xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 6a143a7c)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/objtool.h>
4 #include <linux/percpu.h>
5 
6 #include <asm/debugreg.h>
7 #include <asm/mmu_context.h>
8 
9 #include "cpuid.h"
10 #include "hyperv.h"
11 #include "mmu.h"
12 #include "nested.h"
13 #include "pmu.h"
14 #include "trace.h"
15 #include "vmx.h"
16 #include "x86.h"
17 
18 static bool __read_mostly enable_shadow_vmcs = 1;
19 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
20 
21 static bool __read_mostly nested_early_check = 0;
22 module_param(nested_early_check, bool, S_IRUGO);
23 
24 #define CC(consistency_check)						\
25 ({									\
26 	bool failed = (consistency_check);				\
27 	if (failed)							\
28 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
29 	failed;								\
30 })
31 
32 /*
33  * Hyper-V requires all of these, so mark them as supported even though
34  * they are just treated the same as all-context.
35  */
36 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
37 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
38 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
39 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
40 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
41 
42 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
43 
44 enum {
45 	VMX_VMREAD_BITMAP,
46 	VMX_VMWRITE_BITMAP,
47 	VMX_BITMAP_NR
48 };
49 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
50 
51 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
52 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
53 
54 struct shadow_vmcs_field {
55 	u16	encoding;
56 	u16	offset;
57 };
58 static struct shadow_vmcs_field shadow_read_only_fields[] = {
59 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
60 #include "vmcs_shadow_fields.h"
61 };
62 static int max_shadow_read_only_fields =
63 	ARRAY_SIZE(shadow_read_only_fields);
64 
65 static struct shadow_vmcs_field shadow_read_write_fields[] = {
66 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
67 #include "vmcs_shadow_fields.h"
68 };
69 static int max_shadow_read_write_fields =
70 	ARRAY_SIZE(shadow_read_write_fields);
71 
72 static void init_vmcs_shadow_fields(void)
73 {
74 	int i, j;
75 
76 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
77 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
78 
79 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
80 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
81 		u16 field = entry.encoding;
82 
83 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
84 		    (i + 1 == max_shadow_read_only_fields ||
85 		     shadow_read_only_fields[i + 1].encoding != field + 1))
86 			pr_err("Missing field from shadow_read_only_field %x\n",
87 			       field + 1);
88 
89 		clear_bit(field, vmx_vmread_bitmap);
90 		if (field & 1)
91 #ifdef CONFIG_X86_64
92 			continue;
93 #else
94 			entry.offset += sizeof(u32);
95 #endif
96 		shadow_read_only_fields[j++] = entry;
97 	}
98 	max_shadow_read_only_fields = j;
99 
100 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
101 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
102 		u16 field = entry.encoding;
103 
104 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
105 		    (i + 1 == max_shadow_read_write_fields ||
106 		     shadow_read_write_fields[i + 1].encoding != field + 1))
107 			pr_err("Missing field from shadow_read_write_field %x\n",
108 			       field + 1);
109 
110 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
111 			  field <= GUEST_TR_AR_BYTES,
112 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
113 
114 		/*
115 		 * PML and the preemption timer can be emulated, but the
116 		 * processor cannot vmwrite to fields that don't exist
117 		 * on bare metal.
118 		 */
119 		switch (field) {
120 		case GUEST_PML_INDEX:
121 			if (!cpu_has_vmx_pml())
122 				continue;
123 			break;
124 		case VMX_PREEMPTION_TIMER_VALUE:
125 			if (!cpu_has_vmx_preemption_timer())
126 				continue;
127 			break;
128 		case GUEST_INTR_STATUS:
129 			if (!cpu_has_vmx_apicv())
130 				continue;
131 			break;
132 		default:
133 			break;
134 		}
135 
136 		clear_bit(field, vmx_vmwrite_bitmap);
137 		clear_bit(field, vmx_vmread_bitmap);
138 		if (field & 1)
139 #ifdef CONFIG_X86_64
140 			continue;
141 #else
142 			entry.offset += sizeof(u32);
143 #endif
144 		shadow_read_write_fields[j++] = entry;
145 	}
146 	max_shadow_read_write_fields = j;
147 }
148 
149 /*
150  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
151  * set the success or error code of an emulated VMX instruction (as specified
152  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
153  * instruction.
