preemption_timeout)
2280 {
2281 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2282 
2283 	/*
2284 	 * A timer value of zero is architecturally guaranteed to cause
2285 	 * a VMExit prior to executing any instructions in the guest.
2286 	 */
2287 	if (preemption_timeout == 0) {
2288 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2289 		return;
2290 	}
2291 
2292 	if (vcpu->arch.virtual_tsc_khz == 0)
2293 		return;
2294 
2295 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2296 	preemption_timeout *= 1000000;
2297 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2298 	hrtimer_start(&vmx->nested.preemption_timer,
2299 		      ktime_add_ns(ktime_get(), preemption_timeout),
2300 		      HRTIMER_MODE_ABS_PINNED);
2301 }
2302 
2303 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2304 {
2305 	if (vmx->vcpu.arch.nested_run_pending &&
2306 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2307 		return vmcs12->guest_ia32_efer;
2308 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2309 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2310 	else
2311 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2312 }
2313 
2314 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2315 {
2316 	struct kvm *kvm = vmx->vcpu.kvm;
2317 
2318 	/*
2319 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2320 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2321 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2322 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2323 	 */
2324 	if (vmx->nested.vmcs02_initialized)
2325 		return;
2326 	vmx->nested.vmcs02_initialized = true;
2327 
2328 	if (vmx->ve_info)
2329 		vmcs_write64(VE_INFORMATION_ADDRESS, __pa(vmx->ve_info));
2330 
2331 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2332 	if (cpu_has_vmx_vmfunc())
2333 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2334 
2335 	if (cpu_has_vmx_posted_intr())
2336 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2337 
2338 	if (cpu_has_vmx_msr_bitmap())
2339 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2340 
2341 	/*
2342 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2343 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2344 	 * deal with filtering out L2 GPAs from the buffer.
2345 	 */
2346 	if (enable_pml) {
2347 		vmcs_write64(PML_ADDRESS, 0);
2348 		vmcs_write16(GUEST_PML_INDEX, -1);
2349 	}
2350 
2351 	if (cpu_has_vmx_encls_vmexit())
2352 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2353 
2354 	if (kvm_notify_vmexit_enabled(kvm))
2355 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
2356 
2357 	/*
2358 	 * Set the MSR load/store lists to match L0's settings.  Only the
2359 	 * addresses are constant (for vmcs02), the counts can change based
2360 	 * on L2's behavior, e.g. switching to/from long mode.
2361 	 */
2362 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.val));
2363 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2364 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2365 
2366 	vmx_set_constant_host_state(vmx);
2367 }
2368 
2369 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2370 				      struct vmcs12 *vmcs12)
2371 {
2372 	prepare_vmcs02_constant_state(vmx);
2373 
2374 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2375 
2376 	/*
2377 	 * If VPID is disabled, then guest TLB accesses use VPID=0, i.e. the
2378 	 * same VPID as the host.  Emulate this behavior by using vpid01 for L2
2379 	 * if VPID is disabled in vmcs12.  Note, if VPID is disabled, VM-Enter
2380 	 * and VM-Exit are architecturally required to flush VPID=0, but *only*
2381 	 * VPID=0.  I.e. using vpid02 would be ok (so long as KVM emulates the
2382 	 * required flushes), but doing so would cause KVM to over-flush.  E.g.
2383 	 * if L1 runs L2 X with VPID12=1, then runs L2 Y with VPID12 disabled,
2384 	 * and then runs L2 X again, then KVM can and should retain TLB entries
2385 	 * for VPID12=1.
2386 	 */
2387 	if (enable_vpid) {
2388 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2389 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2390 		else
2391 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2392 	}
2393 }
2394 
2395 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2396 				 struct vmcs12 *vmcs12)
2397 {
2398 	u32 exec_control;
2399 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2400 
2401 	if (vmx->nested.dirty_vmcs12 || nested_vmx_is_evmptr12_valid(vmx))
2402 		prepare_vmcs02_early_rare(vmx, vmcs12);
2403 
2404 	/*
2405 	 * PIN CONTROLS
2406 	 */
2407 	exec_control = __pin_controls_get(vmcs01);
2408 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2409 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2410 
2411 	/* Posted interrupts setting is only taken from vmcs12.  */
2412 	vmx->nested.pi_pending = false;
2413 	if (nested_cpu_has_posted_intr(vmcs12)) {
2414 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2415 	} else {
2416 		vmx->nested.posted_intr_nv = -1;
2417 		exec_control &= ~PIN_BASED_POSTED_INTR;
2418 	}
2419 	pin_controls_set(vmx, exec_control);
2420 
2421 	/*
2422 	 * EXEC CONTROLS
2423 	 */
2424 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2425 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2426 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2427 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2428 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2429 
2430 	if (exec_control & CPU_BASED_TPR_SHADOW)
2431 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2432 #ifdef CONFIG_X86_64
2433 	else
2434 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2435 				CPU_BASED_CR8_STORE_EXITING;
2436 #endif
2437 
2438 	/*
2439 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2440 	 * for I/O port accesses.
2441 	 */
2442 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2443 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2444 
2445 	/*
2446 	 * This bit will be computed in nested_get_vmcs12_pages, because
2447 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2448 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2449 	 * only set/clear this bit.
2450 	 */
2451 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2452 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2453 
2454 	exec_controls_set(vmx, exec_control);
2455 
2456 	/*
2457 	 * SECONDARY EXEC CONTROLS
2458 	 */
2459 	if (cpu_has_secondary_exec_ctrls()) {
2460 		exec_control = __secondary_exec_controls_get(vmcs01);
2461 
2462 		/* Take the following fields only from vmcs12 */
2463 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2464 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2465 				  SECONDARY_EXEC_ENABLE_INVPCID |
2466 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2467 				  SECONDARY_EXEC_ENABLE_XSAVES |
2468 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2469 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2470 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2471 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2472 				  SECONDARY_EXEC_MODE_BASED_EPT_EXEC |
2473 				  SECONDARY_EXEC_DESC);
2474 
2475 		if (nested_cpu_has(vmcs12,
2476 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2477 			exec_control |= vmcs12->secondary_vm_exec_control;
2478 
2479 		/* PML is emulated and never enabled in hardware for L2. */
2480 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2481 
2482 		/* VMCS shadowing for L2 is emulated for now */
2483 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2484 
2485 		/*
2486 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2487 		 * will not have to rewrite the controls just for this bit.
2488 		 */
2489 		if (vmx_umip_emulated() && (vmcs12->guest_cr4 & X86_CR4_UMIP))
2490 			exec_control |= SECONDARY_EXEC_DESC;
2491 
2492 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2493 			vmcs_write16(GUEST_INTR_STATUS,
2494 				vmcs12->guest_intr_status);
2495 
2496 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2497 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2498 
2499 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2500 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2501 
2502 		secondary_exec_controls_set(vmx, exec_control);
2503 	}
2504 
2505 	/*
2506 	 * ENTRY CONTROLS
2507 	 *
2508 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2509 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2510 	 * on the related bits (if supported by the CPU) in the hope that
2511 	 * we can avoid VMWrites during vmx_set_efer().
2512 	 *
2513 	 * Similarly, take vmcs01's PERF_GLOBAL_CTRL in the hope that if KVM is
2514 	 * loading PERF_GLOBAL_CTRL via the VMCS for L1, then KVM will want to
2515 	 * do the same for L2.
2516 	 */
2517 	exec_control = __vm_entry_controls_get(vmcs01);
2518 	exec_control |= (vmcs12->vm_entry_controls &
2519 			 ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
2520 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2521 	if (cpu_has_load_ia32_efer()) {
2522 		if (guest_efer & EFER_LMA)
2523 			exec_control |= VM_ENTRY_IA32E_MODE;
2524 		if (guest_efer != kvm_host.efer)
2525 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2526 	}
2527 	vm_entry_controls_set(vmx, exec_control);
2528 
2529 	/*
2530 	 * EXIT CONTROLS
2531 	 *
2532 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2533 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2534 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2535 	 */
2536 	exec_control = __vm_exit_controls_get(vmcs01);
2537 	if (cpu_has_load_ia32_efer() && guest_efer != kvm_host.efer)
2538 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2539 	else
2540 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2541 	vm_exit_controls_set(vmx, exec_control);
2542 
2543 	/*
2544 	 * Interrupt/Exception Fields
2545 	 */
2546 	if (vmx->vcpu.arch.nested_run_pending) {
2547 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2548 			     vmcs12->vm_entry_intr_info_field);
2549 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2550 			     vmcs12->vm_entry_exception_error_code);
2551 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2552 			     vmcs12->vm_entry_instruction_len);
2553 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2554 			     vmcs12->guest_interruptibility_info);
2555 		vmx->loaded_vmcs->nmi_known_unmasked =
2556 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2557 	} else {
2558 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2559 	}
2560 }
2561 
2562 static void vmcs_read_cet_state(struct kvm_vcpu *vcpu, u64 *s_cet,
2563 				u64 *ssp, u64 *ssp_tbl)
2564 {
2565 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_IBT) ||
2566 	    guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK))
2567 		*s_cet = vmcs_readl(GUEST_S_CET);
2568 
2569 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK)) {
2570 		*ssp = vmcs_readl(GUEST_SSP);
2571 		*ssp_tbl = vmcs_readl(GUEST_INTR_SSP_TABLE);
2572 	}
2573 }
2574 
2575 static void vmcs_write_cet_state(struct kvm_vcpu *vcpu, u64 s_cet,
2576 				 u64 ssp, u64 ssp_tbl)
2577 {
2578 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_IBT) ||
2579 	    guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK))
2580 		vmcs_writel(GUEST_S_CET, s_cet);
2581 
2582 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_SHSTK)) {
2583 		vmcs_writel(GUEST_SSP, ssp);
2584 		vmcs_writel(GUEST_INTR_SSP_TABLE, ssp_tbl);
2585 	}
2586 }
2587 
2588 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2589 {
2590 	struct hv_enlightened_vmcs *hv_evmcs = nested_vmx_evmcs(vmx);
2591 
2592 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2593 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2594 
2595 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2596 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2597 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2598 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2599 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2600 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2601 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2602 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2603 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2604 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2605 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2606 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2607 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2608 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2609 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2610 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2611 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2612 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2613 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2614 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2615 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2616 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2617 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2618 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2619 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2620 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2621 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2622 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2623 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2624 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2625 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2626 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2627 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2628 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2629 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2630 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2631 
2632 		vmx_segment_cache_clear(vmx);
2633 	}
2634 
2635 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2636 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2637 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2638 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2639 			    vmcs12->guest_pending_dbg_exceptions);
2640 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2641 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2642 
2643 		/*
2644 		 * L1 may access the L2's PDPTR, so save them to construct
2645 		 * vmcs12
2646 		 */
2647 		if (enable_ept) {
2648 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2649 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2650 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2651 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2652 		}
2653 
2654 		if (kvm_mpx_supported() && vmx->vcpu.arch.nested_run_pending &&
2655 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2656 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2657 	}
2658 
2659 	if (nested_cpu_has_xsaves(vmcs12))
2660 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2661 
2662 	/*
2663 	 * Whether page-faults are trapped is determined by a combination of
2664 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2665 	 * doesn't care about page faults then we should set all of these to
2666 	 * L1's desires. However, if L0 does care about (some) page faults, it
2667 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2668 	 * simply ask to exit on each and every L2 page fault. This is done by
2669 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2670 	 * Note that below we don't need special code to set EB.PF beyond the
2671 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2672 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2673 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2674 	 */
2675 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2676 		/*
2677 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2678 		 * go ahead and use it?
2679 		 */
2680 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2681 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2682 	} else {
2683 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2684 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2685 	}
2686 
2687 	if (cpu_has_vmx_apicv()) {
2688 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2689 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2690 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2691 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2692 	}
2693 
2694 	/*
2695 	 * If vmcs12 is configured to save TSC on exit via the auto-store list,
2696 	 * append the MSR to vmcs02's auto-store list so that KVM effectively
2697 	 * reads TSC at the time of VM-Exit from L2.  The saved value will be
2698 	 * propagated to vmcs12's list on nested VM-Exit.
2699 	 *
2700 	 * Don't increment the number of MSRs in the vCPU structure, as saving
2701 	 * TSC is specific to this particular incarnation of vmcb02, i.e. must
2702 	 * not bleed into vmcs01.
2703 	 */
2704 	if (nested_msr_store_list_has_msr(&vmx->vcpu, MSR_IA32_TSC) &&
2705 	    !WARN_ON_ONCE(vmx->msr_autostore.nr >= ARRAY_SIZE(vmx->msr_autostore.val))) {
2706 		vmx->nested.tsc_autostore_slot = vmx->msr_autostore.nr;
2707 		vmx->msr_autostore.val[vmx->msr_autostore.nr].index = MSR_IA32_TSC;
2708 
2709 		vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.nr + 1);
2710 	} else {
2711 		vmx->nested.tsc_autostore_slot = -1;
2712 		vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.nr);
2713 	}
2714 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2715 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2716 
2717 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE)
2718 		vmcs_write_cet_state(&vmx->vcpu, vmcs12->guest_s_cet,
2719 				     vmcs12->guest_ssp, vmcs12->guest_ssp_tbl);
2720 
2721 	set_cr4_guest_host_mask(vmx);
2722 }
2723 
2724 /*
2725  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2726  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2727  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2728  * guest in a way that will both be appropriate to L1's requests, and our
2729  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2730  * function also has additional necessary side-effects, like setting various
2731  * vcpu->arch fields.
2732  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2733  * is assigned to entry_failure_code on failure.
2734  */
2735 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2736 			  bool from_vmentry,
2737 			  enum vm_entry_failure_code *entry_failure_code)
2738 {
2739 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2740 	struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
2741 	bool load_guest_pdptrs_vmcs12 = false;
2742 
2743 	if (vmx->nested.dirty_vmcs12 || nested_vmx_is_evmptr12_valid(vmx)) {
2744 		prepare_vmcs02_rare(vmx, vmcs12);
2745 		vmx->nested.dirty_vmcs12 = false;
2746 
2747 		load_guest_pdptrs_vmcs12 = !nested_vmx_is_evmptr12_valid(vmx) ||
2748 			!(evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2749 	}
2750 
2751 	if (vcpu->arch.nested_run_pending &&
2752 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2753 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2754 		vmx_guest_debugctl_write(vcpu, vmcs12->guest_ia32_debugctl &
2755 					       vmx_get_supported_debugctl(vcpu, false));
2756 	} else {
2757 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2758 		vmx_guest_debugctl_write(vcpu, vmx->nested.pre_vmenter_debugctl);
2759 	}
2760 
2761 	if (!vcpu->arch.nested_run_pending ||
2762 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE))
2763 		vmcs_write_cet_state(vcpu, vmx->nested.pre_vmenter_s_cet,
2764 				     vmx->nested.pre_vmenter_ssp,
2765 				     vmx->nested.pre_vmenter_ssp_tbl);
2766 
2767 	if (kvm_mpx_supported() && (!vcpu->arch.nested_run_pending ||
2768 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2769 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.pre_vmenter_bndcfgs);
2770 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2771 
2772 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2773 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2774 	 * trap. Note that CR0.TS also needs updating - we do this later.
2775 	 */
2776 	vmx_update_exception_bitmap(vcpu);
2777 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2778 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2779 
2780 	if (vcpu->arch.nested_run_pending &&
2781 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2782 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2783 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2784 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2785 		vmcs_write64(GUEST_IA32_PAT, vcpu->arch.pat);
2786 	}
2787 
2788 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2789 			vcpu->arch.l1_tsc_offset,
2790 			vmx_get_l2_tsc_offset(vcpu),
2791 			vmx_get_l2_tsc_multiplier(vcpu));
2792 
2793 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2794 			vcpu->arch.l1_tsc_scaling_ratio,
2795 			vmx_get_l2_tsc_multiplier(vcpu));
2796 
2797 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2798 	if (kvm_caps.has_tsc_control)
2799 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2800 
2801 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2802 
2803 	if (nested_cpu_has_ept(vmcs12))
2804 		nested_ept_init_mmu_context(vcpu);
2805 
2806 	/*
2807 	 * Override the CR0/CR4 read shadows after setting the effective guest
2808 	 * CR0/CR4.  The common helpers also set the shadows, but they don't
2809 	 * account for vmcs12's cr0/4_guest_host_mask.
2810 	 */
2811 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2812 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2813 
2814 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2815 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2816 
2817 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2818 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2819 	vmx_set_efer(vcpu, vcpu->arch.efer);
2820 
2821 	/*
2822 	 * Guest state is invalid and unrestricted guest is disabled,
2823 	 * which means L1 attempted VMEntry to L2 with invalid state.
2824 	 * Fail the VMEntry.
2825 	 *
2826 	 * However when force loading the guest state (SMM exit or
2827 	 * loading nested state after migration, it is possible to
2828 	 * have invalid guest state now, which will be later fixed by
2829 	 * restoring L2 register state
2830 	 */
2831 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2832 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2833 		return -EINVAL;
2834 	}
2835 
2836 	/* Shadow page tables on either EPT or shadow page tables. */
2837 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2838 				from_vmentry, entry_failure_code))
2839 		return -EINVAL;
2840 
2841 	/*
2842 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2843 	 * on nested VM-Exit, which can occur without actually running L2 and
2844 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2845 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2846 	 * transition to HLT instead of running L2.
2847 	 */
2848 	if (enable_ept)
2849 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2850 
2851 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2852 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2853 	    is_pae_paging(vcpu)) {
2854 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2855 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2856 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2857 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2858 	}
2859 
2860 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2861 	    kvm_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)) &&
2862 	    WARN_ON_ONCE(__kvm_emulate_msr_write(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2863 						 vmcs12->guest_ia32_perf_global_ctrl))) {
2864 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2865 		return -EINVAL;
2866 	}
2867 
2868 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2869 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2870 
2871 	/*
2872 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2873 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2874 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2875 	 * here.
2876 	 */
2877 	if (nested_vmx_is_evmptr12_valid(vmx))
2878 		evmcs->hv_clean_fields |= HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2879 
2880 	return 0;
2881 }
2882 
2883 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2884 {
2885 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2886 	       nested_cpu_has_virtual_nmis(vmcs12)))
2887 		return -EINVAL;
2888 
2889 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2890 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2891 		return -EINVAL;
2892 
2893 	return 0;
2894 }
2895 
2896 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2897 {
2898 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2899 
2900 	/* Check for memory type validity */
2901 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2902 	case VMX_EPTP_MT_UC:
2903 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2904 			return false;
2905 		break;
2906 	case VMX_EPTP_MT_WB:
2907 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2908 			return false;
2909 		break;
2910 	default:
2911 		return false;
2912 	}
2913 
2914 	/* Page-walk levels validity. */
2915 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2916 	case VMX_EPTP_PWL_5:
2917 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2918 			return false;
2919 		break;
2920 	case VMX_EPTP_PWL_4:
2921 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2922 			return false;
2923 		break;
2924 	default:
2925 		return false;
2926 	}
2927 
2928 	/* Reserved bits should not be set */
2929 	if (CC(!kvm_vcpu_is_legal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2930 		return false;
2931 
2932 	/* AD, if set, should be supported */
2933 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2934 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2935 			return false;
2936 	}
2937 
2938 	return true;
2939 }
2940 
2941 /*
2942  * Checks related to VM-Execution Control Fields
2943  */
2944 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2945                                               struct vmcs12 *vmcs12)
2946 {
2947 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2948 
2949 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2950 				   vmx->nested.msrs.pinbased_ctls_low,
2951 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2952 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2953 				   vmx->nested.msrs.procbased_ctls_low,
2954 				   vmx->nested.msrs.procbased_ctls_high)))
2955 		return -EINVAL;
2956 
2957 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2958 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2959 				   vmx->nested.msrs.secondary_ctls_low,
2960 				   vmx->nested.msrs.secondary_ctls_high)))
2961 		return -EINVAL;
2962 
2963 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2964 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2965 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2966 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2967 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2968 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2969 	    nested_vmx_check_nmi_controls(vmcs12) ||
2970 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2971 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2972 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2973 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2974 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2975 		return -EINVAL;
2976 
2977 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2978 	    nested_cpu_has_save_preemption_timer(vmcs12))
2979 		return -EINVAL;
2980 
2981 	if (nested_cpu_has_ept(vmcs12) &&
2982 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2983 		return -EINVAL;
2984 
2985 	if (nested_cpu_has_vmfunc(vmcs12)) {
2986 		if (CC(vmcs12->vm_function_control &
2987 		       ~vmx->nested.msrs.vmfunc_controls))
2988 			return -EINVAL;
2989 
2990 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2991 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2992 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2993 				return -EINVAL;
2994 		}
2995 	}
2996 
2997 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING) &&
2998 	    CC(!vmcs12->tsc_multiplier))
2999 		return -EINVAL;
3000 
3001 	return 0;
3002 }
3003 
3004 /*
3005  * Checks related to VM-Exit Control Fields
3006  */
3007 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
3008                                          struct vmcs12 *vmcs12)
3009 {
3010 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3011 
3012 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
3013 				    vmx->nested.msrs.exit_ctls_low,
3014 				    vmx->nested.msrs.exit_ctls_high)) ||
3015 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
3016 		return -EINVAL;
3017 
3018 	return 0;
3019 }
3020 
3021 /*
3022  * Checks related to VM-Entry Control Fields
3023  */
3024 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
3025 					  struct vmcs12 *vmcs12)
3026 {
3027 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3028 
3029 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
3030 				    vmx->nested.msrs.entry_ctls_low,
3031 				    vmx->nested.msrs.entry_ctls_high)))
3032 		return -EINVAL;
3033 
3034 	/*
3035 	 * From the Intel SDM, volume 3:
3036 	 * Fields relevant to VM-entry event injection must be set properly.
