xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 7f8256ae)
1 // SPDX-License-Identifier: GPL-2.0
2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3 
4 #include <linux/objtool.h>
5 #include <linux/percpu.h>
6 
7 #include <asm/debugreg.h>
8 #include <asm/mmu_context.h>
9 
10 #include "cpuid.h"
11 #include "hyperv.h"
12 #include "mmu.h"
13 #include "nested.h"
14 #include "pmu.h"
15 #include "sgx.h"
16 #include "trace.h"
17 #include "vmx.h"
18 #include "x86.h"
19 #include "smm.h"
20 
21 static bool __read_mostly enable_shadow_vmcs = 1;
22 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
23 
24 static bool __read_mostly nested_early_check = 0;
25 module_param(nested_early_check, bool, S_IRUGO);
26 
27 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
28 
29 /*
30  * Hyper-V requires all of these, so mark them as supported even though
31  * they are just treated the same as all-context.
32  */
33 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
34 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
35 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
36 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
37 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
38 
39 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
40 
41 enum {
42 	VMX_VMREAD_BITMAP,
43 	VMX_VMWRITE_BITMAP,
44 	VMX_BITMAP_NR
45 };
46 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
47 
48 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
49 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
50 
51 struct shadow_vmcs_field {
52 	u16	encoding;
53 	u16	offset;
54 };
55 static struct shadow_vmcs_field shadow_read_only_fields[] = {
56 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
57 #include "vmcs_shadow_fields.h"
58 };
59 static int max_shadow_read_only_fields =
60 	ARRAY_SIZE(shadow_read_only_fields);
61 
62 static struct shadow_vmcs_field shadow_read_write_fields[] = {
63 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
64 #include "vmcs_shadow_fields.h"
65 };
66 static int max_shadow_read_write_fields =
67 	ARRAY_SIZE(shadow_read_write_fields);
68 
69 static void init_vmcs_shadow_fields(void)
70 {
71 	int i, j;
72 
73 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
74 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
75 
76 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
77 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
78 		u16 field = entry.encoding;
79 
80 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
81 		    (i + 1 == max_shadow_read_only_fields ||
82 		     shadow_read_only_fields[i + 1].encoding != field + 1))
83 			pr_err("Missing field from shadow_read_only_field %x\n",
84 			       field + 1);
85 
86 		clear_bit(field, vmx_vmread_bitmap);
87 		if (field & 1)
88 #ifdef CONFIG_X86_64
89 			continue;
90 #else
91 			entry.offset += sizeof(u32);
92 #endif
93 		shadow_read_only_fields[j++] = entry;
94 	}
95 	max_shadow_read_only_fields = j;
96 
97 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
98 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
99 		u16 field = entry.encoding;
100 
101 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
102 		    (i + 1 == max_shadow_read_write_fields ||
103 		     shadow_read_write_fields[i + 1].encoding != field + 1))
104 			pr_err("Missing field from shadow_read_write_field %x\n",
105 			       field + 1);
106 
107 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
108 			  field <= GUEST_TR_AR_BYTES,
109 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
110 
111 		/*
112 		 * PML and the preemption timer can be emulated, but the
113 		 * processor cannot vmwrite to fields that don't exist
114 		 * on bare metal.
115 		 */
116 		switch (field) {
117 		case GUEST_PML_INDEX:
118 			if (!cpu_has_vmx_pml())
119 				continue;
120 			break;
121 		case VMX_PREEMPTION_TIMER_VALUE:
122 			if (!cpu_has_vmx_preemption_timer())
123 				continue;
124 			break;
125 		case GUEST_INTR_STATUS:
126 			if (!cpu_has_vmx_apicv())
127 				continue;
128 			break;
129 		default:
130 			break;
131 		}
132 
133 		clear_bit(field, vmx_vmwrite_bitmap);
134 		clear_bit(field, vmx_vmread_bitmap);
135 		if (field & 1)
136 #ifdef CONFIG_X86_64
137 			continue;
138 #else
139 			entry.offset += sizeof(u32);
140 #endif
141 		shadow_read_write_fields[j++] = entry;
142 	}
143 	max_shadow_read_write_fields = j;
144 }
145 
146 /*
147  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
148  * set the success or error code of an emulated VMX instruction (as specified
149  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
150  * instruction.
151  */
152 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
153 {
154 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
155 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
156 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
157 	return kvm_skip_emulated_instruction(vcpu);
158 }
159 
160 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
161 {
162 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
163 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
164 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
165 			| X86_EFLAGS_CF);
166 	return kvm_skip_emulated_instruction(vcpu);
167 }
168 
169 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
170 				u32 vm_instruction_error)
171 {
172 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
173 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
174 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
175 			| X86_EFLAGS_ZF);
176 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
177 	/*
178 	 * We don't need to force sync to shadow VMCS because
179 	 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
180 	 * fields and thus must be synced.
181 	 */
182 	if (to_vmx(vcpu)->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
183 		to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
184 
185 	return kvm_skip_emulated_instruction(vcpu);
186 }
187 
188 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
189 {
190 	struct vcpu_vmx *vmx = to_vmx(vcpu);
191 
192 	/*
193 	 * failValid writes the error number to the current VMCS, which
194 	 * can't be done if there isn't a current VMCS.
195 	 */
196 	if (vmx->nested.current_vmptr == INVALID_GPA &&
197 	    !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
198 		return nested_vmx_failInvalid(vcpu);
199 
200 	return nested_vmx_failValid(vcpu, vm_instruction_error);
201 }
202 
203 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
204 {
205 	/* TODO: not to reset guest simply here. */
206 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
207 	pr_debug_ratelimited("nested vmx abort, indicator %d\n", indicator);
208 }
209 
210 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
211 {
212 	return fixed_bits_valid(control, low, high);
213 }
214 
215 static inline u64 vmx_control_msr(u32 low, u32 high)
216 {
217 	return low | ((u64)high << 32);
218 }
219 
220 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
221 {
222 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
223 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
224 	vmx->nested.need_vmcs12_to_shadow_sync = false;
225 }
226 
227 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
228 {
229 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
230 	struct vcpu_vmx *vmx = to_vmx(vcpu);
231 
232 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
233 		kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
234 		vmx->nested.hv_evmcs = NULL;
235 	}
236 
237 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
238 
239 	if (hv_vcpu) {
240 		hv_vcpu->nested.pa_page_gpa = INVALID_GPA;
241 		hv_vcpu->nested.vm_id = 0;
242 		hv_vcpu->nested.vp_id = 0;
243 	}
244 }
245 
246 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
247 				     struct loaded_vmcs *prev)
248 {
249 	struct vmcs_host_state *dest, *src;
250 
251 	if (unlikely(!vmx->guest_state_loaded))
252 		return;
253 
254 	src = &prev->host_state;
255 	dest = &vmx->loaded_vmcs->host_state;
256 
257 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
258 	dest->ldt_sel = src->ldt_sel;
259 #ifdef CONFIG_X86_64
260 	dest->ds_sel = src->ds_sel;
261 	dest->es_sel = src->es_sel;
262 #endif
263 }
264 
265 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
266 {
267 	struct vcpu_vmx *vmx = to_vmx(vcpu);
268 	struct loaded_vmcs *prev;
269 	int cpu;
270 
271 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
272 		return;
273 
274 	cpu = get_cpu();
275 	prev = vmx->loaded_vmcs;
276 	vmx->loaded_vmcs = vmcs;
277 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
278 	vmx_sync_vmcs_host_state(vmx, prev);
279 	put_cpu();
280 
281 	vcpu->arch.regs_avail = ~VMX_REGS_LAZY_LOAD_SET;
282 
283 	/*
284 	 * All lazily updated registers will be reloaded from VMCS12 on both
285 	 * vmentry and vmexit.
286 	 */
287 	vcpu->arch.regs_dirty = 0;
288 }
289 
290 /*
291  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
292  * just stops using VMX.
293  */
294 static void free_nested(struct kvm_vcpu *vcpu)
295 {
296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
297 
298 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
299 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
300 
301 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
302 		return;
303 
304 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
305 
306 	vmx->nested.vmxon = false;
307 	vmx->nested.smm.vmxon = false;
308 	vmx->nested.vmxon_ptr = INVALID_GPA;
309 	free_vpid(vmx->nested.vpid02);
310 	vmx->nested.posted_intr_nv = -1;
311 	vmx->nested.current_vmptr = INVALID_GPA;
312 	if (enable_shadow_vmcs) {
313 		vmx_disable_shadow_vmcs(vmx);
314 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
315 		free_vmcs(vmx->vmcs01.shadow_vmcs);
316 		vmx->vmcs01.shadow_vmcs = NULL;
317 	}
318 	kfree(vmx->nested.cached_vmcs12);
319 	vmx->nested.cached_vmcs12 = NULL;
320 	kfree(vmx->nested.cached_shadow_vmcs12);
321 	vmx->nested.cached_shadow_vmcs12 = NULL;
322 	/*
323 	 * Unpin physical memory we referred to in the vmcs02.  The APIC access
324 	 * page's backing page (yeah, confusing) shouldn't actually be accessed,
325 	 * and if it is written, the contents are irrelevant.
326 	 */
327 	kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
328 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
329 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
330 	vmx->nested.pi_desc = NULL;
331 
332 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
333 
334 	nested_release_evmcs(vcpu);
335 
336 	free_loaded_vmcs(&vmx->nested.vmcs02);
337 }
338 
339 /*
340  * Ensure that the current vmcs of the logical processor is the
341  * vmcs01 of the vcpu before calling free_nested().
342  */
343 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
344 {
345 	vcpu_load(vcpu);
346 	vmx_leave_nested(vcpu);
347 	vcpu_put(vcpu);
348 }
349 
350 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
351 
352 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
353 {
354 	return VALID_PAGE(root_hpa) &&
355 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
356 }
357 
358 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
359 				       gpa_t addr)
360 {
361 	unsigned long roots = 0;
362 	uint i;
363 	struct kvm_mmu_root_info *cached_root;
364 
365 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
366 
367 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
368 		cached_root = &vcpu->arch.mmu->prev_roots[i];
369 
370 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
371 					    eptp))
372 			roots |= KVM_MMU_ROOT_PREVIOUS(i);
373 	}
374 	if (roots)
375 		kvm_mmu_invalidate_addr(vcpu, vcpu->arch.mmu, addr, roots);
376 }
377 
378 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
379 		struct x86_exception *fault)
380 {
381 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
382 	struct vcpu_vmx *vmx = to_vmx(vcpu);
383 	u32 vm_exit_reason;
384 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
385 
386 	if (vmx->nested.pml_full) {
387 		vm_exit_reason = EXIT_REASON_PML_FULL;
388 		vmx->nested.pml_full = false;
389 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
390 	} else {
391 		if (fault->error_code & PFERR_RSVD_MASK)
392 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
393 		else
394 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
395 
396 		/*
397 		 * Although the caller (kvm_inject_emulated_page_fault) would
398 		 * have already synced the faulting address in the shadow EPT
399 		 * tables for the current EPTP12, we also need to sync it for
400 		 * any other cached EPTP02s based on the same EP4TA, since the
401 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
402 		 */
403 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
404 					   fault->address);
405 	}
406 
407 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
408 	vmcs12->guest_physical_address = fault->address;
409 }
410 
411 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
412 {
413 	struct vcpu_vmx *vmx = to_vmx(vcpu);
414 	bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
415 	int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
416 
417 	kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
418 				nested_ept_ad_enabled(vcpu),
419 				nested_ept_get_eptp(vcpu));
420 }
421 
422 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
423 {
424 	WARN_ON(mmu_is_nested(vcpu));
425 
426 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
427 	nested_ept_new_eptp(vcpu);
428 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
429 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
430 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
431 
432 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
433 }
434 
435 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
436 {
437 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
438 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
439 }
440 
441 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
442 					    u16 error_code)
443 {
444 	bool inequality, bit;
445 
446 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
447 	inequality =
448 		(error_code & vmcs12->page_fault_error_code_mask) !=
449 		 vmcs12->page_fault_error_code_match;
450 	return inequality ^ bit;
451 }
452 
453 static bool nested_vmx_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
454 					   u32 error_code)
455 {
456 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
457 
458 	/*
459 	 * Drop bits 31:16 of the error code when performing the #PF mask+match
460 	 * check.  All VMCS fields involved are 32 bits, but Intel CPUs never
461 	 * set bits 31:16 and VMX disallows setting bits 31:16 in the injected
462 	 * error code.  Including the to-be-dropped bits in the check might
463 	 * result in an "impossible" or missed exit from L1's perspective.
464 	 */
465 	if (vector == PF_VECTOR)
466 		return nested_vmx_is_page_fault_vmexit(vmcs12, (u16)error_code);
467 
468 	return (vmcs12->exception_bitmap & (1u << vector));
469 }
470 
471 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
472 					       struct vmcs12 *vmcs12)
473 {
474 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
475 		return 0;
476 
477 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
478 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
479 		return -EINVAL;
480 
481 	return 0;
482 }
483 
484 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
485 						struct vmcs12 *vmcs12)
486 {
487 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
488 		return 0;
489 
490 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
491 		return -EINVAL;
492 
493 	return 0;
494 }
495 
496 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
497 						struct vmcs12 *vmcs12)
498 {
499 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
500 		return 0;
501 
502 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
503 		return -EINVAL;
504 
505 	return 0;
506 }
507 
508 /*
509  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
510  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
511  * only the "disable intercept" case needs to be handled.
512  */
513 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
514 							unsigned long *msr_bitmap_l0,
515 							u32 msr, int type)
516 {
517 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
518 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
519 
520 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
521 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
522 }
523 
524 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
525 {
526 	int msr;
527 
528 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
529 		unsigned word = msr / BITS_PER_LONG;
530 
531 		msr_bitmap[word] = ~0;
532 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
533 	}
534 }
535 
536 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
537 static inline									\
538 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
539 					 unsigned long *msr_bitmap_l1,		\
540 					 unsigned long *msr_bitmap_l0, u32 msr)	\
541 {										\
542 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
543 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
544 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
545 	else									\
546 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
547 }
548 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
549 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
550 
551 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
552 						    unsigned long *msr_bitmap_l1,
553 						    unsigned long *msr_bitmap_l0,
554 						    u32 msr, int types)
555 {
556 	if (types & MSR_TYPE_R)
557 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
558 						  msr_bitmap_l0, msr);
559 	if (types & MSR_TYPE_W)
560 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
561 						   msr_bitmap_l0, msr);
562 }
563 
564 /*
565  * Merge L0's and L1's MSR bitmap, return false to indicate that
566  * we do not use the hardware.
567  */
568 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
569 						 struct vmcs12 *vmcs12)
570 {
571 	struct vcpu_vmx *vmx = to_vmx(vcpu);
572 	int msr;
573 	unsigned long *msr_bitmap_l1;
574 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
575 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
576 	struct kvm_host_map *map = &vmx->nested.msr_bitmap_map;
577 
578 	/* Nothing to do if the MSR bitmap is not in use.  */
579 	if (!cpu_has_vmx_msr_bitmap() ||
580 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
581 		return false;
582 
583 	/*
584 	 * MSR bitmap update can be skipped when:
585 	 * - MSR bitmap for L1 hasn't changed.
586 	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
587 	 *   before.
588 	 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
589 	 *   and tells KVM (L0) there were no changes in MSR bitmap for L2.
590 	 */
591 	if (!vmx->nested.force_msr_bitmap_recalc && evmcs &&
592 	    evmcs->hv_enlightenments_control.msr_bitmap &&
593 	    evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
594 		return true;
595 
596 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
597 		return false;
598 
599 	msr_bitmap_l1 = (unsigned long *)map->hva;
600 
601 	/*
602 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
603 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
604 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
605 	 */
606 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
607 
608 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
609 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
610 			/*
611 			 * L0 need not intercept reads for MSRs between 0x800
612 			 * and 0x8ff, it just lets the processor take the value
613 			 * from the virtual-APIC page; take those 256 bits
614 			 * directly from the L1 bitmap.
615 			 */
616 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
617 				unsigned word = msr / BITS_PER_LONG;
618 
619 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
620 			}
621 		}
622 
623 		nested_vmx_disable_intercept_for_x2apic_msr(
624 			msr_bitmap_l1, msr_bitmap_l0,
625 			X2APIC_MSR(APIC_TASKPRI),
626 			MSR_TYPE_R | MSR_TYPE_W);
627 
628 		if (nested_cpu_has_vid(vmcs12)) {
629 			nested_vmx_disable_intercept_for_x2apic_msr(
630 				msr_bitmap_l1, msr_bitmap_l0,
631 				X2APIC_MSR(APIC_EOI),
632 				MSR_TYPE_W);
633 			nested_vmx_disable_intercept_for_x2apic_msr(
634 				msr_bitmap_l1, msr_bitmap_l0,
635 				X2APIC_MSR(APIC_SELF_IPI),
636 				MSR_TYPE_W);
637 		}
638 	}
639 
640 	/*
641 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
642 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
643 	 */
644 #ifdef CONFIG_X86_64
645 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
646 					 MSR_FS_BASE, MSR_TYPE_RW);
647 
648 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
649 					 MSR_GS_BASE, MSR_TYPE_RW);
650 
651 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
652 					 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
653 #endif
654 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
655 					 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
656 
657 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
658 					 MSR_IA32_PRED_CMD, MSR_TYPE_W);
659 
660 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
661 					 MSR_IA32_FLUSH_CMD, MSR_TYPE_W);
662 
663 	kvm_vcpu_unmap(vcpu, &vmx->nested.msr_bitmap_map, false);
664 
665 	vmx->nested.force_msr_bitmap_recalc = false;
666 
667 	return true;
668 }
669 
670 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
671 				       struct vmcs12 *vmcs12)
672 {
673 	struct vcpu_vmx *vmx = to_vmx(vcpu);
674 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
675 
676 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
677 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
678 		return;
679 
680 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
681 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
682 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
683 		return;
684 
685 	kvm_read_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
686 			      VMCS12_SIZE);
687 }
688 
689 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
690 					      struct vmcs12 *vmcs12)
691 {
692 	struct vcpu_vmx *vmx = to_vmx(vcpu);
693 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
694 
695 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
696 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
697 		return;
698 
699 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
700 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
701 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
702 		return;
703 
704 	kvm_write_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
705 			       VMCS12_SIZE);
706 }
707 
708 /*
709  * In nested virtualization, check if L1 has set
710  * VM_EXIT_ACK_INTR_ON_EXIT
711  */
712 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
713 {
714 	return get_vmcs12(vcpu)->vm_exit_controls &
715 		VM_EXIT_ACK_INTR_ON_EXIT;
716 }
717 
718 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
719 					  struct vmcs12 *vmcs12)
720 {
721 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
722 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
723 		return -EINVAL;
724 	else
725 		return 0;
726 }
727 
728 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
729 					   struct vmcs12 *vmcs12)
730 {
731 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
732 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
733 	    !nested_cpu_has_vid(vmcs12) &&
734 	    !nested_cpu_has_posted_intr(vmcs12))
735 		return 0;
736 
737 	/*
738 	 * If virtualize x2apic mode is enabled,
739 	 * virtualize apic access must be disabled.
740 	 */
741 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
742 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
743 		return -EINVAL;
744 
745 	/*
746 	 * If virtual interrupt delivery is enabled,
747 	 * we must exit on external interrupts.
748 	 */
749 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
750 		return -EINVAL;
751 
752 	/*
753 	 * bits 15:8 should be zero in posted_intr_nv,
754 	 * the descriptor address has been already checked
755 	 * in nested_get_vmcs12_pages.
756 	 *
757 	 * bits 5:0 of posted_intr_desc_addr should be zero.
758 	 */
759 	if (nested_cpu_has_posted_intr(vmcs12) &&
760 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
761 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
762 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
763 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
764 		return -EINVAL;
765 
766 	/* tpr shadow is needed by all apicv features. */
767 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
768 		return -EINVAL;
769 
770 	return 0;
771 }
772 
773 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
774 				       u32 count, u64 addr)
775 {
776 	if (count == 0)
777 		return 0;
778 
779 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
780 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
781 		return -EINVAL;
782 
783 	return 0;
784 }
785 
786 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
787 						     struct vmcs12 *vmcs12)
788 {
789 	if (CC(nested_vmx_check_msr_switch(vcpu,
790 					   vmcs12->vm_exit_msr_load_count,
791 					   vmcs12->vm_exit_msr_load_addr)) ||
792 	    CC(nested_vmx_check_msr_switch(vcpu,
793 					   vmcs12->vm_exit_msr_store_count,
794 					   vmcs12->vm_exit_msr_store_addr)))
795 		return -EINVAL;
796 
797 	return 0;
798 }
799 
800 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
801                                                       struct vmcs12 *vmcs12)
802 {
803 	if (CC(nested_vmx_check_msr_switch(vcpu,
804 					   vmcs12->vm_entry_msr_load_count,
805 					   vmcs12->vm_entry_msr_load_addr)))
806                 return -EINVAL;
807 
808 	return 0;
809 }
810 
811 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
812 					 struct vmcs12 *vmcs12)
813 {
814 	if (!nested_cpu_has_pml(vmcs12))
815 		return 0;
816 
817 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
818 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
819 		return -EINVAL;
820 
821 	return 0;
822 }
823 
824 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
825 							struct vmcs12 *vmcs12)
826 {
827 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
828 	       !nested_cpu_has_ept(vmcs12)))
829 		return -EINVAL;
830 	return 0;
831 }
832 
833 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
834 							 struct vmcs12 *vmcs12)
835 {
836 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
837 	       !nested_cpu_has_ept(vmcs12)))
838 		return -EINVAL;
839 	return 0;
840 }
841 
842 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
843 						 struct vmcs12 *vmcs12)
844 {
845 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
846 		return 0;
847 
848 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
849 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
850 		return -EINVAL;
851 
852 	return 0;
853 }
854 
855 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
856 				       struct vmx_msr_entry *e)
857 {
858 	/* x2APIC MSR accesses are not allowed */
859 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
860 		return -EINVAL;
861 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
862 	    CC(e->index == MSR_IA32_UCODE_REV))
863 		return -EINVAL;
864 	if (CC(e->reserved != 0))
865 		return -EINVAL;
866 	return 0;
867 }
868 
869 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
870 				     struct vmx_msr_entry *e)
871 {
872 	if (CC(e->index == MSR_FS_BASE) ||
873 	    CC(e->index == MSR_GS_BASE) ||
874 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
875 	    nested_vmx_msr_check_common(vcpu, e))
876 		return -EINVAL;
877 	return 0;
878 }
879 
880 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
881 				      struct vmx_msr_entry *e)
882 {
883 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
884 	    nested_vmx_msr_check_common(vcpu, e))
885 		return -EINVAL;
886 	return 0;
887 }
888 
889 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
890 {
891 	struct vcpu_vmx *vmx = to_vmx(vcpu);
892 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
893 				       vmx->nested.msrs.misc_high);
894 
895 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
896 }
897 
898 /*
899  * Load guest's/host's msr at nested entry/exit.
900  * return 0 for success, entry index for failure.
901  *
902  * One of the failure modes for MSR load/store is when a list exceeds the
903  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
904  * as possible, process all valid entries before failing rather than precheck
905  * for a capacity violation.
906  */
907 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
908 {
909 	u32 i;
910 	struct vmx_msr_entry e;
911 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
912 
913 	for (i = 0; i < count; i++) {
914 		if (unlikely(i >= max_msr_list_size))
915 			goto fail;
916 
917 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
918 					&e, sizeof(e))) {
919 			pr_debug_ratelimited(
920 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
921 				__func__, i, gpa + i * sizeof(e));
922 			goto fail;
923 		}
924 		if (nested_vmx_load_msr_check(vcpu, &e)) {
925 			pr_debug_ratelimited(
926 				"%s check failed (%u, 0x%x, 0x%x)\n",
927 				__func__, i, e.index, e.reserved);
928 			goto fail;
929 		}
930 		if (kvm_set_msr(vcpu, e.index, e.value)) {
931 			pr_debug_ratelimited(
932 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
933 				__func__, i, e.index, e.value);
934 			goto fail;
935 		}
936 	}
937 	return 0;
938 fail:
939 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
940 	return i + 1;
941 }
942 
943 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
944 					    u32 msr_index,
945 					    u64 *data)
946 {
947 	struct vcpu_vmx *vmx = to_vmx(vcpu);
948 
949 	/*
950 	 * If the L0 hypervisor stored a more accurate value for the TSC that
951 	 * does not include the time taken for emulation of the L2->L1
952 	 * VM-exit in L0, use the more accurate value.