154  */
155 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
156 {
157 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
158 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
159 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
160 	return kvm_skip_emulated_instruction(vcpu);
161 }
162 
163 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
164 {
165 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
166 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
167 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
168 			| X86_EFLAGS_CF);
169 	return kvm_skip_emulated_instruction(vcpu);
170 }
171 
172 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
173 				u32 vm_instruction_error)
174 {
175 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
176 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
177 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
178 			| X86_EFLAGS_ZF);
179 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
180 	/*
181 	 * We don't need to force a shadow sync because
182 	 * VM_INSTRUCTION_ERROR is not shadowed
183 	 */
184 	return kvm_skip_emulated_instruction(vcpu);
185 }
186 
187 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
188 {
189 	struct vcpu_vmx *vmx = to_vmx(vcpu);
190 
191 	/*
192 	 * failValid writes the error number to the current VMCS, which
193 	 * can't be done if there isn't a current VMCS.
194 	 */
195 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
196 		return nested_vmx_failInvalid(vcpu);
197 
198 	return nested_vmx_failValid(vcpu, vm_instruction_error);
199 }
200 
201 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
202 {
203 	/* TODO: not to reset guest simply here. */
204 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
205 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
206 }
207 
208 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
209 {
210 	return fixed_bits_valid(control, low, high);
211 }
212 
213 static inline u64 vmx_control_msr(u32 low, u32 high)
214 {
215 	return low | ((u64)high << 32);
216 }
217 
218 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
219 {
220 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
221 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
222 	vmx->nested.need_vmcs12_to_shadow_sync = false;
223 }
224 
225 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
226 {
227 	struct vcpu_vmx *vmx = to_vmx(vcpu);
228 
229 	if (!vmx->nested.hv_evmcs)
230 		return;
231 
232 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
233 	vmx->nested.hv_evmcs_vmptr = 0;
234 	vmx->nested.hv_evmcs = NULL;
235 }
236 
237 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
238 				     struct loaded_vmcs *prev)
239 {
240 	struct vmcs_host_state *dest, *src;
241 
242 	if (unlikely(!vmx->guest_state_loaded))
243 		return;
244 
245 	src = &prev->host_state;
246 	dest = &vmx->loaded_vmcs->host_state;
247 
248 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
249 	dest->ldt_sel = src->ldt_sel;
250 #ifdef CONFIG_X86_64
251 	dest->ds_sel = src->ds_sel;
252 	dest->es_sel = src->es_sel;
253 #endif
254 }
255 
256 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
257 {
258 	struct vcpu_vmx *vmx = to_vmx(vcpu);
259 	struct loaded_vmcs *prev;
260 	int cpu;
261 
262 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
263 		return;
264 
265 	cpu = get_cpu();
266 	prev = vmx->loaded_vmcs;
267 	vmx->loaded_vmcs = vmcs;
268 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
269 	vmx_sync_vmcs_host_state(vmx, prev);
270 	put_cpu();
271 
272 	vmx_register_cache_reset(vcpu);
273 }
274 
275 /*
276  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
277  * just stops using VMX.