3037 	 * These fields are the VM-entry interruption-information field, the
3038 	 * VM-entry exception error code, and the VM-entry instruction length.
3039 	 */
3040 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
3041 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
3042 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
3043 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
3044 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
3045 		bool urg = nested_cpu_has2(vmcs12,
3046 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
3047 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
3048 
3049 		/* VM-entry interruption-info field: interruption type */
3050 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
3051 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
3052 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
3053 			return -EINVAL;
3054 
3055 		/* VM-entry interruption-info field: vector */
3056 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
3057 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
3058 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
3059 			return -EINVAL;
3060 
3061 		/*
3062 		 * Cannot deliver error code in real mode or if the interrupt
3063 		 * type is not hardware exception. For other cases, do the
3064 		 * consistency check only if the vCPU doesn't enumerate
3065 		 * VMX_BASIC_NO_HW_ERROR_CODE_CC.
3066 		 */
3067 		if (!prot_mode || intr_type != INTR_TYPE_HARD_EXCEPTION) {
3068 			if (CC(has_error_code))
3069 				return -EINVAL;
3070 		} else if (!nested_cpu_has_no_hw_errcode_cc(vcpu)) {
3071 			if (CC(has_error_code != x86_exception_has_error_code(vector)))
3072 				return -EINVAL;
3073 		}
3074 
3075 		/* VM-entry exception error code */
3076 		if (CC(has_error_code &&
3077 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
3078 			return -EINVAL;
3079 
3080 		/* VM-entry interruption-info field: reserved bits */
3081 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
3082 			return -EINVAL;
3083 
3084 		/* VM-entry instruction length */
3085 		switch (intr_type) {
3086 		case INTR_TYPE_SOFT_EXCEPTION:
3087 		case INTR_TYPE_SOFT_INTR:
3088 		case INTR_TYPE_PRIV_SW_EXCEPTION:
3089 			if (CC(vmcs12->vm_entry_instruction_len > X86_MAX_INSTRUCTION_LENGTH) ||
3090 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
3091 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
3092 				return -EINVAL;
3093 		}
3094 	}
3095 
3096 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
3097 		return -EINVAL;
3098 
3099 	return 0;
3100 }
3101 
3102 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
3103 				     struct vmcs12 *vmcs12)
3104 {
3105 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
3106 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
3107 	    nested_check_vm_entry_controls(vcpu, vmcs12))
3108 		return -EINVAL;
3109 
3110 #ifdef CONFIG_KVM_HYPERV
3111 	if (guest_cpu_cap_has_evmcs(vcpu))
3112 		return nested_evmcs_check_controls(vmcs12);
3113 #endif
3114 
3115 	return 0;
3116 }
3117 
3118 static int nested_vmx_check_controls_late(struct kvm_vcpu *vcpu,
3119 					  struct vmcs12 *vmcs12)
3120 {
3121 	void *vapic = to_vmx(vcpu)->nested.virtual_apic_map.hva;
3122 	u32 vtpr = vapic ? (*(u32 *)(vapic + APIC_TASKPRI)) >> 4 : 0;
3123 
3124 	/*
3125 	 * Don't bother with the consistency checks if KVM isn't configured to
3126 	 * WARN on missed consistency checks, as KVM needs to rely on hardware
3127 	 * to fully detect an illegal vTPR vs. TRP Threshold combination due to
3128 	 * the vTPR being writable by L1 at all times (it's an in-memory value,
3129 	 * not a VMCS field).  I.e. even if the check passes now, it might fail
3130 	 * at the actual VM-Enter.
3131 	 *
3132 	 * Keying off the module param also allows treating an invalid vAPIC
3133 	 * mapping as a consistency check failure without increasing the risk
3134 	 * of breaking a "real" VM.
3135 	 */
3136 	if (!warn_on_missed_cc)
3137 		return 0;
3138 
3139 	if ((exec_controls_get(to_vmx(vcpu)) & CPU_BASED_TPR_SHADOW) &&
3140 	    nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW) &&
3141 	    !nested_cpu_has_vid(vmcs12) &&
3142 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
3143 	    (CC(!vapic) ||
3144 	     CC((vmcs12->tpr_threshold & GENMASK(3, 0)) > (vtpr & GENMASK(3, 0)))))
3145 		return -EINVAL;
3146 
3147 	return 0;
3148 }
3149 
3150 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
3151 				       struct vmcs12 *vmcs12)
3152 {
3153 #ifdef CONFIG_X86_64
3154 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
3155 		!!(vcpu->arch.efer & EFER_LMA)))
3156 		return -EINVAL;
3157 #endif
3158 	return 0;
3159 }
3160 
3161 static bool is_l1_noncanonical_address_on_vmexit(u64 la, struct vmcs12 *vmcs12)
3162 {
3163 	/*
3164 	 * Check that the given linear address is canonical after a VM exit
3165 	 * from L2, based on HOST_CR4.LA57 value that will be loaded for L1.
3166 	 */
3167 	u8 l1_address_bits_on_exit = (vmcs12->host_cr4 & X86_CR4_LA57) ? 57 : 48;
3168 
3169 	return !__is_canonical_address(la, l1_address_bits_on_exit);
3170 }
3171 
3172 static int nested_vmx_check_cet_state_common(struct kvm_vcpu *vcpu, u64 s_cet,
3173 					     u64 ssp, u64 ssp_tbl)
3174 {
3175 	if (CC(!kvm_is_valid_u_s_cet(vcpu, s_cet)) || CC(!IS_ALIGNED(ssp, 4)) ||
3176 	    CC(is_noncanonical_msr_address(ssp_tbl, vcpu)))
3177 		return -EINVAL;
3178 
3179 	return 0;
3180 }
3181 
3182 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
3183 				       struct vmcs12 *vmcs12)
3184 {
3185 	bool ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
3186 
3187 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
3188 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
3189 	    CC(!kvm_vcpu_is_legal_cr3(vcpu, vmcs12->host_cr3)))
3190 		return -EINVAL;
3191 
3192 	if (CC(vmcs12->host_cr4 & X86_CR4_CET && !(vmcs12->host_cr0 & X86_CR0_WP)))
3193 		return -EINVAL;
3194 
3195 	if (CC(is_noncanonical_msr_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
3196 	    CC(is_noncanonical_msr_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
3197 		return -EINVAL;
3198 
3199 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
3200 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
3201 		return -EINVAL;
3202 
3203 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3204 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3205 					   vmcs12->host_ia32_perf_global_ctrl)))
3206 		return -EINVAL;
3207 
3208 	if (ia32e) {
3209 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
3210 			return -EINVAL;
3211 	} else {
3212 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
3213 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
3214 		    CC((vmcs12->host_rip) >> 32))
3215 			return -EINVAL;
3216 	}
3217 
3218 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3219 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3220 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3221 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3222 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3223 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3224 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3225 	    CC(vmcs12->host_cs_selector == 0) ||
3226 	    CC(vmcs12->host_tr_selector == 0) ||
3227 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
3228 		return -EINVAL;
3229 
3230 	if (CC(is_noncanonical_base_address(vmcs12->host_fs_base, vcpu)) ||
3231 	    CC(is_noncanonical_base_address(vmcs12->host_gs_base, vcpu)) ||
3232 	    CC(is_noncanonical_base_address(vmcs12->host_gdtr_base, vcpu)) ||
3233 	    CC(is_noncanonical_base_address(vmcs12->host_idtr_base, vcpu)) ||
3234 	    CC(is_noncanonical_base_address(vmcs12->host_tr_base, vcpu)) ||
3235 	    CC(is_l1_noncanonical_address_on_vmexit(vmcs12->host_rip, vmcs12)))
3236 		return -EINVAL;
3237 
3238 	/*
3239 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
3240 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
3241 	 * the values of the LMA and LME bits in the field must each be that of
3242 	 * the host address-space size VM-exit control.
3243 	 */
3244 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
3245 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
3246 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
3247 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
3248 			return -EINVAL;
3249 	}
3250 
3251 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_CET_STATE) {
3252 		if (nested_vmx_check_cet_state_common(vcpu, vmcs12->host_s_cet,
3253 						      vmcs12->host_ssp,
3254 						      vmcs12->host_ssp_tbl))
3255 			return -EINVAL;
3256 
3257 		/*
3258 		 * IA32_S_CET and SSP must be canonical if the host will
3259 		 * enter 64-bit mode after VM-exit; otherwise, higher
3260 		 * 32-bits must be all 0s.
3261 		 */
3262 		if (ia32e) {
3263 			if (CC(is_noncanonical_msr_address(vmcs12->host_s_cet, vcpu)) ||
3264 			    CC(is_noncanonical_msr_address(vmcs12->host_ssp, vcpu)))
3265 				return -EINVAL;
3266 		} else {
3267 			if (CC(vmcs12->host_s_cet >> 32) || CC(vmcs12->host_ssp >> 32))
3268 				return -EINVAL;
3269 		}
3270 	}
3271 
3272 	return 0;
3273 }
3274 
3275 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
3276 					  struct vmcs12 *vmcs12)
3277 {
3278 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3279 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
3280 	struct vmcs_hdr hdr;
3281 
3282 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
3283 		return 0;
3284 
3285 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
3286 		return -EINVAL;
3287 
3288 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
3289 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
3290 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
3291                 return -EINVAL;
3292 
3293 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
3294 					    offsetof(struct vmcs12, hdr),
3295 					    sizeof(hdr))))
3296 		return -EINVAL;
3297 
3298 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
3299 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
3300 		return -EINVAL;
3301 
3302 	return 0;
3303 }
3304 
3305 /*
3306  * Checks related to Guest Non-register State
3307  */
3308 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
3309 {
3310 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
3311 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
3312 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
3313 		return -EINVAL;
3314 
3315 	return 0;
3316 }
3317 
3318 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
3319 					struct vmcs12 *vmcs12,
3320 					enum vm_entry_failure_code *entry_failure_code)
3321 {
3322 	bool ia32e = !!(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE);
3323 
3324 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
3325 
3326 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3327 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3328 		return -EINVAL;
3329 
3330 	if (CC(vmcs12->guest_cr4 & X86_CR4_CET && !(vmcs12->guest_cr0 & X86_CR0_WP)))
3331 		return -EINVAL;
3332 
3333 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
3334 		u64 debugctl = vmcs12->guest_ia32_debugctl;
3335 
3336 		/*
3337 		 * FREEZE_IN_SMM is not virtualized, but allow L1 to set it in
3338 		 * vmcs12's DEBUGCTL under a quirk for backwards compatibility.
3339 		 * Note that the quirk only relaxes the consistency check.  The
3340 		 * vmcc02 bit is still under the control of the host.  In
3341 		 * particular, if a host administrator decides to clear the bit,
3342 		 * then L1 has no say in the matter.
3343 		 */
3344 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_VMCS12_ALLOW_FREEZE_IN_SMM))
3345 			debugctl &= ~DEBUGCTLMSR_FREEZE_IN_SMM;
3346 
3347 		if (CC(!kvm_dr7_valid(vmcs12->guest_dr7)) ||
3348 		    CC(!vmx_is_valid_debugctl(vcpu, debugctl, false)))
3349 			return -EINVAL;
3350 	}
3351 
3352 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3353 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3354 		return -EINVAL;
3355 
3356 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3357 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3358 		return -EINVAL;
3359 	}
3360 
3361 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3362 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3363 					   vmcs12->guest_ia32_perf_global_ctrl)))
3364 		return -EINVAL;
3365 
3366 	if (CC((vmcs12->guest_cr0 & (X86_CR0_PG | X86_CR0_PE)) == X86_CR0_PG))
3367 		return -EINVAL;
3368 
3369 	if (CC(ia32e && !(vmcs12->guest_cr4 & X86_CR4_PAE)) ||
3370 	    CC(ia32e && !(vmcs12->guest_cr0 & X86_CR0_PG)))
3371 		return -EINVAL;
3372 
3373 	/*
3374 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3375 	 * are performed on the field for the IA32_EFER MSR:
3376 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3377 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3378 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3379 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3380 	 *   CR0.PG) is 1.
3381 	 */
3382 	if (vcpu->arch.nested_run_pending &&
3383 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3384 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3385 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3386 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3387 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3388 			return -EINVAL;
3389 	}
3390 
3391 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3392 	    (CC(is_noncanonical_msr_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3393 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3394 		return -EINVAL;
3395 
3396 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE) {
3397 		if (nested_vmx_check_cet_state_common(vcpu, vmcs12->guest_s_cet,
3398 						      vmcs12->guest_ssp,
3399 						      vmcs12->guest_ssp_tbl))
3400 			return -EINVAL;
3401 
3402 		/*
3403 		 * Guest SSP must have 63:N bits identical, rather than
3404 		 * be canonical (i.e., 63:N-1 bits identical), where N is
3405 		 * the CPU's maximum linear-address width. Similar to
3406 		 * is_noncanonical_msr_address(), use the host's
3407 		 * linear-address width.
3408 		 */
3409 		if (CC(!__is_canonical_address(vmcs12->guest_ssp, max_host_virt_addr_bits() + 1)))
3410 			return -EINVAL;
3411 	}
3412 
3413 	if (nested_check_guest_non_reg_state(vmcs12))
3414 		return -EINVAL;
3415 
3416 	return 0;
3417 }
3418 
3419 #ifdef CONFIG_KVM_HYPERV
3420 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3421 {
3422 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3423 
3424 	/*
3425 	 * hv_evmcs may end up being not mapped after migration (when
3426 	 * L2 was running), map it here to make sure vmcs12 changes are
3427 	 * properly reflected.
3428 	 */
3429 	if (guest_cpu_cap_has_evmcs(vcpu) &&
3430 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3431 		enum nested_evmptrld_status evmptrld_status =
3432 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3433 
3434 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3435 		    evmptrld_status == EVMPTRLD_ERROR)
3436 			return false;
3437 
3438 		/*
3439 		 * Post migration VMCS12 always provides the most actual
3440 		 * information, copy it to eVMCS upon entry.
3441 		 */
3442 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3443 	}
3444 
3445 	return true;
3446 }
3447 #endif
3448 
3449 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3450 {
3451 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3452 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3453 	struct kvm_host_map *map;
3454 
3455 	if (!vcpu->arch.pdptrs_from_userspace &&
3456 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3457 		/*
3458 		 * Reload the guest's PDPTRs since after a migration
3459 		 * the guest CR3 might be restored prior to setting the nested
3460 		 * state which can lead to a load of wrong PDPTRs.
3461 		 */
3462 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3463 			return false;
3464 	}
3465 
3466 
3467 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3468 		map = &vmx->nested.apic_access_page_map;
3469 
3470 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->apic_access_addr), map)) {
3471 			vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(map->pfn));
3472 		} else {
3473 			pr_debug_ratelimited("%s: no backing for APIC-access address in vmcs12\n",
3474 					     __func__);
3475 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3476 			vcpu->run->internal.suberror =
3477 				KVM_INTERNAL_ERROR_EMULATION;
3478 			vcpu->run->internal.ndata = 0;
3479 			return false;
3480 		}
3481 	}
3482 
3483 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3484 		map = &vmx->nested.virtual_apic_map;
3485 
3486 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3487 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3488 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3489 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3490 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3491 			/*
3492 			 * The processor will never use the TPR shadow, simply
3493 			 * clear the bit from the execution control.  Such a
3494 			 * configuration is useless, but it happens in tests.
3495 			 * For any other configuration, failing the vm entry is
3496 			 * _not_ what the processor does but it's basically the
3497 			 * only possibility we have.
3498 			 */
3499 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3500 		} else {
3501 			/*
3502 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3503 			 * force VM-Entry to fail.
3504 			 */
3505 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3506 		}
3507 	}
3508 
3509 	if (nested_cpu_has_posted_intr(vmcs12)) {
3510 		map = &vmx->nested.pi_desc_map;
3511 
3512 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3513 			vmx->nested.pi_desc =
3514 				(struct pi_desc *)(((void *)map->hva) +
3515 				offset_in_page(vmcs12->posted_intr_desc_addr));
3516 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3517 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3518 		} else {
3519 			/*
3520 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3521 			 * access the contents of the VMCS12 posted interrupt
3522 			 * descriptor. (Note that KVM may do this when it
3523 			 * should not, per the architectural specification.)
3524 			 */
3525 			vmx->nested.pi_desc = NULL;
3526 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3527 		}
3528 	}
3529 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3530 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3531 	else
3532 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3533 
3534 	return true;
3535 }
3536 
3537 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3538 {
3539 #ifdef CONFIG_KVM_HYPERV
3540 	/*
3541 	 * Note: nested_get_evmcs_page() also updates 'vp_assist_page' copy
3542 	 * in 'struct kvm_vcpu_hv' in case eVMCS is in use, this is mandatory
3543 	 * to make nested_evmcs_l2_tlb_flush_enabled() work correctly post
3544 	 * migration.
3545 	 */
3546 	if (!nested_get_evmcs_page(vcpu)) {
3547 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3548 				     __func__);
3549 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3550 		vcpu->run->internal.suberror =
3551 			KVM_INTERNAL_ERROR_EMULATION;
3552 		vcpu->run->internal.ndata = 0;
3553 
3554 		return false;
3555 	}
3556 #endif
3557 
3558 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3559 		return false;
3560 
3561 	return true;
3562 }
3563 
3564 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3565 {
3566 	struct vmcs12 *vmcs12;
3567 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3568 	gpa_t dst;
3569 
3570 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3571 		return 0;
3572 
3573 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3574 		return 1;
3575 
3576 	/*
3577 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3578 	 * set is already checked as part of A/D emulation.
3579 	 */
3580 	vmcs12 = get_vmcs12(vcpu);
3581 	if (!nested_cpu_has_pml(vmcs12))
3582 		return 0;
3583 
3584 	if (vmcs12->guest_pml_index >= PML_LOG_NR_ENTRIES) {
3585 		vmx->nested.pml_full = true;
3586 		return 1;
3587 	}
3588 
3589 	gpa &= ~0xFFFull;
3590 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3591 
3592 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3593 				 offset_in_page(dst), sizeof(gpa)))
3594 		return 0;
3595 
3596 	vmcs12->guest_pml_index--;
3597 
3598 	return 0;
3599 }
3600 
3601 /*
3602  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3603  * for running VMX instructions (except VMXON, whose prerequisites are
3604  * slightly different). It also specifies what exception to inject otherwise.
3605  * Note that many of these exceptions have priority over VM exits, so they
3606  * don't have to be checked again here.
3607  */
3608 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3609 {
3610 	if (!to_vmx(vcpu)->nested.vmxon) {
3611 		kvm_queue_exception(vcpu, UD_VECTOR);
3612 		return 0;
3613 	}
3614 
3615 	if (vmx_get_cpl(vcpu)) {
3616 		kvm_inject_gp(vcpu, 0);
3617 		return 0;
3618 	}
3619 
3620 	return 1;
3621 }
3622 
3623 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3624 				   struct vmcs12 *vmcs12);
3625 
3626 /*
3627  * If from_vmentry is false, this is being called from state restore (either RSM
3628  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3629  *
3630  * Returns:
3631  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3632  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3633  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3634  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3635  */
3636 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3637 							bool from_vmentry)
3638 {
3639 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3640 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3641 	enum vm_entry_failure_code entry_failure_code;
3642 	union vmx_exit_reason exit_reason = {
3643 		.basic = EXIT_REASON_INVALID_STATE,
3644 		.failed_vmentry = 1,
3645 	};
3646 	u32 failed_index;
3647 
3648 	trace_kvm_nested_vmenter(kvm_rip_read(vcpu),
3649 				 vmx->nested.current_vmptr,
3650 				 vmcs12->guest_rip,
3651 				 vmcs12->guest_intr_status,
3652 				 vmcs12->vm_entry_intr_info_field,
3653 				 vmcs12->secondary_vm_exec_control & SECONDARY_EXEC_ENABLE_EPT,
3654 				 vmcs12->ept_pointer,
3655 				 vmcs12->guest_cr3,
3656 				 KVM_ISA_VMX);
3657 
3658 	kvm_service_local_tlb_flush_requests(vcpu);
3659 
3660 	if (!vcpu->arch.nested_run_pending ||
3661 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3662 		vmx->nested.pre_vmenter_debugctl = vmx_guest_debugctl_read();
3663 	if (kvm_mpx_supported() &&
3664 	    (!vcpu->arch.nested_run_pending ||
3665 	     !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
3666 		vmx->nested.pre_vmenter_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3667 
3668 	if (!vcpu->arch.nested_run_pending ||
3669 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_CET_STATE))
3670 		vmcs_read_cet_state(vcpu, &vmx->nested.pre_vmenter_s_cet,
3671 				    &vmx->nested.pre_vmenter_ssp,
3672 				    &vmx->nested.pre_vmenter_ssp_tbl);
3673 
3674 	/*
3675 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled.  In the
3676 	 * event of a "late" VM-Fail, i.e. a VM-Fail detected by hardware but
3677 	 * not KVM, KVM must unwind its software model to the pre-VM-Entry host
3678 	 * state.  When EPT is disabled, GUEST_CR3 holds KVM's shadow CR3, not
3679 	 * L1's "real" CR3, which causes nested_vmx_restore_host_state() to
3680 	 * corrupt vcpu->arch.cr3.  Stuffing vmcs01.GUEST_CR3 results in the
3681 	 * unwind naturally setting arch.cr3 to the correct value.  Smashing
3682 	 * vmcs01.GUEST_CR3 is safe because nested VM-Exits, and the unwind,
3683 	 * reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is guaranteed to be
3684 	 * overwritten with a shadow CR3 prior to re-entering L1.