953 	 */
954 	if (msr_index == MSR_IA32_TSC) {
955 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
956 						    MSR_IA32_TSC);
957 
958 		if (i >= 0) {
959 			u64 val = vmx->msr_autostore.guest.val[i].value;
960 
961 			*data = kvm_read_l1_tsc(vcpu, val);
962 			return true;
963 		}
964 	}
965 
966 	if (kvm_get_msr(vcpu, msr_index, data)) {
967 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
968 			msr_index);
969 		return false;
970 	}
971 	return true;
972 }
973 
974 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
975 				     struct vmx_msr_entry *e)
976 {
977 	if (kvm_vcpu_read_guest(vcpu,
978 				gpa + i * sizeof(*e),
979 				e, 2 * sizeof(u32))) {
980 		pr_debug_ratelimited(
981 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
982 			__func__, i, gpa + i * sizeof(*e));
983 		return false;
984 	}
985 	if (nested_vmx_store_msr_check(vcpu, e)) {
986 		pr_debug_ratelimited(
987 			"%s check failed (%u, 0x%x, 0x%x)\n",
988 			__func__, i, e->index, e->reserved);
989 		return false;
990 	}
991 	return true;
992 }
993 
994 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
995 {
996 	u64 data;
997 	u32 i;
998 	struct vmx_msr_entry e;
999 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1000 
1001 	for (i = 0; i < count; i++) {
1002 		if (unlikely(i >= max_msr_list_size))
1003 			return -EINVAL;
1004 
1005 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1006 			return -EINVAL;
1007 
1008 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1009 			return -EINVAL;
1010 
1011 		if (kvm_vcpu_write_guest(vcpu,
1012 					 gpa + i * sizeof(e) +
1013 					     offsetof(struct vmx_msr_entry, value),
1014 					 &data, sizeof(data))) {
1015 			pr_debug_ratelimited(
1016 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1017 				__func__, i, e.index, data);
1018 			return -EINVAL;
1019 		}
1020 	}
1021 	return 0;
1022 }
1023 
1024 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1025 {
1026 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1027 	u32 count = vmcs12->vm_exit_msr_store_count;
1028 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1029 	struct vmx_msr_entry e;
1030 	u32 i;
1031 
1032 	for (i = 0; i < count; i++) {
1033 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1034 			return false;
1035 
1036 		if (e.index == msr_index)
1037 			return true;
1038 	}
1039 	return false;
1040 }
1041 
1042 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1043 					   u32 msr_index)
1044 {
1045 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1046 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1047 	bool in_vmcs12_store_list;
1048 	int msr_autostore_slot;
1049 	bool in_autostore_list;
1050 	int last;
1051 
1052 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1053 	in_autostore_list = msr_autostore_slot >= 0;
1054 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1055 
1056 	if (in_vmcs12_store_list && !in_autostore_list) {
1057 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1058 			/*
1059 			 * Emulated VMEntry does not fail here.  Instead a less
1060 			 * accurate value will be returned by
1061 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1062 			 * instead of reading the value from the vmcs02 VMExit
1063 			 * MSR-store area.
1064 			 */
1065 			pr_warn_ratelimited(
1066 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1067 				msr_index);
1068 			return;
1069 		}
1070 		last = autostore->nr++;
1071 		autostore->val[last].index = msr_index;
1072 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1073 		last = --autostore->nr;
1074 		autostore->val[msr_autostore_slot] = autostore->val[last];
1075 	}
1076 }
1077 
1078 /*
1079  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1080  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1081  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1082  * @entry_failure_code.
1083  */
1084 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1085 			       bool nested_ept, bool reload_pdptrs,
1086 			       enum vm_entry_failure_code *entry_failure_code)
1087 {
1088 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1089 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1090 		return -EINVAL;
1091 	}
1092 
1093 	/*
1094 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1095 	 * must not be dereferenced.
1096 	 */
1097 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1098 	    CC(!load_pdptrs(vcpu, cr3))) {
1099 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1100 		return -EINVAL;
1101 	}
1102 
1103 	vcpu->arch.cr3 = cr3;
1104 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1105 
1106 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1107 	kvm_init_mmu(vcpu);
1108 
1109 	if (!nested_ept)
1110 		kvm_mmu_new_pgd(vcpu, cr3);
1111 
1112 	return 0;
1113 }
1114 
1115 /*
1116  * Returns if KVM is able to config CPU to tag TLB entries
1117  * populated by L2 differently than TLB entries populated
1118  * by L1.
1119  *
1120  * If L0 uses EPT, L1 and L2 run with different EPTP because
1121  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1122  * are tagged with different EPTP.
1123  *
1124  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1125  * with different VPID (L1 entries are tagged with vmx->vpid
1126  * while L2 entries are tagged with vmx->nested.vpid02).
1127  */
1128 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1129 {
1130 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1131 
1132 	return enable_ept ||
1133 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1134 }
1135 
1136 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1137 					    struct vmcs12 *vmcs12,
1138 					    bool is_vmenter)
1139 {
1140 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1141 
1142 	/*
1143 	 * KVM_REQ_HV_TLB_FLUSH flushes entries from either L1's VP_ID or
1144 	 * L2's VP_ID upon request from the guest. Make sure we check for
1145 	 * pending entries in the right FIFO upon L1/L2 transition as these
1146 	 * requests are put by other vCPUs asynchronously.
1147 	 */
1148 	if (to_hv_vcpu(vcpu) && enable_ept)
1149 		kvm_make_request(KVM_REQ_HV_TLB_FLUSH, vcpu);
1150 
1151 	/*
1152 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1153 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1154 	 * full TLB flush from the guest's perspective.  This is required even
1155 	 * if VPID is disabled in the host as KVM may need to synchronize the
1156 	 * MMU in response to the guest TLB flush.
1157 	 *
1158 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1159 	 * EPT is a special snowflake, as guest-physical mappings aren't
1160 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1161 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1162 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1163 	 * those mappings.
1164 	 */
1165 	if (!nested_cpu_has_vpid(vmcs12)) {
1166 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1167 		return;
1168 	}
1169 
1170 	/* L2 should never have a VPID if VPID is disabled. */
1171 	WARN_ON(!enable_vpid);
1172 
1173 	/*
1174 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1175 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1176 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1177 	 * that the new vpid12 has never been used and thus represents a new
1178 	 * guest ASID that cannot have entries in the TLB.
1179 	 */
1180 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1181 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1182 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1183 		return;
1184 	}
1185 
1186 	/*
1187 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1188 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1189 	 * KVM was unable to allocate a VPID for L2, flush the current context
1190 	 * as the effective ASID is common to both L1 and L2.
1191 	 */
1192 	if (!nested_has_guest_tlb_tag(vcpu))
1193 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1194 }
1195 
1196 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1197 {
1198 	superset &= mask;
1199 	subset &= mask;
1200 
1201 	return (superset | subset) == superset;
1202 }
1203 
1204 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1205 {
1206 	const u64 feature_and_reserved =
1207 		/* feature (except bit 48; see below) */
1208 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1209 		/* reserved */
1210 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1211 	u64 vmx_basic = vmcs_config.nested.basic;
1212 
1213 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1214 		return -EINVAL;
1215 
1216 	/*
1217 	 * KVM does not emulate a version of VMX that constrains physical
1218 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1219 	 */
1220 	if (data & BIT_ULL(48))
1221 		return -EINVAL;
1222 
1223 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1224 	    vmx_basic_vmcs_revision_id(data))
1225 		return -EINVAL;
1226 
1227 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1228 		return -EINVAL;
1229 
1230 	vmx->nested.msrs.basic = data;
1231 	return 0;
1232 }
1233 
1234 static void vmx_get_control_msr(struct nested_vmx_msrs *msrs, u32 msr_index,
1235 				u32 **low, u32 **high)
1236 {
1237 	switch (msr_index) {
1238 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1239 		*low = &msrs->pinbased_ctls_low;
1240 		*high = &msrs->pinbased_ctls_high;
1241 		break;
1242 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1243 		*low = &msrs->procbased_ctls_low;
1244 		*high = &msrs->procbased_ctls_high;
1245 		break;
1246 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1247 		*low = &msrs->exit_ctls_low;
1248 		*high = &msrs->exit_ctls_high;
1249 		break;
1250 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1251 		*low = &msrs->entry_ctls_low;
1252 		*high = &msrs->entry_ctls_high;
1253 		break;
1254 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1255 		*low = &msrs->secondary_ctls_low;
1256 		*high = &msrs->secondary_ctls_high;
1257 		break;
1258 	default:
1259 		BUG();
1260 	}
1261 }
1262 
1263 static int
1264 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1265 {
1266 	u32 *lowp, *highp;
1267 	u64 supported;
1268 
1269 	vmx_get_control_msr(&vmcs_config.nested, msr_index, &lowp, &highp);
1270 
1271 	supported = vmx_control_msr(*lowp, *highp);
1272 
1273 	/* Check must-be-1 bits are still 1. */
1274 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1275 		return -EINVAL;
1276 
1277 	/* Check must-be-0 bits are still 0. */
1278 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1279 		return -EINVAL;
1280 
1281 	vmx_get_control_msr(&vmx->nested.msrs, msr_index, &lowp, &highp);
1282 	*lowp = data;
1283 	*highp = data >> 32;
1284 	return 0;
1285 }
1286 
1287 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1288 {
1289 	const u64 feature_and_reserved_bits =
1290 		/* feature */
1291 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1292 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1293 		/* reserved */
1294 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1295 	u64 vmx_misc = vmx_control_msr(vmcs_config.nested.misc_low,
1296 				       vmcs_config.nested.misc_high);
1297 
1298 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1299 		return -EINVAL;
1300 
1301 	if ((vmx->nested.msrs.pinbased_ctls_high &
1302 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1303 	    vmx_misc_preemption_timer_rate(data) !=
1304 	    vmx_misc_preemption_timer_rate(vmx_misc))
1305 		return -EINVAL;
1306 
1307 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1308 		return -EINVAL;
1309 
1310 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1311 		return -EINVAL;
1312 
1313 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1314 		return -EINVAL;
1315 
1316 	vmx->nested.msrs.misc_low = data;
1317 	vmx->nested.msrs.misc_high = data >> 32;
1318 
1319 	return 0;
1320 }
1321 
1322 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1323 {
1324 	u64 vmx_ept_vpid_cap = vmx_control_msr(vmcs_config.nested.ept_caps,
1325 					       vmcs_config.nested.vpid_caps);
1326 
1327 	/* Every bit is either reserved or a feature bit. */
1328 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1329 		return -EINVAL;
1330 
1331 	vmx->nested.msrs.ept_caps = data;
1332 	vmx->nested.msrs.vpid_caps = data >> 32;
1333 	return 0;
1334 }
1335 
1336 static u64 *vmx_get_fixed0_msr(struct nested_vmx_msrs *msrs, u32 msr_index)
1337 {
1338 	switch (msr_index) {
1339 	case MSR_IA32_VMX_CR0_FIXED0:
1340 		return &msrs->cr0_fixed0;
1341 	case MSR_IA32_VMX_CR4_FIXED0:
1342 		return &msrs->cr4_fixed0;
1343 	default:
1344 		BUG();
1345 	}
1346 }
1347 
1348 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1349 {
1350 	const u64 *msr = vmx_get_fixed0_msr(&vmcs_config.nested, msr_index);
1351 
1352 	/*
1353 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1354 	 * must be 1 in the restored value.
1355 	 */
1356 	if (!is_bitwise_subset(data, *msr, -1ULL))
1357 		return -EINVAL;
1358 
1359 	*vmx_get_fixed0_msr(&vmx->nested.msrs, msr_index) = data;
1360 	return 0;
1361 }
1362 
1363 /*
1364  * Called when userspace is restoring VMX MSRs.
1365  *
1366  * Returns 0 on success, non-0 otherwise.
1367  */
1368 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1369 {
1370 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1371 
1372 	/*
1373 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1374 	 * is in VMX operation.
1375 	 */
1376 	if (vmx->nested.vmxon)
1377 		return -EBUSY;
1378 
1379 	switch (msr_index) {
1380 	case MSR_IA32_VMX_BASIC:
1381 		return vmx_restore_vmx_basic(vmx, data);
1382 	case MSR_IA32_VMX_PINBASED_CTLS:
1383 	case MSR_IA32_VMX_PROCBASED_CTLS:
1384 	case MSR_IA32_VMX_EXIT_CTLS:
1385 	case MSR_IA32_VMX_ENTRY_CTLS:
1386 		/*
1387 		 * The "non-true" VMX capability MSRs are generated from the
1388 		 * "true" MSRs, so we do not support restoring them directly.
1389 		 *
1390 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1391 		 * should restore the "true" MSRs with the must-be-1 bits
1392 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1393 		 * DEFAULT SETTINGS".
1394 		 */
1395 		return -EINVAL;
1396 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1397 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1398 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1399 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1400 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1401 		return vmx_restore_control_msr(vmx, msr_index, data);
1402 	case MSR_IA32_VMX_MISC:
1403 		return vmx_restore_vmx_misc(vmx, data);
1404 	case MSR_IA32_VMX_CR0_FIXED0:
1405 	case MSR_IA32_VMX_CR4_FIXED0:
1406 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1407 	case MSR_IA32_VMX_CR0_FIXED1:
1408 	case MSR_IA32_VMX_CR4_FIXED1:
1409 		/*
1410 		 * These MSRs are generated based on the vCPU's CPUID, so we
1411 		 * do not support restoring them directly.
1412 		 */
1413 		return -EINVAL;
1414 	case MSR_IA32_VMX_EPT_VPID_CAP:
1415 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1416 	case MSR_IA32_VMX_VMCS_ENUM:
1417 		vmx->nested.msrs.vmcs_enum = data;
1418 		return 0;
1419 	case MSR_IA32_VMX_VMFUNC:
1420 		if (data & ~vmcs_config.nested.vmfunc_controls)
1421 			return -EINVAL;
1422 		vmx->nested.msrs.vmfunc_controls = data;
1423 		return 0;
1424 	default:
1425 		/*
1426 		 * The rest of the VMX capability MSRs do not support restore.
1427 		 */
1428 		return -EINVAL;
1429 	}
1430 }
1431 
1432 /* Returns 0 on success, non-0 otherwise. */
1433 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1434 {
1435 	switch (msr_index) {
1436 	case MSR_IA32_VMX_BASIC:
1437 		*pdata = msrs->basic;
1438 		break;
1439 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1440 	case MSR_IA32_VMX_PINBASED_CTLS:
1441 		*pdata = vmx_control_msr(
1442 			msrs->pinbased_ctls_low,
1443 			msrs->pinbased_ctls_high);
1444 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1445 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1446 		break;
1447 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1448 	case MSR_IA32_VMX_PROCBASED_CTLS:
1449 		*pdata = vmx_control_msr(
1450 			msrs->procbased_ctls_low,
1451 			msrs->procbased_ctls_high);
1452 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1453 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1454 		break;
1455 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1456 	case MSR_IA32_VMX_EXIT_CTLS:
1457 		*pdata = vmx_control_msr(
1458 			msrs->exit_ctls_low,
1459 			msrs->exit_ctls_high);
1460 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1461 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1462 		break;
1463 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1464 	case MSR_IA32_VMX_ENTRY_CTLS:
1465 		*pdata = vmx_control_msr(
1466 			msrs->entry_ctls_low,
1467 			msrs->entry_ctls_high);
1468 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1469 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1470 		break;
1471 	case MSR_IA32_VMX_MISC:
1472 		*pdata = vmx_control_msr(
1473 			msrs->misc_low,
1474 			msrs->misc_high);
1475 		break;
1476 	case MSR_IA32_VMX_CR0_FIXED0:
1477 		*pdata = msrs->cr0_fixed0;
1478 		break;
1479 	case MSR_IA32_VMX_CR0_FIXED1:
1480 		*pdata = msrs->cr0_fixed1;
1481 		break;
1482 	case MSR_IA32_VMX_CR4_FIXED0:
1483 		*pdata = msrs->cr4_fixed0;
1484 		break;
1485 	case MSR_IA32_VMX_CR4_FIXED1:
1486 		*pdata = msrs->cr4_fixed1;
1487 		break;
1488 	case MSR_IA32_VMX_VMCS_ENUM:
1489 		*pdata = msrs->vmcs_enum;
1490 		break;
1491 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1492 		*pdata = vmx_control_msr(
1493 			msrs->secondary_ctls_low,
1494 			msrs->secondary_ctls_high);
1495 		break;
1496 	case MSR_IA32_VMX_EPT_VPID_CAP:
1497 		*pdata = msrs->ept_caps |
1498 			((u64)msrs->vpid_caps << 32);
1499 		break;
1500 	case MSR_IA32_VMX_VMFUNC:
1501 		*pdata = msrs->vmfunc_controls;
1502 		break;
1503 	default:
1504 		return 1;
1505 	}
1506 
1507 	return 0;
1508 }
1509 
1510 /*
1511  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1512  * been modified by the L1 guest.  Note, "writable" in this context means
1513  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1514  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1515  * VM-exit information fields (which are actually writable if the vCPU is
1516  * configured to support "VMWRITE to any supported field in the VMCS").
1517  */
1518 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1519 {
1520 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1521 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1522 	struct shadow_vmcs_field field;
1523 	unsigned long val;
1524 	int i;
1525 
1526 	if (WARN_ON(!shadow_vmcs))
1527 		return;
1528 
1529 	preempt_disable();
1530 
1531 	vmcs_load(shadow_vmcs);
1532 
1533 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1534 		field = shadow_read_write_fields[i];
1535 		val = __vmcs_readl(field.encoding);
1536 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1537 	}
1538 
1539 	vmcs_clear(shadow_vmcs);
1540 	vmcs_load(vmx->loaded_vmcs->vmcs);
1541 
1542 	preempt_enable();
1543 }
1544 
1545 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1546 {
1547 	const struct shadow_vmcs_field *fields[] = {
1548 		shadow_read_write_fields,
1549 		shadow_read_only_fields
1550 	};
1551 	const int max_fields[] = {
1552 		max_shadow_read_write_fields,
1553 		max_shadow_read_only_fields
1554 	};
1555 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1556 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1557 	struct shadow_vmcs_field field;
1558 	unsigned long val;
1559 	int i, q;
1560 
1561 	if (WARN_ON(!shadow_vmcs))
1562 		return;
1563 
1564 	vmcs_load(shadow_vmcs);
1565 
1566 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1567 		for (i = 0; i < max_fields[q]; i++) {
1568 			field = fields[q][i];
1569 			val = vmcs12_read_any(vmcs12, field.encoding,
1570 					      field.offset);
1571 			__vmcs_writel(field.encoding, val);
1572 		}
1573 	}
1574 
1575 	vmcs_clear(shadow_vmcs);
1576 	vmcs_load(vmx->loaded_vmcs->vmcs);
1577 }
1578 
1579 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1580 {
1581 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1582 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1583 	struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(&vmx->vcpu);
1584 
1585 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1586 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1587 	vmcs12->guest_rip = evmcs->guest_rip;
1588 
1589 	if (unlikely(!(hv_clean_fields &
1590 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_ENLIGHTENMENTSCONTROL))) {
1591 		hv_vcpu->nested.pa_page_gpa = evmcs->partition_assist_page;
1592 		hv_vcpu->nested.vm_id = evmcs->hv_vm_id;
1593 		hv_vcpu->nested.vp_id = evmcs->hv_vp_id;
1594 	}
1595 
1596 	if (unlikely(!(hv_clean_fields &
1597 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1598 		vmcs12->guest_rsp = evmcs->guest_rsp;
1599 		vmcs12->guest_rflags = evmcs->guest_rflags;
1600 		vmcs12->guest_interruptibility_info =
1601 			evmcs->guest_interruptibility_info;
1602 		/*
1603 		 * Not present in struct vmcs12:
1604 		 * vmcs12->guest_ssp = evmcs->guest_ssp;
1605 		 */
1606 	}
1607 
1608 	if (unlikely(!(hv_clean_fields &
1609 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1610 		vmcs12->cpu_based_vm_exec_control =
1611 			evmcs->cpu_based_vm_exec_control;
1612 	}
1613 
1614 	if (unlikely(!(hv_clean_fields &
1615 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1616 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1617 	}
1618 
1619 	if (unlikely(!(hv_clean_fields &
1620 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1621 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1622 	}
1623 
1624 	if (unlikely(!(hv_clean_fields &
1625 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1626 		vmcs12->vm_entry_intr_info_field =
1627 			evmcs->vm_entry_intr_info_field;
1628 		vmcs12->vm_entry_exception_error_code =
1629 			evmcs->vm_entry_exception_error_code;
1630 		vmcs12->vm_entry_instruction_len =
1631 			evmcs->vm_entry_instruction_len;
1632 	}
1633 
1634 	if (unlikely(!(hv_clean_fields &
1635 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1636 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1637 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1638 		vmcs12->host_cr0 = evmcs->host_cr0;
1639 		vmcs12->host_cr3 = evmcs->host_cr3;
1640 		vmcs12->host_cr4 = evmcs->host_cr4;
1641 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1642 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1643 		vmcs12->host_rip = evmcs->host_rip;
1644 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1645 		vmcs12->host_es_selector = evmcs->host_es_selector;
1646 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1647 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1648 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1649 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1650 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1651 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1652 		vmcs12->host_ia32_perf_global_ctrl = evmcs->host_ia32_perf_global_ctrl;
1653 		/*
1654 		 * Not present in struct vmcs12:
1655 		 * vmcs12->host_ia32_s_cet = evmcs->host_ia32_s_cet;
1656 		 * vmcs12->host_ssp = evmcs->host_ssp;
1657 		 * vmcs12->host_ia32_int_ssp_table_addr = evmcs->host_ia32_int_ssp_table_addr;
1658 		 */
1659 	}
1660 
1661 	if (unlikely(!(hv_clean_fields &
1662 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1663 		vmcs12->pin_based_vm_exec_control =
1664 			evmcs->pin_based_vm_exec_control;
1665 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1666 		vmcs12->secondary_vm_exec_control =
1667 			evmcs->secondary_vm_exec_control;
1668 	}
1669 
1670 	if (unlikely(!(hv_clean_fields &
1671 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1672 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1673 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1674 	}
1675 
1676 	if (unlikely(!(hv_clean_fields &
1677 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1678 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1679 	}
1680 
1681 	if (unlikely(!(hv_clean_fields &
1682 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1683 		vmcs12->guest_es_base = evmcs->guest_es_base;
1684 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1685 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1686 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1687 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1688 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1689 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1690 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1691 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1692 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1693 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1694 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1695 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1696 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1697 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1698 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1699 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1700 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1701 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1702 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1703 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1704 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1705 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1706 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1707 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1708 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1709 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1710 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1711 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1712 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1713 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1714 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1715 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1716 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1717 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1718 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1719 	}
1720 
1721 	if (unlikely(!(hv_clean_fields &
1722 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1723 		vmcs12->tsc_offset = evmcs->tsc_offset;
1724 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1725 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1726 		vmcs12->encls_exiting_bitmap = evmcs->encls_exiting_bitmap;
1727 		vmcs12->tsc_multiplier = evmcs->tsc_multiplier;
1728 	}
1729 
1730 	if (unlikely(!(hv_clean_fields &
1731 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1732 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1733 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1734 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1735 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1736 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1737 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1738 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1739 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1740 	}
1741 
1742 	if (unlikely(!(hv_clean_fields &
1743 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1744 		vmcs12->host_fs_base = evmcs->host_fs_base;
1745 		vmcs12->host_gs_base = evmcs->host_gs_base;
1746 		vmcs12->host_tr_base = evmcs->host_tr_base;
1747 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1748 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1749 		vmcs12->host_rsp = evmcs->host_rsp;
1750 	}
1751 
1752 	if (unlikely(!(hv_clean_fields &
1753 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1754 		vmcs12->ept_pointer = evmcs->ept_pointer;
1755 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1756 	}
1757 
1758 	if (unlikely(!(hv_clean_fields &
1759 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1760 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1761 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1762 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1763 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1764 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1765 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1766 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1767 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1768 		vmcs12->guest_pending_dbg_exceptions =
1769 			evmcs->guest_pending_dbg_exceptions;
1770 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1771 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1772 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1773 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1774 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1775 		vmcs12->guest_ia32_perf_global_ctrl = evmcs->guest_ia32_perf_global_ctrl;
1776 		/*
1777 		 * Not present in struct vmcs12:
1778 		 * vmcs12->guest_ia32_s_cet = evmcs->guest_ia32_s_cet;
1779 		 * vmcs12->guest_ia32_lbr_ctl = evmcs->guest_ia32_lbr_ctl;
1780 		 * vmcs12->guest_ia32_int_ssp_table_addr = evmcs->guest_ia32_int_ssp_table_addr;
1781 		 */
1782 	}
1783 
1784 	/*
1785 	 * Not used?