278  */
279 static void free_nested(struct kvm_vcpu *vcpu)
280 {
281 	struct vcpu_vmx *vmx = to_vmx(vcpu);
282 
283 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
284 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
285 
286 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
287 		return;
288 
289 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
290 
291 	vmx->nested.vmxon = false;
292 	vmx->nested.smm.vmxon = false;
293 	free_vpid(vmx->nested.vpid02);
294 	vmx->nested.posted_intr_nv = -1;
295 	vmx->nested.current_vmptr = -1ull;
296 	if (enable_shadow_vmcs) {
297 		vmx_disable_shadow_vmcs(vmx);
298 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
299 		free_vmcs(vmx->vmcs01.shadow_vmcs);
300 		vmx->vmcs01.shadow_vmcs = NULL;
301 	}
302 	kfree(vmx->nested.cached_vmcs12);
303 	vmx->nested.cached_vmcs12 = NULL;
304 	kfree(vmx->nested.cached_shadow_vmcs12);
305 	vmx->nested.cached_shadow_vmcs12 = NULL;
306 	/* Unpin physical memory we referred to in the vmcs02 */
307 	if (vmx->nested.apic_access_page) {
308 		kvm_release_page_clean(vmx->nested.apic_access_page);
309 		vmx->nested.apic_access_page = NULL;
310 	}
311 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
312 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
313 	vmx->nested.pi_desc = NULL;
314 
315 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
316 
317 	nested_release_evmcs(vcpu);
318 
319 	free_loaded_vmcs(&vmx->nested.vmcs02);
320 }
321 
322 /*
323  * Ensure that the current vmcs of the logical processor is the
324  * vmcs01 of the vcpu before calling free_nested().
325  */
326 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
327 {
328 	vcpu_load(vcpu);
329 	vmx_leave_nested(vcpu);
330 	vcpu_put(vcpu);
331 }
332 
333 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
334 		struct x86_exception *fault)
335 {
336 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
337 	struct vcpu_vmx *vmx = to_vmx(vcpu);
338 	u32 vm_exit_reason;
339 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
340 
341 	if (vmx->nested.pml_full) {
342 		vm_exit_reason = EXIT_REASON_PML_FULL;
343 		vmx->nested.pml_full = false;
344 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
345 	} else if (fault->error_code & PFERR_RSVD_MASK)
346 		vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
347 	else
348 		vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
349 
350 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
351 	vmcs12->guest_physical_address = fault->address;
352 }
353 
354 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
355 {
356 	WARN_ON(mmu_is_nested(vcpu));
357 
358 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
359 	kvm_init_shadow_ept_mmu(vcpu,
360 			to_vmx(vcpu)->nested.msrs.ept_caps &
361 			VMX_EPT_EXECUTE_ONLY_BIT,
362 			nested_ept_ad_enabled(vcpu),
363 			nested_ept_get_eptp(vcpu));
364 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
365 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
366 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
367 
368 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
369 }
370 
371 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
372 {
373 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
374 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
375 }
376 
377 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
378 					    u16 error_code)
379 {
380 	bool inequality, bit;
381 
382 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
383 	inequality =
384 		(error_code & vmcs12->page_fault_error_code_mask) !=
385 		 vmcs12->page_fault_error_code_match;
386 	return inequality ^ bit;
387 }
388 
389 
390 /*
391  * KVM wants to inject page-faults which it got to the guest. This function
392  * checks whether in a nested guest, we need to inject them to L1 or L2.