3685 	 */
3686 	if (!enable_ept)
3687 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3688 
3689 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3690 
3691 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3692 
3693 	if (from_vmentry) {
3694 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3695 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3696 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3697 		}
3698 
3699 		if (nested_vmx_check_controls_late(vcpu, vmcs12)) {
3700 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3701 			return NVMX_VMENTRY_VMFAIL;
3702 		}
3703 
3704 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3705 						 &entry_failure_code)) {
3706 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3707 			vmcs12->exit_qualification = entry_failure_code;
3708 			goto vmentry_fail_vmexit;
3709 		}
3710 	}
3711 
3712 	enter_guest_mode(vcpu);
3713 
3714 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3715 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3716 		vmcs12->exit_qualification = entry_failure_code;
3717 		goto vmentry_fail_vmexit_guest_mode;
3718 	}
3719 
3720 	if (from_vmentry) {
3721 		failed_index = nested_vmx_load_msr(vcpu,
3722 						   vmcs12->vm_entry_msr_load_addr,
3723 						   vmcs12->vm_entry_msr_load_count);
3724 		if (failed_index) {
3725 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3726 			vmcs12->exit_qualification = failed_index;
3727 			goto vmentry_fail_vmexit_guest_mode;
3728 		}
3729 	} else {
3730 		/*
3731 		 * The MMU is not initialized to point at the right entities yet and
3732 		 * "get pages" would need to read data from the guest (i.e. we will
3733 		 * need to perform gpa to hpa translation). Request a call
3734 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3735 		 * have already been set at vmentry time and should not be reset.
3736 		 */
3737 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3738 	}
3739 
3740 	/*
3741 	 * Re-evaluate pending events if L1 had a pending IRQ/NMI/INIT/SIPI
3742 	 * when it executed VMLAUNCH/VMRESUME, as entering non-root mode can
3743 	 * effectively unblock various events, e.g. INIT/SIPI cause VM-Exit
3744 	 * unconditionally.  Take care to pull data from vmcs01 as appropriate,
3745 	 * e.g. when checking for interrupt windows, as vmcs02 is now loaded.
3746 	 */
3747 	if ((__exec_controls_get(&vmx->vmcs01) & (CPU_BASED_INTR_WINDOW_EXITING |
3748 						  CPU_BASED_NMI_WINDOW_EXITING)) ||
3749 	    kvm_apic_has_pending_init_or_sipi(vcpu) ||
3750 	    kvm_apic_has_interrupt(vcpu))
3751 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3752 
3753 	/*
3754 	 * Do not start the preemption timer hrtimer until after we know
3755 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3756 	 * the timer.
3757 	 */
3758 	vmx->nested.preemption_timer_expired = false;
3759 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3760 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3761 		vmx_start_preemption_timer(vcpu, timer_value);
3762 	}
3763 
3764 	/*
3765 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3766 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3767 	 * returned as far as L1 is concerned. It will only return (and set
3768 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3769 	 */
3770 	return NVMX_VMENTRY_SUCCESS;
3771 
3772 	/*
3773 	 * A failed consistency check that leads to a VMExit during L1's
3774 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3775 	 * 26.7 "VM-entry failures during or after loading guest state".
3776 	 */
3777 vmentry_fail_vmexit_guest_mode:
3778 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3779 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3780 	leave_guest_mode(vcpu);
3781 
3782 vmentry_fail_vmexit:
3783 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3784 
3785 	if (!from_vmentry)
3786 		return NVMX_VMENTRY_VMEXIT;
3787 
3788 	load_vmcs12_host_state(vcpu, vmcs12);
3789 	vmcs12->vm_exit_reason = exit_reason.full;
3790 	if (enable_shadow_vmcs || nested_vmx_is_evmptr12_valid(vmx))
3791 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3792 	return NVMX_VMENTRY_VMEXIT;
3793 }
3794 
3795 /*
3796  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3797  * for running an L2 nested guest.
3798  */
3799 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3800 {
3801 	struct vmcs12 *vmcs12;
3802 	enum nvmx_vmentry_status status;
3803 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3804 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3805 	enum nested_evmptrld_status evmptrld_status;
3806 
3807 	if (!nested_vmx_check_permission(vcpu))
3808 		return 1;
3809 
3810 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3811 	if (evmptrld_status == EVMPTRLD_ERROR) {
3812 		kvm_queue_exception(vcpu, UD_VECTOR);
3813 		return 1;
3814 	}
3815 
3816 	kvm_pmu_branch_retired(vcpu);
3817 
3818 	if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3819 		return nested_vmx_failInvalid(vcpu);
3820 
3821 	if (CC(!nested_vmx_is_evmptr12_valid(vmx) &&
3822 	       vmx->nested.current_vmptr == INVALID_GPA))
3823 		return nested_vmx_failInvalid(vcpu);
3824 
3825 	vmcs12 = get_vmcs12(vcpu);
3826 
3827 	/*
3828 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3829 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3830 	 * rather than RFLAGS.ZF, and no error number is stored to the
3831 	 * VM-instruction error field.
3832 	 */
3833 	if (CC(vmcs12->hdr.shadow_vmcs))
3834 		return nested_vmx_failInvalid(vcpu);
3835 
3836 	if (nested_vmx_is_evmptr12_valid(vmx)) {
3837 		struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
3838 
3839 		copy_enlightened_to_vmcs12(vmx, evmcs->hv_clean_fields);
3840 		/* Enlightened VMCS doesn't have launch state */
3841 		vmcs12->launch_state = !launch;
3842 	} else if (enable_shadow_vmcs) {
3843 		copy_shadow_to_vmcs12(vmx);
3844 	}
3845 
3846 	/*
3847 	 * The nested entry process starts with enforcing various prerequisites
3848 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3849 	 * they fail: As the SDM explains, some conditions should cause the
3850 	 * instruction to fail, while others will cause the instruction to seem
3851 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3852 	 * To speed up the normal (success) code path, we should avoid checking
3853 	 * for misconfigurations which will anyway be caught by the processor
3854 	 * when using the merged vmcs02.
3855 	 */
3856 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3857 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3858 
3859 	if (CC(vmcs12->launch_state == launch))
3860 		return nested_vmx_fail(vcpu,
3861 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3862 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3863 
3864 	if (nested_vmx_check_controls(vcpu, vmcs12))
3865 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3866 
3867 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3868 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3869 
3870 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3871 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3872 
3873 	/*
3874 	 * We're finally done with prerequisite checking, and can start with
3875 	 * the nested entry.
3876 	 */
3877 	vcpu->arch.nested_run_pending = KVM_NESTED_RUN_PENDING;
3878 	vmx->nested.has_preemption_timer_deadline = false;
3879 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3880 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3881 		goto vmentry_failed;
3882 
3883 	/* Hide L1D cache contents from the nested guest.  */
3884 	kvm_request_l1tf_flush_l1d();
3885 
3886 	/*
3887 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3888 	 * also be used as part of restoring nVMX state for
3889 	 * snapshot restore (migration).
3890 	 *
3891 	 * In this flow, it is assumed that vmcs12 cache was
3892 	 * transferred as part of captured nVMX state and should
3893 	 * therefore not be read from guest memory (which may not
3894 	 * exist on destination host yet).
3895 	 */
3896 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3897 
3898 	switch (vmcs12->guest_activity_state) {
3899 	case GUEST_ACTIVITY_HLT:
3900 		/*
3901 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3902 		 * awakened by event injection or by an NMI-window VM-exit or
3903 		 * by an interrupt-window VM-exit, halt the vcpu.
3904 		 */
3905 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3906 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3907 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3908 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3909 			vcpu->arch.nested_run_pending = 0;
3910 			return kvm_emulate_halt_noskip(vcpu);
3911 		}
3912 		break;
3913 	case GUEST_ACTIVITY_WAIT_SIPI:
3914 		vcpu->arch.nested_run_pending = 0;
3915 		kvm_set_mp_state(vcpu, KVM_MP_STATE_INIT_RECEIVED);
3916 		break;
3917 	default:
3918 		break;
3919 	}
3920 
3921 	return 1;
3922 
3923 vmentry_failed:
3924 	vcpu->arch.nested_run_pending = 0;
3925 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3926 		return 0;
3927 	if (status == NVMX_VMENTRY_VMEXIT)
3928 		return 1;
3929 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3930 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3931 }
3932 
3933 /*
3934  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3935  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3936  * This function returns the new value we should put in vmcs12.guest_cr0.
3937  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3938  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3939  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3940  *     didn't trap the bit, because if L1 did, so would L0).
3941  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3942  *     been modified by L2, and L1 knows it. So just leave the old value of
3943  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3944  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3945  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3946  *     changed these bits, and therefore they need to be updated, but L0
3947  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3948  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3949  */
3950 static inline unsigned long
3951 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3952 {
3953 	return
3954 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3955 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3956 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3957 			vcpu->arch.cr0_guest_owned_bits));
3958 }
3959 
3960 static inline unsigned long
3961 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3962 {
3963 	return
3964 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3965 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3966 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3967 			vcpu->arch.cr4_guest_owned_bits));
3968 }
3969 
3970 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3971 				      struct vmcs12 *vmcs12,
3972 				      u32 vm_exit_reason, u32 exit_intr_info)
3973 {
3974 	u32 idt_vectoring;
3975 	unsigned int nr;
3976 
3977 	/*
3978 	 * Per the SDM, VM-Exits due to double and triple faults are never
3979 	 * considered to occur during event delivery, even if the double/triple
3980 	 * fault is the result of an escalating vectoring issue.
3981 	 *
3982 	 * Note, the SDM qualifies the double fault behavior with "The original
3983 	 * event results in a double-fault exception".  It's unclear why the
3984 	 * qualification exists since exits due to double fault can occur only
3985 	 * while vectoring a different exception (injected events are never
3986 	 * subject to interception), i.e. there's _always_ an original event.
3987 	 *
3988 	 * The SDM also uses NMI as a confusing example for the "original event
3989 	 * causes the VM exit directly" clause.  NMI isn't special in any way,
3990 	 * the same rule applies to all events that cause an exit directly.
3991 	 * NMI is an odd choice for the example because NMIs can only occur on
3992 	 * instruction boundaries, i.e. they _can't_ occur during vectoring.
3993 	 */
3994 	if ((u16)vm_exit_reason == EXIT_REASON_TRIPLE_FAULT ||
3995 	    ((u16)vm_exit_reason == EXIT_REASON_EXCEPTION_NMI &&
3996 	     is_double_fault(exit_intr_info))) {
3997 		vmcs12->idt_vectoring_info_field = 0;
3998 	} else if (vcpu->arch.exception.injected) {
3999 		nr = vcpu->arch.exception.vector;
4000 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
4001 
4002 		if (kvm_exception_is_soft(nr)) {
4003 			vmcs12->vm_exit_instruction_len =
4004 				vcpu->arch.event_exit_inst_len;
4005 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
4006 		} else
4007 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
4008 
4009 		if (vcpu->arch.exception.has_error_code) {
4010 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
4011 			vmcs12->idt_vectoring_error_code =
4012 				vcpu->arch.exception.error_code;
4013 		}
4014 
4015 		vmcs12->idt_vectoring_info_field = idt_vectoring;
4016 	} else if (vcpu->arch.nmi_injected) {
4017 		vmcs12->idt_vectoring_info_field =
4018 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
4019 	} else if (vcpu->arch.interrupt.injected) {
4020 		nr = vcpu->arch.interrupt.nr;
4021 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
4022 
4023 		if (vcpu->arch.interrupt.soft) {
4024 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
4025 			vmcs12->vm_entry_instruction_len =
4026 				vcpu->arch.event_exit_inst_len;
4027 		} else
4028 			idt_vectoring |= INTR_TYPE_EXT_INTR;
4029 
4030 		vmcs12->idt_vectoring_info_field = idt_vectoring;
4031 	} else {
4032 		vmcs12->idt_vectoring_info_field = 0;
4033 	}
4034 }
4035 
4036 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4037 {
4038 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4039 	int max_irr;
4040 	void *vapic_page;
4041 	u16 status;
4042 
4043 	if (!vmx->nested.pi_pending)
4044 		return 0;
4045 
4046 	if (!vmx->nested.pi_desc)
4047 		goto mmio_needed;
4048 
4049 	vmx->nested.pi_pending = false;
4050 
4051 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4052 		return 0;
4053 
4054 	max_irr = pi_find_highest_vector(vmx->nested.pi_desc);
4055 	if (max_irr > 0) {
4056 		vapic_page = vmx->nested.virtual_apic_map.hva;
4057 		if (!vapic_page)
4058 			goto mmio_needed;
4059 
4060 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
4061 			vapic_page, &max_irr);
4062 		status = vmcs_read16(GUEST_INTR_STATUS);
4063 		if ((u8)max_irr > ((u8)status & 0xff)) {
4064 			status &= ~0xff;
4065 			status |= (u8)max_irr;
4066 			vmcs_write16(GUEST_INTR_STATUS, status);
4067 		}
4068 	}
4069 
4070 	kvm_vcpu_map_mark_dirty(vcpu, &vmx->nested.virtual_apic_map);
4071 	kvm_vcpu_map_mark_dirty(vcpu, &vmx->nested.pi_desc_map);
4072 	return 0;
4073 
4074 mmio_needed:
4075 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
4076 	return -ENXIO;
4077 }
4078 
4079 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu)
4080 {
4081 	struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
4082 	u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
4083 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4084 	unsigned long exit_qual;
4085 
4086 	if (ex->has_payload) {
4087 		exit_qual = ex->payload;
4088 	} else if (ex->vector == PF_VECTOR) {
4089 		exit_qual = vcpu->arch.cr2;
4090 	} else if (ex->vector == DB_VECTOR) {
4091 		exit_qual = vcpu->arch.dr6;
4092 		exit_qual &= ~DR6_BT;
4093 		exit_qual ^= DR6_ACTIVE_LOW;
4094 	} else {
4095 		exit_qual = 0;
4096 	}
4097 
4098 	/*
4099 	 * Unlike AMD's Paged Real Mode, which reports an error code on #PF
4100 	 * VM-Exits even if the CPU is in Real Mode, Intel VMX never sets the
4101 	 * "has error code" flags on VM-Exit if the CPU is in Real Mode.
4102 	 */
4103 	if (ex->has_error_code && is_protmode(vcpu)) {
4104 		/*
4105 		 * Intel CPUs do not generate error codes with bits 31:16 set,
4106 		 * and more importantly VMX disallows setting bits 31:16 in the
4107 		 * injected error code for VM-Entry.  Drop the bits to mimic
4108 		 * hardware and avoid inducing failure on nested VM-Entry if L1
4109 		 * chooses to inject the exception back to L2.  AMD CPUs _do_
4110 		 * generate "full" 32-bit error codes, so KVM allows userspace
4111 		 * to inject exception error codes with bits 31:16 set.
4112 		 */
4113 		vmcs12->vm_exit_intr_error_code = (u16)ex->error_code;
4114 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
4115 	}
4116 
4117 	if (kvm_exception_is_soft(ex->vector))
4118 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
4119 	else
4120 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
4121 
4122 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
4123 	    vmx_get_nmi_mask(vcpu))
4124 		intr_info |= INTR_INFO_UNBLOCK_NMI;
4125 
4126 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
4127 }
4128 
4129 /*
4130  * Returns true if a debug trap is (likely) pending delivery.  Infer the class
4131  * of a #DB (trap-like vs. fault-like) from the exception payload (to-be-DR6).
4132  * Using the payload is flawed because code breakpoints (fault-like) and data
4133  * breakpoints (trap-like) set the same bits in DR6 (breakpoint detected), i.e.
4134  * this will return false positives if a to-be-injected code breakpoint #DB is
4135  * pending (from KVM's perspective, but not "pending" across an instruction
4136  * boundary).  ICEBP, a.k.a. INT1, is also not reflected here even though it
4137  * too is trap-like.
4138  *
4139  * KVM "works" despite these flaws as ICEBP isn't currently supported by the
4140  * emulator, Monitor Trap Flag is not marked pending on intercepted #DBs (the
4141  * #DB has already happened), and MTF isn't marked pending on code breakpoints
4142  * from the emulator (because such #DBs are fault-like and thus don't trigger
4143  * actions that fire on instruction retire).
4144  */
4145 static unsigned long vmx_get_pending_dbg_trap(struct kvm_queued_exception *ex)
4146 {
4147 	if (!ex->pending || ex->vector != DB_VECTOR)
4148 		return 0;
4149 
4150 	/* General Detect #DBs are always fault-like. */
4151 	return ex->payload & ~DR6_BD;
4152 }
4153 
4154 /*
4155  * Returns true if there's a pending #DB exception that is lower priority than
4156  * a pending Monitor Trap Flag VM-Exit.  TSS T-flag #DBs are not emulated by
4157  * KVM, but could theoretically be injected by userspace.  Note, this code is
4158  * imperfect, see above.
4159  */
4160 static bool vmx_is_low_priority_db_trap(struct kvm_queued_exception *ex)
4161 {
4162 	return vmx_get_pending_dbg_trap(ex) & ~DR6_BT;
4163 }
4164 
4165 /*
4166  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
4167  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
4168  * represents these debug traps with a payload that is said to be compatible
4169  * with the 'pending debug exceptions' field, write the payload to the VMCS
4170  * field if a VM-exit is delivered before the debug trap.
4171  */
4172 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
4173 {
4174 	unsigned long pending_dbg;
4175 
4176 	pending_dbg = vmx_get_pending_dbg_trap(&vcpu->arch.exception);
4177 	if (pending_dbg)
4178 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, pending_dbg);
4179 }
4180 
4181 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
4182 {
4183 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
4184 	       to_vmx(vcpu)->nested.preemption_timer_expired;
4185 }
4186 
4187 static bool vmx_has_nested_events(struct kvm_vcpu *vcpu, bool for_injection)
4188 {
4189 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4190 	void *vapic = vmx->nested.virtual_apic_map.hva;
4191 	int max_irr, vppr;
4192 
4193 	if (nested_vmx_preemption_timer_pending(vcpu) ||
4194 	    vmx->nested.mtf_pending)
4195 		return true;
4196 
4197 	/*
4198 	 * Virtual Interrupt Delivery doesn't require manual injection.  Either
4199 	 * the interrupt is already in GUEST_RVI and will be recognized by CPU
4200 	 * at VM-Entry, or there is a KVM_REQ_EVENT pending and KVM will move
4201 	 * the interrupt from the PIR to RVI prior to entering the guest.
4202 	 */
4203 	if (for_injection)
4204 		return false;
4205 
4206 	if (!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
4207 	    __vmx_interrupt_blocked(vcpu))
4208 		return false;
4209 
4210 	if (!vapic)
4211 		return false;
4212 
4213 	vppr = *((u32 *)(vapic + APIC_PROCPRI));
4214 
4215 	max_irr = vmx_get_rvi();
4216 	if ((max_irr & 0xf0) > (vppr & 0xf0))
4217 		return true;
4218 
4219 	if (vmx->nested.pi_pending && vmx->nested.pi_desc &&
4220 	    pi_test_on(vmx->nested.pi_desc)) {
4221 		max_irr = pi_find_highest_vector(vmx->nested.pi_desc);
4222 		if (max_irr > 0 && (max_irr & 0xf0) > (vppr & 0xf0))
4223 			return true;
4224 	}
4225 
4226 	return false;
4227 }
4228 
4229 /*
4230  * Per the Intel SDM's table "Priority Among Concurrent Events", with minor
4231  * edits to fill in missing examples, e.g. #DB due to split-lock accesses,
4232  * and less minor edits to splice in the priority of VMX Non-Root specific
4233  * events, e.g. MTF and NMI/INTR-window exiting.