1786 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1787 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1788 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1789 	 * vmcs12->page_fault_error_code_mask =
1790 	 *		evmcs->page_fault_error_code_mask;
1791 	 * vmcs12->page_fault_error_code_match =
1792 	 *		evmcs->page_fault_error_code_match;
1793 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1794 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1795 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1796 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1797 	 */
1798 
1799 	/*
1800 	 * Read only fields:
1801 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1802 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1803 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1804 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1805 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1806 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1807 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1808 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1809 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1810 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1811 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1812 	 *
1813 	 * Not present in struct vmcs12:
1814 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1815 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1816 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1817 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1818 	 */
1819 
1820 	return;
1821 }
1822 
1823 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1824 {
1825 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1826 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1827 
1828 	/*
1829 	 * Should not be changed by KVM:
1830 	 *
1831 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1832 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1833 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1834 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1835 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1836 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1837 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1838 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1839 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1840 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1841 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1842 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1843 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1844 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1845 	 * evmcs->host_rip = vmcs12->host_rip;
1846 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1847 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1848 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1849 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1850 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1851 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1852 	 * evmcs->host_rsp = vmcs12->host_rsp;
1853 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1854 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1855 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1856 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1857 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1858 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1859 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1860 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1861 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1862 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1863 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1864 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1865 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1866 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1867 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1868 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1869 	 * evmcs->page_fault_error_code_mask =
1870 	 *		vmcs12->page_fault_error_code_mask;
1871 	 * evmcs->page_fault_error_code_match =
1872 	 *		vmcs12->page_fault_error_code_match;
1873 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1874 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1875 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1876 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1877 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1878 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1879 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1880 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1881 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1882 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1883 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1884 	 * evmcs->guest_ia32_perf_global_ctrl = vmcs12->guest_ia32_perf_global_ctrl;
1885 	 * evmcs->host_ia32_perf_global_ctrl = vmcs12->host_ia32_perf_global_ctrl;
1886 	 * evmcs->encls_exiting_bitmap = vmcs12->encls_exiting_bitmap;
1887 	 * evmcs->tsc_multiplier = vmcs12->tsc_multiplier;
1888 	 *
1889 	 * Not present in struct vmcs12:
1890 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1891 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1892 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1893 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1894 	 * evmcs->host_ia32_s_cet = vmcs12->host_ia32_s_cet;
1895 	 * evmcs->host_ssp = vmcs12->host_ssp;
1896 	 * evmcs->host_ia32_int_ssp_table_addr = vmcs12->host_ia32_int_ssp_table_addr;
1897 	 * evmcs->guest_ia32_s_cet = vmcs12->guest_ia32_s_cet;
1898 	 * evmcs->guest_ia32_lbr_ctl = vmcs12->guest_ia32_lbr_ctl;
1899 	 * evmcs->guest_ia32_int_ssp_table_addr = vmcs12->guest_ia32_int_ssp_table_addr;
1900 	 * evmcs->guest_ssp = vmcs12->guest_ssp;
1901 	 */
1902 
1903 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1904 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1905 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1906 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1907 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1908 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1909 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1910 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1911 
1912 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1913 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1914 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1915 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1916 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1917 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1918 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1919 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1920 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1921 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1922 
1923 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1924 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1925 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1926 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1927 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1928 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1929 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1930 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1931 
1932 	evmcs->guest_es_base = vmcs12->guest_es_base;
1933 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1934 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1935 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1936 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1937 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1938 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1939 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1940 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1941 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1942 
1943 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1944 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1945 
1946 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1947 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1948 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1949 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1950 
1951 	evmcs->guest_pending_dbg_exceptions =
1952 		vmcs12->guest_pending_dbg_exceptions;
1953 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1954 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1955 
1956 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1957 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1958 
1959 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1960 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1961 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1962 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1963 
1964 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1965 
1966 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1967 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1968 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1969 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1970 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1971 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1972 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1973 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1974 
1975 	evmcs->exit_qualification = vmcs12->exit_qualification;
1976 
1977 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1978 	evmcs->guest_rsp = vmcs12->guest_rsp;
1979 	evmcs->guest_rflags = vmcs12->guest_rflags;
1980 
1981 	evmcs->guest_interruptibility_info =
1982 		vmcs12->guest_interruptibility_info;
1983 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1984 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1985 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1986 	evmcs->vm_entry_exception_error_code =
1987 		vmcs12->vm_entry_exception_error_code;
1988 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1989 
1990 	evmcs->guest_rip = vmcs12->guest_rip;
1991 
1992 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1993 
1994 	return;
1995 }
1996 
1997 /*
1998  * This is an equivalent of the nested hypervisor executing the vmptrld
1999  * instruction.
2000  */
2001 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
2002 	struct kvm_vcpu *vcpu, bool from_launch)
2003 {
2004 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2005 	bool evmcs_gpa_changed = false;
2006 	u64 evmcs_gpa;
2007 
2008 	if (likely(!guest_cpuid_has_evmcs(vcpu)))
2009 		return EVMPTRLD_DISABLED;
2010 
2011 	evmcs_gpa = nested_get_evmptr(vcpu);
2012 	if (!evmptr_is_valid(evmcs_gpa)) {
2013 		nested_release_evmcs(vcpu);
2014 		return EVMPTRLD_DISABLED;
2015 	}
2016 
2017 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
2018 		vmx->nested.current_vmptr = INVALID_GPA;
2019 
2020 		nested_release_evmcs(vcpu);
2021 
2022 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2023 				 &vmx->nested.hv_evmcs_map))
2024 			return EVMPTRLD_ERROR;
2025 
2026 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2027 
2028 		/*
2029 		 * Currently, KVM only supports eVMCS version 1
2030 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2031 		 * value to first u32 field of eVMCS which should specify eVMCS
2032 		 * VersionNumber.
2033 		 *
2034 		 * Guest should be aware of supported eVMCS versions by host by
2035 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2036 		 * expected to set this CPUID leaf according to the value
2037 		 * returned in vmcs_version from nested_enable_evmcs().
2038 		 *
2039 		 * However, it turns out that Microsoft Hyper-V fails to comply
2040 		 * to their own invented interface: When Hyper-V use eVMCS, it
2041 		 * just sets first u32 field of eVMCS to revision_id specified
2042 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2043 		 * which is one of the supported versions specified in
2044 		 * CPUID.0x4000000A.EAX[0:15].
2045 		 *
2046 		 * To overcome Hyper-V bug, we accept here either a supported
2047 		 * eVMCS version or VMCS12 revision_id as valid values for first
2048 		 * u32 field of eVMCS.
2049 		 */
2050 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2051 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2052 			nested_release_evmcs(vcpu);
2053 			return EVMPTRLD_VMFAIL;
2054 		}
2055 
2056 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2057 
2058 		evmcs_gpa_changed = true;
2059 		/*
2060 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2061 		 * reloaded from guest's memory (read only fields, fields not
2062 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2063 		 * are no leftovers.
2064 		 */
2065 		if (from_launch) {
2066 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2067 			memset(vmcs12, 0, sizeof(*vmcs12));
2068 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2069 		}
2070 
2071 	}
2072 
2073 	/*
2074 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2075 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2076 	 */
2077 	if (from_launch || evmcs_gpa_changed) {
2078 		vmx->nested.hv_evmcs->hv_clean_fields &=
2079 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2080 
2081 		vmx->nested.force_msr_bitmap_recalc = true;
2082 	}
2083 
2084 	return EVMPTRLD_SUCCEEDED;
2085 }
2086 
2087 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2088 {
2089 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2090 
2091 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2092 		copy_vmcs12_to_enlightened(vmx);
2093 	else
2094 		copy_vmcs12_to_shadow(vmx);
2095 
2096 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2097 }
2098 
2099 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2100 {
2101 	struct vcpu_vmx *vmx =
2102 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2103 
2104 	vmx->nested.preemption_timer_expired = true;
2105 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2106 	kvm_vcpu_kick(&vmx->vcpu);
2107 
2108 	return HRTIMER_NORESTART;
2109 }
2110 
2111 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2112 {
2113 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2114 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2115 
2116 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2117 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2118 
2119 	if (!vmx->nested.has_preemption_timer_deadline) {
2120 		vmx->nested.preemption_timer_deadline =
2121 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2122 		vmx->nested.has_preemption_timer_deadline = true;
2123 	}
2124 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2125 }
2126 
2127 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2128 					u64 preemption_timeout)
2129 {
2130 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2131 
2132 	/*
2133 	 * A timer value of zero is architecturally guaranteed to cause
2134 	 * a VMExit prior to executing any instructions in the guest.
2135 	 */
2136 	if (preemption_timeout == 0) {
2137 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2138 		return;
2139 	}
2140 
2141 	if (vcpu->arch.virtual_tsc_khz == 0)
2142 		return;
2143 
2144 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2145 	preemption_timeout *= 1000000;
2146 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2147 	hrtimer_start(&vmx->nested.preemption_timer,
2148 		      ktime_add_ns(ktime_get(), preemption_timeout),
2149 		      HRTIMER_MODE_ABS_PINNED);
2150 }
2151 
2152 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2153 {
2154 	if (vmx->nested.nested_run_pending &&
2155 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2156 		return vmcs12->guest_ia32_efer;
2157 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2158 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2159 	else
2160 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2161 }
2162 
2163 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2164 {
2165 	struct kvm *kvm = vmx->vcpu.kvm;
2166 
2167 	/*
2168 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2169 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2170 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2171 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2172 	 */
2173 	if (vmx->nested.vmcs02_initialized)
2174 		return;
2175 	vmx->nested.vmcs02_initialized = true;
2176 
2177 	/*
2178 	 * We don't care what the EPTP value is we just need to guarantee
2179 	 * it's valid so we don't get a false positive when doing early
2180 	 * consistency checks.
2181 	 */
2182 	if (enable_ept && nested_early_check)
2183 		vmcs_write64(EPT_POINTER,
2184 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2185 
2186 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2187 	if (cpu_has_vmx_vmfunc())
2188 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2189 
2190 	if (cpu_has_vmx_posted_intr())
2191 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2192 
2193 	if (cpu_has_vmx_msr_bitmap())
2194 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2195 
2196 	/*
2197 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2198 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2199 	 * deal with filtering out L2 GPAs from the buffer.
2200 	 */
2201 	if (enable_pml) {
2202 		vmcs_write64(PML_ADDRESS, 0);
2203 		vmcs_write16(GUEST_PML_INDEX, -1);
2204 	}
2205 
2206 	if (cpu_has_vmx_encls_vmexit())
2207 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2208 
2209 	if (kvm_notify_vmexit_enabled(kvm))
2210 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
2211 
2212 	/*
2213 	 * Set the MSR load/store lists to match L0's settings.  Only the
2214 	 * addresses are constant (for vmcs02), the counts can change based
2215 	 * on L2's behavior, e.g. switching to/from long mode.
2216 	 */
2217 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2218 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2219 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2220 
2221 	vmx_set_constant_host_state(vmx);
2222 }
2223 
2224 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2225 				      struct vmcs12 *vmcs12)
2226 {
2227 	prepare_vmcs02_constant_state(vmx);
2228 
2229 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2230 
2231 	if (enable_vpid) {
2232 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2233 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2234 		else
2235 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2236 	}
2237 }
2238 
2239 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2240 				 struct vmcs12 *vmcs12)
2241 {
2242 	u32 exec_control;
2243 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2244 
2245 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2246 		prepare_vmcs02_early_rare(vmx, vmcs12);
2247 
2248 	/*
2249 	 * PIN CONTROLS
2250 	 */
2251 	exec_control = __pin_controls_get(vmcs01);
2252 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2253 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2254 
2255 	/* Posted interrupts setting is only taken from vmcs12.  */
2256 	vmx->nested.pi_pending = false;
2257 	if (nested_cpu_has_posted_intr(vmcs12))
2258 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2259 	else
2260 		exec_control &= ~PIN_BASED_POSTED_INTR;
2261 	pin_controls_set(vmx, exec_control);
2262 
2263 	/*
2264 	 * EXEC CONTROLS
2265 	 */
2266 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2267 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2268 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2269 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2270 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2271 
2272 	vmx->nested.l1_tpr_threshold = -1;
2273 	if (exec_control & CPU_BASED_TPR_SHADOW)
2274 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2275 #ifdef CONFIG_X86_64
2276 	else
2277 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2278 				CPU_BASED_CR8_STORE_EXITING;
2279 #endif
2280 
2281 	/*
2282 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2283 	 * for I/O port accesses.
2284 	 */
2285 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2286 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2287 
2288 	/*
2289 	 * This bit will be computed in nested_get_vmcs12_pages, because
2290 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2291 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2292 	 * only set/clear this bit.
2293 	 */
2294 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2295 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2296 
2297 	exec_controls_set(vmx, exec_control);
2298 
2299 	/*
2300 	 * SECONDARY EXEC CONTROLS
2301 	 */
2302 	if (cpu_has_secondary_exec_ctrls()) {
2303 		exec_control = __secondary_exec_controls_get(vmcs01);
2304 
2305 		/* Take the following fields only from vmcs12 */
2306 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2307 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2308 				  SECONDARY_EXEC_ENABLE_INVPCID |
2309 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2310 				  SECONDARY_EXEC_XSAVES |
2311 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2312 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2313 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2314 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2315 				  SECONDARY_EXEC_DESC);
2316 
2317 		if (nested_cpu_has(vmcs12,
2318 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2319 			exec_control |= vmcs12->secondary_vm_exec_control;
2320 
2321 		/* PML is emulated and never enabled in hardware for L2. */
2322 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2323 
2324 		/* VMCS shadowing for L2 is emulated for now */
2325 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2326 
2327 		/*
2328 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2329 		 * will not have to rewrite the controls just for this bit.
2330 		 */
2331 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2332 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2333 			exec_control |= SECONDARY_EXEC_DESC;
2334 
2335 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2336 			vmcs_write16(GUEST_INTR_STATUS,
2337 				vmcs12->guest_intr_status);
2338 
2339 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2340 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2341 
2342 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2343 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2344 
2345 		secondary_exec_controls_set(vmx, exec_control);
2346 	}
2347 
2348 	/*
2349 	 * ENTRY CONTROLS
2350 	 *
2351 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2352 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2353 	 * on the related bits (if supported by the CPU) in the hope that
2354 	 * we can avoid VMWrites during vmx_set_efer().
2355 	 *
2356 	 * Similarly, take vmcs01's PERF_GLOBAL_CTRL in the hope that if KVM is
2357 	 * loading PERF_GLOBAL_CTRL via the VMCS for L1, then KVM will want to
2358 	 * do the same for L2.
2359 	 */
2360 	exec_control = __vm_entry_controls_get(vmcs01);
2361 	exec_control |= (vmcs12->vm_entry_controls &
2362 			 ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
2363 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2364 	if (cpu_has_load_ia32_efer()) {
2365 		if (guest_efer & EFER_LMA)
2366 			exec_control |= VM_ENTRY_IA32E_MODE;
2367 		if (guest_efer != host_efer)
2368 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2369 	}
2370 	vm_entry_controls_set(vmx, exec_control);
2371 
2372 	/*
2373 	 * EXIT CONTROLS
2374 	 *
2375 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2376 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2377 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2378 	 */
2379 	exec_control = __vm_exit_controls_get(vmcs01);
2380 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2381 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2382 	else
2383 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2384 	vm_exit_controls_set(vmx, exec_control);
2385 
2386 	/*
2387 	 * Interrupt/Exception Fields
2388 	 */
2389 	if (vmx->nested.nested_run_pending) {
2390 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2391 			     vmcs12->vm_entry_intr_info_field);
2392 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2393 			     vmcs12->vm_entry_exception_error_code);
2394 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2395 			     vmcs12->vm_entry_instruction_len);
2396 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2397 			     vmcs12->guest_interruptibility_info);
2398 		vmx->loaded_vmcs->nmi_known_unmasked =
2399 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2400 	} else {
2401 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2402 	}
2403 }
2404 
2405 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2406 {
2407 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2408 
2409 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2410 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2411 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2412 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2413 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2414 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2415 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2416 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2417 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2418 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2419 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2420 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2421 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2422 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2423 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2424 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2425 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2426 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2427 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2428 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2429 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2430 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2431 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2432 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2433 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2434 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2435 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2436 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2437 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2438 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2439 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2440 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2441 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2442 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2443 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2444 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2445 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2446 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2447 
2448 		vmx->segment_cache.bitmask = 0;
2449 	}
2450 
2451 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2452 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2453 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2454 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2455 			    vmcs12->guest_pending_dbg_exceptions);
2456 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2457 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2458 
2459 		/*
2460 		 * L1 may access the L2's PDPTR, so save them to construct
2461 		 * vmcs12
2462 		 */
2463 		if (enable_ept) {
2464 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2465 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2466 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2467 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2468 		}
2469 
2470 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2471 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2472 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2473 	}
2474 
2475 	if (nested_cpu_has_xsaves(vmcs12))
2476 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2477 
2478 	/*
2479 	 * Whether page-faults are trapped is determined by a combination of
2480 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2481 	 * doesn't care about page faults then we should set all of these to
2482 	 * L1's desires. However, if L0 does care about (some) page faults, it
2483 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2484 	 * simply ask to exit on each and every L2 page fault. This is done by
2485 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2486 	 * Note that below we don't need special code to set EB.PF beyond the
2487 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2488 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2489 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2490 	 */
2491 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2492 		/*
2493 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2494 		 * go ahead and use it?
2495 		 */
2496 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2497 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2498 	} else {
2499 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2500 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2501 	}
2502 
2503 	if (cpu_has_vmx_apicv()) {
2504 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2505 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2506 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2507 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2508 	}
2509 
2510 	/*
2511 	 * Make sure the msr_autostore list is up to date before we set the
2512 	 * count in the vmcs02.
2513 	 */
2514 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2515 
2516 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2517 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2518 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2519 
2520 	set_cr4_guest_host_mask(vmx);
2521 }
2522 
2523 /*
2524  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2525  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2526  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2527  * guest in a way that will both be appropriate to L1's requests, and our
2528  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2529  * function also has additional necessary side-effects, like setting various
2530  * vcpu->arch fields.
2531  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2532  * is assigned to entry_failure_code on failure.
2533  */
2534 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2535 			  bool from_vmentry,
2536 			  enum vm_entry_failure_code *entry_failure_code)
2537 {
2538 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2539 	bool load_guest_pdptrs_vmcs12 = false;
2540 
2541 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2542 		prepare_vmcs02_rare(vmx, vmcs12);
2543 		vmx->nested.dirty_vmcs12 = false;
2544 
2545 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2546 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2547 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2548 	}
2549 
2550 	if (vmx->nested.nested_run_pending &&
2551 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2552 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2553 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2554 	} else {
2555 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2556 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.pre_vmenter_debugctl);
2557 	}
2558 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2559 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2560 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.pre_vmenter_bndcfgs);
2561 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2562 
2563 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2564 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2565 	 * trap. Note that CR0.TS also needs updating - we do this later.
2566 	 */
2567 	vmx_update_exception_bitmap(vcpu);
2568 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2569 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2570 
2571 	if (vmx->nested.nested_run_pending &&
2572 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2573 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2574 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2575 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2576 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2577 	}
2578 
2579 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2580 			vcpu->arch.l1_tsc_offset,
2581 			vmx_get_l2_tsc_offset(vcpu),
2582 			vmx_get_l2_tsc_multiplier(vcpu));
2583 
2584 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2585 			vcpu->arch.l1_tsc_scaling_ratio,
2586 			vmx_get_l2_tsc_multiplier(vcpu));
2587 
2588 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2589 	if (kvm_caps.has_tsc_control)
2590 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2591 
2592 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2593 
2594 	if (nested_cpu_has_ept(vmcs12))
2595 		nested_ept_init_mmu_context(vcpu);
2596 
2597 	/*
2598 	 * Override the CR0/CR4 read shadows after setting the effective guest
2599 	 * CR0/CR4.  The common helpers also set the shadows, but they don't
2600 	 * account for vmcs12's cr0/4_guest_host_mask.
2601 	 */
2602 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2603 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2604 
2605 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2606 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2607 
2608 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2609 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2610 	vmx_set_efer(vcpu, vcpu->arch.efer);
2611 
2612 	/*
2613 	 * Guest state is invalid and unrestricted guest is disabled,
2614 	 * which means L1 attempted VMEntry to L2 with invalid state.
2615 	 * Fail the VMEntry.
2616 	 *
2617 	 * However when force loading the guest state (SMM exit or
2618 	 * loading nested state after migration, it is possible to
2619 	 * have invalid guest state now, which will be later fixed by
2620 	 * restoring L2 register state
2621 	 */
2622 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2623 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2624 		return -EINVAL;
2625 	}
2626 
2627 	/* Shadow page tables on either EPT or shadow page tables. */
2628 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2629 				from_vmentry, entry_failure_code))
2630 		return -EINVAL;
2631 
2632 	/*
2633 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2634 	 * on nested VM-Exit, which can occur without actually running L2 and
2635 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2636 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2637 	 * transition to HLT instead of running L2.
2638 	 */
2639 	if (enable_ept)
2640 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2641 
2642 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2643 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2644 	    is_pae_paging(vcpu)) {
2645 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2646 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2647 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2648 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2649 	}
2650 
2651 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2652 	    intel_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)) &&
2653 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2654 				     vmcs12->guest_ia32_perf_global_ctrl))) {
2655 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2656 		return -EINVAL;
2657 	}
2658 
2659 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2660 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2661 
2662 	/*
2663 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2664 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2665 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2666 	 * here.
2667 	 */
2668 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2669 		vmx->nested.hv_evmcs->hv_clean_fields |=
2670 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2671 
2672 	return 0;
2673 }
2674 
2675 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2676 {
2677 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2678 	       nested_cpu_has_virtual_nmis(vmcs12)))
2679 		return -EINVAL;
2680 
2681 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2682 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2683 		return -EINVAL;
2684 
2685 	return 0;
2686 }
2687 
2688 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2689 {
2690 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2691 
2692 	/* Check for memory type validity */
2693 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2694 	case VMX_EPTP_MT_UC:
2695 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2696 			return false;
2697 		break;
2698 	case VMX_EPTP_MT_WB:
2699 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2700 			return false;
2701 		break;
2702 	default:
2703 		return false;
2704 	}
2705 
2706 	/* Page-walk levels validity. */
2707 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2708 	case VMX_EPTP_PWL_5:
2709 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2710 			return false;
2711 		break;
2712 	case VMX_EPTP_PWL_4:
2713 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2714 			return false;
2715 		break;
2716 	default:
2717 		return false;
2718 	}
2719 
2720 	/* Reserved bits should not be set */
2721 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2722 		return false;
2723 
2724 	/* AD, if set, should be supported */
2725 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2726 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2727 			return false;
2728 	}
2729 
2730 	return true;
2731 }
2732 
2733 /*
2734  * Checks related to VM-Execution Control Fields
2735  */
2736 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2737                                               struct vmcs12 *vmcs12)
2738 {
2739 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2740 
2741 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2742 				   vmx->nested.msrs.pinbased_ctls_low,
2743 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2744 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2745 				   vmx->nested.msrs.procbased_ctls_low,
2746 				   vmx->nested.msrs.procbased_ctls_high)))
2747 		return -EINVAL;
2748 
2749 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2750 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2751 				   vmx->nested.msrs.secondary_ctls_low,
2752 				   vmx->nested.msrs.secondary_ctls_high)))
2753 		return -EINVAL;
2754 
2755 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2756 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2757 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2758 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2759 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2760 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2761 	    nested_vmx_check_nmi_controls(vmcs12) ||
2762 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2763 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2764 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2765 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2766 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2767 		return -EINVAL;
2768 
2769 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2770 	    nested_cpu_has_save_preemption_timer(vmcs12))
2771 		return -EINVAL;
2772 
2773 	if (nested_cpu_has_ept(vmcs12) &&
2774 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2775 		return -EINVAL;
2776 
2777 	if (nested_cpu_has_vmfunc(vmcs12)) {
2778 		if (CC(vmcs12->vm_function_control &
2779 		       ~vmx->nested.msrs.vmfunc_controls))
2780 			return -EINVAL;
2781 
2782 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2783 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2784 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2785 				return -EINVAL;
2786 		}
2787 	}
2788 
2789 	return 0;
2790 }
2791 
2792 /*
2793  * Checks related to VM-Exit Control Fields
2794  */
2795 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2796                                          struct vmcs12 *vmcs12)
2797 {
2798 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2799 
2800 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2801 				    vmx->nested.msrs.exit_ctls_low,
2802 				    vmx->nested.msrs.exit_ctls_high)) ||
2803 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2804 		return -EINVAL;
2805 
2806 	return 0;
2807 }
2808 
2809 /*
2810  * Checks related to VM-Entry Control Fields
2811  */
2812 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2813 					  struct vmcs12 *vmcs12)
2814 {
2815 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2816 
2817 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2818 				    vmx->nested.msrs.entry_ctls_low,
2819 				    vmx->nested.msrs.entry_ctls_high)))
2820 		return -EINVAL;
2821 
2822 	/*
2823 	 * From the Intel SDM, volume 3:
2824 	 * Fields relevant to VM-entry event injection must be set properly.
2825 	 * These fields are the VM-entry interruption-information field, the
2826 	 * VM-entry exception error code, and the VM-entry instruction length.