393  */
394 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
395 {
396 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
397 	unsigned int nr = vcpu->arch.exception.nr;
398 	bool has_payload = vcpu->arch.exception.has_payload;
399 	unsigned long payload = vcpu->arch.exception.payload;
400 
401 	if (nr == PF_VECTOR) {
402 		if (vcpu->arch.exception.nested_apf) {
403 			*exit_qual = vcpu->arch.apf.nested_apf_token;
404 			return 1;
405 		}
406 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
407 						    vcpu->arch.exception.error_code)) {
408 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
409 			return 1;
410 		}
411 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
412 		if (nr == DB_VECTOR) {
413 			if (!has_payload) {
414 				payload = vcpu->arch.dr6;
415 				payload &= ~DR6_BT;
416 				payload ^= DR6_ACTIVE_LOW;
417 			}
418 			*exit_qual = payload;
419 		} else
420 			*exit_qual = 0;
421 		return 1;
422 	}
423 
424 	return 0;
425 }
426 
427 
428 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
429 		struct x86_exception *fault)
430 {
431 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
432 
433 	WARN_ON(!is_guest_mode(vcpu));
434 
435 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
436 		!to_vmx(vcpu)->nested.nested_run_pending) {
437 		vmcs12->vm_exit_intr_error_code = fault->error_code;
438 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
439 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
440 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
441 				  fault->address);
442 	} else {
443 		kvm_inject_page_fault(vcpu, fault);
444 	}
445 }
446 
447 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
448 					       struct vmcs12 *vmcs12)
449 {
450 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
451 		return 0;
452 
453 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
454 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
455 		return -EINVAL;
456 
457 	return 0;
458 }
459 
460 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
461 						struct vmcs12 *vmcs12)
462 {
463 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
464 		return 0;
465 
466 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
467 		return -EINVAL;
468 
469 	return 0;
470 }
471 
472 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
473 						struct vmcs12 *vmcs12)
474 {
475 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
476 		return 0;
477 
478 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
479 		return -EINVAL;
480 
481 	return 0;
482 }
483 
484 /*
485  * Check if MSR is intercepted for L01 MSR bitmap.
486  */
487 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
488 {
489 	unsigned long *msr_bitmap;
490 	int f = sizeof(unsigned long);
491 
492 	if (!cpu_has_vmx_msr_bitmap())
493 		return true;
494 
495 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
496 
497 	if (msr <= 0x1fff) {
498 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
499 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
500 		msr &= 0x1fff;
501 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
502 	}
503 
504 	return true;
505 }
506 
507 /*
508  * If a msr is allowed by L0, we should check whether it is allowed by L1.
509  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
510  */
511 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
512 					       unsigned long *msr_bitmap_nested,
513 					       u32 msr, int type)
514 {
515 	int f = sizeof(unsigned long);
516 
517 	/*
518 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
519 	 * have the write-low and read-high bitmap offsets the wrong way round.
520 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
521 	 */
522 	if (msr <= 0x1fff) {
523 		if (type & MSR_TYPE_R &&
524 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
525 			/* read-low */
526 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
527 
528 		if (type & MSR_TYPE_W &&
529 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
530 			/* write-low */
531 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
532 
533 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
534 		msr &= 0x1fff;
535 		if (type & MSR_TYPE_R &&
536 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
537 			/* read-high */
538 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
539 
540 		if (type & MSR_TYPE_W &&
541 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
542 			/* write-high */
543 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
544 
545 	}
546 }
547 
548 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
549 {
550 	int msr;
551 
552 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
553 		unsigned word = msr / BITS_PER_LONG;
554 
555 		msr_bitmap[word] = ~0;
556 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
557 	}
558 }
559 
560 /*
561  * Merge L0's and L1's MSR bitmap, return false to indicate that
562  * we do not use the hardware.
563  */
564 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
565 						 struct vmcs12 *vmcs12)
566 {
567 	int msr;
568 	unsigned long *msr_bitmap_l1;
569 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
570 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
571 
572 	/* Nothing to do if the MSR bitmap is not in use.  */
573 	if (!cpu_has_vmx_msr_bitmap() ||
574 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
575 		return false;
576 
577 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
578 		return false;
579 
580 	msr_bitmap_l1 = (unsigned long *)map->hva;
581 
582 	/*
583 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
584 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
585 	 * the x2APIC MSR range and selectively disable them below.
586 	 */
587 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
588 
589 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
590 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
591 			/*
592 			 * L0 need not intercept reads for MSRs between 0x800
593 			 * and 0x8ff, it just lets the processor take the value
594 			 * from the virtual-APIC page; take those 256 bits
595 			 * directly from the L1 bitmap.