4234  *
4235  * 1 Hardware Reset and Machine Checks
4236  *	- RESET
4237  *	- Machine Check
4238  *
4239  * 2 Trap on Task Switch
4240  *	- T flag in TSS is set (on task switch)
4241  *
4242  * 3 External Hardware Interventions
4243  *	- FLUSH
4244  *	- STOPCLK
4245  *	- SMI
4246  *	- INIT
4247  *
4248  * 3.5 Monitor Trap Flag (MTF) VM-exit[1]
4249  *
4250  * 4 Traps on Previous Instruction
4251  *	- Breakpoints
4252  *	- Trap-class Debug Exceptions (#DB due to TF flag set, data/I-O
4253  *	  breakpoint, or #DB due to a split-lock access)
4254  *
4255  * 4.3	VMX-preemption timer expired VM-exit
4256  *
4257  * 4.6	NMI-window exiting VM-exit[2]
4258  *
4259  * 5 Nonmaskable Interrupts (NMI)
4260  *
4261  * 5.5 Interrupt-window exiting VM-exit and Virtual-interrupt delivery
4262  *
4263  * 6 Maskable Hardware Interrupts
4264  *
4265  * 7 Code Breakpoint Fault
4266  *
4267  * 8 Faults from Fetching Next Instruction
4268  *	- Code-Segment Limit Violation
4269  *	- Code Page Fault
4270  *	- Control protection exception (missing ENDBRANCH at target of indirect
4271  *					call or jump)
4272  *
4273  * 9 Faults from Decoding Next Instruction
4274  *	- Instruction length > 15 bytes
4275  *	- Invalid Opcode
4276  *	- Coprocessor Not Available
4277  *
4278  *10 Faults on Executing Instruction
4279  *	- Overflow
4280  *	- Bound error
4281  *	- Invalid TSS
4282  *	- Segment Not Present
4283  *	- Stack fault
4284  *	- General Protection
4285  *	- Data Page Fault
4286  *	- Alignment Check
4287  *	- x86 FPU Floating-point exception
4288  *	- SIMD floating-point exception
4289  *	- Virtualization exception
4290  *	- Control protection exception
4291  *
4292  * [1] Per the "Monitor Trap Flag" section: System-management interrupts (SMIs),
4293  *     INIT signals, and higher priority events take priority over MTF VM exits.
4294  *     MTF VM exits take priority over debug-trap exceptions and lower priority
4295  *     events.
4296  *
4297  * [2] Debug-trap exceptions and higher priority events take priority over VM exits
4298  *     caused by the VMX-preemption timer.  VM exits caused by the VMX-preemption
4299  *     timer take priority over VM exits caused by the "NMI-window exiting"
4300  *     VM-execution control and lower priority events.
4301  *
4302  * [3] Debug-trap exceptions and higher priority events take priority over VM exits
4303  *     caused by "NMI-window exiting".  VM exits caused by this control take
4304  *     priority over non-maskable interrupts (NMIs) and lower priority events.
4305  *
4306  * [4] Virtual-interrupt delivery has the same priority as that of VM exits due to
4307  *     the 1-setting of the "interrupt-window exiting" VM-execution control.  Thus,
4308  *     non-maskable interrupts (NMIs) and higher priority events take priority over
4309  *     delivery of a virtual interrupt; delivery of a virtual interrupt takes
4310  *     priority over external interrupts and lower priority events.
4311  */
4312 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
4313 {
4314 	struct kvm_lapic *apic = vcpu->arch.apic;
4315 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4316 	/*
4317 	 * Only a pending nested run blocks a pending exception.  If there is a
4318 	 * previously injected event, the pending exception occurred while said
4319 	 * event was being delivered and thus needs to be handled.
4320 	 */
4321 	bool block_nested_exceptions = vcpu->arch.nested_run_pending;
4322 	/*
4323 	 * Events that don't require injection, i.e. that are virtualized by
4324 	 * hardware, aren't blocked by a pending VM-Enter as KVM doesn't need
4325 	 * to regain control in order to deliver the event, and hardware will
4326 	 * handle event ordering, e.g. with respect to injected exceptions.
4327 	 *
4328 	 * But, new events (not exceptions) are only recognized at instruction
4329 	 * boundaries.  If an event needs reinjection, then KVM is handling a
4330 	 * VM-Exit that occurred _during_ instruction execution; new events,
4331 	 * irrespective of whether or not they're injected, are blocked until
4332 	 * the instruction completes.
4333 	 */
4334 	bool block_non_injected_events = kvm_event_needs_reinjection(vcpu);
4335 	/*
4336 	 * Inject events are blocked by nested VM-Enter, as KVM is responsible
4337 	 * for managing priority between concurrent events, i.e. KVM needs to
4338 	 * wait until after VM-Enter completes to deliver injected events.
4339 	 */
4340 	bool block_nested_events = block_nested_exceptions ||
4341 				   block_non_injected_events;
4342 
4343 	if (lapic_in_kernel(vcpu) &&
4344 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
4345 		if (block_nested_events)
4346 			return -EBUSY;
4347 		nested_vmx_update_pending_dbg(vcpu);
4348 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
4349 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
4350 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
4351 
4352 		/* MTF is discarded if the vCPU is in WFS. */
4353 		vmx->nested.mtf_pending = false;
4354 		return 0;
4355 	}
4356 
4357 	if (lapic_in_kernel(vcpu) &&
4358 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
4359 		if (block_nested_events)
4360 			return -EBUSY;
4361 
4362 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
4363 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
4364 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
4365 						apic->sipi_vector & 0xFFUL);
4366 			return 0;
4367 		}
4368 		/* Fallthrough, the SIPI is completely ignored. */
4369 	}
4370 
4371 	/*
4372 	 * Process exceptions that are higher priority than Monitor Trap Flag:
4373 	 * fault-like exceptions, TSS T flag #DB (not emulated by KVM, but
4374 	 * could theoretically come in from userspace), and ICEBP (INT1).
4375 	 *
4376 	 * TODO: SMIs have higher priority than MTF and trap-like #DBs (except
4377 	 * for TSS T flag #DBs).  KVM also doesn't save/restore pending MTF
4378 	 * across SMI/RSM as it should; that needs to be addressed in order to
4379 	 * prioritize SMI over MTF and trap-like #DBs.
4380 	 */
4381 	if (vcpu->arch.exception_vmexit.pending &&
4382 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception_vmexit)) {
4383 		if (block_nested_exceptions)
4384 			return -EBUSY;
4385 
4386 		nested_vmx_inject_exception_vmexit(vcpu);
4387 		return 0;
4388 	}
4389 
4390 	if (vcpu->arch.exception.pending &&
4391 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception)) {
4392 		if (block_nested_exceptions)
4393 			return -EBUSY;
4394 		goto no_vmexit;
4395 	}
4396 
4397 	if (vmx->nested.mtf_pending) {
4398 		if (block_nested_events)
4399 			return -EBUSY;
4400 		nested_vmx_update_pending_dbg(vcpu);
4401 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
4402 		return 0;
4403 	}
4404 
4405 	if (vcpu->arch.exception_vmexit.pending) {
4406 		if (block_nested_exceptions)
4407 			return -EBUSY;
4408 
4409 		nested_vmx_inject_exception_vmexit(vcpu);
4410 		return 0;
4411 	}
4412 
4413 	if (vcpu->arch.exception.pending) {
4414 		if (block_nested_exceptions)
4415 			return -EBUSY;
4416 		goto no_vmexit;
4417 	}
4418 
4419 	if (nested_vmx_preemption_timer_pending(vcpu)) {
4420 		if (block_nested_events)
4421 			return -EBUSY;
4422 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
4423 		return 0;
4424 	}
4425 
4426 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
4427 		if (block_nested_events)
4428 			return -EBUSY;
4429 		goto no_vmexit;
4430 	}
4431 
4432 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
4433 		if (block_nested_events)
4434 			return -EBUSY;
4435 		if (!nested_exit_on_nmi(vcpu))
4436 			goto no_vmexit;
4437 
4438 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
4439 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
4440 				  INTR_INFO_VALID_MASK, 0);
4441 		/*
4442 		 * The NMI-triggered VM exit counts as injection:
4443 		 * clear this one and block further NMIs.
4444 		 */
4445 		vcpu->arch.nmi_pending = 0;
4446 		vmx_set_nmi_mask(vcpu, true);
4447 		return 0;
4448 	}
4449 
4450 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
4451 		int irq;
4452 
4453 		if (!nested_exit_on_intr(vcpu)) {
4454 			if (block_nested_events)
4455 				return -EBUSY;
4456 
4457 			goto no_vmexit;
4458 		}
4459 
4460 		if (!nested_exit_intr_ack_set(vcpu)) {
4461 			if (block_nested_events)
4462 				return -EBUSY;
4463 
4464 			nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
4465 			return 0;
4466 		}
4467 
4468 		irq = kvm_cpu_get_extint(vcpu);
4469 		if (irq != -1) {
4470 			if (block_nested_events)
4471 				return -EBUSY;
4472 
4473 			nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT,
4474 					  INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR | irq, 0);
4475 			return 0;
4476 		}
4477 
4478 		irq = kvm_apic_has_interrupt(vcpu);
4479 		if (WARN_ON_ONCE(irq < 0))
4480 			goto no_vmexit;
4481 
4482 		/*
4483 		 * If the IRQ is L2's PI notification vector, process posted
4484 		 * interrupts for L2 instead of injecting VM-Exit, as the
4485 		 * detection/morphing architecturally occurs when the IRQ is
4486 		 * delivered to the CPU.  Note, only interrupts that are routed
4487 		 * through the local APIC trigger posted interrupt processing,
4488 		 * and enabling posted interrupts requires ACK-on-exit.
4489 		 */
4490 		if (irq == vmx->nested.posted_intr_nv) {
4491 			/*
4492 			 * Nested posted interrupts are delivered via RVI, i.e.
4493 			 * aren't injected by KVM, and so can be queued even if
4494 			 * manual event injection is disallowed.
4495 			 */
4496 			if (block_non_injected_events)
4497 				return -EBUSY;
4498 
4499 			vmx->nested.pi_pending = true;
4500 			kvm_apic_clear_irr(vcpu, irq);
4501 			goto no_vmexit;
4502 		}
4503 
4504 		if (block_nested_events)
4505 			return -EBUSY;
4506 
4507 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT,
4508 				  INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR | irq, 0);
4509 
4510 		/*
4511 		 * ACK the interrupt _after_ emulating VM-Exit, as the IRQ must
4512 		 * be marked as in-service in vmcs01.GUEST_INTERRUPT_STATUS.SVI
4513 		 * if APICv is active.
4514 		 */
4515 		kvm_apic_ack_interrupt(vcpu, irq);
4516 		return 0;
4517 	}
4518 
4519 no_vmexit:
4520 	return vmx_complete_nested_posted_interrupt(vcpu);
4521 }
4522 
4523 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
4524 {
4525 	ktime_t remaining =
4526 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
4527 	u64 value;
4528 
4529 	if (ktime_to_ns(remaining) <= 0)
4530 		return 0;
4531 
4532 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
4533 	do_div(value, 1000000);
4534 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
4535 }
4536 
4537 static bool is_vmcs12_ext_field(unsigned long field)
4538 {
4539 	switch (field) {
4540 	case GUEST_ES_SELECTOR:
4541 	case GUEST_CS_SELECTOR:
4542 	case GUEST_SS_SELECTOR:
4543 	case GUEST_DS_SELECTOR:
4544 	case GUEST_FS_SELECTOR:
4545 	case GUEST_GS_SELECTOR:
4546 	case GUEST_LDTR_SELECTOR:
4547 	case GUEST_TR_SELECTOR:
4548 	case GUEST_ES_LIMIT:
4549 	case GUEST_CS_LIMIT:
4550 	case GUEST_SS_LIMIT:
4551 	case GUEST_DS_LIMIT:
4552 	case GUEST_FS_LIMIT:
4553 	case GUEST_GS_LIMIT:
4554 	case GUEST_LDTR_LIMIT:
4555 	case GUEST_TR_LIMIT:
4556 	case GUEST_GDTR_LIMIT:
4557 	case GUEST_IDTR_LIMIT:
4558 	case GUEST_ES_AR_BYTES:
4559 	case GUEST_DS_AR_BYTES:
4560 	case GUEST_FS_AR_BYTES:
4561 	case GUEST_GS_AR_BYTES:
4562 	case GUEST_LDTR_AR_BYTES:
4563 	case GUEST_TR_AR_BYTES:
4564 	case GUEST_ES_BASE:
4565 	case GUEST_CS_BASE:
4566 	case GUEST_SS_BASE:
4567 	case GUEST_DS_BASE:
4568 	case GUEST_FS_BASE:
4569 	case GUEST_GS_BASE:
4570 	case GUEST_LDTR_BASE:
4571 	case GUEST_TR_BASE:
4572 	case GUEST_GDTR_BASE:
4573 	case GUEST_IDTR_BASE:
4574 	case GUEST_PENDING_DBG_EXCEPTIONS:
4575 	case GUEST_BNDCFGS:
4576 		return true;
4577 	default:
4578 		break;
4579 	}
4580 
4581 	return false;
4582 }
4583 
4584 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4585 				       struct vmcs12 *vmcs12)
4586 {
4587 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4588 
4589 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4590 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4591 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4592 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4593 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4594 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4595 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4596 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4597 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4598 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4599 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4600 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4601 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4602 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4603 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4604 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4605 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4606 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4607 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4608 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4609 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4610 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4611 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4612 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4613 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4614 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4615 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4616 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4617 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4618 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4619 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4620 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4621 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4622 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4623 	vmcs12->guest_pending_dbg_exceptions =
4624 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4625 
4626 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4627 }
4628 
4629 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4630 				       struct vmcs12 *vmcs12)
4631 {
4632 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4633 	int cpu;
4634 
4635 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4636 		return;
4637 
4638 
4639 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4640 
4641 	cpu = get_cpu();
4642 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4643 	vmx_vcpu_load_vmcs(vcpu, cpu);
4644 
4645 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4646 
4647 	vmx->loaded_vmcs = &vmx->vmcs01;
4648 	vmx_vcpu_load_vmcs(vcpu, cpu);
4649 	put_cpu();
4650 }
4651 
4652 /*
4653  * Update the guest state fields of vmcs12 to reflect changes that
4654  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4655  * VM-entry controls is also updated, since this is really a guest
4656  * state bit.)
4657  */
4658 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4659 {
4660 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4661 
4662 	if (nested_vmx_is_evmptr12_valid(vmx))
4663 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4664 
4665 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4666 		!nested_vmx_is_evmptr12_valid(vmx);
4667 
4668 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4669 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4670 
4671 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4672 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4673 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4674 
4675 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4676 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4677 
4678 	vmcs12->guest_interruptibility_info =
4679 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4680 
4681 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4682 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4683 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4684 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4685 	else
4686 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4687 
4688 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4689 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4690 	    !vcpu->arch.nested_run_pending)
4691 		vmcs12->vmx_preemption_timer_value =
4692 			vmx_get_preemption_timer_value(vcpu);
4693 
4694 	/*
4695 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4696 	 * own CR3 without exiting. If it has changed it, we must keep it.
4697 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4698 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4699 	 *
4700 	 * Additionally, restore L2's PDPTR to vmcs12.
4701 	 */
4702 	if (enable_ept) {
4703 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4704 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4705 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4706 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4707 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4708 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4709 		}
4710 	}
4711 
4712 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4713 
4714 	if (nested_cpu_has_vid(vmcs12))
4715 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4716 
4717 	vmcs12->vm_entry_controls =
4718 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4719 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4720 
4721 	/*
4722 	 * Note!  Save DR7, but intentionally don't grab DEBUGCTL from vmcs02.
4723 	 * Writes to DEBUGCTL that aren't intercepted by L1 are immediately
4724 	 * propagated to vmcs12 (see vmx_set_msr()), as the value loaded into
4725 	 * vmcs02 doesn't strictly track vmcs12.
4726 	 */
4727 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4728 		vmcs12->guest_dr7 = vcpu->arch.dr7;
4729 
4730 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4731 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4732 
4733 	vmcs_read_cet_state(&vmx->vcpu, &vmcs12->guest_s_cet,
4734 			    &vmcs12->guest_ssp,
4735 			    &vmcs12->guest_ssp_tbl);
4736 }
4737 
4738 /*
4739  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4740  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4741  * and this function updates it to reflect the changes to the guest state while
4742  * L2 was running (and perhaps made some exits which were handled directly by L0
4743  * without going back to L1), and to reflect the exit reason.
4744  * Note that we do not have to copy here all VMCS fields, just those that
4745  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4746  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4747  * which already writes to vmcs12 directly.
4748  */
4749 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4750 			   u32 vm_exit_reason, u32 exit_intr_info,
4751 			   unsigned long exit_qualification, u32 exit_insn_len)
4752 {
4753 	/* update exit information fields: */
4754 	vmcs12->vm_exit_reason = vm_exit_reason;
4755 	if (vmx_get_exit_reason(vcpu).enclave_mode)
4756 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4757 	vmcs12->exit_qualification = exit_qualification;
4758 
4759 	/*
4760 	 * On VM-Exit due to a failed VM-Entry, the VMCS isn't marked launched
4761 	 * and only EXIT_REASON and EXIT_QUALIFICATION are updated, all other
4762 	 * exit info fields are unmodified.
4763 	 */
4764 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4765 		vmcs12->launch_state = 1;
4766 
4767 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4768 		 * instead of reading the real value. */
4769 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4770 
4771 		/*
4772 		 * Transfer the event that L0 or L1 may wanted to inject into
4773 		 * L2 to IDT_VECTORING_INFO_FIELD.
4774 		 */
4775 		vmcs12_save_pending_event(vcpu, vmcs12,
4776 					  vm_exit_reason, exit_intr_info);
4777 
4778 		vmcs12->vm_exit_intr_info = exit_intr_info;
4779 		vmcs12->vm_exit_instruction_len = exit_insn_len;
4780 		vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4781 
4782 		/*
4783 		 * According to spec, there's no need to store the guest's
4784 		 * MSRs if the exit is due to a VM-entry failure that occurs
4785 		 * during or after loading the guest state. Since this exit
4786 		 * does not fall in that category, we need to save the MSRs.
4787 		 */
4788 		if (nested_vmx_store_msr(vcpu,
4789 					 vmcs12->vm_exit_msr_store_addr,
4790 					 vmcs12->vm_exit_msr_store_count))
4791 			nested_vmx_abort(vcpu,
4792 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4793 	}
4794 }
4795 
4796 /*
4797  * A part of what we need to when the nested L2 guest exits and we want to
4798  * run its L1 parent, is to reset L1's guest state to the host state specified
4799  * in vmcs12.
4800  * This function is to be called not only on normal nested exit, but also on
4801  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4802  * Failures During or After Loading Guest State").
4803  * This function should be called when the active VMCS is L1's (vmcs01).
4804  */
4805 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4806 				   struct vmcs12 *vmcs12)
4807 {
4808 	enum vm_entry_failure_code ignored;
4809 	struct kvm_segment seg;
4810 
4811 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4812 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4813 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4814 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4815 	else
4816 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4817 	vmx_set_efer(vcpu, vcpu->arch.efer);
4818 
4819 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4820 	kvm_rip_write(vcpu, vmcs12->host_rip);
4821 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4822 	vmx_set_interrupt_shadow(vcpu, 0);
4823 
4824 	/*
4825 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4826 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4827 	 *
4828 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4829 	 * (KVM doesn't change it);
4830 	 */
4831 	vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4832 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4833 
4834 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4835 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4836 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4837 
4838 	nested_ept_uninit_mmu_context(vcpu);
4839 
4840 	/*
4841 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4842 	 * couldn't have changed.
4843 	 */
4844 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4845 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4846 
4847 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4848 
4849 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4850 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4851 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4852 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4853 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4854 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4855 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4856 
4857 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4858 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4859 		vmcs_write64(GUEST_BNDCFGS, 0);
4860 
4861 	/*
4862 	 * Load CET state from host state if VM_EXIT_LOAD_CET_STATE is set.
4863 	 * otherwise CET state should be retained across VM-exit, i.e.,
4864 	 * guest values should be propagated from vmcs12 to vmcs01.