2827 	 */
2828 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2829 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2830 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2831 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2832 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2833 		bool should_have_error_code;
2834 		bool urg = nested_cpu_has2(vmcs12,
2835 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2836 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2837 
2838 		/* VM-entry interruption-info field: interruption type */
2839 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2840 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2841 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2842 			return -EINVAL;
2843 
2844 		/* VM-entry interruption-info field: vector */
2845 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2846 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2847 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2848 			return -EINVAL;
2849 
2850 		/* VM-entry interruption-info field: deliver error code */
2851 		should_have_error_code =
2852 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2853 			x86_exception_has_error_code(vector);
2854 		if (CC(has_error_code != should_have_error_code))
2855 			return -EINVAL;
2856 
2857 		/* VM-entry exception error code */
2858 		if (CC(has_error_code &&
2859 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2860 			return -EINVAL;
2861 
2862 		/* VM-entry interruption-info field: reserved bits */
2863 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2864 			return -EINVAL;
2865 
2866 		/* VM-entry instruction length */
2867 		switch (intr_type) {
2868 		case INTR_TYPE_SOFT_EXCEPTION:
2869 		case INTR_TYPE_SOFT_INTR:
2870 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2871 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2872 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2873 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2874 				return -EINVAL;
2875 		}
2876 	}
2877 
2878 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2879 		return -EINVAL;
2880 
2881 	return 0;
2882 }
2883 
2884 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2885 				     struct vmcs12 *vmcs12)
2886 {
2887 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2888 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2889 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2890 		return -EINVAL;
2891 
2892 	if (guest_cpuid_has_evmcs(vcpu))
2893 		return nested_evmcs_check_controls(vmcs12);
2894 
2895 	return 0;
2896 }
2897 
2898 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
2899 				       struct vmcs12 *vmcs12)
2900 {
2901 #ifdef CONFIG_X86_64
2902 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
2903 		!!(vcpu->arch.efer & EFER_LMA)))
2904 		return -EINVAL;
2905 #endif
2906 	return 0;
2907 }
2908 
2909 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2910 				       struct vmcs12 *vmcs12)
2911 {
2912 	bool ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
2913 
2914 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2915 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2916 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2917 		return -EINVAL;
2918 
2919 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2920 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2921 		return -EINVAL;
2922 
2923 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2924 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2925 		return -EINVAL;
2926 
2927 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2928 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2929 					   vmcs12->host_ia32_perf_global_ctrl)))
2930 		return -EINVAL;
2931 
2932 	if (ia32e) {
2933 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2934 			return -EINVAL;
2935 	} else {
2936 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2937 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2938 		    CC((vmcs12->host_rip) >> 32))
2939 			return -EINVAL;
2940 	}
2941 
2942 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2943 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2944 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2945 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2946 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2947 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2948 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2949 	    CC(vmcs12->host_cs_selector == 0) ||
2950 	    CC(vmcs12->host_tr_selector == 0) ||
2951 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2952 		return -EINVAL;
2953 
2954 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2955 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2956 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2957 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2958 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2959 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2960 		return -EINVAL;
2961 
2962 	/*
2963 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2964 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2965 	 * the values of the LMA and LME bits in the field must each be that of
2966 	 * the host address-space size VM-exit control.
2967 	 */
2968 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2969 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2970 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2971 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2972 			return -EINVAL;
2973 	}
2974 
2975 	return 0;
2976 }
2977 
2978 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2979 					  struct vmcs12 *vmcs12)
2980 {
2981 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2982 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
2983 	struct vmcs_hdr hdr;
2984 
2985 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
2986 		return 0;
2987 
2988 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2989 		return -EINVAL;
2990 
2991 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
2992 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
2993 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
2994                 return -EINVAL;
2995 
2996 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
2997 					    offsetof(struct vmcs12, hdr),
2998 					    sizeof(hdr))))
2999 		return -EINVAL;
3000 
3001 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
3002 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
3003 		return -EINVAL;
3004 
3005 	return 0;
3006 }
3007 
3008 /*
3009  * Checks related to Guest Non-register State
3010  */
3011 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
3012 {
3013 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
3014 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
3015 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
3016 		return -EINVAL;
3017 
3018 	return 0;
3019 }
3020 
3021 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
3022 					struct vmcs12 *vmcs12,
3023 					enum vm_entry_failure_code *entry_failure_code)
3024 {
3025 	bool ia32e = !!(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE);
3026 
3027 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
3028 
3029 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3030 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3031 		return -EINVAL;
3032 
3033 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
3034 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
3035 		return -EINVAL;
3036 
3037 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3038 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3039 		return -EINVAL;
3040 
3041 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3042 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3043 		return -EINVAL;
3044 	}
3045 
3046 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3047 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3048 					   vmcs12->guest_ia32_perf_global_ctrl)))
3049 		return -EINVAL;
3050 
3051 	if (CC((vmcs12->guest_cr0 & (X86_CR0_PG | X86_CR0_PE)) == X86_CR0_PG))
3052 		return -EINVAL;
3053 
3054 	if (CC(ia32e && !(vmcs12->guest_cr4 & X86_CR4_PAE)) ||
3055 	    CC(ia32e && !(vmcs12->guest_cr0 & X86_CR0_PG)))
3056 		return -EINVAL;
3057 
3058 	/*
3059 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3060 	 * are performed on the field for the IA32_EFER MSR:
3061 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3062 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3063 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3064 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3065 	 *   CR0.PG) is 1.
3066 	 */
3067 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3068 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3069 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3070 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3071 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3072 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3073 			return -EINVAL;
3074 	}
3075 
3076 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3077 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3078 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3079 		return -EINVAL;
3080 
3081 	if (nested_check_guest_non_reg_state(vmcs12))
3082 		return -EINVAL;
3083 
3084 	return 0;
3085 }
3086 
3087 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3088 {
3089 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3090 	unsigned long cr3, cr4;
3091 	bool vm_fail;
3092 
3093 	if (!nested_early_check)
3094 		return 0;
3095 
3096 	if (vmx->msr_autoload.host.nr)
3097 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3098 	if (vmx->msr_autoload.guest.nr)
3099 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3100 
3101 	preempt_disable();
3102 
3103 	vmx_prepare_switch_to_guest(vcpu);
3104 
3105 	/*
3106 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3107 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3108 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3109 	 * there is no need to preserve other bits or save/restore the field.
3110 	 */
3111 	vmcs_writel(GUEST_RFLAGS, 0);
3112 
3113 	cr3 = __get_current_cr3_fast();
3114 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3115 		vmcs_writel(HOST_CR3, cr3);
3116 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3117 	}
3118 
3119 	cr4 = cr4_read_shadow();
3120 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3121 		vmcs_writel(HOST_CR4, cr4);
3122 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3123 	}
3124 
3125 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3126 				 __vmx_vcpu_run_flags(vmx));
3127 
3128 	if (vmx->msr_autoload.host.nr)
3129 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3130 	if (vmx->msr_autoload.guest.nr)
3131 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3132 
3133 	if (vm_fail) {
3134 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3135 
3136 		preempt_enable();
3137 
3138 		trace_kvm_nested_vmenter_failed(
3139 			"early hardware check VM-instruction error: ", error);
3140 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3141 		return 1;
3142 	}
3143 
3144 	/*
3145 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3146 	 */
3147 	if (hw_breakpoint_active())
3148 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3149 	local_irq_enable();
3150 	preempt_enable();
3151 
3152 	/*
3153 	 * A non-failing VMEntry means we somehow entered guest mode with
3154 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3155 	 * is no telling what memory has been modified or what state has
3156 	 * been exposed to unknown code.  Hitting this all but guarantees
3157 	 * a (very critical) hardware issue.
3158 	 */
3159 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3160 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3161 
3162 	return 0;
3163 }
3164 
3165 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3166 {
3167 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3168 
3169 	/*
3170 	 * hv_evmcs may end up being not mapped after migration (when
3171 	 * L2 was running), map it here to make sure vmcs12 changes are
3172 	 * properly reflected.
3173 	 */
3174 	if (guest_cpuid_has_evmcs(vcpu) &&
3175 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3176 		enum nested_evmptrld_status evmptrld_status =
3177 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3178 
3179 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3180 		    evmptrld_status == EVMPTRLD_ERROR)
3181 			return false;
3182 
3183 		/*
3184 		 * Post migration VMCS12 always provides the most actual
3185 		 * information, copy it to eVMCS upon entry.
3186 		 */
3187 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3188 	}
3189 
3190 	return true;
3191 }
3192 
3193 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3194 {
3195 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3196 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3197 	struct kvm_host_map *map;
3198 
3199 	if (!vcpu->arch.pdptrs_from_userspace &&
3200 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3201 		/*
3202 		 * Reload the guest's PDPTRs since after a migration
3203 		 * the guest CR3 might be restored prior to setting the nested
3204 		 * state which can lead to a load of wrong PDPTRs.
3205 		 */
3206 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3207 			return false;
3208 	}
3209 
3210 
3211 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3212 		map = &vmx->nested.apic_access_page_map;
3213 
3214 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->apic_access_addr), map)) {
3215 			vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(map->pfn));
3216 		} else {
3217 			pr_debug_ratelimited("%s: no backing for APIC-access address in vmcs12\n",
3218 					     __func__);
3219 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3220 			vcpu->run->internal.suberror =
3221 				KVM_INTERNAL_ERROR_EMULATION;
3222 			vcpu->run->internal.ndata = 0;
3223 			return false;
3224 		}
3225 	}
3226 
3227 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3228 		map = &vmx->nested.virtual_apic_map;
3229 
3230 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3231 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3232 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3233 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3234 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3235 			/*
3236 			 * The processor will never use the TPR shadow, simply
3237 			 * clear the bit from the execution control.  Such a
3238 			 * configuration is useless, but it happens in tests.
3239 			 * For any other configuration, failing the vm entry is
3240 			 * _not_ what the processor does but it's basically the
3241 			 * only possibility we have.
3242 			 */
3243 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3244 		} else {
3245 			/*
3246 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3247 			 * force VM-Entry to fail.
3248 			 */
3249 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3250 		}
3251 	}
3252 
3253 	if (nested_cpu_has_posted_intr(vmcs12)) {
3254 		map = &vmx->nested.pi_desc_map;
3255 
3256 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3257 			vmx->nested.pi_desc =
3258 				(struct pi_desc *)(((void *)map->hva) +
3259 				offset_in_page(vmcs12->posted_intr_desc_addr));
3260 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3261 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3262 		} else {
3263 			/*
3264 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3265 			 * access the contents of the VMCS12 posted interrupt
3266 			 * descriptor. (Note that KVM may do this when it
3267 			 * should not, per the architectural specification.)
3268 			 */
3269 			vmx->nested.pi_desc = NULL;
3270 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3271 		}
3272 	}
3273 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3274 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3275 	else
3276 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3277 
3278 	return true;
3279 }
3280 
3281 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3282 {
3283 	/*
3284 	 * Note: nested_get_evmcs_page() also updates 'vp_assist_page' copy
3285 	 * in 'struct kvm_vcpu_hv' in case eVMCS is in use, this is mandatory
3286 	 * to make nested_evmcs_l2_tlb_flush_enabled() work correctly post
3287 	 * migration.
3288 	 */
3289 	if (!nested_get_evmcs_page(vcpu)) {
3290 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3291 				     __func__);
3292 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3293 		vcpu->run->internal.suberror =
3294 			KVM_INTERNAL_ERROR_EMULATION;
3295 		vcpu->run->internal.ndata = 0;
3296 
3297 		return false;
3298 	}
3299 
3300 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3301 		return false;
3302 
3303 	return true;
3304 }
3305 
3306 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3307 {
3308 	struct vmcs12 *vmcs12;
3309 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3310 	gpa_t dst;
3311 
3312 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3313 		return 0;
3314 
3315 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3316 		return 1;
3317 
3318 	/*
3319 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3320 	 * set is already checked as part of A/D emulation.
3321 	 */
3322 	vmcs12 = get_vmcs12(vcpu);
3323 	if (!nested_cpu_has_pml(vmcs12))
3324 		return 0;
3325 
3326 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3327 		vmx->nested.pml_full = true;
3328 		return 1;
3329 	}
3330 
3331 	gpa &= ~0xFFFull;
3332 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3333 
3334 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3335 				 offset_in_page(dst), sizeof(gpa)))
3336 		return 0;
3337 
3338 	vmcs12->guest_pml_index--;
3339 
3340 	return 0;
3341 }
3342 
3343 /*
3344  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3345  * for running VMX instructions (except VMXON, whose prerequisites are
3346  * slightly different). It also specifies what exception to inject otherwise.
3347  * Note that many of these exceptions have priority over VM exits, so they
3348  * don't have to be checked again here.
3349  */
3350 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3351 {
3352 	if (!to_vmx(vcpu)->nested.vmxon) {
3353 		kvm_queue_exception(vcpu, UD_VECTOR);
3354 		return 0;
3355 	}
3356 
3357 	if (vmx_get_cpl(vcpu)) {
3358 		kvm_inject_gp(vcpu, 0);
3359 		return 0;
3360 	}
3361 
3362 	return 1;
3363 }
3364 
3365 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3366 {
3367 	u8 rvi = vmx_get_rvi();
3368 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3369 
3370 	return ((rvi & 0xf0) > (vppr & 0xf0));
3371 }
3372 
3373 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3374 				   struct vmcs12 *vmcs12);
3375 
3376 /*
3377  * If from_vmentry is false, this is being called from state restore (either RSM
3378  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3379  *
3380  * Returns:
3381  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3382  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3383  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3384  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3385  */
3386 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3387 							bool from_vmentry)
3388 {
3389 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3390 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3391 	enum vm_entry_failure_code entry_failure_code;
3392 	bool evaluate_pending_interrupts;
3393 	union vmx_exit_reason exit_reason = {
3394 		.basic = EXIT_REASON_INVALID_STATE,
3395 		.failed_vmentry = 1,
3396 	};
3397 	u32 failed_index;
3398 
3399 	trace_kvm_nested_vmenter(kvm_rip_read(vcpu),
3400 				 vmx->nested.current_vmptr,
3401 				 vmcs12->guest_rip,
3402 				 vmcs12->guest_intr_status,
3403 				 vmcs12->vm_entry_intr_info_field,
3404 				 vmcs12->secondary_vm_exec_control & SECONDARY_EXEC_ENABLE_EPT,
3405 				 vmcs12->ept_pointer,
3406 				 vmcs12->guest_cr3,
3407 				 KVM_ISA_VMX);
3408 
3409 	kvm_service_local_tlb_flush_requests(vcpu);
3410 
3411 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3412 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3413 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3414 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3415 	if (!evaluate_pending_interrupts)
3416 		evaluate_pending_interrupts |= kvm_apic_has_pending_init_or_sipi(vcpu);
3417 
3418 	if (!vmx->nested.nested_run_pending ||
3419 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3420 		vmx->nested.pre_vmenter_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3421 	if (kvm_mpx_supported() &&
3422 	    (!vmx->nested.nested_run_pending ||
3423 	     !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
3424 		vmx->nested.pre_vmenter_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3425 
3426 	/*
3427 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3428 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3429 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3430 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3431 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3432 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3433 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3434 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3435 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3436 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3437 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3438 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3439 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3440 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3441 	 */
3442 	if (!enable_ept && !nested_early_check)
3443 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3444 
3445 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3446 
3447 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3448 
3449 	if (from_vmentry) {
3450 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3451 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3452 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3453 		}
3454 
3455 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3456 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3457 			return NVMX_VMENTRY_VMFAIL;
3458 		}
3459 
3460 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3461 						 &entry_failure_code)) {
3462 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3463 			vmcs12->exit_qualification = entry_failure_code;
3464 			goto vmentry_fail_vmexit;
3465 		}
3466 	}
3467 
3468 	enter_guest_mode(vcpu);
3469 
3470 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3471 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3472 		vmcs12->exit_qualification = entry_failure_code;
3473 		goto vmentry_fail_vmexit_guest_mode;
3474 	}
3475 
3476 	if (from_vmentry) {
3477 		failed_index = nested_vmx_load_msr(vcpu,
3478 						   vmcs12->vm_entry_msr_load_addr,
3479 						   vmcs12->vm_entry_msr_load_count);
3480 		if (failed_index) {
3481 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3482 			vmcs12->exit_qualification = failed_index;
3483 			goto vmentry_fail_vmexit_guest_mode;
3484 		}
3485 	} else {
3486 		/*
3487 		 * The MMU is not initialized to point at the right entities yet and
3488 		 * "get pages" would need to read data from the guest (i.e. we will
3489 		 * need to perform gpa to hpa translation). Request a call
3490 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3491 		 * have already been set at vmentry time and should not be reset.
3492 		 */
3493 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3494 	}
3495 
3496 	/*
3497 	 * Re-evaluate pending events if L1 had a pending IRQ/NMI/INIT/SIPI
3498 	 * when it executed VMLAUNCH/VMRESUME, as entering non-root mode can
3499 	 * effectively unblock various events, e.g. INIT/SIPI cause VM-Exit
3500 	 * unconditionally.
3501 	 */
3502 	if (unlikely(evaluate_pending_interrupts))
3503 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3504 
3505 	/*
3506 	 * Do not start the preemption timer hrtimer until after we know
3507 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3508 	 * the timer.
3509 	 */
3510 	vmx->nested.preemption_timer_expired = false;
3511 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3512 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3513 		vmx_start_preemption_timer(vcpu, timer_value);
3514 	}
3515 
3516 	/*
3517 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3518 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3519 	 * returned as far as L1 is concerned. It will only return (and set
3520 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3521 	 */
3522 	return NVMX_VMENTRY_SUCCESS;
3523 
3524 	/*
3525 	 * A failed consistency check that leads to a VMExit during L1's
3526 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3527 	 * 26.7 "VM-entry failures during or after loading guest state".
3528 	 */
3529 vmentry_fail_vmexit_guest_mode:
3530 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3531 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3532 	leave_guest_mode(vcpu);
3533 
3534 vmentry_fail_vmexit:
3535 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3536 
3537 	if (!from_vmentry)
3538 		return NVMX_VMENTRY_VMEXIT;
3539 
3540 	load_vmcs12_host_state(vcpu, vmcs12);
3541 	vmcs12->vm_exit_reason = exit_reason.full;
3542 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3543 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3544 	return NVMX_VMENTRY_VMEXIT;
3545 }
3546 
3547 /*
3548  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3549  * for running an L2 nested guest.
3550  */
3551 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3552 {
3553 	struct vmcs12 *vmcs12;
3554 	enum nvmx_vmentry_status status;
3555 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3556 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3557 	enum nested_evmptrld_status evmptrld_status;
3558 
3559 	if (!nested_vmx_check_permission(vcpu))
3560 		return 1;
3561 
3562 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3563 	if (evmptrld_status == EVMPTRLD_ERROR) {
3564 		kvm_queue_exception(vcpu, UD_VECTOR);
3565 		return 1;
3566 	}
3567 
3568 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
3569 
3570 	if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3571 		return nested_vmx_failInvalid(vcpu);
3572 
3573 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3574 	       vmx->nested.current_vmptr == INVALID_GPA))
3575 		return nested_vmx_failInvalid(vcpu);
3576 
3577 	vmcs12 = get_vmcs12(vcpu);
3578 
3579 	/*
3580 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3581 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3582 	 * rather than RFLAGS.ZF, and no error number is stored to the
3583 	 * VM-instruction error field.
3584 	 */
3585 	if (CC(vmcs12->hdr.shadow_vmcs))
3586 		return nested_vmx_failInvalid(vcpu);
3587 
3588 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3589 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3590 		/* Enlightened VMCS doesn't have launch state */
3591 		vmcs12->launch_state = !launch;
3592 	} else if (enable_shadow_vmcs) {
3593 		copy_shadow_to_vmcs12(vmx);
3594 	}
3595 
3596 	/*
3597 	 * The nested entry process starts with enforcing various prerequisites
3598 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3599 	 * they fail: As the SDM explains, some conditions should cause the
3600 	 * instruction to fail, while others will cause the instruction to seem
3601 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3602 	 * To speed up the normal (success) code path, we should avoid checking
3603 	 * for misconfigurations which will anyway be caught by the processor
3604 	 * when using the merged vmcs02.
3605 	 */
3606 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3607 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3608 
3609 	if (CC(vmcs12->launch_state == launch))
3610 		return nested_vmx_fail(vcpu,
3611 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3612 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3613 
3614 	if (nested_vmx_check_controls(vcpu, vmcs12))
3615 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3616 
3617 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3618 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3619 
3620 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3621 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3622 
3623 	/*
3624 	 * We're finally done with prerequisite checking, and can start with
3625 	 * the nested entry.
3626 	 */
3627 	vmx->nested.nested_run_pending = 1;
3628 	vmx->nested.has_preemption_timer_deadline = false;
3629 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3630 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3631 		goto vmentry_failed;
3632 
3633 	/* Emulate processing of posted interrupts on VM-Enter. */
3634 	if (nested_cpu_has_posted_intr(vmcs12) &&
3635 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3636 		vmx->nested.pi_pending = true;
3637 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3638 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3639 	}
3640 
3641 	/* Hide L1D cache contents from the nested guest.  */
3642 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3643 
3644 	/*
3645 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3646 	 * also be used as part of restoring nVMX state for
3647 	 * snapshot restore (migration).
3648 	 *
3649 	 * In this flow, it is assumed that vmcs12 cache was
3650 	 * transferred as part of captured nVMX state and should
3651 	 * therefore not be read from guest memory (which may not
3652 	 * exist on destination host yet).
3653 	 */
3654 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3655 
3656 	switch (vmcs12->guest_activity_state) {
3657 	case GUEST_ACTIVITY_HLT:
3658 		/*
3659 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3660 		 * awakened by event injection or by an NMI-window VM-exit or
3661 		 * by an interrupt-window VM-exit, halt the vcpu.
3662 		 */
3663 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3664 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3665 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3666 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3667 			vmx->nested.nested_run_pending = 0;
3668 			return kvm_emulate_halt_noskip(vcpu);
3669 		}
3670 		break;
3671 	case GUEST_ACTIVITY_WAIT_SIPI:
3672 		vmx->nested.nested_run_pending = 0;
3673 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3674 		break;
3675 	default:
3676 		break;
3677 	}
3678 
3679 	return 1;
3680 
3681 vmentry_failed:
3682 	vmx->nested.nested_run_pending = 0;
3683 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3684 		return 0;
3685 	if (status == NVMX_VMENTRY_VMEXIT)
3686 		return 1;
3687 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3688 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3689 }
3690 
3691 /*
3692  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3693  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3694  * This function returns the new value we should put in vmcs12.guest_cr0.
3695  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3696  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3697  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3698  *     didn't trap the bit, because if L1 did, so would L0).
3699  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3700  *     been modified by L2, and L1 knows it. So just leave the old value of
3701  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3702  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3703  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3704  *     changed these bits, and therefore they need to be updated, but L0
3705  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3706  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3707  */
3708 static inline unsigned long
3709 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3710 {
3711 	return
3712 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3713 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3714 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3715 			vcpu->arch.cr0_guest_owned_bits));
3716 }
3717 
3718 static inline unsigned long
3719 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3720 {
3721 	return
3722 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3723 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3724 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3725 			vcpu->arch.cr4_guest_owned_bits));
3726 }
3727 
3728 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3729 				      struct vmcs12 *vmcs12,
3730 				      u32 vm_exit_reason, u32 exit_intr_info)
3731 {
3732 	u32 idt_vectoring;
3733 	unsigned int nr;
3734 
3735 	/*
3736 	 * Per the SDM, VM-Exits due to double and triple faults are never
3737 	 * considered to occur during event delivery, even if the double/triple
3738 	 * fault is the result of an escalating vectoring issue.
3739 	 *
3740 	 * Note, the SDM qualifies the double fault behavior with "The original
3741 	 * event results in a double-fault exception".  It's unclear why the
3742 	 * qualification exists since exits due to double fault can occur only
3743 	 * while vectoring a different exception (injected events are never
3744 	 * subject to interception), i.e. there's _always_ an original event.
3745 	 *
3746 	 * The SDM also uses NMI as a confusing example for the "original event
3747 	 * causes the VM exit directly" clause.  NMI isn't special in any way,
3748 	 * the same rule applies to all events that cause an exit directly.
3749 	 * NMI is an odd choice for the example because NMIs can only occur on
3750 	 * instruction boundaries, i.e. they _can't_ occur during vectoring.
3751 	 */
3752 	if ((u16)vm_exit_reason == EXIT_REASON_TRIPLE_FAULT ||
3753 	    ((u16)vm_exit_reason == EXIT_REASON_EXCEPTION_NMI &&
3754 	     is_double_fault(exit_intr_info))) {
3755 		vmcs12->idt_vectoring_info_field = 0;
3756 	} else if (vcpu->arch.exception.injected) {
3757 		nr = vcpu->arch.exception.vector;
3758 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3759 
3760 		if (kvm_exception_is_soft(nr)) {
3761 			vmcs12->vm_exit_instruction_len =
3762 				vcpu->arch.event_exit_inst_len;
3763 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3764 		} else
3765 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3766 
3767 		if (vcpu->arch.exception.has_error_code) {
3768 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3769 			vmcs12->idt_vectoring_error_code =
3770 				vcpu->arch.exception.error_code;
3771 		}
3772 
3773 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3774 	} else if (vcpu->arch.nmi_injected) {
3775 		vmcs12->idt_vectoring_info_field =
3776 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3777 	} else if (vcpu->arch.interrupt.injected) {
3778 		nr = vcpu->arch.interrupt.nr;
3779 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3780 
3781 		if (vcpu->arch.interrupt.soft) {
3782 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3783 			vmcs12->vm_entry_instruction_len =
3784 				vcpu->arch.event_exit_inst_len;
3785 		} else
3786 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3787 
3788 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3789 	} else {
3790 		vmcs12->idt_vectoring_info_field = 0;
3791 	}
3792 }
3793 
3794 
3795 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3796 {
3797 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3798 	gfn_t gfn;
3799 
3800 	/*
3801 	 * Don't need to mark the APIC access page dirty; it is never
3802 	 * written to by the CPU during APIC virtualization.