596 			 */
597 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
598 				unsigned word = msr / BITS_PER_LONG;
599 
600 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
601 			}
602 		}
603 
604 		nested_vmx_disable_intercept_for_msr(
605 			msr_bitmap_l1, msr_bitmap_l0,
606 			X2APIC_MSR(APIC_TASKPRI),
607 			MSR_TYPE_R | MSR_TYPE_W);
608 
609 		if (nested_cpu_has_vid(vmcs12)) {
610 			nested_vmx_disable_intercept_for_msr(
611 				msr_bitmap_l1, msr_bitmap_l0,
612 				X2APIC_MSR(APIC_EOI),
613 				MSR_TYPE_W);
614 			nested_vmx_disable_intercept_for_msr(
615 				msr_bitmap_l1, msr_bitmap_l0,
616 				X2APIC_MSR(APIC_SELF_IPI),
617 				MSR_TYPE_W);
618 		}
619 	}
620 
621 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
622 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
623 					     MSR_FS_BASE, MSR_TYPE_RW);
624 
625 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
626 					     MSR_GS_BASE, MSR_TYPE_RW);
627 
628 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
629 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
630 
631 	/*
632 	 * Checking the L0->L1 bitmap is trying to verify two things:
633 	 *
634 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
635 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
636 	 *    from the L12 MSR bitmap that is too permissive.
637 	 * 2. That L1 or L2s have actually used the MSR. This avoids
638 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
639 	 *    works properly because we only update the L01 MSR bitmap lazily.
640 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
641 	 *    updated to reflect this when L1 (or its L2s) actually write to
642 	 *    the MSR.
643 	 */
644 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
645 		nested_vmx_disable_intercept_for_msr(
646 					msr_bitmap_l1, msr_bitmap_l0,
647 					MSR_IA32_SPEC_CTRL,
648 					MSR_TYPE_R | MSR_TYPE_W);
649 
650 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
651 		nested_vmx_disable_intercept_for_msr(
652 					msr_bitmap_l1, msr_bitmap_l0,
653 					MSR_IA32_PRED_CMD,
654 					MSR_TYPE_W);
655 
656 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
657 
658 	return true;
659 }
660 
661 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
662 				       struct vmcs12 *vmcs12)
663 {
664 	struct kvm_host_map map;
665 	struct vmcs12 *shadow;
666 
667 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
668 	    vmcs12->vmcs_link_pointer == -1ull)
669 		return;
670 
671 	shadow = get_shadow_vmcs12(vcpu);
672 
673 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
674 		return;
675 
676 	memcpy(shadow, map.hva, VMCS12_SIZE);
677 	kvm_vcpu_unmap(vcpu, &map, false);
678 }
679 
680 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
681 					      struct vmcs12 *vmcs12)
682 {
683 	struct vcpu_vmx *vmx = to_vmx(vcpu);
684 
685 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
686 	    vmcs12->vmcs_link_pointer == -1ull)
687 		return;
688 
689 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
690 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
691 }
692 
693 /*
694  * In nested virtualization, check if L1 has set
695  * VM_EXIT_ACK_INTR_ON_EXIT
696  */
697 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
698 {
699 	return get_vmcs12(vcpu)->vm_exit_controls &
700 		VM_EXIT_ACK_INTR_ON_EXIT;
701 }
702 
703 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
704 					  struct vmcs12 *vmcs12)
705 {
706 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
707 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
708 		return -EINVAL;
709 	else
710 		return 0;
711 }
712 
713 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
714 					   struct vmcs12 *vmcs12)
715 {
716 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
717 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
718 	    !nested_cpu_has_vid(vmcs12) &&
719 	    !nested_cpu_has_posted_intr(vmcs12))
720 		return 0;
721 
722 	/*
723 	 * If virtualize x2apic mode is enabled,
724 	 * virtualize apic access must be disabled.
725 	 */
726 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
727 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
728 		return -EINVAL;
729 
730 	/*
731 	 * If virtual interrupt delivery is enabled,
732 	 * we must exit on external interrupts.