4865 	 */
4866 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_CET_STATE)
4867 		vmcs_write_cet_state(vcpu, vmcs12->host_s_cet, vmcs12->host_ssp,
4868 				     vmcs12->host_ssp_tbl);
4869 	else
4870 		vmcs_write_cet_state(vcpu, vmcs12->guest_s_cet, vmcs12->guest_ssp,
4871 				     vmcs12->guest_ssp_tbl);
4872 
4873 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4874 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4875 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4876 	}
4877 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
4878 	    kvm_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)))
4879 		WARN_ON_ONCE(__kvm_emulate_msr_write(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4880 						     vmcs12->host_ia32_perf_global_ctrl));
4881 
4882 	/* Set L1 segment info according to Intel SDM
4883 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4884 	seg = (struct kvm_segment) {
4885 		.base = 0,
4886 		.limit = 0xFFFFFFFF,
4887 		.selector = vmcs12->host_cs_selector,
4888 		.type = 11,
4889 		.present = 1,
4890 		.s = 1,
4891 		.g = 1
4892 	};
4893 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4894 		seg.l = 1;
4895 	else
4896 		seg.db = 1;
4897 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4898 	seg = (struct kvm_segment) {
4899 		.base = 0,
4900 		.limit = 0xFFFFFFFF,
4901 		.type = 3,
4902 		.present = 1,
4903 		.s = 1,
4904 		.db = 1,
4905 		.g = 1
4906 	};
4907 	seg.selector = vmcs12->host_ds_selector;
4908 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4909 	seg.selector = vmcs12->host_es_selector;
4910 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4911 	seg.selector = vmcs12->host_ss_selector;
4912 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4913 	seg.selector = vmcs12->host_fs_selector;
4914 	seg.base = vmcs12->host_fs_base;
4915 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4916 	seg.selector = vmcs12->host_gs_selector;
4917 	seg.base = vmcs12->host_gs_base;
4918 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4919 	seg = (struct kvm_segment) {
4920 		.base = vmcs12->host_tr_base,
4921 		.limit = 0x67,
4922 		.selector = vmcs12->host_tr_selector,
4923 		.type = 11,
4924 		.present = 1
4925 	};
4926 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4927 
4928 	memset(&seg, 0, sizeof(seg));
4929 	seg.unusable = 1;
4930 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4931 
4932 	kvm_set_dr(vcpu, 7, 0x400);
4933 	vmx_guest_debugctl_write(vcpu, 0);
4934 
4935 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4936 				vmcs12->vm_exit_msr_load_count))
4937 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4938 
4939 	to_vt(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4940 }
4941 
4942 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4943 {
4944 	struct vmx_uret_msr *efer_msr;
4945 	unsigned int i;
4946 
4947 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4948 		return vmcs_read64(GUEST_IA32_EFER);
4949 
4950 	if (cpu_has_load_ia32_efer())
4951 		return kvm_host.efer;
4952 
4953 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4954 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4955 			return vmx->msr_autoload.guest.val[i].value;
4956 	}
4957 
4958 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4959 	if (efer_msr)
4960 		return efer_msr->data;
4961 
4962 	return kvm_host.efer;
4963 }
4964 
4965 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4966 {
4967 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4968 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4969 	struct vmx_msr_entry g, h;
4970 	gpa_t gpa;
4971 	u32 i, j;
4972 
4973 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4974 
4975 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4976 		/*
4977 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4978 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4979 		 * and vcpu->arch.dr7 is not squirreled away before the
4980 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4981 		 */
4982 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4983 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4984 		else
4985 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4986 	}
4987 
4988 	/* Reload DEBUGCTL to ensure vmcs01 has a fresh FREEZE_IN_SMM value. */
4989 	vmx_reload_guest_debugctl(vcpu);
4990 
4991 	/*
4992 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4993 	 * handle a variety of side effects to KVM's software model.
4994 	 */
4995 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4996 
4997 	vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4998 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4999 
5000 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
5001 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
5002 
5003 	nested_ept_uninit_mmu_context(vcpu);
5004 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
5005 	kvm_register_mark_available(vcpu, VCPU_REG_CR3);
5006 
5007 	/*
5008 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
5009 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
5010 	 * VMFail, like everything else we just need to ensure our
5011 	 * software model is up-to-date.
5012 	 */
5013 	if (enable_ept && is_pae_paging(vcpu))
5014 		ept_save_pdptrs(vcpu);
5015 
5016 	kvm_mmu_reset_context(vcpu);
5017 
5018 	/*
5019 	 * This nasty bit of open coding is a compromise between blindly
5020 	 * loading L1's MSRs using the exit load lists (incorrect emulation
5021 	 * of VMFail), leaving the nested VM's MSRs in the software model
5022 	 * (incorrect behavior) and snapshotting the modified MSRs (too
5023 	 * expensive since the lists are unbound by hardware).  For each
5024 	 * MSR that was (prematurely) loaded from the nested VMEntry load
5025 	 * list, reload it from the exit load list if it exists and differs
5026 	 * from the guest value.  The intent is to stuff host state as
5027 	 * silently as possible, not to fully process the exit load list.
5028 	 */
5029 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
5030 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
5031 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
5032 			pr_debug_ratelimited(
5033 				"%s read MSR index failed (%u, 0x%08llx)\n",
5034 				__func__, i, gpa);
5035 			goto vmabort;
5036 		}
5037 
5038 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
5039 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
5040 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
5041 				pr_debug_ratelimited(
5042 					"%s read MSR failed (%u, 0x%08llx)\n",
5043 					__func__, j, gpa);
5044 				goto vmabort;
5045 			}
5046 			if (h.index != g.index)
5047 				continue;
5048 			if (h.value == g.value)
5049 				break;
5050 
5051 			if (nested_vmx_load_msr_check(vcpu, &h)) {
5052 				pr_debug_ratelimited(
5053 					"%s check failed (%u, 0x%x, 0x%x)\n",
5054 					__func__, j, h.index, h.reserved);
5055 				goto vmabort;
5056 			}
5057 
5058 			if (kvm_emulate_msr_write(vcpu, h.index, h.value)) {
5059 				pr_debug_ratelimited(
5060 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
5061 					__func__, j, h.index, h.value);
5062 				goto vmabort;
5063 			}
5064 		}
5065 	}
5066 
5067 	return;
5068 
5069 vmabort:
5070 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
5071 }
5072 
5073 /*
5074  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
5075  * and modify vmcs12 to make it see what it would expect to see there if
5076  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
5077  */
5078 void __nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
5079 			 u32 exit_intr_info, unsigned long exit_qualification,
5080 			 u32 exit_insn_len)
5081 {
5082 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5083 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5084 
5085 	/* Pending MTF traps are discarded on VM-Exit. */
5086 	vmx->nested.mtf_pending = false;
5087 
5088 	/* trying to cancel vmlaunch/vmresume is a bug */
5089 	kvm_warn_on_nested_run_pending(vcpu);
5090 
5091 #ifdef CONFIG_KVM_HYPERV
5092 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
5093 		/*
5094 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
5095 		 * Enlightened VMCS after migration and we still need to
5096 		 * do that when something is forcing L2->L1 exit prior to
5097 		 * the first L2 run.
5098 		 */
5099 		(void)nested_get_evmcs_page(vcpu);
5100 	}
5101 #endif
5102 
5103 	/* Service pending TLB flush requests for L2 before switching to L1. */
5104 	kvm_service_local_tlb_flush_requests(vcpu);
5105 
5106 	/*
5107 	 * VCPU_REG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
5108 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
5109 	 * up-to-date before switching to L1.
5110 	 */
5111 	if (enable_ept && is_pae_paging(vcpu))
5112 		vmx_ept_load_pdptrs(vcpu);
5113 
5114 	leave_guest_mode(vcpu);
5115 
5116 	if (nested_cpu_has_preemption_timer(vmcs12))
5117 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
5118 
5119 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
5120 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
5121 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
5122 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
5123 	}
5124 
5125 	if (likely(!vmx->fail)) {
5126 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
5127 
5128 		if (vm_exit_reason != -1)
5129 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
5130 				       exit_intr_info, exit_qualification,
5131 				       exit_insn_len);
5132 
5133 		/*
5134 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
5135 		 * also be used to capture vmcs12 cache as part of
5136 		 * capturing nVMX state for snapshot (migration).
5137 		 *
5138 		 * Otherwise, this flush will dirty guest memory at a
5139 		 * point it is already assumed by user-space to be
5140 		 * immutable.
5141 		 */
5142 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
5143 	} else {
5144 		/*
5145 		 * The only expected VM-instruction error is "VM entry with
5146 		 * invalid control field(s)." Anything else indicates a
5147 		 * problem with L0.
5148 		 */
5149 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
5150 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
5151 
5152 		/* VM-Fail at VM-Entry means KVM missed a consistency check. */
5153 		WARN_ON_ONCE(warn_on_missed_cc);
5154 	}
5155 
5156 	/*
5157 	 * Drop events/exceptions that were queued for re-injection to L2
5158 	 * (picked up via vmx_complete_interrupts()), as well as exceptions
5159 	 * that were pending for L2.  Note, this must NOT be hoisted above
5160 	 * prepare_vmcs12(), events/exceptions queued for re-injection need to
5161 	 * be captured in vmcs12 (see vmcs12_save_pending_event()).
5162 	 */
5163 	vcpu->arch.nmi_injected = false;
5164 	kvm_clear_exception_queue(vcpu);
5165 	kvm_clear_interrupt_queue(vcpu);
5166 
5167 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
5168 
5169 	kvm_nested_vmexit_handle_ibrs(vcpu);
5170 
5171 	/*
5172 	 * Update any VMCS fields that might have changed while vmcs02 was the
5173 	 * active VMCS.  The tracking is per-vCPU, not per-VMCS.
5174 	 */
5175 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.nr);
5176 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
5177 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
5178 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
5179 	if (kvm_caps.has_tsc_control)
5180 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
5181 
5182 	nested_put_vmcs12_pages(vcpu);
5183 
5184 	if ((vm_exit_reason != -1) &&
5185 	    (enable_shadow_vmcs || nested_vmx_is_evmptr12_valid(vmx)))
5186 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5187 
5188 	/* in case we halted in L2 */
5189 	kvm_set_mp_state(vcpu, KVM_MP_STATE_RUNNABLE);
5190 
5191 	if (likely(!vmx->fail)) {
5192 		if (vm_exit_reason != -1)
5193 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
5194 						       vmcs12->exit_qualification,
5195 						       vmcs12->idt_vectoring_info_field,
5196 						       vmcs12->vm_exit_intr_info,
5197 						       vmcs12->vm_exit_intr_error_code,
5198 						       KVM_ISA_VMX);
5199 
5200 		load_vmcs12_host_state(vcpu, vmcs12);
5201 
5202 		/*
5203 		 * Process events if an injectable IRQ or NMI is pending, even
5204 		 * if the event is blocked (RFLAGS.IF is cleared on VM-Exit).
5205 		 * If an event became pending while L2 was active, KVM needs to
5206 		 * either inject the event or request an IRQ/NMI window.  SMIs
5207 		 * don't need to be processed as SMM is mutually exclusive with
5208 		 * non-root mode.  INIT/SIPI don't need to be checked as INIT
5209 		 * is blocked post-VMXON, and SIPIs are ignored.
5210 		 */
5211 		if (kvm_cpu_has_injectable_intr(vcpu) || vcpu->arch.nmi_pending)
5212 			kvm_make_request(KVM_REQ_EVENT, vcpu);
5213 		return;
5214 	}
5215 
5216 	/*
5217 	 * After an early L2 VM-entry failure, we're now back
5218 	 * in L1 which thinks it just finished a VMLAUNCH or
5219 	 * VMRESUME instruction, so we need to set the failure
5220 	 * flag and the VM-instruction error field of the VMCS
5221 	 * accordingly, and skip the emulated instruction.
5222 	 */
5223 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
5224 
5225 	/*
5226 	 * Restore L1's host state to KVM's software model.  We're here
5227 	 * because a consistency check was caught by hardware, which
5228 	 * means some amount of guest state has been propagated to KVM's
5229 	 * model and needs to be unwound to the host's state.
5230 	 */
5231 	nested_vmx_restore_host_state(vcpu);
5232 
5233 	vmx->fail = 0;
5234 }
5235 
5236 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
5237 {
5238 	kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5239 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
5240 }
5241 
5242 /*
5243  * Decode the memory-address operand of a vmx instruction, as recorded on an
5244  * exit caused by such an instruction (run by a guest hypervisor).
5245  * On success, returns 0. When the operand is invalid, returns 1 and throws
5246  * #UD, #GP, or #SS.
5247  */
5248 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
5249 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
5250 {
5251 	gva_t off;
5252 	bool exn;
5253 	struct kvm_segment s;
5254 
5255 	/*
5256 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5257 	 * Execution", on an exit, vmx_instruction_info holds most of the
5258 	 * addressing components of the operand. Only the displacement part
5259 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5260 	 * For how an actual address is calculated from all these components,
5261 	 * refer to Vol. 1, "Operand Addressing".
5262 	 */
5263 	int  scaling = vmx_instruction_info & 3;
5264 	int  addr_size = (vmx_instruction_info >> 7) & 7;
5265 	bool is_reg = vmx_instruction_info & (1u << 10);
5266 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
5267 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5268 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5269 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5270 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5271 
5272 	if (is_reg) {
5273 		kvm_queue_exception(vcpu, UD_VECTOR);
5274 		return 1;
5275 	}
5276 
5277 	/* Addr = segment_base + offset */
5278 	/* offset = base + [index * scale] + displacement */
5279 	off = exit_qualification; /* holds the displacement */
5280 	if (addr_size == 1)
5281 		off = (gva_t)sign_extend64(off, 31);
5282 	else if (addr_size == 0)
5283 		off = (gva_t)sign_extend64(off, 15);
5284 	if (base_is_valid)
5285 		off += kvm_register_read(vcpu, base_reg);
5286 	if (index_is_valid)
5287 		off += kvm_register_read(vcpu, index_reg) << scaling;
5288 	vmx_get_segment(vcpu, &s, seg_reg);
5289 
5290 	/*
5291 	 * The effective address, i.e. @off, of a memory operand is truncated
5292 	 * based on the address size of the instruction.  Note that this is
5293 	 * the *effective address*, i.e. the address prior to accounting for
5294 	 * the segment's base.
5295 	 */
5296 	if (addr_size == 1) /* 32 bit */
5297 		off &= 0xffffffff;
5298 	else if (addr_size == 0) /* 16 bit */
5299 		off &= 0xffff;
5300 
5301 	/* Checks for #GP/#SS exceptions. */
5302 	exn = false;
5303 	if (is_long_mode(vcpu)) {
5304 		/*
5305 		 * The virtual/linear address is never truncated in 64-bit
5306 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
5307 		 * address when using FS/GS with a non-zero base.
5308 		 */
5309 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
5310 			*ret = s.base + off;
5311 		else
5312 			*ret = off;
5313 
5314 		*ret = vmx_get_untagged_addr(vcpu, *ret, 0);
5315 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
5316 		 * non-canonical form. This is the only check on the memory
5317 		 * destination for long mode!
5318 		 */
5319 		exn = is_noncanonical_address(*ret, vcpu, 0);
5320 	} else {
5321 		/*
5322 		 * When not in long mode, the virtual/linear address is
5323 		 * unconditionally truncated to 32 bits regardless of the
5324 		 * address size.
5325 		 */
5326 		*ret = (s.base + off) & 0xffffffff;
5327 
5328 		/* Protected mode: apply checks for segment validity in the
5329 		 * following order:
5330 		 * - segment type check (#GP(0) may be thrown)
5331 		 * - usability check (#GP(0)/#SS(0))
5332 		 * - limit check (#GP(0)/#SS(0))
5333 		 */
5334 		if (wr)
5335 			/* #GP(0) if the destination operand is located in a
5336 			 * read-only data segment or any code segment.
5337 			 */
5338 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
5339 		else
5340 			/* #GP(0) if the source operand is located in an
5341 			 * execute-only code segment
5342 			 */
5343 			exn = ((s.type & 0xa) == 8);
5344 		if (exn) {
5345 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5346 			return 1;
5347 		}
5348 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
5349 		 */
5350 		exn = (s.unusable != 0);
5351 
5352 		/*
5353 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
5354 		 * outside the segment limit.  All CPUs that support VMX ignore
5355 		 * limit checks for flat segments, i.e. segments with base==0,
5356 		 * limit==0xffffffff and of type expand-up data or code.
5357 		 */
5358 		if (!(s.base == 0 && s.limit == 0xffffffff &&
5359 		     ((s.type & 8) || !(s.type & 4))))
5360 			exn = exn || ((u64)off + len - 1 > s.limit);
5361 	}
5362 	if (exn) {
5363 		kvm_queue_exception_e(vcpu,
5364 				      seg_reg == VCPU_SREG_SS ?
5365 						SS_VECTOR : GP_VECTOR,
5366 				      0);
5367 		return 1;
5368 	}
5369 
5370 	return 0;
5371 }
5372 
5373 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
5374 				int *ret)
5375 {
5376 	gva_t gva;
5377 	struct x86_exception e;
5378 	int r;
5379 
5380 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5381 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
5382 				sizeof(*vmpointer), &gva)) {
5383 		*ret = 1;
5384 		return -EINVAL;
5385 	}
5386 
5387 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
5388 	if (r != X86EMUL_CONTINUE) {
5389 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
5390 		return -EINVAL;
5391 	}
5392 
5393 	return 0;
5394 }
5395 
5396 /*
5397  * Allocate a shadow VMCS and associate it with the currently loaded
5398  * VMCS, unless such a shadow VMCS already exists. The newly allocated
5399  * VMCS is also VMCLEARed, so that it is ready for use.
5400  */
5401 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
5402 {
5403 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5404 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
5405 
5406 	/*
5407 	 * KVM allocates a shadow VMCS only when L1 executes VMXON and frees it
5408 	 * when L1 executes VMXOFF or the vCPU is forced out of nested
5409 	 * operation.  VMXON faults if the CPU is already post-VMXON, so it
5410 	 * should be impossible to already have an allocated shadow VMCS.  KVM
5411 	 * doesn't support virtualization of VMCS shadowing, so vmcs01 should
5412 	 * always be the loaded VMCS.
5413 	 */
5414 	if (WARN_ON(loaded_vmcs != &vmx->vmcs01 || loaded_vmcs->shadow_vmcs))
5415 		return loaded_vmcs->shadow_vmcs;
5416 
5417 	loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
5418 	if (loaded_vmcs->shadow_vmcs)
5419 		vmcs_clear(loaded_vmcs->shadow_vmcs);
5420 
5421 	return loaded_vmcs->shadow_vmcs;
5422 }
5423 
5424 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
5425 {
5426 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5427 	int r;
5428 
5429 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
5430 	if (r < 0)
5431 		goto out_vmcs02;
5432 
5433 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5434 	if (!vmx->nested.cached_vmcs12)
5435 		goto out_cached_vmcs12;
5436 
5437 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
5438 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5439 	if (!vmx->nested.cached_shadow_vmcs12)
5440 		goto out_cached_shadow_vmcs12;
5441 
5442 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
5443 		goto out_shadow_vmcs;
5444 
5445 	hrtimer_setup(&vmx->nested.preemption_timer, vmx_preemption_timer_fn, CLOCK_MONOTONIC,
5446 		      HRTIMER_MODE_ABS_PINNED);
5447 
5448 	vmx->nested.vpid02 = allocate_vpid();
5449 
5450 	vmx->nested.vmcs02_initialized = false;
5451 	vmx->nested.vmxon = true;
5452 
5453 	if (vmx_pt_mode_is_host_guest()) {
5454 		vmx->pt_desc.guest.ctl = 0;
5455 		pt_update_intercept_for_msr(vcpu);
5456 	}
5457 
5458 	return 0;
5459 
5460 out_shadow_vmcs:
5461 	kfree(vmx->nested.cached_shadow_vmcs12);
5462 
5463 out_cached_shadow_vmcs12:
5464 	kfree(vmx->nested.cached_vmcs12);
5465 
5466 out_cached_vmcs12:
5467 	free_loaded_vmcs(&vmx->nested.vmcs02);
5468 
5469 out_vmcs02:
5470 	return -ENOMEM;
5471 }
5472 
5473 /* Emulate the VMXON instruction. */
5474 static int handle_vmxon(struct kvm_vcpu *vcpu)
5475 {
5476 	int ret;
5477 	gpa_t vmptr;
5478 	uint32_t revision;
5479 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5480 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
5481 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
5482 
5483 	/*
5484 	 * Manually check CR4.VMXE checks, KVM must force CR4.VMXE=1 to enter
5485 	 * the guest and so cannot rely on hardware to perform the check,
5486 	 * which has higher priority than VM-Exit (see Intel SDM's pseudocode
5487 	 * for VMXON).
5488 	 *
5489 	 * Rely on hardware for the other pre-VM-Exit checks, CR0.PE=1, !VM86
5490 	 * and !COMPATIBILITY modes.  For an unrestricted guest, KVM doesn't
5491 	 * force any of the relevant guest state.  For a restricted guest, KVM
5492 	 * does force CR0.PE=1, but only to also force VM86 in order to emulate
5493 	 * Real Mode, and so there's no need to check CR0.PE manually.
5494 	 */
5495 	if (!kvm_is_cr4_bit_set(vcpu, X86_CR4_VMXE)) {
5496 		kvm_queue_exception(vcpu, UD_VECTOR);
5497 		return 1;
5498 	}
5499 
5500 	/*
5501 	 * The CPL is checked for "not in VMX operation" and for "in VMX root",
5502 	 * and has higher priority than the VM-Fail due to being post-VMXON,
5503 	 * i.e. VMXON #GPs outside of VMX non-root if CPL!=0.  In VMX non-root,
5504 	 * VMXON causes VM-Exit and KVM unconditionally forwards VMXON VM-Exits
5505 	 * from L2 to L1, i.e. there's no need to check for the vCPU being in
5506 	 * VMX non-root.
5507 	 *
5508 	 * Forwarding the VM-Exit unconditionally, i.e. without performing the
5509 	 * #UD checks (see above), is functionally ok because KVM doesn't allow
5510 	 * L1 to run L2 without CR4.VMXE=0, and because KVM never modifies L2's
5511 	 * CR0 or CR4, i.e. it's L2's responsibility to emulate #UDs that are
5512 	 * missed by hardware due to shadowing CR0 and/or CR4.
5513 	 */
5514 	if (vmx_get_cpl(vcpu)) {
5515 		kvm_inject_gp(vcpu, 0);
5516 		return 1;
5517 	}
5518 
5519 	if (vmx->nested.vmxon)
5520 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5521 
5522 	/*
5523 	 * Invalid CR0/CR4 generates #GP.  These checks are performed if and
5524 	 * only if the vCPU isn't already in VMX operation, i.e. effectively
5525 	 * have lower priority than the VM-Fail above.