3803 	 */
3804 
3805 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3806 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3807 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3808 	}
3809 
3810 	if (nested_cpu_has_posted_intr(vmcs12)) {
3811 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3812 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3813 	}
3814 }
3815 
3816 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3817 {
3818 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3819 	int max_irr;
3820 	void *vapic_page;
3821 	u16 status;
3822 
3823 	if (!vmx->nested.pi_pending)
3824 		return 0;
3825 
3826 	if (!vmx->nested.pi_desc)
3827 		goto mmio_needed;
3828 
3829 	vmx->nested.pi_pending = false;
3830 
3831 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3832 		return 0;
3833 
3834 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3835 	if (max_irr != 256) {
3836 		vapic_page = vmx->nested.virtual_apic_map.hva;
3837 		if (!vapic_page)
3838 			goto mmio_needed;
3839 
3840 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3841 			vapic_page, &max_irr);
3842 		status = vmcs_read16(GUEST_INTR_STATUS);
3843 		if ((u8)max_irr > ((u8)status & 0xff)) {
3844 			status &= ~0xff;
3845 			status |= (u8)max_irr;
3846 			vmcs_write16(GUEST_INTR_STATUS, status);
3847 		}
3848 	}
3849 
3850 	nested_mark_vmcs12_pages_dirty(vcpu);
3851 	return 0;
3852 
3853 mmio_needed:
3854 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3855 	return -ENXIO;
3856 }
3857 
3858 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu)
3859 {
3860 	struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
3861 	u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
3862 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3863 	unsigned long exit_qual;
3864 
3865 	if (ex->has_payload) {
3866 		exit_qual = ex->payload;
3867 	} else if (ex->vector == PF_VECTOR) {
3868 		exit_qual = vcpu->arch.cr2;
3869 	} else if (ex->vector == DB_VECTOR) {
3870 		exit_qual = vcpu->arch.dr6;
3871 		exit_qual &= ~DR6_BT;
3872 		exit_qual ^= DR6_ACTIVE_LOW;
3873 	} else {
3874 		exit_qual = 0;
3875 	}
3876 
3877 	/*
3878 	 * Unlike AMD's Paged Real Mode, which reports an error code on #PF
3879 	 * VM-Exits even if the CPU is in Real Mode, Intel VMX never sets the
3880 	 * "has error code" flags on VM-Exit if the CPU is in Real Mode.
3881 	 */
3882 	if (ex->has_error_code && is_protmode(vcpu)) {
3883 		/*
3884 		 * Intel CPUs do not generate error codes with bits 31:16 set,
3885 		 * and more importantly VMX disallows setting bits 31:16 in the
3886 		 * injected error code for VM-Entry.  Drop the bits to mimic
3887 		 * hardware and avoid inducing failure on nested VM-Entry if L1
3888 		 * chooses to inject the exception back to L2.  AMD CPUs _do_
3889 		 * generate "full" 32-bit error codes, so KVM allows userspace
3890 		 * to inject exception error codes with bits 31:16 set.
3891 		 */
3892 		vmcs12->vm_exit_intr_error_code = (u16)ex->error_code;
3893 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3894 	}
3895 
3896 	if (kvm_exception_is_soft(ex->vector))
3897 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3898 	else
3899 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3900 
3901 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3902 	    vmx_get_nmi_mask(vcpu))
3903 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3904 
3905 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3906 }
3907 
3908 /*
3909  * Returns true if a debug trap is (likely) pending delivery.  Infer the class
3910  * of a #DB (trap-like vs. fault-like) from the exception payload (to-be-DR6).
3911  * Using the payload is flawed because code breakpoints (fault-like) and data
3912  * breakpoints (trap-like) set the same bits in DR6 (breakpoint detected), i.e.
3913  * this will return false positives if a to-be-injected code breakpoint #DB is
3914  * pending (from KVM's perspective, but not "pending" across an instruction
3915  * boundary).  ICEBP, a.k.a. INT1, is also not reflected here even though it
3916  * too is trap-like.
3917  *
3918  * KVM "works" despite these flaws as ICEBP isn't currently supported by the
3919  * emulator, Monitor Trap Flag is not marked pending on intercepted #DBs (the
3920  * #DB has already happened), and MTF isn't marked pending on code breakpoints
3921  * from the emulator (because such #DBs are fault-like and thus don't trigger
3922  * actions that fire on instruction retire).
3923  */
3924 static unsigned long vmx_get_pending_dbg_trap(struct kvm_queued_exception *ex)
3925 {
3926 	if (!ex->pending || ex->vector != DB_VECTOR)
3927 		return 0;
3928 
3929 	/* General Detect #DBs are always fault-like. */
3930 	return ex->payload & ~DR6_BD;
3931 }
3932 
3933 /*
3934  * Returns true if there's a pending #DB exception that is lower priority than
3935  * a pending Monitor Trap Flag VM-Exit.  TSS T-flag #DBs are not emulated by
3936  * KVM, but could theoretically be injected by userspace.  Note, this code is
3937  * imperfect, see above.
3938  */
3939 static bool vmx_is_low_priority_db_trap(struct kvm_queued_exception *ex)
3940 {
3941 	return vmx_get_pending_dbg_trap(ex) & ~DR6_BT;
3942 }
3943 
3944 /*
3945  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3946  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3947  * represents these debug traps with a payload that is said to be compatible
3948  * with the 'pending debug exceptions' field, write the payload to the VMCS
3949  * field if a VM-exit is delivered before the debug trap.
3950  */
3951 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3952 {
3953 	unsigned long pending_dbg;
3954 
3955 	pending_dbg = vmx_get_pending_dbg_trap(&vcpu->arch.exception);
3956 	if (pending_dbg)
3957 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, pending_dbg);
3958 }
3959 
3960 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3961 {
3962 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3963 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3964 }
3965 
3966 static bool vmx_has_nested_events(struct kvm_vcpu *vcpu)
3967 {
3968 	return nested_vmx_preemption_timer_pending(vcpu) ||
3969 	       to_vmx(vcpu)->nested.mtf_pending;
3970 }
3971 
3972 /*
3973  * Per the Intel SDM's table "Priority Among Concurrent Events", with minor
3974  * edits to fill in missing examples, e.g. #DB due to split-lock accesses,
3975  * and less minor edits to splice in the priority of VMX Non-Root specific
3976  * events, e.g. MTF and NMI/INTR-window exiting.
3977  *
3978  * 1 Hardware Reset and Machine Checks
3979  *	- RESET
3980  *	- Machine Check
3981  *
3982  * 2 Trap on Task Switch
3983  *	- T flag in TSS is set (on task switch)
3984  *
3985  * 3 External Hardware Interventions
3986  *	- FLUSH
3987  *	- STOPCLK
3988  *	- SMI
3989  *	- INIT
3990  *
3991  * 3.5 Monitor Trap Flag (MTF) VM-exit[1]
3992  *
3993  * 4 Traps on Previous Instruction
3994  *	- Breakpoints
3995  *	- Trap-class Debug Exceptions (#DB due to TF flag set, data/I-O
3996  *	  breakpoint, or #DB due to a split-lock access)
3997  *
3998  * 4.3	VMX-preemption timer expired VM-exit
3999  *
4000  * 4.6	NMI-window exiting VM-exit[2]
4001  *
4002  * 5 Nonmaskable Interrupts (NMI)
4003  *
4004  * 5.5 Interrupt-window exiting VM-exit and Virtual-interrupt delivery
4005  *
4006  * 6 Maskable Hardware Interrupts
4007  *
4008  * 7 Code Breakpoint Fault
4009  *
4010  * 8 Faults from Fetching Next Instruction
4011  *	- Code-Segment Limit Violation
4012  *	- Code Page Fault
4013  *	- Control protection exception (missing ENDBRANCH at target of indirect
4014  *					call or jump)
4015  *
4016  * 9 Faults from Decoding Next Instruction
4017  *	- Instruction length > 15 bytes
4018  *	- Invalid Opcode
4019  *	- Coprocessor Not Available
4020  *
4021  *10 Faults on Executing Instruction
4022  *	- Overflow
4023  *	- Bound error
4024  *	- Invalid TSS
4025  *	- Segment Not Present
4026  *	- Stack fault
4027  *	- General Protection
4028  *	- Data Page Fault
4029  *	- Alignment Check
4030  *	- x86 FPU Floating-point exception
4031  *	- SIMD floating-point exception
4032  *	- Virtualization exception
4033  *	- Control protection exception
4034  *
4035  * [1] Per the "Monitor Trap Flag" section: System-management interrupts (SMIs),
4036  *     INIT signals, and higher priority events take priority over MTF VM exits.
4037  *     MTF VM exits take priority over debug-trap exceptions and lower priority
4038  *     events.
4039  *
4040  * [2] Debug-trap exceptions and higher priority events take priority over VM exits
4041  *     caused by the VMX-preemption timer.  VM exits caused by the VMX-preemption
4042  *     timer take priority over VM exits caused by the "NMI-window exiting"
4043  *     VM-execution control and lower priority events.
4044  *
4045  * [3] Debug-trap exceptions and higher priority events take priority over VM exits
4046  *     caused by "NMI-window exiting".  VM exits caused by this control take
4047  *     priority over non-maskable interrupts (NMIs) and lower priority events.
4048  *
4049  * [4] Virtual-interrupt delivery has the same priority as that of VM exits due to
4050  *     the 1-setting of the "interrupt-window exiting" VM-execution control.  Thus,
4051  *     non-maskable interrupts (NMIs) and higher priority events take priority over
4052  *     delivery of a virtual interrupt; delivery of a virtual interrupt takes
4053  *     priority over external interrupts and lower priority events.
4054  */
4055 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
4056 {
4057 	struct kvm_lapic *apic = vcpu->arch.apic;
4058 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4059 	/*
4060 	 * Only a pending nested run blocks a pending exception.  If there is a
4061 	 * previously injected event, the pending exception occurred while said
4062 	 * event was being delivered and thus needs to be handled.
4063 	 */
4064 	bool block_nested_exceptions = vmx->nested.nested_run_pending;
4065 	/*
4066 	 * New events (not exceptions) are only recognized at instruction
4067 	 * boundaries.  If an event needs reinjection, then KVM is handling a
4068 	 * VM-Exit that occurred _during_ instruction execution; new events are
4069 	 * blocked until the instruction completes.
4070 	 */
4071 	bool block_nested_events = block_nested_exceptions ||
4072 				   kvm_event_needs_reinjection(vcpu);
4073 
4074 	if (lapic_in_kernel(vcpu) &&
4075 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
4076 		if (block_nested_events)
4077 			return -EBUSY;
4078 		nested_vmx_update_pending_dbg(vcpu);
4079 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
4080 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
4081 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
4082 
4083 		/* MTF is discarded if the vCPU is in WFS. */
4084 		vmx->nested.mtf_pending = false;
4085 		return 0;
4086 	}
4087 
4088 	if (lapic_in_kernel(vcpu) &&
4089 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
4090 		if (block_nested_events)
4091 			return -EBUSY;
4092 
4093 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
4094 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
4095 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
4096 						apic->sipi_vector & 0xFFUL);
4097 			return 0;
4098 		}
4099 		/* Fallthrough, the SIPI is completely ignored. */
4100 	}
4101 
4102 	/*
4103 	 * Process exceptions that are higher priority than Monitor Trap Flag:
4104 	 * fault-like exceptions, TSS T flag #DB (not emulated by KVM, but
4105 	 * could theoretically come in from userspace), and ICEBP (INT1).
4106 	 *
4107 	 * TODO: SMIs have higher priority than MTF and trap-like #DBs (except
4108 	 * for TSS T flag #DBs).  KVM also doesn't save/restore pending MTF
4109 	 * across SMI/RSM as it should; that needs to be addressed in order to
4110 	 * prioritize SMI over MTF and trap-like #DBs.
4111 	 */
4112 	if (vcpu->arch.exception_vmexit.pending &&
4113 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception_vmexit)) {
4114 		if (block_nested_exceptions)
4115 			return -EBUSY;
4116 
4117 		nested_vmx_inject_exception_vmexit(vcpu);
4118 		return 0;
4119 	}
4120 
4121 	if (vcpu->arch.exception.pending &&
4122 	    !vmx_is_low_priority_db_trap(&vcpu->arch.exception)) {
4123 		if (block_nested_exceptions)
4124 			return -EBUSY;
4125 		goto no_vmexit;
4126 	}
4127 
4128 	if (vmx->nested.mtf_pending) {
4129 		if (block_nested_events)
4130 			return -EBUSY;
4131 		nested_vmx_update_pending_dbg(vcpu);
4132 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
4133 		return 0;
4134 	}
4135 
4136 	if (vcpu->arch.exception_vmexit.pending) {
4137 		if (block_nested_exceptions)
4138 			return -EBUSY;
4139 
4140 		nested_vmx_inject_exception_vmexit(vcpu);
4141 		return 0;
4142 	}
4143 
4144 	if (vcpu->arch.exception.pending) {
4145 		if (block_nested_exceptions)
4146 			return -EBUSY;
4147 		goto no_vmexit;
4148 	}
4149 
4150 	if (nested_vmx_preemption_timer_pending(vcpu)) {
4151 		if (block_nested_events)
4152 			return -EBUSY;
4153 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
4154 		return 0;
4155 	}
4156 
4157 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
4158 		if (block_nested_events)
4159 			return -EBUSY;
4160 		goto no_vmexit;
4161 	}
4162 
4163 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
4164 		if (block_nested_events)
4165 			return -EBUSY;
4166 		if (!nested_exit_on_nmi(vcpu))
4167 			goto no_vmexit;
4168 
4169 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
4170 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
4171 				  INTR_INFO_VALID_MASK, 0);
4172 		/*
4173 		 * The NMI-triggered VM exit counts as injection:
4174 		 * clear this one and block further NMIs.
4175 		 */
4176 		vcpu->arch.nmi_pending = 0;
4177 		vmx_set_nmi_mask(vcpu, true);
4178 		return 0;
4179 	}
4180 
4181 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
4182 		if (block_nested_events)
4183 			return -EBUSY;
4184 		if (!nested_exit_on_intr(vcpu))
4185 			goto no_vmexit;
4186 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
4187 		return 0;
4188 	}
4189 
4190 no_vmexit:
4191 	return vmx_complete_nested_posted_interrupt(vcpu);
4192 }
4193 
4194 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
4195 {
4196 	ktime_t remaining =
4197 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
4198 	u64 value;
4199 
4200 	if (ktime_to_ns(remaining) <= 0)
4201 		return 0;
4202 
4203 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
4204 	do_div(value, 1000000);
4205 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
4206 }
4207 
4208 static bool is_vmcs12_ext_field(unsigned long field)
4209 {
4210 	switch (field) {
4211 	case GUEST_ES_SELECTOR:
4212 	case GUEST_CS_SELECTOR:
4213 	case GUEST_SS_SELECTOR:
4214 	case GUEST_DS_SELECTOR:
4215 	case GUEST_FS_SELECTOR:
4216 	case GUEST_GS_SELECTOR:
4217 	case GUEST_LDTR_SELECTOR:
4218 	case GUEST_TR_SELECTOR:
4219 	case GUEST_ES_LIMIT:
4220 	case GUEST_CS_LIMIT:
4221 	case GUEST_SS_LIMIT:
4222 	case GUEST_DS_LIMIT:
4223 	case GUEST_FS_LIMIT:
4224 	case GUEST_GS_LIMIT:
4225 	case GUEST_LDTR_LIMIT:
4226 	case GUEST_TR_LIMIT:
4227 	case GUEST_GDTR_LIMIT:
4228 	case GUEST_IDTR_LIMIT:
4229 	case GUEST_ES_AR_BYTES:
4230 	case GUEST_DS_AR_BYTES:
4231 	case GUEST_FS_AR_BYTES:
4232 	case GUEST_GS_AR_BYTES:
4233 	case GUEST_LDTR_AR_BYTES:
4234 	case GUEST_TR_AR_BYTES:
4235 	case GUEST_ES_BASE:
4236 	case GUEST_CS_BASE:
4237 	case GUEST_SS_BASE:
4238 	case GUEST_DS_BASE:
4239 	case GUEST_FS_BASE:
4240 	case GUEST_GS_BASE:
4241 	case GUEST_LDTR_BASE:
4242 	case GUEST_TR_BASE:
4243 	case GUEST_GDTR_BASE:
4244 	case GUEST_IDTR_BASE:
4245 	case GUEST_PENDING_DBG_EXCEPTIONS:
4246 	case GUEST_BNDCFGS:
4247 		return true;
4248 	default:
4249 		break;
4250 	}
4251 
4252 	return false;
4253 }
4254 
4255 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4256 				       struct vmcs12 *vmcs12)
4257 {
4258 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4259 
4260 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4261 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4262 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4263 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4264 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4265 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4266 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4267 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4268 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4269 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4270 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4271 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4272 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4273 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4274 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4275 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4276 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4277 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4278 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4279 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4280 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4281 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4282 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4283 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4284 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4285 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4286 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4287 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4288 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4289 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4290 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4291 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4292 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4293 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4294 	vmcs12->guest_pending_dbg_exceptions =
4295 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4296 
4297 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4298 }
4299 
4300 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4301 				       struct vmcs12 *vmcs12)
4302 {
4303 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4304 	int cpu;
4305 
4306 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4307 		return;
4308 
4309 
4310 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4311 
4312 	cpu = get_cpu();
4313 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4314 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4315 
4316 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4317 
4318 	vmx->loaded_vmcs = &vmx->vmcs01;
4319 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4320 	put_cpu();
4321 }
4322 
4323 /*
4324  * Update the guest state fields of vmcs12 to reflect changes that
4325  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4326  * VM-entry controls is also updated, since this is really a guest
4327  * state bit.)
4328  */
4329 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4330 {
4331 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4332 
4333 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4334 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4335 
4336 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4337 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4338 
4339 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4340 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4341 
4342 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4343 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4344 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4345 
4346 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4347 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4348 
4349 	vmcs12->guest_interruptibility_info =
4350 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4351 
4352 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4353 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4354 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4355 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4356 	else
4357 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4358 
4359 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4360 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4361 	    !vmx->nested.nested_run_pending)
4362 		vmcs12->vmx_preemption_timer_value =
4363 			vmx_get_preemption_timer_value(vcpu);
4364 
4365 	/*
4366 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4367 	 * own CR3 without exiting. If it has changed it, we must keep it.
4368 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4369 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4370 	 *
4371 	 * Additionally, restore L2's PDPTR to vmcs12.
4372 	 */
4373 	if (enable_ept) {
4374 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4375 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4376 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4377 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4378 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4379 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4380 		}
4381 	}
4382 
4383 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4384 
4385 	if (nested_cpu_has_vid(vmcs12))
4386 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4387 
4388 	vmcs12->vm_entry_controls =
4389 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4390 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4391 
4392 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4393 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4394 
4395 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4396 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4397 }
4398 
4399 /*
4400  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4401  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4402  * and this function updates it to reflect the changes to the guest state while
4403  * L2 was running (and perhaps made some exits which were handled directly by L0
4404  * without going back to L1), and to reflect the exit reason.
4405  * Note that we do not have to copy here all VMCS fields, just those that
4406  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4407  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4408  * which already writes to vmcs12 directly.
4409  */
4410 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4411 			   u32 vm_exit_reason, u32 exit_intr_info,
4412 			   unsigned long exit_qualification)
4413 {
4414 	/* update exit information fields: */
4415 	vmcs12->vm_exit_reason = vm_exit_reason;
4416 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4417 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4418 	vmcs12->exit_qualification = exit_qualification;
4419 
4420 	/*
4421 	 * On VM-Exit due to a failed VM-Entry, the VMCS isn't marked launched
4422 	 * and only EXIT_REASON and EXIT_QUALIFICATION are updated, all other
4423 	 * exit info fields are unmodified.
4424 	 */
4425 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4426 		vmcs12->launch_state = 1;
4427 
4428 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4429 		 * instead of reading the real value. */
4430 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4431 
4432 		/*
4433 		 * Transfer the event that L0 or L1 may wanted to inject into
4434 		 * L2 to IDT_VECTORING_INFO_FIELD.
4435 		 */
4436 		vmcs12_save_pending_event(vcpu, vmcs12,
4437 					  vm_exit_reason, exit_intr_info);
4438 
4439 		vmcs12->vm_exit_intr_info = exit_intr_info;
4440 		vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4441 		vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4442 
4443 		/*
4444 		 * According to spec, there's no need to store the guest's
4445 		 * MSRs if the exit is due to a VM-entry failure that occurs
4446 		 * during or after loading the guest state. Since this exit
4447 		 * does not fall in that category, we need to save the MSRs.
4448 		 */
4449 		if (nested_vmx_store_msr(vcpu,
4450 					 vmcs12->vm_exit_msr_store_addr,
4451 					 vmcs12->vm_exit_msr_store_count))
4452 			nested_vmx_abort(vcpu,
4453 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4454 	}
4455 }
4456 
4457 /*
4458  * A part of what we need to when the nested L2 guest exits and we want to
4459  * run its L1 parent, is to reset L1's guest state to the host state specified
4460  * in vmcs12.
4461  * This function is to be called not only on normal nested exit, but also on
4462  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4463  * Failures During or After Loading Guest State").
4464  * This function should be called when the active VMCS is L1's (vmcs01).
4465  */
4466 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4467 				   struct vmcs12 *vmcs12)
4468 {
4469 	enum vm_entry_failure_code ignored;
4470 	struct kvm_segment seg;
4471 
4472 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4473 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4474 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4475 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4476 	else
4477 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4478 	vmx_set_efer(vcpu, vcpu->arch.efer);
4479 
4480 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4481 	kvm_rip_write(vcpu, vmcs12->host_rip);
4482 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4483 	vmx_set_interrupt_shadow(vcpu, 0);
4484 
4485 	/*
4486 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4487 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4488 	 *
4489 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4490 	 * (KVM doesn't change it);
4491 	 */
4492 	vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4493 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4494 
4495 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4496 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4497 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4498 
4499 	nested_ept_uninit_mmu_context(vcpu);
4500 
4501 	/*
4502 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4503 	 * couldn't have changed.
4504 	 */
4505 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4506 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4507 
4508 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4509 
4510 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4511 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4512 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4513 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4514 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4515 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4516 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4517 
4518 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4519 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4520 		vmcs_write64(GUEST_BNDCFGS, 0);
4521 
4522 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4523 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4524 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4525 	}
4526 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
4527 	    intel_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)))
4528 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4529 					 vmcs12->host_ia32_perf_global_ctrl));
4530 
4531 	/* Set L1 segment info according to Intel SDM
4532 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4533 	seg = (struct kvm_segment) {
4534 		.base = 0,
4535 		.limit = 0xFFFFFFFF,
4536 		.selector = vmcs12->host_cs_selector,
4537 		.type = 11,
4538 		.present = 1,
4539 		.s = 1,
4540 		.g = 1
4541 	};
4542 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4543 		seg.l = 1;
4544 	else
4545 		seg.db = 1;
4546 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4547 	seg = (struct kvm_segment) {
4548 		.base = 0,
4549 		.limit = 0xFFFFFFFF,
4550 		.type = 3,
4551 		.present = 1,
4552 		.s = 1,
4553 		.db = 1,
4554 		.g = 1
4555 	};
4556 	seg.selector = vmcs12->host_ds_selector;
4557 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4558 	seg.selector = vmcs12->host_es_selector;
4559 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4560 	seg.selector = vmcs12->host_ss_selector;
4561 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4562 	seg.selector = vmcs12->host_fs_selector;
4563 	seg.base = vmcs12->host_fs_base;
4564 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4565 	seg.selector = vmcs12->host_gs_selector;
4566 	seg.base = vmcs12->host_gs_base;
4567 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4568 	seg = (struct kvm_segment) {
4569 		.base = vmcs12->host_tr_base,
4570 		.limit = 0x67,
4571 		.selector = vmcs12->host_tr_selector,
4572 		.type = 11,
4573 		.present = 1
4574 	};
4575 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4576 
4577 	memset(&seg, 0, sizeof(seg));
4578 	seg.unusable = 1;
4579 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4580 
4581 	kvm_set_dr(vcpu, 7, 0x400);
4582 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4583 
4584 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4585 				vmcs12->vm_exit_msr_load_count))
4586 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4587 
4588 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4589 }
4590 
4591 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4592 {
4593 	struct vmx_uret_msr *efer_msr;
4594 	unsigned int i;
4595 
4596 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4597 		return vmcs_read64(GUEST_IA32_EFER);
4598 
4599 	if (cpu_has_load_ia32_efer())
4600 		return host_efer;
4601 
4602 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4603 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4604 			return vmx->msr_autoload.guest.val[i].value;
4605 	}
4606 
4607 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4608 	if (efer_msr)
4609 		return efer_msr->data;
4610 
4611 	return host_efer;
4612 }
4613 
4614 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4615 {
4616 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4617 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4618 	struct vmx_msr_entry g, h;
4619 	gpa_t gpa;
4620 	u32 i, j;
4621 
4622 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4623 
4624 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4625 		/*
4626 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4627 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4628 		 * and vcpu->arch.dr7 is not squirreled away before the
4629 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4630 		 */
4631 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4632 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4633 		else
4634 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4635 	}
4636 
4637 	/*
4638 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4639 	 * handle a variety of side effects to KVM's software model.
4640 	 */
4641 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4642 
4643 	vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4644 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4645 
4646 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4647 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4648 
4649 	nested_ept_uninit_mmu_context(vcpu);
4650 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4651 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4652 
4653 	/*
4654 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4655 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4656 	 * VMFail, like everything else we just need to ensure our
4657 	 * software model is up-to-date.