733 	 */
734 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
735 		return -EINVAL;
736 
737 	/*
738 	 * bits 15:8 should be zero in posted_intr_nv,
739 	 * the descriptor address has been already checked
740 	 * in nested_get_vmcs12_pages.
741 	 *
742 	 * bits 5:0 of posted_intr_desc_addr should be zero.
743 	 */
744 	if (nested_cpu_has_posted_intr(vmcs12) &&
745 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
746 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
747 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
748 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
749 		return -EINVAL;
750 
751 	/* tpr shadow is needed by all apicv features. */
752 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
753 		return -EINVAL;
754 
755 	return 0;
756 }
757 
758 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
759 				       u32 count, u64 addr)
760 {
761 	if (count == 0)
762 		return 0;
763 
764 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
765 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
766 		return -EINVAL;
767 
768 	return 0;
769 }
770 
771 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
772 						     struct vmcs12 *vmcs12)
773 {
774 	if (CC(nested_vmx_check_msr_switch(vcpu,
775 					   vmcs12->vm_exit_msr_load_count,
776 					   vmcs12->vm_exit_msr_load_addr)) ||
777 	    CC(nested_vmx_check_msr_switch(vcpu,
778 					   vmcs12->vm_exit_msr_store_count,
779 					   vmcs12->vm_exit_msr_store_addr)))
780 		return -EINVAL;
781 
782 	return 0;
783 }
784 
785 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
786                                                       struct vmcs12 *vmcs12)
787 {
788 	if (CC(nested_vmx_check_msr_switch(vcpu,
789 					   vmcs12->vm_entry_msr_load_count,
790 					   vmcs12->vm_entry_msr_load_addr)))
791                 return -EINVAL;
792 
793 	return 0;
794 }
795 
796 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
797 					 struct vmcs12 *vmcs12)
798 {
799 	if (!nested_cpu_has_pml(vmcs12))
800 		return 0;
801 
802 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
803 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
804 		return -EINVAL;
805 
806 	return 0;
807 }
808 
809 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
810 							struct vmcs12 *vmcs12)
811 {
812 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
813 	       !nested_cpu_has_ept(vmcs12)))
814 		return -EINVAL;
815 	return 0;
816 }
817 
818 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
819 							 struct vmcs12 *vmcs12)
820 {
821 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
822 	       !nested_cpu_has_ept(vmcs12)))
823 		return -EINVAL;
824 	return 0;
825 }
826 
827 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
828 						 struct vmcs12 *vmcs12)
829 {
830 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
831 		return 0;
832 
833 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
834 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
835 		return -EINVAL;
836 
837 	return 0;
838 }
839 
840 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
841 				       struct vmx_msr_entry *e)
842 {
843 	/* x2APIC MSR accesses are not allowed */
844 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
845 		return -EINVAL;
846 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
847 	    CC(e->index == MSR_IA32_UCODE_REV))
848 		return -EINVAL;
849 	if (CC(e->reserved != 0))
850 		return -EINVAL;
851 	return 0;
852 }
853 
854 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
855 				     struct vmx_msr_entry *e)
856 {
857 	if (CC(e->index == MSR_FS_BASE) ||
858 	    CC(e->index == MSR_GS_BASE) ||
859 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
860 	    nested_vmx_msr_check_common(vcpu, e))
861 		return -EINVAL;
862 	return 0;
863 }
864 
865 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
866 				      struct vmx_msr_entry *e)
867 {
868 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
869 	    nested_vmx_msr_check_common(vcpu, e))
870 		return -EINVAL;
871 	return 0;
872 }
873 
874 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
875 {
876 	struct vcpu_vmx *vmx = to_vmx(vcpu);
877 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
878 				       vmx->nested.msrs.misc_high);
879 
880 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
881 }
882 
883 /*
884  * Load guest's/host's msr at nested entry/exit.
885  * return 0 for success, entry index for failure.