5526 	 */
5527 	if (!nested_host_cr0_valid(vcpu, kvm_read_cr0(vcpu)) ||
5528 	    !nested_host_cr4_valid(vcpu, kvm_read_cr4(vcpu))) {
5529 		kvm_inject_gp(vcpu, 0);
5530 		return 1;
5531 	}
5532 
5533 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5534 			!= VMXON_NEEDED_FEATURES) {
5535 		kvm_inject_gp(vcpu, 0);
5536 		return 1;
5537 	}
5538 
5539 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
5540 		return ret;
5541 
5542 	/*
5543 	 * SDM 3: 24.11.5
5544 	 * The first 4 bytes of VMXON region contain the supported
5545 	 * VMCS revision identifier
5546 	 *
5547 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
5548 	 * which replaces physical address width with 32
5549 	 */
5550 	if (!page_address_valid(vcpu, vmptr))
5551 		return nested_vmx_failInvalid(vcpu);
5552 
5553 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
5554 	    revision != VMCS12_REVISION)
5555 		return nested_vmx_failInvalid(vcpu);
5556 
5557 	vmx->nested.vmxon_ptr = vmptr;
5558 	ret = enter_vmx_operation(vcpu);
5559 	if (ret)
5560 		return ret;
5561 
5562 	return nested_vmx_succeed(vcpu);
5563 }
5564 
5565 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
5566 {
5567 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5568 
5569 	if (vmx->nested.current_vmptr == INVALID_GPA)
5570 		return;
5571 
5572 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
5573 
5574 	if (enable_shadow_vmcs) {
5575 		/* copy to memory all shadowed fields in case
5576 		   they were modified */
5577 		copy_shadow_to_vmcs12(vmx);
5578 		vmx_disable_shadow_vmcs(vmx);
5579 	}
5580 	vmx->nested.posted_intr_nv = -1;
5581 
5582 	/* Flush VMCS12 to guest memory */
5583 	kvm_vcpu_write_guest_page(vcpu,
5584 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
5585 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5586 
5587 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5588 
5589 	vmx->nested.current_vmptr = INVALID_GPA;
5590 }
5591 
5592 /* Emulate the VMXOFF instruction */
5593 static int handle_vmxoff(struct kvm_vcpu *vcpu)
5594 {
5595 	if (!nested_vmx_check_permission(vcpu))
5596 		return 1;
5597 
5598 	free_nested(vcpu);
5599 
5600 	if (kvm_apic_has_pending_init_or_sipi(vcpu))
5601 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5602 
5603 	return nested_vmx_succeed(vcpu);
5604 }
5605 
5606 /* Emulate the VMCLEAR instruction */
5607 static int handle_vmclear(struct kvm_vcpu *vcpu)
5608 {
5609 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5610 	u32 zero = 0;
5611 	gpa_t vmptr;
5612 	int r;
5613 
5614 	if (!nested_vmx_check_permission(vcpu))
5615 		return 1;
5616 
5617 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5618 		return r;
5619 
5620 	if (!page_address_valid(vcpu, vmptr))
5621 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5622 
5623 	if (vmptr == vmx->nested.vmxon_ptr)
5624 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5625 
5626 	if (likely(!nested_evmcs_handle_vmclear(vcpu, vmptr))) {
5627 		if (vmptr == vmx->nested.current_vmptr)
5628 			nested_release_vmcs12(vcpu);
5629 
5630 		/*
5631 		 * Silently ignore memory errors on VMCLEAR, Intel's pseudocode
5632 		 * for VMCLEAR includes a "ensure that data for VMCS referenced
5633 		 * by the operand is in memory" clause that guards writes to
5634 		 * memory, i.e. doing nothing for I/O is architecturally valid.
5635 		 *
5636 		 * FIXME: Suppress failures if and only if no memslot is found,
5637 		 * i.e. exit to userspace if __copy_to_user() fails.
5638 		 */
5639 		(void)kvm_vcpu_write_guest(vcpu,
5640 					   vmptr + offsetof(struct vmcs12,
5641 							    launch_state),
5642 					   &zero, sizeof(zero));
5643 	}
5644 
5645 	return nested_vmx_succeed(vcpu);
5646 }
5647 
5648 /* Emulate the VMLAUNCH instruction */
5649 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5650 {
5651 	return nested_vmx_run(vcpu, true);
5652 }
5653 
5654 /* Emulate the VMRESUME instruction */
5655 static int handle_vmresume(struct kvm_vcpu *vcpu)
5656 {
5657 
5658 	return nested_vmx_run(vcpu, false);
5659 }
5660 
5661 static int handle_vmread(struct kvm_vcpu *vcpu)
5662 {
5663 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5664 						    : get_vmcs12(vcpu);
5665 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5666 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5667 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5668 	struct x86_exception e;
5669 	unsigned long field;
5670 	u64 value;
5671 	gva_t gva = 0;
5672 	short offset;
5673 	int len, r;
5674 
5675 	if (!nested_vmx_check_permission(vcpu))
5676 		return 1;
5677 
5678 	/* Decode instruction info and find the field to read */
5679 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5680 
5681 	if (!nested_vmx_is_evmptr12_valid(vmx)) {
5682 		/*
5683 		 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5684 		 * any VMREAD sets the ALU flags for VMfailInvalid.
5685 		 */
5686 		if (vmx->nested.current_vmptr == INVALID_GPA ||
5687 		    (is_guest_mode(vcpu) &&
5688 		     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5689 			return nested_vmx_failInvalid(vcpu);
5690 
5691 		offset = get_vmcs12_field_offset(field);
5692 		if (offset < 0)
5693 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5694 
5695 		if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5696 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5697 
5698 		/* Read the field, zero-extended to a u64 value */
5699 		value = vmcs12_read_any(vmcs12, field, offset);
5700 	} else {
5701 		/*
5702 		 * Hyper-V TLFS (as of 6.0b) explicitly states, that while an
5703 		 * enlightened VMCS is active VMREAD/VMWRITE instructions are
5704 		 * unsupported. Unfortunately, certain versions of Windows 11
5705 		 * don't comply with this requirement which is not enforced in
5706 		 * genuine Hyper-V. Allow VMREAD from an enlightened VMCS as a
5707 		 * workaround, as misbehaving guests will panic on VM-Fail.
5708 		 * Note, enlightened VMCS is incompatible with shadow VMCS so
5709 		 * all VMREADs from L2 should go to L1.
5710 		 */
5711 		if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5712 			return nested_vmx_failInvalid(vcpu);
5713 
5714 		offset = evmcs_field_offset(field, NULL);
5715 		if (offset < 0)
5716 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5717 
5718 		/* Read the field, zero-extended to a u64 value */
5719 		value = evmcs_read_any(nested_vmx_evmcs(vmx), field, offset);
5720 	}
5721 
5722 	/*
5723 	 * Now copy part of this value to register or memory, as requested.
5724 	 * Note that the number of bits actually copied is 32 or 64 depending
5725 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5726 	 */
5727 	if (instr_info & BIT(10)) {
5728 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5729 	} else {
5730 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5731 		if (get_vmx_mem_address(vcpu, exit_qualification,
5732 					instr_info, true, len, &gva))
5733 			return 1;
5734 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5735 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5736 		if (r != X86EMUL_CONTINUE)
5737 			return kvm_handle_memory_failure(vcpu, r, &e);
5738 	}
5739 
5740 	return nested_vmx_succeed(vcpu);
5741 }
5742 
5743 static bool is_shadow_field_rw(unsigned long field)
5744 {
5745 	switch (field) {
5746 #define SHADOW_FIELD_RW(x, y) case x:
5747 #include "vmcs_shadow_fields.h"
5748 		return true;
5749 	default:
5750 		break;
5751 	}
5752 	return false;
5753 }
5754 
5755 static bool is_shadow_field_ro(unsigned long field)
5756 {
5757 	switch (field) {
5758 #define SHADOW_FIELD_RO(x, y) case x:
5759 #include "vmcs_shadow_fields.h"
5760 		return true;
5761 	default:
5762 		break;
5763 	}
5764 	return false;
5765 }
5766 
5767 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5768 {
5769 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5770 						    : get_vmcs12(vcpu);
5771 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5772 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5773 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5774 	struct x86_exception e;
5775 	unsigned long field;
5776 	short offset;
5777 	gva_t gva;
5778 	int len, r;
5779 
5780 	/*
5781 	 * The value to write might be 32 or 64 bits, depending on L1's long
5782 	 * mode, and eventually we need to write that into a field of several
5783 	 * possible lengths. The code below first zero-extends the value to 64
5784 	 * bit (value), and then copies only the appropriate number of
5785 	 * bits into the vmcs12 field.
5786 	 */
5787 	u64 value = 0;
5788 
5789 	if (!nested_vmx_check_permission(vcpu))
5790 		return 1;
5791 
5792 	/*
5793 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5794 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5795 	 */
5796 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5797 	    (is_guest_mode(vcpu) &&
5798 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5799 		return nested_vmx_failInvalid(vcpu);
5800 
5801 	if (instr_info & BIT(10))
5802 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5803 	else {
5804 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5805 		if (get_vmx_mem_address(vcpu, exit_qualification,
5806 					instr_info, false, len, &gva))
5807 			return 1;
5808 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5809 		if (r != X86EMUL_CONTINUE)
5810 			return kvm_handle_memory_failure(vcpu, r, &e);
5811 	}
5812 
5813 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5814 
5815 	offset = get_vmcs12_field_offset(field);
5816 	if (offset < 0)
5817 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5818 
5819 	/*
5820 	 * If the vCPU supports "VMWRITE to any supported field in the
5821 	 * VMCS," then the "read-only" fields are actually read/write.
5822 	 */
5823 	if (vmcs_field_readonly(field) &&
5824 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5825 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5826 
5827 	/*
5828 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5829 	 * vmcs12, else we may crush a field or consume a stale value.
5830 	 */
5831 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5832 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5833 
5834 	/*
5835 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5836 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5837 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5838 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5839 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5840 	 * the stripped down value, L2 sees the full value as stored by KVM).
5841 	 */
5842 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5843 		value &= 0x1f0ff;
5844 
5845 	vmcs12_write_any(vmcs12, field, offset, value);
5846 
5847 	/*
5848 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5849 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5850 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5851 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5852 	 */
5853 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5854 		/*
5855 		 * L1 can read these fields without exiting, ensure the
5856 		 * shadow VMCS is up-to-date.
5857 		 */
5858 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5859 			preempt_disable();
5860 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5861 
5862 			__vmcs_writel(field, value);
5863 
5864 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5865 			vmcs_load(vmx->loaded_vmcs->vmcs);
5866 			preempt_enable();
5867 		}
5868 		vmx->nested.dirty_vmcs12 = true;
5869 	}
5870 
5871 	return nested_vmx_succeed(vcpu);
5872 }
5873 
5874 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5875 {
5876 	vmx->nested.current_vmptr = vmptr;
5877 	if (enable_shadow_vmcs) {
5878 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5879 		vmcs_write64(VMCS_LINK_POINTER,
5880 			     __pa(vmx->vmcs01.shadow_vmcs));
5881 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5882 	}
5883 	vmx->nested.dirty_vmcs12 = true;
5884 	vmx->nested.force_msr_bitmap_recalc = true;
5885 }
5886 
5887 /* Emulate the VMPTRLD instruction */
5888 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5889 {
5890 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5891 	gpa_t vmptr;
5892 	int r;
5893 
5894 	if (!nested_vmx_check_permission(vcpu))
5895 		return 1;
5896 
5897 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5898 		return r;
5899 
5900 	if (!page_address_valid(vcpu, vmptr))
5901 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5902 
5903 	if (vmptr == vmx->nested.vmxon_ptr)
5904 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5905 
5906 	/* Forbid normal VMPTRLD if Enlightened version was used */
5907 	if (nested_vmx_is_evmptr12_valid(vmx))
5908 		return 1;
5909 
5910 	if (vmx->nested.current_vmptr != vmptr) {
5911 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5912 		struct vmcs_hdr hdr;
5913 
5914 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5915 			/*
5916 			 * Reads from an unbacked page return all 1s,
5917 			 * which means that the 32 bits located at the
5918 			 * given physical address won't match the required
5919 			 * VMCS12_REVISION identifier.
5920 			 */
5921 			return nested_vmx_fail(vcpu,
5922 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5923 		}
5924 
5925 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5926 						 offsetof(struct vmcs12, hdr),
5927 						 sizeof(hdr))) {
5928 			return nested_vmx_fail(vcpu,
5929 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5930 		}
5931 
5932 		if (hdr.revision_id != VMCS12_REVISION ||
5933 		    (hdr.shadow_vmcs &&
5934 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5935 			return nested_vmx_fail(vcpu,
5936 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5937 		}
5938 
5939 		nested_release_vmcs12(vcpu);
5940 
5941 		/*
5942 		 * Load VMCS12 from guest memory since it is not already
5943 		 * cached.
5944 		 */
5945 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5946 					  VMCS12_SIZE)) {
5947 			return nested_vmx_fail(vcpu,
5948 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5949 		}
5950 
5951 		set_current_vmptr(vmx, vmptr);
5952 	}
5953 
5954 	return nested_vmx_succeed(vcpu);
5955 }
5956 
5957 /* Emulate the VMPTRST instruction */
5958 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5959 {
5960 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5961 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5962 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5963 	struct x86_exception e;
5964 	gva_t gva;
5965 	int r;
5966 
5967 	if (!nested_vmx_check_permission(vcpu))
5968 		return 1;
5969 
5970 	if (unlikely(nested_vmx_is_evmptr12_valid(to_vmx(vcpu))))
5971 		return 1;
5972 
5973 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5974 				true, sizeof(gpa_t), &gva))
5975 		return 1;
5976 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5977 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5978 					sizeof(gpa_t), &e);
5979 	if (r != X86EMUL_CONTINUE)
5980 		return kvm_handle_memory_failure(vcpu, r, &e);
5981 
5982 	return nested_vmx_succeed(vcpu);
5983 }
5984 
5985 /* Emulate the INVEPT instruction */
5986 static int handle_invept(struct kvm_vcpu *vcpu)
5987 {
5988 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5989 	u32 vmx_instruction_info, types;
5990 	unsigned long type, roots_to_free;
5991 	struct kvm_mmu *mmu;
5992 	gva_t gva;
5993 	struct x86_exception e;
5994 	struct {
5995 		u64 eptp, gpa;
5996 	} operand;
5997 	int i, r, gpr_index;
5998 
5999 	if (!(vmx->nested.msrs.secondary_ctls_high &
6000 	      SECONDARY_EXEC_ENABLE_EPT) ||
6001 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
6002 		kvm_queue_exception(vcpu, UD_VECTOR);
6003 		return 1;
6004 	}
6005 
6006 	if (!nested_vmx_check_permission(vcpu))
6007 		return 1;
6008 
6009 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6010 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
6011 	type = kvm_register_read(vcpu, gpr_index);
6012 
6013 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6014 
6015 	if (type >= 32 || !(types & (1 << type)))
6016 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6017 
6018 	/* According to the Intel VMX instruction reference, the memory
6019 	 * operand is read even if it isn't needed (e.g., for type==global)
6020 	 */
6021 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
6022 			vmx_instruction_info, false, sizeof(operand), &gva))
6023 		return 1;
6024 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
6025 	if (r != X86EMUL_CONTINUE)
6026 		return kvm_handle_memory_failure(vcpu, r, &e);
6027 
6028 	/*
6029 	 * Nested EPT roots are always held through guest_mmu,
6030 	 * not root_mmu.
6031 	 */
6032 	mmu = &vcpu->arch.guest_mmu;
6033 
6034 	switch (type) {
6035 	case VMX_EPT_EXTENT_CONTEXT:
6036 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
6037 			return nested_vmx_fail(vcpu,
6038 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6039 
6040 		roots_to_free = 0;
6041 		if (nested_ept_root_matches(mmu->root.hpa, mmu->root.pgd,
6042 					    operand.eptp))
6043 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
6044 
6045 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
6046 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
6047 						    mmu->prev_roots[i].pgd,
6048 						    operand.eptp))
6049 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
6050 		}
6051 		break;
6052 	case VMX_EPT_EXTENT_GLOBAL:
6053 		roots_to_free = KVM_MMU_ROOTS_ALL;
6054 		break;
6055 	default:
6056 		BUG();
6057 		break;
6058 	}
6059 
6060 	if (roots_to_free)
6061 		kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
6062 
6063 	return nested_vmx_succeed(vcpu);
6064 }
6065 
6066 static int handle_invvpid(struct kvm_vcpu *vcpu)
6067 {
6068 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6069 	u32 vmx_instruction_info;
6070 	unsigned long type, types;
6071 	gva_t gva;
6072 	struct x86_exception e;
6073 	struct {
6074 		u64 vpid;
6075 		u64 gla;
6076 	} operand;
6077 	u16 vpid02;
6078 	int r, gpr_index;
6079 
6080 	if (!(vmx->nested.msrs.secondary_ctls_high &
6081 	      SECONDARY_EXEC_ENABLE_VPID) ||
6082 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
6083 		kvm_queue_exception(vcpu, UD_VECTOR);
6084 		return 1;
6085 	}
6086 
6087 	if (!nested_vmx_check_permission(vcpu))
6088 		return 1;
6089 
6090 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6091 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
6092 	type = kvm_register_read(vcpu, gpr_index);
6093 
6094 	types = (vmx->nested.msrs.vpid_caps &
6095 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
6096 
6097 	if (type >= 32 || !(types & (1 << type)))
6098 		return nested_vmx_fail(vcpu,
6099 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6100 
6101 	/* according to the intel vmx instruction reference, the memory
6102 	 * operand is read even if it isn't needed (e.g., for type==global)
6103 	 */
6104 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
6105 			vmx_instruction_info, false, sizeof(operand), &gva))
6106 		return 1;
6107 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
6108 	if (r != X86EMUL_CONTINUE)
6109 		return kvm_handle_memory_failure(vcpu, r, &e);
6110 
6111 	if (operand.vpid >> 16)
6112 		return nested_vmx_fail(vcpu,
6113 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6114 
6115 	/*
6116 	 * Always flush the effective vpid02, i.e. never flush the current VPID
6117 	 * and never explicitly flush vpid01.  INVVPID targets a VPID, not a
6118 	 * VMCS, and so whether or not the current vmcs12 has VPID enabled is
6119 	 * irrelevant (and there may not be a loaded vmcs12).
6120 	 */
6121 	vpid02 = nested_get_vpid02(vcpu);
6122 	switch (type) {
6123 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
6124 		/*
6125 		 * LAM doesn't apply to addresses that are inputs to TLB
6126 		 * invalidation.
6127 		 */
6128 		if (!operand.vpid ||
6129 		    is_noncanonical_invlpg_address(operand.gla, vcpu))
6130 			return nested_vmx_fail(vcpu,
6131 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6132 		vpid_sync_vcpu_addr(vpid02, operand.gla);
6133 		break;
6134 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
6135 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
6136 		if (!operand.vpid)
6137 			return nested_vmx_fail(vcpu,
6138 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6139 		vpid_sync_context(vpid02);
6140 		break;
6141 	case VMX_VPID_EXTENT_ALL_CONTEXT:
6142 		vpid_sync_context(vpid02);
6143 		break;
6144 	default:
6145 		WARN_ON_ONCE(1);
6146 		return kvm_skip_emulated_instruction(vcpu);
6147 	}
6148 
6149 	/*
6150 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
6151 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
6152 	 * roots as VPIDs are not tracked in the MMU role.
6153 	 *
6154 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
6155 	 * an MMU when EPT is disabled.
6156 	 *
6157 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
6158 	 */
6159 	if (!enable_ept)
6160 		kvm_mmu_free_guest_mode_roots(vcpu->kvm, &vcpu->arch.root_mmu);
6161 
6162 	return nested_vmx_succeed(vcpu);
6163 }
6164 
6165 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
6166 				     struct vmcs12 *vmcs12)
6167 {
6168 	u32 index = kvm_rcx_read(vcpu);
6169 	u64 new_eptp;
6170 
6171 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
6172 		return 1;
6173 	if (index >= VMFUNC_EPTP_ENTRIES)
6174 		return 1;
6175 
6176 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
6177 				     &new_eptp, index * 8, 8))
6178 		return 1;
6179 
6180 	/*
6181 	 * If the (L2) guest does a vmfunc to the currently
6182 	 * active ept pointer, we don't have to do anything else
6183 	 */
6184 	if (vmcs12->ept_pointer != new_eptp) {
6185 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
6186 			return 1;
6187 
6188 		vmcs12->ept_pointer = new_eptp;
6189 		nested_ept_new_eptp(vcpu);
6190 
6191 		if (!nested_cpu_has_vpid(vmcs12))
6192 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
6193 	}
6194 
6195 	return 0;
6196 }
6197 
6198 static int handle_vmfunc(struct kvm_vcpu *vcpu)
6199 {
6200 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6201 	struct vmcs12 *vmcs12;
6202 	u32 function = kvm_rax_read(vcpu);
6203 
6204 	/*
6205 	 * VMFUNC should never execute cleanly while L1 is active; KVM supports
6206 	 * VMFUNC for nested VMs, but not for L1.