4658 	 */
4659 	if (enable_ept && is_pae_paging(vcpu))
4660 		ept_save_pdptrs(vcpu);
4661 
4662 	kvm_mmu_reset_context(vcpu);
4663 
4664 	/*
4665 	 * This nasty bit of open coding is a compromise between blindly
4666 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4667 	 * of VMFail), leaving the nested VM's MSRs in the software model
4668 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4669 	 * expensive since the lists are unbound by hardware).  For each
4670 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4671 	 * list, reload it from the exit load list if it exists and differs
4672 	 * from the guest value.  The intent is to stuff host state as
4673 	 * silently as possible, not to fully process the exit load list.
4674 	 */
4675 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4676 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4677 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4678 			pr_debug_ratelimited(
4679 				"%s read MSR index failed (%u, 0x%08llx)\n",
4680 				__func__, i, gpa);
4681 			goto vmabort;
4682 		}
4683 
4684 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4685 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4686 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4687 				pr_debug_ratelimited(
4688 					"%s read MSR failed (%u, 0x%08llx)\n",
4689 					__func__, j, gpa);
4690 				goto vmabort;
4691 			}
4692 			if (h.index != g.index)
4693 				continue;
4694 			if (h.value == g.value)
4695 				break;
4696 
4697 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4698 				pr_debug_ratelimited(
4699 					"%s check failed (%u, 0x%x, 0x%x)\n",
4700 					__func__, j, h.index, h.reserved);
4701 				goto vmabort;
4702 			}
4703 
4704 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4705 				pr_debug_ratelimited(
4706 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4707 					__func__, j, h.index, h.value);
4708 				goto vmabort;
4709 			}
4710 		}
4711 	}
4712 
4713 	return;
4714 
4715 vmabort:
4716 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4717 }
4718 
4719 /*
4720  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4721  * and modify vmcs12 to make it see what it would expect to see there if
4722  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4723  */
4724 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4725 		       u32 exit_intr_info, unsigned long exit_qualification)
4726 {
4727 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4728 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4729 
4730 	/* Pending MTF traps are discarded on VM-Exit. */
4731 	vmx->nested.mtf_pending = false;
4732 
4733 	/* trying to cancel vmlaunch/vmresume is a bug */
4734 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4735 
4736 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4737 		/*
4738 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4739 		 * Enlightened VMCS after migration and we still need to
4740 		 * do that when something is forcing L2->L1 exit prior to
4741 		 * the first L2 run.
4742 		 */
4743 		(void)nested_get_evmcs_page(vcpu);
4744 	}
4745 
4746 	/* Service pending TLB flush requests for L2 before switching to L1. */
4747 	kvm_service_local_tlb_flush_requests(vcpu);
4748 
4749 	/*
4750 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4751 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4752 	 * up-to-date before switching to L1.
4753 	 */
4754 	if (enable_ept && is_pae_paging(vcpu))
4755 		vmx_ept_load_pdptrs(vcpu);
4756 
4757 	leave_guest_mode(vcpu);
4758 
4759 	if (nested_cpu_has_preemption_timer(vmcs12))
4760 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4761 
4762 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4763 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4764 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4765 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4766 	}
4767 
4768 	if (likely(!vmx->fail)) {
4769 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4770 
4771 		if (vm_exit_reason != -1)
4772 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4773 				       exit_intr_info, exit_qualification);
4774 
4775 		/*
4776 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4777 		 * also be used to capture vmcs12 cache as part of
4778 		 * capturing nVMX state for snapshot (migration).
4779 		 *
4780 		 * Otherwise, this flush will dirty guest memory at a
4781 		 * point it is already assumed by user-space to be
4782 		 * immutable.
4783 		 */
4784 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4785 	} else {
4786 		/*
4787 		 * The only expected VM-instruction error is "VM entry with
4788 		 * invalid control field(s)." Anything else indicates a
4789 		 * problem with L0.  And we should never get here with a
4790 		 * VMFail of any type if early consistency checks are enabled.
4791 		 */
4792 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4793 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4794 		WARN_ON_ONCE(nested_early_check);
4795 	}
4796 
4797 	/*
4798 	 * Drop events/exceptions that were queued for re-injection to L2
4799 	 * (picked up via vmx_complete_interrupts()), as well as exceptions
4800 	 * that were pending for L2.  Note, this must NOT be hoisted above
4801 	 * prepare_vmcs12(), events/exceptions queued for re-injection need to
4802 	 * be captured in vmcs12 (see vmcs12_save_pending_event()).
4803 	 */
4804 	vcpu->arch.nmi_injected = false;
4805 	kvm_clear_exception_queue(vcpu);
4806 	kvm_clear_interrupt_queue(vcpu);
4807 
4808 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4809 
4810 	/*
4811 	 * If IBRS is advertised to the vCPU, KVM must flush the indirect
4812 	 * branch predictors when transitioning from L2 to L1, as L1 expects
4813 	 * hardware (KVM in this case) to provide separate predictor modes.
4814 	 * Bare metal isolates VMX root (host) from VMX non-root (guest), but
4815 	 * doesn't isolate different VMCSs, i.e. in this case, doesn't provide
4816 	 * separate modes for L2 vs L1.
4817 	 */
4818 	if (guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
4819 		indirect_branch_prediction_barrier();
4820 
4821 	/* Update any VMCS fields that might have changed while L2 ran */
4822 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4823 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4824 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4825 	if (kvm_caps.has_tsc_control)
4826 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4827 
4828 	if (vmx->nested.l1_tpr_threshold != -1)
4829 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4830 
4831 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4832 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4833 		vmx_set_virtual_apic_mode(vcpu);
4834 	}
4835 
4836 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4837 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4838 		vmx_update_cpu_dirty_logging(vcpu);
4839 	}
4840 
4841 	/* Unpin physical memory we referred to in vmcs02 */
4842 	kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
4843 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4844 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4845 	vmx->nested.pi_desc = NULL;
4846 
4847 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4848 		vmx->nested.reload_vmcs01_apic_access_page = false;
4849 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4850 	}
4851 
4852 	if (vmx->nested.update_vmcs01_apicv_status) {
4853 		vmx->nested.update_vmcs01_apicv_status = false;
4854 		kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
4855 	}
4856 
4857 	if ((vm_exit_reason != -1) &&
4858 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4859 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4860 
4861 	/* in case we halted in L2 */
4862 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4863 
4864 	if (likely(!vmx->fail)) {
4865 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4866 		    nested_exit_intr_ack_set(vcpu)) {
4867 			int irq = kvm_cpu_get_interrupt(vcpu);
4868 			WARN_ON(irq < 0);
4869 			vmcs12->vm_exit_intr_info = irq |
4870 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4871 		}
4872 
4873 		if (vm_exit_reason != -1)
4874 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4875 						       vmcs12->exit_qualification,
4876 						       vmcs12->idt_vectoring_info_field,
4877 						       vmcs12->vm_exit_intr_info,
4878 						       vmcs12->vm_exit_intr_error_code,
4879 						       KVM_ISA_VMX);
4880 
4881 		load_vmcs12_host_state(vcpu, vmcs12);
4882 
4883 		return;
4884 	}
4885 
4886 	/*
4887 	 * After an early L2 VM-entry failure, we're now back
4888 	 * in L1 which thinks it just finished a VMLAUNCH or
4889 	 * VMRESUME instruction, so we need to set the failure
4890 	 * flag and the VM-instruction error field of the VMCS
4891 	 * accordingly, and skip the emulated instruction.
4892 	 */
4893 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4894 
4895 	/*
4896 	 * Restore L1's host state to KVM's software model.  We're here
4897 	 * because a consistency check was caught by hardware, which
4898 	 * means some amount of guest state has been propagated to KVM's
4899 	 * model and needs to be unwound to the host's state.
4900 	 */
4901 	nested_vmx_restore_host_state(vcpu);
4902 
4903 	vmx->fail = 0;
4904 }
4905 
4906 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4907 {
4908 	kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4909 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4910 }
4911 
4912 /*
4913  * Decode the memory-address operand of a vmx instruction, as recorded on an
4914  * exit caused by such an instruction (run by a guest hypervisor).
4915  * On success, returns 0. When the operand is invalid, returns 1 and throws
4916  * #UD, #GP, or #SS.
4917  */
4918 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4919 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4920 {
4921 	gva_t off;
4922 	bool exn;
4923 	struct kvm_segment s;
4924 
4925 	/*
4926 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4927 	 * Execution", on an exit, vmx_instruction_info holds most of the
4928 	 * addressing components of the operand. Only the displacement part
4929 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4930 	 * For how an actual address is calculated from all these components,
4931 	 * refer to Vol. 1, "Operand Addressing".
4932 	 */
4933 	int  scaling = vmx_instruction_info & 3;
4934 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4935 	bool is_reg = vmx_instruction_info & (1u << 10);
4936 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4937 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4938 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4939 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4940 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4941 
4942 	if (is_reg) {
4943 		kvm_queue_exception(vcpu, UD_VECTOR);
4944 		return 1;
4945 	}
4946 
4947 	/* Addr = segment_base + offset */
4948 	/* offset = base + [index * scale] + displacement */
4949 	off = exit_qualification; /* holds the displacement */
4950 	if (addr_size == 1)
4951 		off = (gva_t)sign_extend64(off, 31);
4952 	else if (addr_size == 0)
4953 		off = (gva_t)sign_extend64(off, 15);
4954 	if (base_is_valid)
4955 		off += kvm_register_read(vcpu, base_reg);
4956 	if (index_is_valid)
4957 		off += kvm_register_read(vcpu, index_reg) << scaling;
4958 	vmx_get_segment(vcpu, &s, seg_reg);
4959 
4960 	/*
4961 	 * The effective address, i.e. @off, of a memory operand is truncated
4962 	 * based on the address size of the instruction.  Note that this is
4963 	 * the *effective address*, i.e. the address prior to accounting for
4964 	 * the segment's base.
4965 	 */
4966 	if (addr_size == 1) /* 32 bit */
4967 		off &= 0xffffffff;
4968 	else if (addr_size == 0) /* 16 bit */
4969 		off &= 0xffff;
4970 
4971 	/* Checks for #GP/#SS exceptions. */
4972 	exn = false;
4973 	if (is_long_mode(vcpu)) {
4974 		/*
4975 		 * The virtual/linear address is never truncated in 64-bit
4976 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4977 		 * address when using FS/GS with a non-zero base.
4978 		 */
4979 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4980 			*ret = s.base + off;
4981 		else
4982 			*ret = off;
4983 
4984 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4985 		 * non-canonical form. This is the only check on the memory
4986 		 * destination for long mode!
4987 		 */
4988 		exn = is_noncanonical_address(*ret, vcpu);
4989 	} else {
4990 		/*
4991 		 * When not in long mode, the virtual/linear address is
4992 		 * unconditionally truncated to 32 bits regardless of the
4993 		 * address size.
4994 		 */
4995 		*ret = (s.base + off) & 0xffffffff;
4996 
4997 		/* Protected mode: apply checks for segment validity in the
4998 		 * following order:
4999 		 * - segment type check (#GP(0) may be thrown)
5000 		 * - usability check (#GP(0)/#SS(0))
5001 		 * - limit check (#GP(0)/#SS(0))
5002 		 */
5003 		if (wr)
5004 			/* #GP(0) if the destination operand is located in a
5005 			 * read-only data segment or any code segment.
5006 			 */
5007 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
5008 		else
5009 			/* #GP(0) if the source operand is located in an
5010 			 * execute-only code segment
5011 			 */
5012 			exn = ((s.type & 0xa) == 8);
5013 		if (exn) {
5014 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5015 			return 1;
5016 		}
5017 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
5018 		 */
5019 		exn = (s.unusable != 0);
5020 
5021 		/*
5022 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
5023 		 * outside the segment limit.  All CPUs that support VMX ignore
5024 		 * limit checks for flat segments, i.e. segments with base==0,
5025 		 * limit==0xffffffff and of type expand-up data or code.
5026 		 */
5027 		if (!(s.base == 0 && s.limit == 0xffffffff &&
5028 		     ((s.type & 8) || !(s.type & 4))))
5029 			exn = exn || ((u64)off + len - 1 > s.limit);
5030 	}
5031 	if (exn) {
5032 		kvm_queue_exception_e(vcpu,
5033 				      seg_reg == VCPU_SREG_SS ?
5034 						SS_VECTOR : GP_VECTOR,
5035 				      0);
5036 		return 1;
5037 	}
5038 
5039 	return 0;
5040 }
5041 
5042 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
5043 				int *ret)
5044 {
5045 	gva_t gva;
5046 	struct x86_exception e;
5047 	int r;
5048 
5049 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5050 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
5051 				sizeof(*vmpointer), &gva)) {
5052 		*ret = 1;
5053 		return -EINVAL;
5054 	}
5055 
5056 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
5057 	if (r != X86EMUL_CONTINUE) {
5058 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
5059 		return -EINVAL;
5060 	}
5061 
5062 	return 0;
5063 }
5064 
5065 /*
5066  * Allocate a shadow VMCS and associate it with the currently loaded
5067  * VMCS, unless such a shadow VMCS already exists. The newly allocated
5068  * VMCS is also VMCLEARed, so that it is ready for use.
5069  */
5070 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
5071 {
5072 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5073 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
5074 
5075 	/*
5076 	 * KVM allocates a shadow VMCS only when L1 executes VMXON and frees it
5077 	 * when L1 executes VMXOFF or the vCPU is forced out of nested
5078 	 * operation.  VMXON faults if the CPU is already post-VMXON, so it
5079 	 * should be impossible to already have an allocated shadow VMCS.  KVM
5080 	 * doesn't support virtualization of VMCS shadowing, so vmcs01 should
5081 	 * always be the loaded VMCS.
5082 	 */
5083 	if (WARN_ON(loaded_vmcs != &vmx->vmcs01 || loaded_vmcs->shadow_vmcs))
5084 		return loaded_vmcs->shadow_vmcs;
5085 
5086 	loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
5087 	if (loaded_vmcs->shadow_vmcs)
5088 		vmcs_clear(loaded_vmcs->shadow_vmcs);
5089 
5090 	return loaded_vmcs->shadow_vmcs;
5091 }
5092 
5093 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
5094 {
5095 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5096 	int r;
5097 
5098 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
5099 	if (r < 0)
5100 		goto out_vmcs02;
5101 
5102 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5103 	if (!vmx->nested.cached_vmcs12)
5104 		goto out_cached_vmcs12;
5105 
5106 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
5107 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5108 	if (!vmx->nested.cached_shadow_vmcs12)
5109 		goto out_cached_shadow_vmcs12;
5110 
5111 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
5112 		goto out_shadow_vmcs;
5113 
5114 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
5115 		     HRTIMER_MODE_ABS_PINNED);
5116 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
5117 
5118 	vmx->nested.vpid02 = allocate_vpid();
5119 
5120 	vmx->nested.vmcs02_initialized = false;
5121 	vmx->nested.vmxon = true;
5122 
5123 	if (vmx_pt_mode_is_host_guest()) {
5124 		vmx->pt_desc.guest.ctl = 0;
5125 		pt_update_intercept_for_msr(vcpu);
5126 	}
5127 
5128 	return 0;
5129 
5130 out_shadow_vmcs:
5131 	kfree(vmx->nested.cached_shadow_vmcs12);
5132 
5133 out_cached_shadow_vmcs12:
5134 	kfree(vmx->nested.cached_vmcs12);
5135 
5136 out_cached_vmcs12:
5137 	free_loaded_vmcs(&vmx->nested.vmcs02);
5138 
5139 out_vmcs02:
5140 	return -ENOMEM;
5141 }
5142 
5143 /* Emulate the VMXON instruction. */
5144 static int handle_vmxon(struct kvm_vcpu *vcpu)
5145 {
5146 	int ret;
5147 	gpa_t vmptr;
5148 	uint32_t revision;
5149 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5150 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
5151 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
5152 
5153 	/*
5154 	 * Manually check CR4.VMXE checks, KVM must force CR4.VMXE=1 to enter
5155 	 * the guest and so cannot rely on hardware to perform the check,
5156 	 * which has higher priority than VM-Exit (see Intel SDM's pseudocode
5157 	 * for VMXON).
5158 	 *
5159 	 * Rely on hardware for the other pre-VM-Exit checks, CR0.PE=1, !VM86
5160 	 * and !COMPATIBILITY modes.  For an unrestricted guest, KVM doesn't
5161 	 * force any of the relevant guest state.  For a restricted guest, KVM
5162 	 * does force CR0.PE=1, but only to also force VM86 in order to emulate
5163 	 * Real Mode, and so there's no need to check CR0.PE manually.
5164 	 */
5165 	if (!kvm_is_cr4_bit_set(vcpu, X86_CR4_VMXE)) {
5166 		kvm_queue_exception(vcpu, UD_VECTOR);
5167 		return 1;
5168 	}
5169 
5170 	/*
5171 	 * The CPL is checked for "not in VMX operation" and for "in VMX root",
5172 	 * and has higher priority than the VM-Fail due to being post-VMXON,
5173 	 * i.e. VMXON #GPs outside of VMX non-root if CPL!=0.  In VMX non-root,
5174 	 * VMXON causes VM-Exit and KVM unconditionally forwards VMXON VM-Exits
5175 	 * from L2 to L1, i.e. there's no need to check for the vCPU being in
5176 	 * VMX non-root.
5177 	 *
5178 	 * Forwarding the VM-Exit unconditionally, i.e. without performing the
5179 	 * #UD checks (see above), is functionally ok because KVM doesn't allow
5180 	 * L1 to run L2 without CR4.VMXE=0, and because KVM never modifies L2's
5181 	 * CR0 or CR4, i.e. it's L2's responsibility to emulate #UDs that are
5182 	 * missed by hardware due to shadowing CR0 and/or CR4.
5183 	 */
5184 	if (vmx_get_cpl(vcpu)) {
5185 		kvm_inject_gp(vcpu, 0);
5186 		return 1;
5187 	}
5188 
5189 	if (vmx->nested.vmxon)
5190 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5191 
5192 	/*
5193 	 * Invalid CR0/CR4 generates #GP.  These checks are performed if and
5194 	 * only if the vCPU isn't already in VMX operation, i.e. effectively
5195 	 * have lower priority than the VM-Fail above.
5196 	 */
5197 	if (!nested_host_cr0_valid(vcpu, kvm_read_cr0(vcpu)) ||
5198 	    !nested_host_cr4_valid(vcpu, kvm_read_cr4(vcpu))) {
5199 		kvm_inject_gp(vcpu, 0);
5200 		return 1;
5201 	}
5202 
5203 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5204 			!= VMXON_NEEDED_FEATURES) {
5205 		kvm_inject_gp(vcpu, 0);
5206 		return 1;
5207 	}
5208 
5209 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
5210 		return ret;
5211 
5212 	/*
5213 	 * SDM 3: 24.11.5
5214 	 * The first 4 bytes of VMXON region contain the supported
5215 	 * VMCS revision identifier
5216 	 *
5217 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
5218 	 * which replaces physical address width with 32
5219 	 */
5220 	if (!page_address_valid(vcpu, vmptr))
5221 		return nested_vmx_failInvalid(vcpu);
5222 
5223 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
5224 	    revision != VMCS12_REVISION)
5225 		return nested_vmx_failInvalid(vcpu);
5226 
5227 	vmx->nested.vmxon_ptr = vmptr;
5228 	ret = enter_vmx_operation(vcpu);
5229 	if (ret)
5230 		return ret;
5231 
5232 	return nested_vmx_succeed(vcpu);
5233 }
5234 
5235 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
5236 {
5237 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5238 
5239 	if (vmx->nested.current_vmptr == INVALID_GPA)
5240 		return;
5241 
5242 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
5243 
5244 	if (enable_shadow_vmcs) {
5245 		/* copy to memory all shadowed fields in case
5246 		   they were modified */
5247 		copy_shadow_to_vmcs12(vmx);
5248 		vmx_disable_shadow_vmcs(vmx);
5249 	}
5250 	vmx->nested.posted_intr_nv = -1;
5251 
5252 	/* Flush VMCS12 to guest memory */
5253 	kvm_vcpu_write_guest_page(vcpu,
5254 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
5255 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5256 
5257 	kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5258 
5259 	vmx->nested.current_vmptr = INVALID_GPA;
5260 }
5261 
5262 /* Emulate the VMXOFF instruction */
5263 static int handle_vmxoff(struct kvm_vcpu *vcpu)
5264 {
5265 	if (!nested_vmx_check_permission(vcpu))
5266 		return 1;
5267 
5268 	free_nested(vcpu);
5269 
5270 	if (kvm_apic_has_pending_init_or_sipi(vcpu))
5271 		kvm_make_request(KVM_REQ_EVENT, vcpu);
5272 
5273 	return nested_vmx_succeed(vcpu);
5274 }
5275 
5276 /* Emulate the VMCLEAR instruction */
5277 static int handle_vmclear(struct kvm_vcpu *vcpu)
5278 {
5279 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5280 	u32 zero = 0;
5281 	gpa_t vmptr;
5282 	int r;
5283 
5284 	if (!nested_vmx_check_permission(vcpu))
5285 		return 1;
5286 
5287 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5288 		return r;
5289 
5290 	if (!page_address_valid(vcpu, vmptr))
5291 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5292 
5293 	if (vmptr == vmx->nested.vmxon_ptr)
5294 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5295 
5296 	/*
5297 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
5298 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
5299 	 * way to distinguish it from VMCS12) and we must not corrupt it by
5300 	 * writing to the non-existent 'launch_state' field. The area doesn't
5301 	 * have to be the currently active EVMCS on the calling CPU and there's
5302 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5303 	 * state. It is possible that the area will stay mapped as
5304 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5305 	 */
5306 	if (likely(!guest_cpuid_has_evmcs(vcpu) ||
5307 		   !evmptr_is_valid(nested_get_evmptr(vcpu)))) {
5308 		if (vmptr == vmx->nested.current_vmptr)
5309 			nested_release_vmcs12(vcpu);
5310 
5311 		/*
5312 		 * Silently ignore memory errors on VMCLEAR, Intel's pseudocode
5313 		 * for VMCLEAR includes a "ensure that data for VMCS referenced
5314 		 * by the operand is in memory" clause that guards writes to
5315 		 * memory, i.e. doing nothing for I/O is architecturally valid.
5316 		 *
5317 		 * FIXME: Suppress failures if and only if no memslot is found,
5318 		 * i.e. exit to userspace if __copy_to_user() fails.
5319 		 */
5320 		(void)kvm_vcpu_write_guest(vcpu,
5321 					   vmptr + offsetof(struct vmcs12,
5322 							    launch_state),
5323 					   &zero, sizeof(zero));
5324 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5325 		nested_release_evmcs(vcpu);
5326 	}
5327 
5328 	return nested_vmx_succeed(vcpu);
5329 }
5330 
5331 /* Emulate the VMLAUNCH instruction */
5332 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5333 {
5334 	return nested_vmx_run(vcpu, true);
5335 }
5336 
5337 /* Emulate the VMRESUME instruction */
5338 static int handle_vmresume(struct kvm_vcpu *vcpu)
5339 {
5340 
5341 	return nested_vmx_run(vcpu, false);
5342 }
5343 
5344 static int handle_vmread(struct kvm_vcpu *vcpu)
5345 {
5346 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5347 						    : get_vmcs12(vcpu);
5348 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5349 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5350 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5351 	struct x86_exception e;
5352 	unsigned long field;
5353 	u64 value;
5354 	gva_t gva = 0;
5355 	short offset;
5356 	int len, r;
5357 
5358 	if (!nested_vmx_check_permission(vcpu))
5359 		return 1;
5360 
5361 	/* Decode instruction info and find the field to read */
5362 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5363 
5364 	if (!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
5365 		/*
5366 		 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5367 		 * any VMREAD sets the ALU flags for VMfailInvalid.
5368 		 */
5369 		if (vmx->nested.current_vmptr == INVALID_GPA ||
5370 		    (is_guest_mode(vcpu) &&
5371 		     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5372 			return nested_vmx_failInvalid(vcpu);
5373 
5374 		offset = get_vmcs12_field_offset(field);
5375 		if (offset < 0)
5376 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5377 
5378 		if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5379 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5380 
5381 		/* Read the field, zero-extended to a u64 value */
5382 		value = vmcs12_read_any(vmcs12, field, offset);
5383 	} else {
5384 		/*
5385 		 * Hyper-V TLFS (as of 6.0b) explicitly states, that while an
5386 		 * enlightened VMCS is active VMREAD/VMWRITE instructions are
5387 		 * unsupported. Unfortunately, certain versions of Windows 11
5388 		 * don't comply with this requirement which is not enforced in
5389 		 * genuine Hyper-V. Allow VMREAD from an enlightened VMCS as a
5390 		 * workaround, as misbehaving guests will panic on VM-Fail.
5391 		 * Note, enlightened VMCS is incompatible with shadow VMCS so
5392 		 * all VMREADs from L2 should go to L1.
5393 		 */
5394 		if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5395 			return nested_vmx_failInvalid(vcpu);
5396 
5397 		offset = evmcs_field_offset(field, NULL);
5398 		if (offset < 0)
5399 			return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5400 
5401 		/* Read the field, zero-extended to a u64 value */
5402 		value = evmcs_read_any(vmx->nested.hv_evmcs, field, offset);
5403 	}
5404 
5405 	/*
5406 	 * Now copy part of this value to register or memory, as requested.