886  *
887  * One of the failure modes for MSR load/store is when a list exceeds the
888  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
889  * as possible, process all valid entries before failing rather than precheck
890  * for a capacity violation.
891  */
892 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
893 {
894 	u32 i;
895 	struct vmx_msr_entry e;
896 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
897 
898 	for (i = 0; i < count; i++) {
899 		if (unlikely(i >= max_msr_list_size))
900 			goto fail;
901 
902 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
903 					&e, sizeof(e))) {
904 			pr_debug_ratelimited(
905 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
906 				__func__, i, gpa + i * sizeof(e));
907 			goto fail;
908 		}
909 		if (nested_vmx_load_msr_check(vcpu, &e)) {
910 			pr_debug_ratelimited(
911 				"%s check failed (%u, 0x%x, 0x%x)\n",
912 				__func__, i, e.index, e.reserved);
913 			goto fail;
914 		}
915 		if (kvm_set_msr(vcpu, e.index, e.value)) {
916 			pr_debug_ratelimited(
917 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
918 				__func__, i, e.index, e.value);
919 			goto fail;
920 		}
921 	}
922 	return 0;
923 fail:
924 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
925 	return i + 1;
926 }
927 
928 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
929 					    u32 msr_index,
930 					    u64 *data)
931 {
932 	struct vcpu_vmx *vmx = to_vmx(vcpu);
933 
934 	/*
935 	 * If the L0 hypervisor stored a more accurate value for the TSC that
936 	 * does not include the time taken for emulation of the L2->L1
937 	 * VM-exit in L0, use the more accurate value.
938 	 */
939 	if (msr_index == MSR_IA32_TSC) {
940 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
941 						    MSR_IA32_TSC);
942 
943 		if (i >= 0) {
944 			u64 val = vmx->msr_autostore.guest.val[i].value;
945 
946 			*data = kvm_read_l1_tsc(vcpu, val);
947 			return true;
948 		}
949 	}
950 
951 	if (kvm_get_msr(vcpu, msr_index, data)) {
952 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
953 			msr_index);
954 		return false;
955 	}
956 	return true;
957 }
958 
959 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
960 				     struct vmx_msr_entry *e)
961 {
962 	if (kvm_vcpu_read_guest(vcpu,
963 				gpa + i * sizeof(*e),
964 				e, 2 * sizeof(u32))) {
965 		pr_debug_ratelimited(
966 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
967 			__func__, i, gpa + i * sizeof(*e));
968 		return false;
969 	}
970 	if (nested_vmx_store_msr_check(vcpu, e)) {
971 		pr_debug_ratelimited(
972 			"%s check failed (%u, 0x%x, 0x%x)\n",
973 			__func__, i, e->index, e->reserved);
974 		return false;
975 	}
976 	return true;
977 }
978 
979 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
980 {
981 	u64 data;
982 	u32 i;
983 	struct vmx_msr_entry e;
984 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
985 
986 	for (i = 0; i < count; i++) {
987 		if (unlikely(i >= max_msr_list_size))
988 			return -EINVAL;
989 
990 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
991 			return -EINVAL;
992 
993 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
994 			return -EINVAL;
995 
996 		if (kvm_vcpu_write_guest(vcpu,
997 					 gpa + i * sizeof(e) +
998 					     offsetof(struct vmx_msr_entry, value),
999 					 &data, sizeof(data))) {
1000 			pr_debug_ratelimited(
1001 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1002 				__func__, i, e.index, data);
1003 			return -EINVAL;
1004 		}
1005 	}
1006 	return 0;
1007 }
1008 
1009 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1010 {
1011 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1012 	u32 count = vmcs12->vm_exit_msr_store_count;
1013 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1014 	struct vmx_msr_entry e;
1015 	u32 i;
1016 
1017 	for (i = 0; i < count; i++) {
1018 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1019 			return false;
1020 
1021 		if (e.index == msr_index)
1022 			return true;
1023 	}
1024 	return false;
1025 }
1026 
1027 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1028 					   u32 msr_index)
1029 {
1030 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1031 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1032 	bool in_vmcs12_store_list;
1033 	int msr_autostore_slot;
1034 	bool in_autostore_list;
1035 	int last;
1036 
1037 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1038 	in_autostore_list = msr_autostore_slot >= 0;
1039 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1040 
1041 	if (in_vmcs12_store_list && !in_autostore_list) {
1042 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1043 			/*
1044 			 * Emulated VMEntry does not fail here.  Instead a less
1045 			 * accurate value will be returned by
1046 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1047 			 * instead of reading the value from the vmcs02 VMExit
1048 			 * MSR-store area.