6207 	 */
6208 	if (WARN_ON_ONCE(!is_guest_mode(vcpu))) {
6209 		kvm_queue_exception(vcpu, UD_VECTOR);
6210 		return 1;
6211 	}
6212 
6213 	vmcs12 = get_vmcs12(vcpu);
6214 
6215 	/*
6216 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
6217 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
6218 	 */
6219 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
6220 		kvm_queue_exception(vcpu, UD_VECTOR);
6221 		return 1;
6222 	}
6223 
6224 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
6225 		goto fail;
6226 
6227 	switch (function) {
6228 	case 0:
6229 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
6230 			goto fail;
6231 		break;
6232 	default:
6233 		goto fail;
6234 	}
6235 	return kvm_skip_emulated_instruction(vcpu);
6236 
6237 fail:
6238 	/*
6239 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
6240 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
6241 	 * EXIT_REASON_VMFUNC as the exit reason.
6242 	 */
6243 	nested_vmx_vmexit(vcpu, vmx->vt.exit_reason.full,
6244 			  vmx_get_intr_info(vcpu),
6245 			  vmx_get_exit_qual(vcpu));
6246 	return 1;
6247 }
6248 
6249 /*
6250  * Return true if an IO instruction with the specified port and size should cause
6251  * a VM-exit into L1.
6252  */
6253 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
6254 				 int size)
6255 {
6256 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6257 	gpa_t bitmap, last_bitmap;
6258 	u8 b;
6259 
6260 	last_bitmap = INVALID_GPA;
6261 	b = -1;
6262 
6263 	while (size > 0) {
6264 		if (port < 0x8000)
6265 			bitmap = vmcs12->io_bitmap_a;
6266 		else if (port < 0x10000)
6267 			bitmap = vmcs12->io_bitmap_b;
6268 		else
6269 			return true;
6270 		bitmap += (port & 0x7fff) / 8;
6271 
6272 		if (last_bitmap != bitmap)
6273 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
6274 				return true;
6275 		if (b & (1 << (port & 7)))
6276 			return true;
6277 
6278 		port++;
6279 		size--;
6280 		last_bitmap = bitmap;
6281 	}
6282 
6283 	return false;
6284 }
6285 
6286 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6287 				       struct vmcs12 *vmcs12)
6288 {
6289 	unsigned long exit_qualification;
6290 	unsigned short port;
6291 	int size;
6292 
6293 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6294 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
6295 
6296 	exit_qualification = vmx_get_exit_qual(vcpu);
6297 
6298 	port = exit_qualification >> 16;
6299 	size = (exit_qualification & 7) + 1;
6300 
6301 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
6302 }
6303 
6304 /*
6305  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
6306  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6307  * disinterest in the current event (read or write a specific MSR) by using an
6308  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6309  */
6310 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6311 					struct vmcs12 *vmcs12,
6312 					union vmx_exit_reason exit_reason)
6313 {
6314 	u32 msr_index;
6315 	gpa_t bitmap;
6316 
6317 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6318 		return true;
6319 
6320 	if (exit_reason.basic == EXIT_REASON_MSR_READ_IMM ||
6321 	    exit_reason.basic == EXIT_REASON_MSR_WRITE_IMM)
6322 		msr_index = vmx_get_exit_qual(vcpu);
6323 	else
6324 		msr_index = kvm_rcx_read(vcpu);
6325 
6326 	/*
6327 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6328 	 * for the four combinations of read/write and low/high MSR numbers.
6329 	 * First we need to figure out which of the four to use:
6330 	 */
6331 	bitmap = vmcs12->msr_bitmap;
6332 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE ||
6333 	    exit_reason.basic == EXIT_REASON_MSR_WRITE_IMM)
6334 		bitmap += 2048;
6335 	if (msr_index >= 0xc0000000) {
6336 		msr_index -= 0xc0000000;
6337 		bitmap += 1024;
6338 	}
6339 
6340 	/* Then read the msr_index'th bit from this bitmap: */
6341 	if (msr_index < 1024*8) {
6342 		unsigned char b;
6343 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
6344 			return true;
6345 		return 1 & (b >> (msr_index & 7));
6346 	} else
6347 		return true; /* let L1 handle the wrong parameter */
6348 }
6349 
6350 /*
6351  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6352  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6353  * intercept (via guest_host_mask etc.) the current event.
6354  */
6355 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6356 	struct vmcs12 *vmcs12)
6357 {
6358 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
6359 	int cr = exit_qualification & 15;
6360 	int reg;
6361 	unsigned long val;
6362 
6363 	switch ((exit_qualification >> 4) & 3) {
6364 	case 0: /* mov to cr */
6365 		reg = (exit_qualification >> 8) & 15;
6366 		val = kvm_register_read(vcpu, reg);
6367 		switch (cr) {
6368 		case 0:
6369 			if (vmcs12->cr0_guest_host_mask &
6370 			    (val ^ vmcs12->cr0_read_shadow))
6371 				return true;
6372 			break;
6373 		case 3:
6374 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6375 				return true;
6376 			break;
6377 		case 4:
6378 			if (vmcs12->cr4_guest_host_mask &
6379 			    (vmcs12->cr4_read_shadow ^ val))
6380 				return true;
6381 			break;
6382 		case 8:
6383 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6384 				return true;
6385 			break;
6386 		}
6387 		break;
6388 	case 2: /* clts */
6389 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6390 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
6391 			return true;
6392 		break;
6393 	case 1: /* mov from cr */
6394 		switch (cr) {
6395 		case 3:
6396 			if (vmcs12->cpu_based_vm_exec_control &
6397 			    CPU_BASED_CR3_STORE_EXITING)
6398 				return true;
6399 			break;
6400 		case 8:
6401 			if (vmcs12->cpu_based_vm_exec_control &
6402 			    CPU_BASED_CR8_STORE_EXITING)
6403 				return true;
6404 			break;
6405 		}
6406 		break;
6407 	case 3: /* lmsw */
6408 		/*
6409 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6410 		 * cr0. Other attempted changes are ignored, with no exit.
6411 		 */
6412 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6413 		if (vmcs12->cr0_guest_host_mask & 0xe &
6414 		    (val ^ vmcs12->cr0_read_shadow))
6415 			return true;
6416 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6417 		    !(vmcs12->cr0_read_shadow & 0x1) &&
6418 		    (val & 0x1))
6419 			return true;
6420 		break;
6421 	}
6422 	return false;
6423 }
6424 
6425 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
6426 					  struct vmcs12 *vmcs12)
6427 {
6428 	u32 encls_leaf;
6429 
6430 	if (!guest_cpu_cap_has(vcpu, X86_FEATURE_SGX) ||
6431 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
6432 		return false;
6433 
6434 	encls_leaf = kvm_rax_read(vcpu);
6435 	if (encls_leaf > 62)
6436 		encls_leaf = 63;
6437 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
6438 }
6439 
6440 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
6441 	struct vmcs12 *vmcs12, gpa_t bitmap)
6442 {
6443 	u32 vmx_instruction_info;
6444 	unsigned long field;
6445 	u8 b;
6446 
6447 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
6448 		return true;
6449 
6450 	/* Decode instruction info and find the field to access */
6451 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6452 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6453 
6454 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
6455 	if (field >> 15)
6456 		return true;
6457 
6458 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
6459 		return true;
6460 
6461 	return 1 & (b >> (field & 7));
6462 }
6463 
6464 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
6465 {
6466 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
6467 
6468 	if (nested_cpu_has_mtf(vmcs12))
6469 		return true;
6470 
6471 	/*
6472 	 * An MTF VM-exit may be injected into the guest by setting the
6473 	 * interruption-type to 7 (other event) and the vector field to 0. Such
6474 	 * is the case regardless of the 'monitor trap flag' VM-execution
6475 	 * control.
6476 	 */
6477 	return entry_intr_info == (INTR_INFO_VALID_MASK
6478 				   | INTR_TYPE_OTHER_EVENT);
6479 }
6480 
6481 /*
6482  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
6483  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
6484  */
6485 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
6486 				     union vmx_exit_reason exit_reason)
6487 {
6488 	u32 intr_info;
6489 
6490 	switch ((u16)exit_reason.basic) {
6491 	case EXIT_REASON_EXCEPTION_NMI:
6492 		intr_info = vmx_get_intr_info(vcpu);
6493 		if (is_nmi(intr_info))
6494 			return true;
6495 		else if (is_page_fault(intr_info))
6496 			return vcpu->arch.apf.host_apf_flags ||
6497 			       vmx_need_pf_intercept(vcpu);
6498 		else if (is_debug(intr_info) &&
6499 			 vcpu->guest_debug &
6500 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6501 			return true;
6502 		else if (is_breakpoint(intr_info) &&
6503 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6504 			return true;
6505 		else if (is_alignment_check(intr_info) &&
6506 			 !vmx_guest_inject_ac(vcpu))
6507 			return true;
6508 		else if (is_ve_fault(intr_info))
6509 			return true;
6510 		return false;
6511 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6512 		return true;
6513 	case EXIT_REASON_MCE_DURING_VMENTRY:
6514 		return true;
6515 	case EXIT_REASON_EPT_VIOLATION:
6516 		/*
6517 		 * L0 always deals with the EPT violation. If nested EPT is
6518 		 * used, and the nested mmu code discovers that the address is
6519 		 * missing in the guest EPT table (EPT12), the EPT violation
6520 		 * will be injected with nested_ept_inject_page_fault()
6521 		 */
6522 		return true;
6523 	case EXIT_REASON_EPT_MISCONFIG:
6524 		/*
6525 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
6526 		 * table (shadow on EPT) or a merged EPT table that L0 built
6527 		 * (EPT on EPT). So any problems with the structure of the
6528 		 * table is L0's fault.
6529 		 */
6530 		return true;
6531 	case EXIT_REASON_PREEMPTION_TIMER:
6532 		return true;
6533 	case EXIT_REASON_PML_FULL:
6534 		/*
6535 		 * PML is emulated for an L1 VMM and should never be enabled in
6536 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
6537 		 */
6538 		return true;
6539 	case EXIT_REASON_VMFUNC:
6540 		/* VM functions are emulated through L2->L0 vmexits. */
6541 		return true;
6542 	case EXIT_REASON_BUS_LOCK:
6543 		/*
6544 		 * At present, bus lock VM exit is never exposed to L1.
6545 		 * Handle L2's bus locks in L0 directly.
6546 		 */
6547 		return true;
6548 #ifdef CONFIG_KVM_HYPERV
6549 	case EXIT_REASON_VMCALL:
6550 		/* Hyper-V L2 TLB flush hypercall is handled by L0 */
6551 		return guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
6552 			nested_evmcs_l2_tlb_flush_enabled(vcpu) &&
6553 			kvm_hv_is_tlb_flush_hcall(vcpu);
6554 #endif
6555 	default:
6556 		break;
6557 	}
6558 	return false;
6559 }
6560 
6561 /*
6562  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
6563  * is_guest_mode (L2).
6564  */
6565 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
6566 				     union vmx_exit_reason exit_reason)
6567 {
6568 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6569 	u32 intr_info;
6570 
6571 	switch ((u16)exit_reason.basic) {
6572 	case EXIT_REASON_EXCEPTION_NMI:
6573 		intr_info = vmx_get_intr_info(vcpu);
6574 		if (is_nmi(intr_info))
6575 			return true;
6576 		else if (is_page_fault(intr_info))
6577 			return true;
6578 		return vmcs12->exception_bitmap &
6579 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
6580 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6581 		return nested_exit_on_intr(vcpu);
6582 	case EXIT_REASON_TRIPLE_FAULT:
6583 		return true;
6584 	case EXIT_REASON_INTERRUPT_WINDOW:
6585 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
6586 	case EXIT_REASON_NMI_WINDOW:
6587 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
6588 	case EXIT_REASON_TASK_SWITCH:
6589 		return true;
6590 	case EXIT_REASON_CPUID:
6591 		return true;
6592 	case EXIT_REASON_HLT:
6593 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6594 	case EXIT_REASON_INVD:
6595 		return true;
6596 	case EXIT_REASON_INVLPG:
6597 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6598 	case EXIT_REASON_RDPMC:
6599 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6600 	case EXIT_REASON_RDRAND:
6601 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
6602 	case EXIT_REASON_RDSEED:
6603 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
6604 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
6605 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6606 	case EXIT_REASON_VMREAD:
6607 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6608 			vmcs12->vmread_bitmap);
6609 	case EXIT_REASON_VMWRITE:
6610 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6611 			vmcs12->vmwrite_bitmap);
6612 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6613 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6614 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
6615 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6616 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
6617 		/*
6618 		 * VMX instructions trap unconditionally. This allows L1 to
6619 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6620 		 */
6621 		return true;
6622 	case EXIT_REASON_CR_ACCESS:
6623 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6624 	case EXIT_REASON_DR_ACCESS:
6625 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6626 	case EXIT_REASON_IO_INSTRUCTION:
6627 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
6628 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6629 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6630 	case EXIT_REASON_MSR_READ:
6631 	case EXIT_REASON_MSR_WRITE:
6632 	case EXIT_REASON_MSR_READ_IMM:
6633 	case EXIT_REASON_MSR_WRITE_IMM:
6634 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6635 	case EXIT_REASON_INVALID_STATE:
6636 		return true;
6637 	case EXIT_REASON_MWAIT_INSTRUCTION:
6638 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6639 	case EXIT_REASON_MONITOR_TRAP_FLAG:
6640 		return nested_vmx_exit_handled_mtf(vmcs12);
6641 	case EXIT_REASON_MONITOR_INSTRUCTION:
6642 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6643 	case EXIT_REASON_PAUSE_INSTRUCTION:
6644 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6645 			nested_cpu_has2(vmcs12,
6646 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6647 	case EXIT_REASON_MCE_DURING_VMENTRY:
6648 		return true;
6649 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6650 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6651 	case EXIT_REASON_APIC_ACCESS:
6652 	case EXIT_REASON_APIC_WRITE:
6653 	case EXIT_REASON_EOI_INDUCED:
6654 		/*
6655 		 * The controls for "virtualize APIC accesses," "APIC-
6656 		 * register virtualization," and "virtual-interrupt
6657 		 * delivery" only come from vmcs12.
6658 		 */
6659 		return true;
6660 	case EXIT_REASON_INVPCID:
6661 		return
6662 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6663 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6664 	case EXIT_REASON_WBINVD:
6665 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6666 	case EXIT_REASON_XSETBV:
6667 		return true;
6668 	case EXIT_REASON_XSAVES:
6669 	case EXIT_REASON_XRSTORS:
6670 		/*
6671 		 * Always forward XSAVES/XRSTORS to L1 as KVM doesn't utilize
6672 		 * XSS-bitmap, and always loads vmcs02 with vmcs12's XSS-bitmap
6673 		 * verbatim, i.e. any exit is due to L1's bitmap.  WARN if
6674 		 * XSAVES isn't enabled, as the CPU is supposed to inject #UD
6675 		 * in that case, before consulting the XSS-bitmap.
6676 		 */
6677 		WARN_ON_ONCE(!nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_XSAVES));
6678 		return true;
6679 	case EXIT_REASON_UMWAIT:
6680 	case EXIT_REASON_TPAUSE:
6681 		return nested_cpu_has2(vmcs12,
6682 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6683 	case EXIT_REASON_ENCLS:
6684 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6685 	case EXIT_REASON_NOTIFY:
6686 		/* Notify VM exit is not exposed to L1 */
6687 		return false;
6688 	case EXIT_REASON_SEAMCALL:
6689 	case EXIT_REASON_TDCALL:
6690 		/*
6691 		 * SEAMCALL and TDCALL unconditionally VM-Exit, but aren't
6692 		 * virtualized by KVM for L1 hypervisors, i.e. L1 should
6693 		 * never want or expect such an exit.
6694 		 */
6695 		return false;
6696 	default:
6697 		return true;
6698 	}
6699 }
6700 
6701 /*
6702  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6703  * reflected into L1.
6704  */
6705 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6706 {
6707 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6708 	union vmx_exit_reason exit_reason = vmx->vt.exit_reason;
6709 	unsigned long exit_qual;
6710 	u32 exit_intr_info;
6711 
6712 	kvm_warn_on_nested_run_pending(vcpu);
6713 
6714 	/*
6715 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6716 	 * has already loaded L2's state.
6717 	 */
6718 	if (unlikely(vmx->fail)) {
6719 		trace_kvm_nested_vmenter_failed(
6720 			"hardware VM-instruction error: ",
6721 			vmcs_read32(VM_INSTRUCTION_ERROR));
6722 		exit_intr_info = 0;
6723 		exit_qual = 0;
6724 		goto reflect_vmexit;
6725 	}
6726 
6727 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6728 
6729 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6730 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6731 		return false;
6732 
6733 	/* If L1 doesn't want the exit, handle it in L0. */
6734 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6735 		return false;
6736 
6737 	/*
6738 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6739 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6740 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6741 	 * interrupts are never reflected to L1 so it's a non-issue.
6742 	 */
6743 	exit_intr_info = vmx_get_intr_info(vcpu);
6744 	if (is_exception_with_error_code(exit_intr_info)) {
6745 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6746 
6747 		vmcs12->vm_exit_intr_error_code =
6748 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6749 	}
6750 	exit_qual = vmx_get_exit_qual(vcpu);
6751 
6752 reflect_vmexit:
6753 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6754 	return true;
6755 }
6756 
6757 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6758 				struct kvm_nested_state __user *user_kvm_nested_state,
6759 				u32 user_data_size)
6760 {
6761 	struct vcpu_vmx *vmx;
6762 	struct vmcs12 *vmcs12;
6763 	struct kvm_nested_state kvm_state = {
6764 		.flags = 0,
6765 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6766 		.size = sizeof(kvm_state),
6767 		.hdr.vmx.flags = 0,
6768 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6769 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6770 		.hdr.vmx.preemption_timer_deadline = 0,
6771 	};
6772 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6773 		&user_kvm_nested_state->data.vmx[0];
6774 
6775 	if (!vcpu)
6776 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6777 
6778 	vmx = to_vmx(vcpu);
6779 	vmcs12 = get_vmcs12(vcpu);
6780 
6781 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_VMX) &&
6782 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6783 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6784 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6785 
6786 		if (vmx_has_valid_vmcs12(vcpu)) {
6787 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6788 
6789 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6790 			if (nested_vmx_is_evmptr12_set(vmx))
6791 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6792 
6793 			if (is_guest_mode(vcpu) &&
6794 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6795 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6796 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6797 		}
6798 
6799 		if (vmx->nested.smm.vmxon)
6800 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6801 
6802 		if (vmx->nested.smm.guest_mode)
6803 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6804 
6805 		if (is_guest_mode(vcpu)) {
6806 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6807 
6808 			if (vcpu->arch.nested_run_pending)
6809 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6810 
6811 			if (vmx->nested.mtf_pending)
6812 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6813 
6814 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6815 			    vmx->nested.has_preemption_timer_deadline) {
6816 				kvm_state.hdr.vmx.flags |=
6817 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6818 				kvm_state.hdr.vmx.preemption_timer_deadline =
6819 					vmx->nested.preemption_timer_deadline;
6820 			}
6821 		}
6822 	}
6823 
6824 	if (user_data_size < kvm_state.size)
6825 		goto out;
6826 
6827 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6828 		return -EFAULT;
6829 
6830 	if (!vmx_has_valid_vmcs12(vcpu))
6831 		goto out;
6832 
6833 	/*
6834 	 * When running L2, the authoritative vmcs12 state is in the
6835 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6836 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6837 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6838 	 * vmcs12 state is in the vmcs12 already.
6839 	 */
6840 	if (is_guest_mode(vcpu)) {
6841 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6842 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6843 	} else  {
6844 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6845 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6846 			if (nested_vmx_is_evmptr12_valid(vmx))
6847 				/*
6848 				 * L1 hypervisor is not obliged to keep eVMCS
6849 				 * clean fields data always up-to-date while
6850 				 * not in guest mode, 'hv_clean_fields' is only
6851 				 * supposed to be actual upon vmentry so we need
6852 				 * to ignore it here and do full copy.
6853 				 */
6854 				copy_enlightened_to_vmcs12(vmx, 0);
6855 			else if (enable_shadow_vmcs)
6856 				copy_shadow_to_vmcs12(vmx);
6857 		}
6858 	}
6859 
6860 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6861 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6862 
6863 	/*
6864 	 * Copy over the full allocated size of vmcs12 rather than just the size
6865 	 * of the struct.
6866 	 */
6867 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6868 		return -EFAULT;
6869 
6870 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6871 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6872 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6873 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6874 			return -EFAULT;
6875 	}
6876 out:
6877 	return kvm_state.size;
6878 }
6879 
6880 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6881 {
6882 	if (is_guest_mode(vcpu)) {
6883 		vcpu->arch.nested_run_pending = 0;
6884 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6885 	}
6886 	free_nested(vcpu);
6887 }
6888 
6889 int nested_vmx_check_restored_vmcs12(struct kvm_vcpu *vcpu)
6890 {
6891 	enum vm_entry_failure_code ignored;
6892 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6893 
6894 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6895 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6896 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6897 
6898 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6899 		    !shadow_vmcs12->hdr.shadow_vmcs)
6900 			return -EINVAL;
6901 	}
6902 
6903 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6904 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6905 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6906 		return -EINVAL;
6907 
6908 	return 0;
6909 }
6910 
6911 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6912 				struct kvm_nested_state __user *user_kvm_nested_state,
6913 				struct kvm_nested_state *kvm_state)
6914 {
6915 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6916 	struct vmcs12 *vmcs12;
6917 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6918 		&user_kvm_nested_state->data.vmx[0];
6919 	int ret;
6920 
6921 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6922 		return -EINVAL;
6923 
6924 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6925 		if (kvm_state->hdr.vmx.smm.flags)
6926 			return -EINVAL;
6927 
6928 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6929 			return -EINVAL;
6930 
6931 		/*
6932 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6933 		 * enable eVMCS capability on vCPU. However, since then
6934 		 * code was changed such that flag signals vmcs12 should
6935 		 * be copied into eVMCS in guest memory.