5407 	 * Note that the number of bits actually copied is 32 or 64 depending
5408 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5409 	 */
5410 	if (instr_info & BIT(10)) {
5411 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5412 	} else {
5413 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5414 		if (get_vmx_mem_address(vcpu, exit_qualification,
5415 					instr_info, true, len, &gva))
5416 			return 1;
5417 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5418 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5419 		if (r != X86EMUL_CONTINUE)
5420 			return kvm_handle_memory_failure(vcpu, r, &e);
5421 	}
5422 
5423 	return nested_vmx_succeed(vcpu);
5424 }
5425 
5426 static bool is_shadow_field_rw(unsigned long field)
5427 {
5428 	switch (field) {
5429 #define SHADOW_FIELD_RW(x, y) case x:
5430 #include "vmcs_shadow_fields.h"
5431 		return true;
5432 	default:
5433 		break;
5434 	}
5435 	return false;
5436 }
5437 
5438 static bool is_shadow_field_ro(unsigned long field)
5439 {
5440 	switch (field) {
5441 #define SHADOW_FIELD_RO(x, y) case x:
5442 #include "vmcs_shadow_fields.h"
5443 		return true;
5444 	default:
5445 		break;
5446 	}
5447 	return false;
5448 }
5449 
5450 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5451 {
5452 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5453 						    : get_vmcs12(vcpu);
5454 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5455 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5456 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5457 	struct x86_exception e;
5458 	unsigned long field;
5459 	short offset;
5460 	gva_t gva;
5461 	int len, r;
5462 
5463 	/*
5464 	 * The value to write might be 32 or 64 bits, depending on L1's long
5465 	 * mode, and eventually we need to write that into a field of several
5466 	 * possible lengths. The code below first zero-extends the value to 64
5467 	 * bit (value), and then copies only the appropriate number of
5468 	 * bits into the vmcs12 field.
5469 	 */
5470 	u64 value = 0;
5471 
5472 	if (!nested_vmx_check_permission(vcpu))
5473 		return 1;
5474 
5475 	/*
5476 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5477 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5478 	 */
5479 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5480 	    (is_guest_mode(vcpu) &&
5481 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5482 		return nested_vmx_failInvalid(vcpu);
5483 
5484 	if (instr_info & BIT(10))
5485 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5486 	else {
5487 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5488 		if (get_vmx_mem_address(vcpu, exit_qualification,
5489 					instr_info, false, len, &gva))
5490 			return 1;
5491 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5492 		if (r != X86EMUL_CONTINUE)
5493 			return kvm_handle_memory_failure(vcpu, r, &e);
5494 	}
5495 
5496 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5497 
5498 	offset = get_vmcs12_field_offset(field);
5499 	if (offset < 0)
5500 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5501 
5502 	/*
5503 	 * If the vCPU supports "VMWRITE to any supported field in the
5504 	 * VMCS," then the "read-only" fields are actually read/write.
5505 	 */
5506 	if (vmcs_field_readonly(field) &&
5507 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5508 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5509 
5510 	/*
5511 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5512 	 * vmcs12, else we may crush a field or consume a stale value.
5513 	 */
5514 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5515 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5516 
5517 	/*
5518 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5519 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5520 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5521 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5522 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5523 	 * the stripped down value, L2 sees the full value as stored by KVM).
5524 	 */
5525 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5526 		value &= 0x1f0ff;
5527 
5528 	vmcs12_write_any(vmcs12, field, offset, value);
5529 
5530 	/*
5531 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5532 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5533 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5534 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5535 	 */
5536 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5537 		/*
5538 		 * L1 can read these fields without exiting, ensure the
5539 		 * shadow VMCS is up-to-date.
5540 		 */
5541 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5542 			preempt_disable();
5543 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5544 
5545 			__vmcs_writel(field, value);
5546 
5547 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5548 			vmcs_load(vmx->loaded_vmcs->vmcs);
5549 			preempt_enable();
5550 		}
5551 		vmx->nested.dirty_vmcs12 = true;
5552 	}
5553 
5554 	return nested_vmx_succeed(vcpu);
5555 }
5556 
5557 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5558 {
5559 	vmx->nested.current_vmptr = vmptr;
5560 	if (enable_shadow_vmcs) {
5561 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5562 		vmcs_write64(VMCS_LINK_POINTER,
5563 			     __pa(vmx->vmcs01.shadow_vmcs));
5564 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5565 	}
5566 	vmx->nested.dirty_vmcs12 = true;
5567 	vmx->nested.force_msr_bitmap_recalc = true;
5568 }
5569 
5570 /* Emulate the VMPTRLD instruction */
5571 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5572 {
5573 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5574 	gpa_t vmptr;
5575 	int r;
5576 
5577 	if (!nested_vmx_check_permission(vcpu))
5578 		return 1;
5579 
5580 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5581 		return r;
5582 
5583 	if (!page_address_valid(vcpu, vmptr))
5584 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5585 
5586 	if (vmptr == vmx->nested.vmxon_ptr)
5587 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5588 
5589 	/* Forbid normal VMPTRLD if Enlightened version was used */
5590 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5591 		return 1;
5592 
5593 	if (vmx->nested.current_vmptr != vmptr) {
5594 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5595 		struct vmcs_hdr hdr;
5596 
5597 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5598 			/*
5599 			 * Reads from an unbacked page return all 1s,
5600 			 * which means that the 32 bits located at the
5601 			 * given physical address won't match the required
5602 			 * VMCS12_REVISION identifier.
5603 			 */
5604 			return nested_vmx_fail(vcpu,
5605 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5606 		}
5607 
5608 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5609 						 offsetof(struct vmcs12, hdr),
5610 						 sizeof(hdr))) {
5611 			return nested_vmx_fail(vcpu,
5612 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5613 		}
5614 
5615 		if (hdr.revision_id != VMCS12_REVISION ||
5616 		    (hdr.shadow_vmcs &&
5617 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5618 			return nested_vmx_fail(vcpu,
5619 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5620 		}
5621 
5622 		nested_release_vmcs12(vcpu);
5623 
5624 		/*
5625 		 * Load VMCS12 from guest memory since it is not already
5626 		 * cached.
5627 		 */
5628 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5629 					  VMCS12_SIZE)) {
5630 			return nested_vmx_fail(vcpu,
5631 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5632 		}
5633 
5634 		set_current_vmptr(vmx, vmptr);
5635 	}
5636 
5637 	return nested_vmx_succeed(vcpu);
5638 }
5639 
5640 /* Emulate the VMPTRST instruction */
5641 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5642 {
5643 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5644 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5645 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5646 	struct x86_exception e;
5647 	gva_t gva;
5648 	int r;
5649 
5650 	if (!nested_vmx_check_permission(vcpu))
5651 		return 1;
5652 
5653 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5654 		return 1;
5655 
5656 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5657 				true, sizeof(gpa_t), &gva))
5658 		return 1;
5659 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5660 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5661 					sizeof(gpa_t), &e);
5662 	if (r != X86EMUL_CONTINUE)
5663 		return kvm_handle_memory_failure(vcpu, r, &e);
5664 
5665 	return nested_vmx_succeed(vcpu);
5666 }
5667 
5668 /* Emulate the INVEPT instruction */
5669 static int handle_invept(struct kvm_vcpu *vcpu)
5670 {
5671 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5672 	u32 vmx_instruction_info, types;
5673 	unsigned long type, roots_to_free;
5674 	struct kvm_mmu *mmu;
5675 	gva_t gva;
5676 	struct x86_exception e;
5677 	struct {
5678 		u64 eptp, gpa;
5679 	} operand;
5680 	int i, r, gpr_index;
5681 
5682 	if (!(vmx->nested.msrs.secondary_ctls_high &
5683 	      SECONDARY_EXEC_ENABLE_EPT) ||
5684 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5685 		kvm_queue_exception(vcpu, UD_VECTOR);
5686 		return 1;
5687 	}
5688 
5689 	if (!nested_vmx_check_permission(vcpu))
5690 		return 1;
5691 
5692 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5693 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5694 	type = kvm_register_read(vcpu, gpr_index);
5695 
5696 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5697 
5698 	if (type >= 32 || !(types & (1 << type)))
5699 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5700 
5701 	/* According to the Intel VMX instruction reference, the memory
5702 	 * operand is read even if it isn't needed (e.g., for type==global)
5703 	 */
5704 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5705 			vmx_instruction_info, false, sizeof(operand), &gva))
5706 		return 1;
5707 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5708 	if (r != X86EMUL_CONTINUE)
5709 		return kvm_handle_memory_failure(vcpu, r, &e);
5710 
5711 	/*
5712 	 * Nested EPT roots are always held through guest_mmu,
5713 	 * not root_mmu.
5714 	 */
5715 	mmu = &vcpu->arch.guest_mmu;
5716 
5717 	switch (type) {
5718 	case VMX_EPT_EXTENT_CONTEXT:
5719 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5720 			return nested_vmx_fail(vcpu,
5721 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5722 
5723 		roots_to_free = 0;
5724 		if (nested_ept_root_matches(mmu->root.hpa, mmu->root.pgd,
5725 					    operand.eptp))
5726 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5727 
5728 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5729 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5730 						    mmu->prev_roots[i].pgd,
5731 						    operand.eptp))
5732 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5733 		}
5734 		break;
5735 	case VMX_EPT_EXTENT_GLOBAL:
5736 		roots_to_free = KVM_MMU_ROOTS_ALL;
5737 		break;
5738 	default:
5739 		BUG();
5740 		break;
5741 	}
5742 
5743 	if (roots_to_free)
5744 		kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
5745 
5746 	return nested_vmx_succeed(vcpu);
5747 }
5748 
5749 static int handle_invvpid(struct kvm_vcpu *vcpu)
5750 {
5751 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5752 	u32 vmx_instruction_info;
5753 	unsigned long type, types;
5754 	gva_t gva;
5755 	struct x86_exception e;
5756 	struct {
5757 		u64 vpid;
5758 		u64 gla;
5759 	} operand;
5760 	u16 vpid02;
5761 	int r, gpr_index;
5762 
5763 	if (!(vmx->nested.msrs.secondary_ctls_high &
5764 	      SECONDARY_EXEC_ENABLE_VPID) ||
5765 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5766 		kvm_queue_exception(vcpu, UD_VECTOR);
5767 		return 1;
5768 	}
5769 
5770 	if (!nested_vmx_check_permission(vcpu))
5771 		return 1;
5772 
5773 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5774 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5775 	type = kvm_register_read(vcpu, gpr_index);
5776 
5777 	types = (vmx->nested.msrs.vpid_caps &
5778 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5779 
5780 	if (type >= 32 || !(types & (1 << type)))
5781 		return nested_vmx_fail(vcpu,
5782 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5783 
5784 	/* according to the intel vmx instruction reference, the memory
5785 	 * operand is read even if it isn't needed (e.g., for type==global)
5786 	 */
5787 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5788 			vmx_instruction_info, false, sizeof(operand), &gva))
5789 		return 1;
5790 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5791 	if (r != X86EMUL_CONTINUE)
5792 		return kvm_handle_memory_failure(vcpu, r, &e);
5793 
5794 	if (operand.vpid >> 16)
5795 		return nested_vmx_fail(vcpu,
5796 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5797 
5798 	vpid02 = nested_get_vpid02(vcpu);
5799 	switch (type) {
5800 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5801 		if (!operand.vpid ||
5802 		    is_noncanonical_address(operand.gla, vcpu))
5803 			return nested_vmx_fail(vcpu,
5804 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5805 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5806 		break;
5807 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5808 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5809 		if (!operand.vpid)
5810 			return nested_vmx_fail(vcpu,
5811 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5812 		vpid_sync_context(vpid02);
5813 		break;
5814 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5815 		vpid_sync_context(vpid02);
5816 		break;
5817 	default:
5818 		WARN_ON_ONCE(1);
5819 		return kvm_skip_emulated_instruction(vcpu);
5820 	}
5821 
5822 	/*
5823 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5824 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5825 	 * roots as VPIDs are not tracked in the MMU role.
5826 	 *
5827 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5828 	 * an MMU when EPT is disabled.
5829 	 *
5830 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5831 	 */
5832 	if (!enable_ept)
5833 		kvm_mmu_free_guest_mode_roots(vcpu->kvm, &vcpu->arch.root_mmu);
5834 
5835 	return nested_vmx_succeed(vcpu);
5836 }
5837 
5838 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5839 				     struct vmcs12 *vmcs12)
5840 {
5841 	u32 index = kvm_rcx_read(vcpu);
5842 	u64 new_eptp;
5843 
5844 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5845 		return 1;
5846 	if (index >= VMFUNC_EPTP_ENTRIES)
5847 		return 1;
5848 
5849 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5850 				     &new_eptp, index * 8, 8))
5851 		return 1;
5852 
5853 	/*
5854 	 * If the (L2) guest does a vmfunc to the currently
5855 	 * active ept pointer, we don't have to do anything else
5856 	 */
5857 	if (vmcs12->ept_pointer != new_eptp) {
5858 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5859 			return 1;
5860 
5861 		vmcs12->ept_pointer = new_eptp;
5862 		nested_ept_new_eptp(vcpu);
5863 
5864 		if (!nested_cpu_has_vpid(vmcs12))
5865 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5866 	}
5867 
5868 	return 0;
5869 }
5870 
5871 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5872 {
5873 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5874 	struct vmcs12 *vmcs12;
5875 	u32 function = kvm_rax_read(vcpu);
5876 
5877 	/*
5878 	 * VMFUNC should never execute cleanly while L1 is active; KVM supports
5879 	 * VMFUNC for nested VMs, but not for L1.
5880 	 */
5881 	if (WARN_ON_ONCE(!is_guest_mode(vcpu))) {
5882 		kvm_queue_exception(vcpu, UD_VECTOR);
5883 		return 1;
5884 	}
5885 
5886 	vmcs12 = get_vmcs12(vcpu);
5887 
5888 	/*
5889 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5890 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5891 	 */
5892 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5893 		kvm_queue_exception(vcpu, UD_VECTOR);
5894 		return 1;
5895 	}
5896 
5897 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5898 		goto fail;
5899 
5900 	switch (function) {
5901 	case 0:
5902 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5903 			goto fail;
5904 		break;
5905 	default:
5906 		goto fail;
5907 	}
5908 	return kvm_skip_emulated_instruction(vcpu);
5909 
5910 fail:
5911 	/*
5912 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5913 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5914 	 * EXIT_REASON_VMFUNC as the exit reason.
5915 	 */
5916 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5917 			  vmx_get_intr_info(vcpu),
5918 			  vmx_get_exit_qual(vcpu));
5919 	return 1;
5920 }
5921 
5922 /*
5923  * Return true if an IO instruction with the specified port and size should cause
5924  * a VM-exit into L1.
5925  */
5926 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5927 				 int size)
5928 {
5929 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5930 	gpa_t bitmap, last_bitmap;
5931 	u8 b;
5932 
5933 	last_bitmap = INVALID_GPA;
5934 	b = -1;
5935 
5936 	while (size > 0) {
5937 		if (port < 0x8000)
5938 			bitmap = vmcs12->io_bitmap_a;
5939 		else if (port < 0x10000)
5940 			bitmap = vmcs12->io_bitmap_b;
5941 		else
5942 			return true;
5943 		bitmap += (port & 0x7fff) / 8;
5944 
5945 		if (last_bitmap != bitmap)
5946 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5947 				return true;
5948 		if (b & (1 << (port & 7)))
5949 			return true;
5950 
5951 		port++;
5952 		size--;
5953 		last_bitmap = bitmap;
5954 	}
5955 
5956 	return false;
5957 }
5958 
5959 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5960 				       struct vmcs12 *vmcs12)
5961 {
5962 	unsigned long exit_qualification;
5963 	unsigned short port;
5964 	int size;
5965 
5966 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5967 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5968 
5969 	exit_qualification = vmx_get_exit_qual(vcpu);
5970 
5971 	port = exit_qualification >> 16;
5972 	size = (exit_qualification & 7) + 1;
5973 
5974 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5975 }
5976 
5977 /*
5978  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5979  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5980  * disinterest in the current event (read or write a specific MSR) by using an
5981  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5982  */
5983 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5984 					struct vmcs12 *vmcs12,
5985 					union vmx_exit_reason exit_reason)
5986 {
5987 	u32 msr_index = kvm_rcx_read(vcpu);
5988 	gpa_t bitmap;
5989 
5990 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5991 		return true;
5992 
5993 	/*
5994 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5995 	 * for the four combinations of read/write and low/high MSR numbers.
5996 	 * First we need to figure out which of the four to use:
5997 	 */
5998 	bitmap = vmcs12->msr_bitmap;
5999 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
6000 		bitmap += 2048;
6001 	if (msr_index >= 0xc0000000) {
6002 		msr_index -= 0xc0000000;
6003 		bitmap += 1024;
6004 	}
6005 
6006 	/* Then read the msr_index'th bit from this bitmap: */
6007 	if (msr_index < 1024*8) {
6008 		unsigned char b;
6009 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
6010 			return true;
6011 		return 1 & (b >> (msr_index & 7));
6012 	} else
6013 		return true; /* let L1 handle the wrong parameter */
6014 }
6015 
6016 /*
6017  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6018  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6019  * intercept (via guest_host_mask etc.) the current event.
6020  */
6021 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6022 	struct vmcs12 *vmcs12)
6023 {
6024 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
6025 	int cr = exit_qualification & 15;
6026 	int reg;
6027 	unsigned long val;
6028 
6029 	switch ((exit_qualification >> 4) & 3) {
6030 	case 0: /* mov to cr */
6031 		reg = (exit_qualification >> 8) & 15;
6032 		val = kvm_register_read(vcpu, reg);
6033 		switch (cr) {
6034 		case 0:
6035 			if (vmcs12->cr0_guest_host_mask &
6036 			    (val ^ vmcs12->cr0_read_shadow))
6037 				return true;
6038 			break;
6039 		case 3:
6040 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6041 				return true;
6042 			break;
6043 		case 4:
6044 			if (vmcs12->cr4_guest_host_mask &
6045 			    (vmcs12->cr4_read_shadow ^ val))
6046 				return true;
6047 			break;
6048 		case 8:
6049 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6050 				return true;
6051 			break;
6052 		}
6053 		break;
6054 	case 2: /* clts */
6055 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6056 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
6057 			return true;
6058 		break;
6059 	case 1: /* mov from cr */
6060 		switch (cr) {
6061 		case 3:
6062 			if (vmcs12->cpu_based_vm_exec_control &
6063 			    CPU_BASED_CR3_STORE_EXITING)
6064 				return true;
6065 			break;
6066 		case 8:
6067 			if (vmcs12->cpu_based_vm_exec_control &
6068 			    CPU_BASED_CR8_STORE_EXITING)
6069 				return true;
6070 			break;
6071 		}
6072 		break;
6073 	case 3: /* lmsw */
6074 		/*
6075 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6076 		 * cr0. Other attempted changes are ignored, with no exit.
6077 		 */
6078 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6079 		if (vmcs12->cr0_guest_host_mask & 0xe &
6080 		    (val ^ vmcs12->cr0_read_shadow))
6081 			return true;
6082 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6083 		    !(vmcs12->cr0_read_shadow & 0x1) &&
6084 		    (val & 0x1))
6085 			return true;
6086 		break;
6087 	}
6088 	return false;
6089 }
6090 
6091 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
6092 					  struct vmcs12 *vmcs12)
6093 {
6094 	u32 encls_leaf;
6095 
6096 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
6097 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
6098 		return false;
6099 
6100 	encls_leaf = kvm_rax_read(vcpu);
6101 	if (encls_leaf > 62)
6102 		encls_leaf = 63;
6103 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
6104 }
6105 
6106 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
6107 	struct vmcs12 *vmcs12, gpa_t bitmap)
6108 {
6109 	u32 vmx_instruction_info;
6110 	unsigned long field;
6111 	u8 b;
6112 
6113 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
6114 		return true;
6115 
6116 	/* Decode instruction info and find the field to access */
6117 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6118 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6119 
6120 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
6121 	if (field >> 15)
6122 		return true;
6123 
6124 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
6125 		return true;
6126 
6127 	return 1 & (b >> (field & 7));
6128 }
6129 
6130 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
6131 {
6132 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
6133 
6134 	if (nested_cpu_has_mtf(vmcs12))
6135 		return true;
6136 
6137 	/*
6138 	 * An MTF VM-exit may be injected into the guest by setting the
6139 	 * interruption-type to 7 (other event) and the vector field to 0. Such
6140 	 * is the case regardless of the 'monitor trap flag' VM-execution
6141 	 * control.
6142 	 */
6143 	return entry_intr_info == (INTR_INFO_VALID_MASK
6144 				   | INTR_TYPE_OTHER_EVENT);
6145 }
6146 
6147 /*
6148  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
6149  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
6150  */
6151 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
6152 				     union vmx_exit_reason exit_reason)
6153 {
6154 	u32 intr_info;
6155 
6156 	switch ((u16)exit_reason.basic) {
6157 	case EXIT_REASON_EXCEPTION_NMI:
6158 		intr_info = vmx_get_intr_info(vcpu);
6159 		if (is_nmi(intr_info))
6160 			return true;
6161 		else if (is_page_fault(intr_info))
6162 			return vcpu->arch.apf.host_apf_flags ||
6163 			       vmx_need_pf_intercept(vcpu);
6164 		else if (is_debug(intr_info) &&
6165 			 vcpu->guest_debug &
6166 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6167 			return true;
6168 		else if (is_breakpoint(intr_info) &&
6169 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6170 			return true;
6171 		else if (is_alignment_check(intr_info) &&
6172 			 !vmx_guest_inject_ac(vcpu))
6173 			return true;
6174 		return false;
6175 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6176 		return true;
6177 	case EXIT_REASON_MCE_DURING_VMENTRY:
6178 		return true;
6179 	case EXIT_REASON_EPT_VIOLATION:
6180 		/*
6181 		 * L0 always deals with the EPT violation. If nested EPT is
6182 		 * used, and the nested mmu code discovers that the address is
6183 		 * missing in the guest EPT table (EPT12), the EPT violation
6184 		 * will be injected with nested_ept_inject_page_fault()
6185 		 */
6186 		return true;
6187 	case EXIT_REASON_EPT_MISCONFIG:
6188 		/*
6189 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
6190 		 * table (shadow on EPT) or a merged EPT table that L0 built
6191 		 * (EPT on EPT). So any problems with the structure of the
6192 		 * table is L0's fault.
6193 		 */
6194 		return true;
6195 	case EXIT_REASON_PREEMPTION_TIMER:
6196 		return true;
6197 	case EXIT_REASON_PML_FULL:
6198 		/*
6199 		 * PML is emulated for an L1 VMM and should never be enabled in
6200 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
6201 		 */
6202 		return true;
6203 	case EXIT_REASON_VMFUNC:
6204 		/* VM functions are emulated through L2->L0 vmexits. */
6205 		return true;
6206 	case EXIT_REASON_BUS_LOCK:
6207 		/*
6208 		 * At present, bus lock VM exit is never exposed to L1.
6209 		 * Handle L2's bus locks in L0 directly.
6210 		 */
6211 		return true;
6212 	case EXIT_REASON_VMCALL:
6213 		/* Hyper-V L2 TLB flush hypercall is handled by L0 */
6214 		return guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
6215 			nested_evmcs_l2_tlb_flush_enabled(vcpu) &&
6216 			kvm_hv_is_tlb_flush_hcall(vcpu);
6217 	default:
6218 		break;
6219 	}
6220 	return false;
6221 }
6222 
6223 /*
6224  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
6225  * is_guest_mode (L2).
6226  */
6227 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
6228 				     union vmx_exit_reason exit_reason)
6229 {
6230 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6231 	u32 intr_info;
6232 
6233 	switch ((u16)exit_reason.basic) {
6234 	case EXIT_REASON_EXCEPTION_NMI:
6235 		intr_info = vmx_get_intr_info(vcpu);
6236 		if (is_nmi(intr_info))
6237 			return true;
6238 		else if (is_page_fault(intr_info))
6239 			return true;
6240 		return vmcs12->exception_bitmap &
6241 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
6242 	case EXIT_REASON_EXTERNAL_INTERRUPT:
6243 		return nested_exit_on_intr(vcpu);
6244 	case EXIT_REASON_TRIPLE_FAULT:
6245 		return true;
6246 	case EXIT_REASON_INTERRUPT_WINDOW:
6247 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
6248 	case EXIT_REASON_NMI_WINDOW:
6249 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
6250 	case EXIT_REASON_TASK_SWITCH:
6251 		return true;
6252 	case EXIT_REASON_CPUID:
6253 		return true;
6254 	case EXIT_REASON_HLT:
6255 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6256 	case EXIT_REASON_INVD:
6257 		return true;
6258 	case EXIT_REASON_INVLPG:
6259 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6260 	case EXIT_REASON_RDPMC:
6261 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6262 	case EXIT_REASON_RDRAND:
6263 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
6264 	case EXIT_REASON_RDSEED:
6265 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
6266 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
6267 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6268 	case EXIT_REASON_VMREAD:
6269 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6270 			vmcs12->vmread_bitmap);
6271 	case EXIT_REASON_VMWRITE:
6272 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6273 			vmcs12->vmwrite_bitmap);
6274 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6275 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6276 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
6277 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6278 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
6279 		/*
6280 		 * VMX instructions trap unconditionally. This allows L1 to
6281 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6282 		 */
6283 		return true;
6284 	case EXIT_REASON_CR_ACCESS:
6285 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6286 	case EXIT_REASON_DR_ACCESS:
6287 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6288 	case EXIT_REASON_IO_INSTRUCTION:
6289 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
6290 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6291 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6292 	case EXIT_REASON_MSR_READ:
6293 	case EXIT_REASON_MSR_WRITE:
6294 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6295 	case EXIT_REASON_INVALID_STATE:
6296 		return true;
6297 	case EXIT_REASON_MWAIT_INSTRUCTION:
6298 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6299 	case EXIT_REASON_MONITOR_TRAP_FLAG:
6300 		return nested_vmx_exit_handled_mtf(vmcs12);
6301 	case EXIT_REASON_MONITOR_INSTRUCTION:
6302 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6303 	case EXIT_REASON_PAUSE_INSTRUCTION:
6304 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6305 			nested_cpu_has2(vmcs12,
6306 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6307 	case EXIT_REASON_MCE_DURING_VMENTRY:
6308 		return true;
6309 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6310 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6311 	case EXIT_REASON_APIC_ACCESS:
6312 	case EXIT_REASON_APIC_WRITE:
6313 	case EXIT_REASON_EOI_INDUCED:
6314 		/*
6315 		 * The controls for "virtualize APIC accesses," "APIC-
6316 		 * register virtualization," and "virtual-interrupt
6317 		 * delivery" only come from vmcs12.