6936 		 *
6937 		 * To preserve backwards compatibility, allow user
6938 		 * to set this flag even when there is no VMXON region.
6939 		 */
6940 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6941 			return -EINVAL;
6942 	} else {
6943 		if (!guest_cpu_cap_has(vcpu, X86_FEATURE_VMX))
6944 			return -EINVAL;
6945 
6946 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6947 			return -EINVAL;
6948 	}
6949 
6950 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6951 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6952 		return -EINVAL;
6953 
6954 	if (kvm_state->hdr.vmx.smm.flags &
6955 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6956 		return -EINVAL;
6957 
6958 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6959 		return -EINVAL;
6960 
6961 	/*
6962 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6963 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6964 	 * must be zero.
6965 	 */
6966 	if (is_smm(vcpu) ?
6967 		(kvm_state->flags &
6968 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6969 		: kvm_state->hdr.vmx.smm.flags)
6970 		return -EINVAL;
6971 
6972 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6973 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6974 		return -EINVAL;
6975 
6976 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6977 	    (!guest_cpu_cap_has(vcpu, X86_FEATURE_VMX) ||
6978 	     !vmx->nested.enlightened_vmcs_enabled))
6979 			return -EINVAL;
6980 
6981 	vmx_leave_nested(vcpu);
6982 
6983 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6984 		return 0;
6985 
6986 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6987 	ret = enter_vmx_operation(vcpu);
6988 	if (ret)
6989 		return ret;
6990 
6991 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6992 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6993 		/* See vmx_has_valid_vmcs12.  */
6994 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6995 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6996 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6997 			return -EINVAL;
6998 		else
6999 			return 0;
7000 	}
7001 
7002 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
7003 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
7004 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
7005 			return -EINVAL;
7006 
7007 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
7008 #ifdef CONFIG_KVM_HYPERV
7009 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
7010 		/*
7011 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
7012 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
7013 		 * restored yet. EVMCS will be mapped from
7014 		 * nested_get_vmcs12_pages().
7015 		 */
7016 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
7017 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
7018 #endif
7019 	} else {
7020 		return -EINVAL;
7021 	}
7022 
7023 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
7024 		vmx->nested.smm.vmxon = true;
7025 		vmx->nested.vmxon = false;
7026 
7027 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
7028 			vmx->nested.smm.guest_mode = true;
7029 	}
7030 
7031 	vmcs12 = get_vmcs12(vcpu);
7032 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
7033 		return -EFAULT;
7034 
7035 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
7036 		return -EINVAL;
7037 
7038 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
7039 		return 0;
7040 
7041 	if (kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING)
7042 		vcpu->arch.nested_run_pending = KVM_NESTED_RUN_PENDING_UNTRUSTED;
7043 	else
7044 		vcpu->arch.nested_run_pending = 0;
7045 
7046 	vmx->nested.mtf_pending =
7047 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
7048 
7049 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
7050 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
7051 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
7052 
7053 		ret = -EINVAL;
7054 		if (kvm_state->size <
7055 		    sizeof(*kvm_state) +
7056 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
7057 			goto error_guest_mode;
7058 
7059 		ret = -EFAULT;
7060 		if (copy_from_user(shadow_vmcs12,
7061 				   user_vmx_nested_state->shadow_vmcs12,
7062 				   sizeof(*shadow_vmcs12)))
7063 			goto error_guest_mode;
7064 	}
7065 
7066 	vmx->nested.has_preemption_timer_deadline = false;
7067 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
7068 		vmx->nested.has_preemption_timer_deadline = true;
7069 		vmx->nested.preemption_timer_deadline =
7070 			kvm_state->hdr.vmx.preemption_timer_deadline;
7071 	}
7072 
7073 	ret = nested_vmx_check_restored_vmcs12(vcpu);
7074 	if (ret < 0)
7075 		goto error_guest_mode;
7076 
7077 	vmx->nested.dirty_vmcs12 = true;
7078 	vmx->nested.force_msr_bitmap_recalc = true;
7079 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
7080 	if (ret)
7081 		goto error_guest_mode;
7082 
7083 	if (vmx->nested.mtf_pending)
7084 		kvm_make_request(KVM_REQ_EVENT, vcpu);
7085 
7086 	return 0;
7087 
7088 error_guest_mode:
7089 	vcpu->arch.nested_run_pending = 0;
7090 	return ret;
7091 }
7092 
7093 void nested_vmx_set_vmcs_shadowing_bitmap(void)
7094 {
7095 	if (enable_shadow_vmcs) {
7096 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
7097 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
7098 	}
7099 }
7100 
7101 static u64 nested_vmx_calc_vmcs_enum_msr(void)
7102 {
7103 	/*
7104 	 * Note these are the so called "index" of the VMCS field encoding, not
7105 	 * the index into vmcs12.
7106 	 */
7107 	unsigned int max_idx, idx;
7108 	int i;
7109 
7110 	/*
7111 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
7112 	 * vmcs12, regardless of whether or not the associated feature is
7113 	 * exposed to L1.  Simply find the field with the highest index.
7114 	 */
7115 	max_idx = 0;
7116 	for (i = 0; i < nr_vmcs12_fields; i++) {
7117 		/* The vmcs12 table is very, very sparsely populated. */
7118 		if (!vmcs12_field_offsets[i])
7119 			continue;
7120 
7121 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
7122 		if (idx > max_idx)
7123 			max_idx = idx;
7124 	}
7125 
7126 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
7127 }
7128 
7129 static void nested_vmx_setup_pinbased_ctls(struct vmcs_config *vmcs_conf,
7130 					   struct nested_vmx_msrs *msrs)
7131 {
7132 	msrs->pinbased_ctls_low =
7133 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
7134 
7135 	msrs->pinbased_ctls_high = vmcs_conf->pin_based_exec_ctrl;
7136 	msrs->pinbased_ctls_high &=
7137 		PIN_BASED_EXT_INTR_MASK |
7138 		PIN_BASED_NMI_EXITING |
7139 		PIN_BASED_VIRTUAL_NMIS |
7140 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
7141 	msrs->pinbased_ctls_high |=
7142 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
7143 		PIN_BASED_VMX_PREEMPTION_TIMER;
7144 }
7145 
7146 static void nested_vmx_setup_exit_ctls(struct vmcs_config *vmcs_conf,
7147 				       struct nested_vmx_msrs *msrs)
7148 {
7149 	msrs->exit_ctls_low =
7150 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
7151 
7152 	msrs->exit_ctls_high = vmcs_conf->vmexit_ctrl;
7153 	msrs->exit_ctls_high &=
7154 #ifdef CONFIG_X86_64
7155 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
7156 #endif
7157 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
7158 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_CET_STATE;
7159 	msrs->exit_ctls_high |=
7160 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
7161 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
7162 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT |
7163 		VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
7164 
7165 	if (!kvm_cpu_cap_has(X86_FEATURE_SHSTK) &&
7166 	    !kvm_cpu_cap_has(X86_FEATURE_IBT))
7167 		msrs->exit_ctls_high &= ~VM_EXIT_LOAD_CET_STATE;
7168 
7169 	/* We support free control of debug control saving. */
7170 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
7171 }
7172 
7173 static void nested_vmx_setup_entry_ctls(struct vmcs_config *vmcs_conf,
7174 					struct nested_vmx_msrs *msrs)
7175 {
7176 	msrs->entry_ctls_low =
7177 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
7178 
7179 	msrs->entry_ctls_high = vmcs_conf->vmentry_ctrl;
7180 	msrs->entry_ctls_high &=
7181 #ifdef CONFIG_X86_64
7182 		VM_ENTRY_IA32E_MODE |
7183 #endif
7184 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
7185 		VM_ENTRY_LOAD_CET_STATE;
7186 	msrs->entry_ctls_high |=
7187 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER |
7188 		 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
7189 
7190 	if (!kvm_cpu_cap_has(X86_FEATURE_SHSTK) &&
7191 	    !kvm_cpu_cap_has(X86_FEATURE_IBT))
7192 		msrs->entry_ctls_high &= ~VM_ENTRY_LOAD_CET_STATE;
7193 
7194 	/* We support free control of debug control loading. */
7195 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
7196 }
7197 
7198 static void nested_vmx_setup_cpubased_ctls(struct vmcs_config *vmcs_conf,
7199 					   struct nested_vmx_msrs *msrs)
7200 {
7201 	msrs->procbased_ctls_low =
7202 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
7203 
7204 	msrs->procbased_ctls_high = vmcs_conf->cpu_based_exec_ctrl;
7205 	msrs->procbased_ctls_high &=
7206 		CPU_BASED_INTR_WINDOW_EXITING |
7207 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
7208 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
7209 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
7210 		CPU_BASED_CR3_STORE_EXITING |
7211 #ifdef CONFIG_X86_64
7212 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
7213 #endif
7214 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
7215 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
7216 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
7217 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
7218 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
7219 	/*
7220 	 * We can allow some features even when not supported by the
7221 	 * hardware. For example, L1 can specify an MSR bitmap - and we
7222 	 * can use it to avoid exits to L1 - even when L0 runs L2
7223 	 * without MSR bitmaps.
7224 	 */
7225 	msrs->procbased_ctls_high |=
7226 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
7227 		CPU_BASED_USE_MSR_BITMAPS;
7228 
7229 	/* We support free control of CR3 access interception. */
7230 	msrs->procbased_ctls_low &=
7231 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
7232 }
7233 
7234 static void nested_vmx_setup_secondary_ctls(u32 ept_caps,
7235 					    struct vmcs_config *vmcs_conf,
7236 					    struct nested_vmx_msrs *msrs)
7237 {
7238 	msrs->secondary_ctls_low = 0;
7239 
7240 	msrs->secondary_ctls_high = vmcs_conf->cpu_based_2nd_exec_ctrl;
7241 	msrs->secondary_ctls_high &=
7242 		SECONDARY_EXEC_DESC |
7243 		SECONDARY_EXEC_ENABLE_RDTSCP |
7244 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7245 		SECONDARY_EXEC_WBINVD_EXITING |
7246 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
7247 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
7248 		SECONDARY_EXEC_RDRAND_EXITING |
7249 		SECONDARY_EXEC_ENABLE_INVPCID |
7250 		SECONDARY_EXEC_ENABLE_VMFUNC |
7251 		SECONDARY_EXEC_RDSEED_EXITING |
7252 		SECONDARY_EXEC_ENABLE_XSAVES |
7253 		SECONDARY_EXEC_TSC_SCALING |
7254 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
7255 
7256 	/*
7257 	 * We can emulate "VMCS shadowing," even if the hardware
7258 	 * doesn't support it.
7259 	 */
7260 	msrs->secondary_ctls_high |=
7261 		SECONDARY_EXEC_SHADOW_VMCS;
7262 
7263 	if (enable_ept) {
7264 		/* nested EPT: emulate EPT also to L1 */
7265 		msrs->secondary_ctls_high |=
7266 			SECONDARY_EXEC_ENABLE_EPT;
7267 		msrs->ept_caps =
7268 			VMX_EPT_PAGE_WALK_4_BIT |
7269 			VMX_EPT_PAGE_WALK_5_BIT |
7270 			VMX_EPTP_WB_BIT |
7271 			VMX_EPT_INVEPT_BIT |
7272 			VMX_EPT_EXECUTE_ONLY_BIT |
7273 			VMX_EPT_ADVANCED_VMEXIT_INFO_BIT;
7274 
7275 		msrs->ept_caps &= ept_caps;
7276 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
7277 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
7278 			VMX_EPT_1GB_PAGE_BIT;
7279 		if (enable_ept_ad_bits) {
7280 			msrs->secondary_ctls_high |=
7281 				SECONDARY_EXEC_ENABLE_PML;
7282 			msrs->ept_caps |= VMX_EPT_AD_BIT;
7283 		}
7284 
7285 		if (enable_mbec)
7286 			msrs->secondary_ctls_high |=
7287 				SECONDARY_EXEC_MODE_BASED_EPT_EXEC;
7288 		/*
7289 		 * Advertise EPTP switching irrespective of hardware support,
7290 		 * KVM emulates it in software so long as VMFUNC is supported.
7291 		 */
7292 		if (cpu_has_vmx_vmfunc())
7293 			msrs->vmfunc_controls = VMX_VMFUNC_EPTP_SWITCHING;
7294 	}
7295 
7296 	/*
7297 	 * Old versions of KVM use the single-context version without
7298 	 * checking for support, so declare that it is supported even
7299 	 * though it is treated as global context.  The alternative is
7300 	 * not failing the single-context invvpid, and it is worse.
7301 	 */
7302 	if (enable_vpid) {
7303 		msrs->secondary_ctls_high |=
7304 			SECONDARY_EXEC_ENABLE_VPID;
7305 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
7306 			VMX_VPID_EXTENT_SUPPORTED_MASK;
7307 	}
7308 
7309 	if (enable_unrestricted_guest)
7310 		msrs->secondary_ctls_high |=
7311 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
7312 
7313 	if (flexpriority_enabled)
7314 		msrs->secondary_ctls_high |=
7315 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7316 
7317 	if (enable_sgx)
7318 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
7319 }
7320 
7321 static void nested_vmx_setup_misc_data(struct vmcs_config *vmcs_conf,
7322 				       struct nested_vmx_msrs *msrs)
7323 {
7324 	msrs->misc_low = (u32)vmcs_conf->misc & VMX_MISC_SAVE_EFER_LMA;
7325 	msrs->misc_low |=
7326 		VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
7327 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
7328 		VMX_MISC_ACTIVITY_HLT |
7329 		VMX_MISC_ACTIVITY_WAIT_SIPI;
7330 	msrs->misc_high = 0;
7331 }
7332 
7333 static void nested_vmx_setup_basic(struct nested_vmx_msrs *msrs)
7334 {
7335 	/*
7336 	 * This MSR reports some information about VMX support. We
7337 	 * should return information about the VMX we emulate for the
7338 	 * guest, and the VMCS structure we give it - not about the
7339 	 * VMX support of the underlying hardware.
7340 	 */
7341 	msrs->basic = vmx_basic_encode_vmcs_info(VMCS12_REVISION, VMCS12_SIZE,
7342 						 X86_MEMTYPE_WB);
7343 
7344 	msrs->basic |= VMX_BASIC_TRUE_CTLS;
7345 	if (cpu_has_vmx_basic_inout())
7346 		msrs->basic |= VMX_BASIC_INOUT;
7347 	if (cpu_has_vmx_basic_no_hw_errcode_cc())
7348 		msrs->basic |= VMX_BASIC_NO_HW_ERROR_CODE_CC;
7349 }
7350 
7351 static void nested_vmx_setup_cr_fixed(struct nested_vmx_msrs *msrs)
7352 {
7353 	/*
7354 	 * These MSRs specify bits which the guest must keep fixed on
7355 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
7356 	 * We picked the standard core2 setting.
7357 	 */
7358 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
7359 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
7360 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
7361 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
7362 
7363 	/* These MSRs specify bits which the guest must keep fixed off. */
7364 	rdmsrq(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
7365 	rdmsrq(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
7366 
7367 	if (vmx_umip_emulated())
7368 		msrs->cr4_fixed1 |= X86_CR4_UMIP;
7369 }
7370 
7371 /*
7372  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
7373  * returned for the various VMX controls MSRs when nested VMX is enabled.
7374  * The same values should also be used to verify that vmcs12 control fields are
7375  * valid during nested entry from L1 to L2.
7376  * Each of these control msrs has a low and high 32-bit half: A low bit is on
7377  * if the corresponding bit in the (32-bit) control field *must* be on, and a
7378  * bit in the high half is on if the corresponding bit in the control field
7379  * may be on. See also vmx_control_verify().
7380  */
7381 void nested_vmx_setup_ctls_msrs(struct vmcs_config *vmcs_conf, u32 ept_caps)
7382 {
7383 	struct nested_vmx_msrs *msrs = &vmcs_conf->nested;
7384 
7385 	/*
7386 	 * Note that as a general rule, the high half of the MSRs (bits in
7387 	 * the control fields which may be 1) should be initialized by the
7388 	 * intersection of the underlying hardware's MSR (i.e., features which
7389 	 * can be supported) and the list of features we want to expose -
7390 	 * because they are known to be properly supported in our code.
7391 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
7392 	 * be set to 0, meaning that L1 may turn off any of these bits. The
7393 	 * reason is that if one of these bits is necessary, it will appear
7394 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
7395 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
7396 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
7397 	 * These rules have exceptions below.
7398 	 */
7399 	nested_vmx_setup_pinbased_ctls(vmcs_conf, msrs);
7400 
7401 	nested_vmx_setup_exit_ctls(vmcs_conf, msrs);
7402 
7403 	nested_vmx_setup_entry_ctls(vmcs_conf, msrs);
7404 
7405 	nested_vmx_setup_cpubased_ctls(vmcs_conf, msrs);
7406 
7407 	nested_vmx_setup_secondary_ctls(ept_caps, vmcs_conf, msrs);
7408 
7409 	nested_vmx_setup_misc_data(vmcs_conf, msrs);
7410 
7411 	nested_vmx_setup_basic(msrs);
7412 
7413 	nested_vmx_setup_cr_fixed(msrs);
7414 
7415 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
7416 }
7417 
7418 void nested_vmx_hardware_unsetup(void)
7419 {
7420 	int i;
7421 
7422 	if (enable_shadow_vmcs) {
7423 		for (i = 0; i < VMX_BITMAP_NR; i++)
7424 			free_page((unsigned long)vmx_bitmap[i]);
7425 	}
7426 }
7427 
7428 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
7429 {
7430 	int i;
7431 
7432 	/*
7433 	 * Note!  The set of supported vmcs12 fields is consumed by both VMX
7434 	 * MSR and shadow VMCS setup.
7435 	 */
7436 	nested_vmx_setup_vmcs12_fields();
7437 
7438 	nested_vmx_setup_ctls_msrs(&vmcs_config, vmx_capability.ept);
7439 
7440 	if (!cpu_has_vmx_shadow_vmcs())
7441 		enable_shadow_vmcs = 0;
7442 	if (enable_shadow_vmcs) {
7443 		for (i = 0; i < VMX_BITMAP_NR; i++) {
7444 			/*
7445 			 * The vmx_bitmap is not tied to a VM and so should
7446 			 * not be charged to a memcg.
7447 			 */
7448 			vmx_bitmap[i] = (unsigned long *)
7449 				__get_free_page(GFP_KERNEL);
7450 			if (!vmx_bitmap[i]) {
7451 				nested_vmx_hardware_unsetup();
7452 				return -ENOMEM;
7453 			}
7454 		}
7455 
7456 		init_vmcs_shadow_fields();
7457 	}
7458 
7459 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
7460 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
7461 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
7462 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
7463 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
7464 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
7465 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
7466 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmxoff;
7467 	exit_handlers[EXIT_REASON_VMON]		= handle_vmxon;
7468 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
7469 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
7470 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
7471 
7472 	return 0;
7473 }
7474 
7475 static gpa_t vmx_translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa,
7476 				      u64 access,
7477 				      struct x86_exception *exception,
7478 				      u64 pte_access)
7479 {
7480 	struct kvm_mmu *mmu = vcpu->arch.mmu;
7481 
7482 	BUG_ON(!mmu_is_nested(vcpu));
7483 
7484 	/*
7485 	 * MBEC differentiates based on the effective U/S bit of
7486 	 * the guest page tables; not the processor CPL.
7487 	 */
7488 	access &= ~PFERR_USER_MASK;
7489 	if ((pte_access & ACC_USER_MASK) && (access & PFERR_GUEST_FINAL_MASK))
7490 		access |= PFERR_USER_MASK;
7491 
7492 	return mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
7493 }
7494 
7495 struct kvm_x86_nested_ops vmx_nested_ops = {
7496 	.leave_nested = vmx_leave_nested,
7497 	.translate_nested_gpa = vmx_translate_nested_gpa,
7498 	.is_exception_vmexit = nested_vmx_is_exception_vmexit,
7499 	.check_events = vmx_check_nested_events,
7500 	.has_events = vmx_has_nested_events,
7501 	.triple_fault = nested_vmx_triple_fault,
7502 	.get_state = vmx_get_nested_state,
7503 	.set_state = vmx_set_nested_state,
7504 	.get_nested_state_pages = vmx_get_nested_state_pages,
7505 	.write_log_dirty = nested_vmx_write_pml_buffer,
7506 #ifdef CONFIG_KVM_HYPERV
7507 	.enable_evmcs = nested_enable_evmcs,
7508 	.get_evmcs_version = nested_get_evmcs_version,
7509 	.hv_inject_synthetic_vmexit_post_tlb_flush = vmx_hv_inject_synthetic_vmexit_post_tlb_flush,
7510 #endif
7511 };
7512