6318 		 */
6319 		return true;
6320 	case EXIT_REASON_INVPCID:
6321 		return
6322 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6323 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6324 	case EXIT_REASON_WBINVD:
6325 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6326 	case EXIT_REASON_XSETBV:
6327 		return true;
6328 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6329 		/*
6330 		 * This should never happen, since it is not possible to
6331 		 * set XSS to a non-zero value---neither in L1 nor in L2.
6332 		 * If if it were, XSS would have to be checked against
6333 		 * the XSS exit bitmap in vmcs12.
6334 		 */
6335 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
6336 	case EXIT_REASON_UMWAIT:
6337 	case EXIT_REASON_TPAUSE:
6338 		return nested_cpu_has2(vmcs12,
6339 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6340 	case EXIT_REASON_ENCLS:
6341 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6342 	case EXIT_REASON_NOTIFY:
6343 		/* Notify VM exit is not exposed to L1 */
6344 		return false;
6345 	default:
6346 		return true;
6347 	}
6348 }
6349 
6350 /*
6351  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6352  * reflected into L1.
6353  */
6354 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6355 {
6356 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6357 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6358 	unsigned long exit_qual;
6359 	u32 exit_intr_info;
6360 
6361 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6362 
6363 	/*
6364 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6365 	 * has already loaded L2's state.
6366 	 */
6367 	if (unlikely(vmx->fail)) {
6368 		trace_kvm_nested_vmenter_failed(
6369 			"hardware VM-instruction error: ",
6370 			vmcs_read32(VM_INSTRUCTION_ERROR));
6371 		exit_intr_info = 0;
6372 		exit_qual = 0;
6373 		goto reflect_vmexit;
6374 	}
6375 
6376 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6377 
6378 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6379 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6380 		return false;
6381 
6382 	/* If L1 doesn't want the exit, handle it in L0. */
6383 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6384 		return false;
6385 
6386 	/*
6387 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6388 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6389 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6390 	 * interrupts are never reflected to L1 so it's a non-issue.
6391 	 */
6392 	exit_intr_info = vmx_get_intr_info(vcpu);
6393 	if (is_exception_with_error_code(exit_intr_info)) {
6394 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6395 
6396 		vmcs12->vm_exit_intr_error_code =
6397 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6398 	}
6399 	exit_qual = vmx_get_exit_qual(vcpu);
6400 
6401 reflect_vmexit:
6402 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6403 	return true;
6404 }
6405 
6406 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6407 				struct kvm_nested_state __user *user_kvm_nested_state,
6408 				u32 user_data_size)
6409 {
6410 	struct vcpu_vmx *vmx;
6411 	struct vmcs12 *vmcs12;
6412 	struct kvm_nested_state kvm_state = {
6413 		.flags = 0,
6414 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6415 		.size = sizeof(kvm_state),
6416 		.hdr.vmx.flags = 0,
6417 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6418 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6419 		.hdr.vmx.preemption_timer_deadline = 0,
6420 	};
6421 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6422 		&user_kvm_nested_state->data.vmx[0];
6423 
6424 	if (!vcpu)
6425 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6426 
6427 	vmx = to_vmx(vcpu);
6428 	vmcs12 = get_vmcs12(vcpu);
6429 
6430 	if (nested_vmx_allowed(vcpu) &&
6431 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6432 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6433 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6434 
6435 		if (vmx_has_valid_vmcs12(vcpu)) {
6436 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6437 
6438 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6439 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6440 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6441 
6442 			if (is_guest_mode(vcpu) &&
6443 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6444 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6445 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6446 		}
6447 
6448 		if (vmx->nested.smm.vmxon)
6449 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6450 
6451 		if (vmx->nested.smm.guest_mode)
6452 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6453 
6454 		if (is_guest_mode(vcpu)) {
6455 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6456 
6457 			if (vmx->nested.nested_run_pending)
6458 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6459 
6460 			if (vmx->nested.mtf_pending)
6461 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6462 
6463 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6464 			    vmx->nested.has_preemption_timer_deadline) {
6465 				kvm_state.hdr.vmx.flags |=
6466 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6467 				kvm_state.hdr.vmx.preemption_timer_deadline =
6468 					vmx->nested.preemption_timer_deadline;
6469 			}
6470 		}
6471 	}
6472 
6473 	if (user_data_size < kvm_state.size)
6474 		goto out;
6475 
6476 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6477 		return -EFAULT;
6478 
6479 	if (!vmx_has_valid_vmcs12(vcpu))
6480 		goto out;
6481 
6482 	/*
6483 	 * When running L2, the authoritative vmcs12 state is in the
6484 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6485 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6486 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6487 	 * vmcs12 state is in the vmcs12 already.
6488 	 */
6489 	if (is_guest_mode(vcpu)) {
6490 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6491 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6492 	} else  {
6493 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6494 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6495 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6496 				/*
6497 				 * L1 hypervisor is not obliged to keep eVMCS
6498 				 * clean fields data always up-to-date while
6499 				 * not in guest mode, 'hv_clean_fields' is only
6500 				 * supposed to be actual upon vmentry so we need
6501 				 * to ignore it here and do full copy.
6502 				 */
6503 				copy_enlightened_to_vmcs12(vmx, 0);
6504 			else if (enable_shadow_vmcs)
6505 				copy_shadow_to_vmcs12(vmx);
6506 		}
6507 	}
6508 
6509 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6510 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6511 
6512 	/*
6513 	 * Copy over the full allocated size of vmcs12 rather than just the size
6514 	 * of the struct.
6515 	 */
6516 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6517 		return -EFAULT;
6518 
6519 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6520 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6521 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6522 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6523 			return -EFAULT;
6524 	}
6525 out:
6526 	return kvm_state.size;
6527 }
6528 
6529 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6530 {
6531 	if (is_guest_mode(vcpu)) {
6532 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6533 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6534 	}
6535 	free_nested(vcpu);
6536 }
6537 
6538 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6539 				struct kvm_nested_state __user *user_kvm_nested_state,
6540 				struct kvm_nested_state *kvm_state)
6541 {
6542 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6543 	struct vmcs12 *vmcs12;
6544 	enum vm_entry_failure_code ignored;
6545 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6546 		&user_kvm_nested_state->data.vmx[0];
6547 	int ret;
6548 
6549 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6550 		return -EINVAL;
6551 
6552 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6553 		if (kvm_state->hdr.vmx.smm.flags)
6554 			return -EINVAL;
6555 
6556 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6557 			return -EINVAL;
6558 
6559 		/*
6560 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6561 		 * enable eVMCS capability on vCPU. However, since then
6562 		 * code was changed such that flag signals vmcs12 should
6563 		 * be copied into eVMCS in guest memory.
6564 		 *
6565 		 * To preserve backwards compatability, allow user
6566 		 * to set this flag even when there is no VMXON region.
6567 		 */
6568 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6569 			return -EINVAL;
6570 	} else {
6571 		if (!nested_vmx_allowed(vcpu))
6572 			return -EINVAL;
6573 
6574 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6575 			return -EINVAL;
6576 	}
6577 
6578 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6579 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6580 		return -EINVAL;
6581 
6582 	if (kvm_state->hdr.vmx.smm.flags &
6583 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6584 		return -EINVAL;
6585 
6586 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6587 		return -EINVAL;
6588 
6589 	/*
6590 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6591 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6592 	 * must be zero.
6593 	 */
6594 	if (is_smm(vcpu) ?
6595 		(kvm_state->flags &
6596 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6597 		: kvm_state->hdr.vmx.smm.flags)
6598 		return -EINVAL;
6599 
6600 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6601 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6602 		return -EINVAL;
6603 
6604 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6605 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6606 			return -EINVAL;
6607 
6608 	vmx_leave_nested(vcpu);
6609 
6610 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6611 		return 0;
6612 
6613 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6614 	ret = enter_vmx_operation(vcpu);
6615 	if (ret)
6616 		return ret;
6617 
6618 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6619 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6620 		/* See vmx_has_valid_vmcs12.  */
6621 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6622 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6623 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6624 			return -EINVAL;
6625 		else
6626 			return 0;
6627 	}
6628 
6629 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6630 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6631 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6632 			return -EINVAL;
6633 
6634 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6635 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6636 		/*
6637 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6638 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6639 		 * restored yet. EVMCS will be mapped from
6640 		 * nested_get_vmcs12_pages().
6641 		 */
6642 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6643 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6644 	} else {
6645 		return -EINVAL;
6646 	}
6647 
6648 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6649 		vmx->nested.smm.vmxon = true;
6650 		vmx->nested.vmxon = false;
6651 
6652 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6653 			vmx->nested.smm.guest_mode = true;
6654 	}
6655 
6656 	vmcs12 = get_vmcs12(vcpu);
6657 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6658 		return -EFAULT;
6659 
6660 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6661 		return -EINVAL;
6662 
6663 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6664 		return 0;
6665 
6666 	vmx->nested.nested_run_pending =
6667 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6668 
6669 	vmx->nested.mtf_pending =
6670 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6671 
6672 	ret = -EINVAL;
6673 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6674 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6675 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6676 
6677 		if (kvm_state->size <
6678 		    sizeof(*kvm_state) +
6679 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6680 			goto error_guest_mode;
6681 
6682 		if (copy_from_user(shadow_vmcs12,
6683 				   user_vmx_nested_state->shadow_vmcs12,
6684 				   sizeof(*shadow_vmcs12))) {
6685 			ret = -EFAULT;
6686 			goto error_guest_mode;
6687 		}
6688 
6689 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6690 		    !shadow_vmcs12->hdr.shadow_vmcs)
6691 			goto error_guest_mode;
6692 	}
6693 
6694 	vmx->nested.has_preemption_timer_deadline = false;
6695 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6696 		vmx->nested.has_preemption_timer_deadline = true;
6697 		vmx->nested.preemption_timer_deadline =
6698 			kvm_state->hdr.vmx.preemption_timer_deadline;
6699 	}
6700 
6701 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6702 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6703 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6704 		goto error_guest_mode;
6705 
6706 	vmx->nested.dirty_vmcs12 = true;
6707 	vmx->nested.force_msr_bitmap_recalc = true;
6708 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6709 	if (ret)
6710 		goto error_guest_mode;
6711 
6712 	if (vmx->nested.mtf_pending)
6713 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6714 
6715 	return 0;
6716 
6717 error_guest_mode:
6718 	vmx->nested.nested_run_pending = 0;
6719 	return ret;
6720 }
6721 
6722 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6723 {
6724 	if (enable_shadow_vmcs) {
6725 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6726 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6727 	}
6728 }
6729 
6730 /*
6731  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6732  * that madness to get the encoding for comparison.
6733  */
6734 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6735 
6736 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6737 {
6738 	/*
6739 	 * Note these are the so called "index" of the VMCS field encoding, not
6740 	 * the index into vmcs12.
6741 	 */
6742 	unsigned int max_idx, idx;
6743 	int i;
6744 
6745 	/*
6746 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6747 	 * vmcs12, regardless of whether or not the associated feature is
6748 	 * exposed to L1.  Simply find the field with the highest index.
6749 	 */
6750 	max_idx = 0;
6751 	for (i = 0; i < nr_vmcs12_fields; i++) {
6752 		/* The vmcs12 table is very, very sparsely populated. */
6753 		if (!vmcs12_field_offsets[i])
6754 			continue;
6755 
6756 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6757 		if (idx > max_idx)
6758 			max_idx = idx;
6759 	}
6760 
6761 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6762 }
6763 
6764 static void nested_vmx_setup_pinbased_ctls(struct vmcs_config *vmcs_conf,
6765 					   struct nested_vmx_msrs *msrs)
6766 {
6767 	msrs->pinbased_ctls_low =
6768 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6769 
6770 	msrs->pinbased_ctls_high = vmcs_conf->pin_based_exec_ctrl;
6771 	msrs->pinbased_ctls_high &=
6772 		PIN_BASED_EXT_INTR_MASK |
6773 		PIN_BASED_NMI_EXITING |
6774 		PIN_BASED_VIRTUAL_NMIS |
6775 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6776 	msrs->pinbased_ctls_high |=
6777 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6778 		PIN_BASED_VMX_PREEMPTION_TIMER;
6779 }
6780 
6781 static void nested_vmx_setup_exit_ctls(struct vmcs_config *vmcs_conf,
6782 				       struct nested_vmx_msrs *msrs)
6783 {
6784 	msrs->exit_ctls_low =
6785 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6786 
6787 	msrs->exit_ctls_high = vmcs_conf->vmexit_ctrl;
6788 	msrs->exit_ctls_high &=
6789 #ifdef CONFIG_X86_64
6790 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6791 #endif
6792 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6793 		VM_EXIT_CLEAR_BNDCFGS;
6794 	msrs->exit_ctls_high |=
6795 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6796 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6797 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT |
6798 		VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6799 
6800 	/* We support free control of debug control saving. */
6801 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6802 }
6803 
6804 static void nested_vmx_setup_entry_ctls(struct vmcs_config *vmcs_conf,
6805 					struct nested_vmx_msrs *msrs)
6806 {
6807 	msrs->entry_ctls_low =
6808 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6809 
6810 	msrs->entry_ctls_high = vmcs_conf->vmentry_ctrl;
6811 	msrs->entry_ctls_high &=
6812 #ifdef CONFIG_X86_64
6813 		VM_ENTRY_IA32E_MODE |
6814 #endif
6815 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
6816 	msrs->entry_ctls_high |=
6817 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER |
6818 		 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
6819 
6820 	/* We support free control of debug control loading. */
6821 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6822 }
6823 
6824 static void nested_vmx_setup_cpubased_ctls(struct vmcs_config *vmcs_conf,
6825 					   struct nested_vmx_msrs *msrs)
6826 {
6827 	msrs->procbased_ctls_low =
6828 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6829 
6830 	msrs->procbased_ctls_high = vmcs_conf->cpu_based_exec_ctrl;
6831 	msrs->procbased_ctls_high &=
6832 		CPU_BASED_INTR_WINDOW_EXITING |
6833 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6834 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6835 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6836 		CPU_BASED_CR3_STORE_EXITING |
6837 #ifdef CONFIG_X86_64
6838 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6839 #endif
6840 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6841 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6842 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6843 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6844 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6845 	/*
6846 	 * We can allow some features even when not supported by the
6847 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6848 	 * can use it to avoid exits to L1 - even when L0 runs L2
6849 	 * without MSR bitmaps.
6850 	 */
6851 	msrs->procbased_ctls_high |=
6852 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6853 		CPU_BASED_USE_MSR_BITMAPS;
6854 
6855 	/* We support free control of CR3 access interception. */
6856 	msrs->procbased_ctls_low &=
6857 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6858 }
6859 
6860 static void nested_vmx_setup_secondary_ctls(u32 ept_caps,
6861 					    struct vmcs_config *vmcs_conf,
6862 					    struct nested_vmx_msrs *msrs)
6863 {
6864 	msrs->secondary_ctls_low = 0;
6865 
6866 	msrs->secondary_ctls_high = vmcs_conf->cpu_based_2nd_exec_ctrl;
6867 	msrs->secondary_ctls_high &=
6868 		SECONDARY_EXEC_DESC |
6869 		SECONDARY_EXEC_ENABLE_RDTSCP |
6870 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6871 		SECONDARY_EXEC_WBINVD_EXITING |
6872 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6873 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6874 		SECONDARY_EXEC_RDRAND_EXITING |
6875 		SECONDARY_EXEC_ENABLE_INVPCID |
6876 		SECONDARY_EXEC_ENABLE_VMFUNC |
6877 		SECONDARY_EXEC_RDSEED_EXITING |
6878 		SECONDARY_EXEC_XSAVES |
6879 		SECONDARY_EXEC_TSC_SCALING |
6880 		SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
6881 
6882 	/*
6883 	 * We can emulate "VMCS shadowing," even if the hardware
6884 	 * doesn't support it.
6885 	 */
6886 	msrs->secondary_ctls_high |=
6887 		SECONDARY_EXEC_SHADOW_VMCS;
6888 
6889 	if (enable_ept) {
6890 		/* nested EPT: emulate EPT also to L1 */
6891 		msrs->secondary_ctls_high |=
6892 			SECONDARY_EXEC_ENABLE_EPT;
6893 		msrs->ept_caps =
6894 			VMX_EPT_PAGE_WALK_4_BIT |
6895 			VMX_EPT_PAGE_WALK_5_BIT |
6896 			VMX_EPTP_WB_BIT |
6897 			VMX_EPT_INVEPT_BIT |
6898 			VMX_EPT_EXECUTE_ONLY_BIT;
6899 
6900 		msrs->ept_caps &= ept_caps;
6901 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6902 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6903 			VMX_EPT_1GB_PAGE_BIT;
6904 		if (enable_ept_ad_bits) {
6905 			msrs->secondary_ctls_high |=
6906 				SECONDARY_EXEC_ENABLE_PML;
6907 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6908 		}
6909 
6910 		/*
6911 		 * Advertise EPTP switching irrespective of hardware support,
6912 		 * KVM emulates it in software so long as VMFUNC is supported.
6913 		 */
6914 		if (cpu_has_vmx_vmfunc())
6915 			msrs->vmfunc_controls = VMX_VMFUNC_EPTP_SWITCHING;
6916 	}
6917 
6918 	/*
6919 	 * Old versions of KVM use the single-context version without
6920 	 * checking for support, so declare that it is supported even
6921 	 * though it is treated as global context.  The alternative is
6922 	 * not failing the single-context invvpid, and it is worse.
6923 	 */
6924 	if (enable_vpid) {
6925 		msrs->secondary_ctls_high |=
6926 			SECONDARY_EXEC_ENABLE_VPID;
6927 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6928 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6929 	}
6930 
6931 	if (enable_unrestricted_guest)
6932 		msrs->secondary_ctls_high |=
6933 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6934 
6935 	if (flexpriority_enabled)
6936 		msrs->secondary_ctls_high |=
6937 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6938 
6939 	if (enable_sgx)
6940 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6941 }
6942 
6943 static void nested_vmx_setup_misc_data(struct vmcs_config *vmcs_conf,
6944 				       struct nested_vmx_msrs *msrs)
6945 {
6946 	msrs->misc_low = (u32)vmcs_conf->misc & VMX_MISC_SAVE_EFER_LMA;
6947 	msrs->misc_low |=
6948 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6949 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6950 		VMX_MISC_ACTIVITY_HLT |
6951 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6952 	msrs->misc_high = 0;
6953 }
6954 
6955 static void nested_vmx_setup_basic(struct nested_vmx_msrs *msrs)
6956 {
6957 	/*
6958 	 * This MSR reports some information about VMX support. We
6959 	 * should return information about the VMX we emulate for the
6960 	 * guest, and the VMCS structure we give it - not about the
6961 	 * VMX support of the underlying hardware.
6962 	 */
6963 	msrs->basic =
6964 		VMCS12_REVISION |
6965 		VMX_BASIC_TRUE_CTLS |
6966 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6967 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6968 
6969 	if (cpu_has_vmx_basic_inout())
6970 		msrs->basic |= VMX_BASIC_INOUT;
6971 }
6972 
6973 static void nested_vmx_setup_cr_fixed(struct nested_vmx_msrs *msrs)
6974 {
6975 	/*
6976 	 * These MSRs specify bits which the guest must keep fixed on
6977 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6978 	 * We picked the standard core2 setting.
6979 	 */
6980 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6981 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6982 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6983 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6984 
6985 	/* These MSRs specify bits which the guest must keep fixed off. */
6986 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6987 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6988 
6989 	if (vmx_umip_emulated())
6990 		msrs->cr4_fixed1 |= X86_CR4_UMIP;
6991 }
6992 
6993 /*
6994  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6995  * returned for the various VMX controls MSRs when nested VMX is enabled.
6996  * The same values should also be used to verify that vmcs12 control fields are
6997  * valid during nested entry from L1 to L2.
6998  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6999  * if the corresponding bit in the (32-bit) control field *must* be on, and a
7000  * bit in the high half is on if the corresponding bit in the control field
7001  * may be on. See also vmx_control_verify().
7002  */
7003 void nested_vmx_setup_ctls_msrs(struct vmcs_config *vmcs_conf, u32 ept_caps)
7004 {
7005 	struct nested_vmx_msrs *msrs = &vmcs_conf->nested;
7006 
7007 	/*
7008 	 * Note that as a general rule, the high half of the MSRs (bits in
7009 	 * the control fields which may be 1) should be initialized by the
7010 	 * intersection of the underlying hardware's MSR (i.e., features which
7011 	 * can be supported) and the list of features we want to expose -
7012 	 * because they are known to be properly supported in our code.
7013 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
7014 	 * be set to 0, meaning that L1 may turn off any of these bits. The
7015 	 * reason is that if one of these bits is necessary, it will appear
7016 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
7017 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
7018 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
7019 	 * These rules have exceptions below.
7020 	 */
7021 	nested_vmx_setup_pinbased_ctls(vmcs_conf, msrs);
7022 
7023 	nested_vmx_setup_exit_ctls(vmcs_conf, msrs);
7024 
7025 	nested_vmx_setup_entry_ctls(vmcs_conf, msrs);
7026 
7027 	nested_vmx_setup_cpubased_ctls(vmcs_conf, msrs);
7028 
7029 	nested_vmx_setup_secondary_ctls(ept_caps, vmcs_conf, msrs);
7030 
7031 	nested_vmx_setup_misc_data(vmcs_conf, msrs);
7032 
7033 	nested_vmx_setup_basic(msrs);
7034 
7035 	nested_vmx_setup_cr_fixed(msrs);
7036 
7037 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
7038 }
7039 
7040 void nested_vmx_hardware_unsetup(void)
7041 {
7042 	int i;
7043 
7044 	if (enable_shadow_vmcs) {
7045 		for (i = 0; i < VMX_BITMAP_NR; i++)
7046 			free_page((unsigned long)vmx_bitmap[i]);
7047 	}
7048 }
7049 
7050 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
7051 {
7052 	int i;
7053 
7054 	if (!cpu_has_vmx_shadow_vmcs())
7055 		enable_shadow_vmcs = 0;
7056 	if (enable_shadow_vmcs) {
7057 		for (i = 0; i < VMX_BITMAP_NR; i++) {
7058 			/*
7059 			 * The vmx_bitmap is not tied to a VM and so should
7060 			 * not be charged to a memcg.
7061 			 */
7062 			vmx_bitmap[i] = (unsigned long *)
7063 				__get_free_page(GFP_KERNEL);
7064 			if (!vmx_bitmap[i]) {
7065 				nested_vmx_hardware_unsetup();
7066 				return -ENOMEM;
7067 			}
7068 		}
7069 
7070 		init_vmcs_shadow_fields();
7071 	}
7072 
7073 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
7074 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
7075 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
7076 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
7077 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
7078 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
7079 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
7080 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmxoff;
7081 	exit_handlers[EXIT_REASON_VMON]		= handle_vmxon;
7082 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
7083 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
7084 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
7085 
7086 	return 0;
7087 }
7088 
7089 struct kvm_x86_nested_ops vmx_nested_ops = {
7090 	.leave_nested = vmx_leave_nested,
7091 	.is_exception_vmexit = nested_vmx_is_exception_vmexit,
7092 	.check_events = vmx_check_nested_events,
7093 	.has_events = vmx_has_nested_events,
7094 	.triple_fault = nested_vmx_triple_fault,
7095 	.get_state = vmx_get_nested_state,
7096 	.set_state = vmx_set_nested_state,
7097 	.get_nested_state_pages = vmx_get_nested_state_pages,
7098 	.write_log_dirty = nested_vmx_write_pml_buffer,
7099 	.enable_evmcs = nested_enable_evmcs,
7100 	.get_evmcs_version = nested_get_evmcs_version,
7101 	.hv_inject_synthetic_vmexit_post_tlb_flush = vmx_hv_inject_synthetic_vmexit_post_tlb_flush,
7102 };
7103