1049 			 */
1050 			pr_warn_ratelimited(
1051 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1052 				msr_index);
1053 			return;
1054 		}
1055 		last = autostore->nr++;
1056 		autostore->val[last].index = msr_index;
1057 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1058 		last = --autostore->nr;
1059 		autostore->val[msr_autostore_slot] = autostore->val[last];
1060 	}
1061 }
1062 
1063 /*
1064  * Returns true if the MMU needs to be sync'd on nested VM-Enter/VM-Exit.
1065  * tl;dr: the MMU needs a sync if L0 is using shadow paging and L1 didn't
1066  * enable VPID for L2 (implying it expects a TLB flush on VMX transitions).
1067  * Here's why.
1068  *
1069  * If EPT is enabled by L0 a sync is never needed:
1070  * - if it is disabled by L1, then L0 is not shadowing L1 or L2 PTEs, there
1071  *   cannot be unsync'd SPTEs for either L1 or L2.
1072  *
1073  * - if it is also enabled by L1, then L0 doesn't need to sync on VM-Enter
1074  *   VM-Enter as VM-Enter isn't required to invalidate guest-physical mappings
1075  *   (irrespective of VPID), i.e. L1 can't rely on the (virtual) CPU to flush
1076  *   stale guest-physical mappings for L2 from the TLB.  And as above, L0 isn't
1077  *   shadowing L1 PTEs so there are no unsync'd SPTEs to sync on VM-Exit.
1078  *
1079  * If EPT is disabled by L0:
1080  * - if VPID is enabled by L1 (for L2), the situation is similar to when L1
1081  *   enables EPT: L0 doesn't need to sync as VM-Enter and VM-Exit aren't
1082  *   required to invalidate linear mappings (EPT is disabled so there are
1083  *   no combined or guest-physical mappings), i.e. L1 can't rely on the
1084  *   (virtual) CPU to flush stale linear mappings for either L2 or itself (L1).
1085  *
1086  * - however if VPID is disabled by L1, then a sync is needed as L1 expects all
1087  *   linear mappings (EPT is disabled so there are no combined or guest-physical
1088  *   mappings) to be invalidated on both VM-Enter and VM-Exit.
1089  *
1090  * Note, this logic is subtly different than nested_has_guest_tlb_tag(), which
1091  * additionally checks that L2 has been assigned a VPID (when EPT is disabled).
1092  * Whether or not L2 has been assigned a VPID by L0 is irrelevant with respect
1093  * to L1's expectations, e.g. L0 needs to invalidate hardware TLB entries if L2
1094  * doesn't have a unique VPID to prevent reusing L1's entries (assuming L1 has
1095  * been assigned a VPID), but L0 doesn't need to do a MMU sync because L1
1096  * doesn't expect stale (virtual) TLB entries to be flushed, i.e. L1 doesn't
1097  * know that L0 will flush the TLB and so L1 will do INVVPID as needed to flush
1098  * stale TLB entries, at which point L0 will sync L2's MMU.
1099  */
1100 static bool nested_vmx_transition_mmu_sync(struct kvm_vcpu *vcpu)
1101 {
1102 	return !enable_ept && !nested_cpu_has_vpid(get_vmcs12(vcpu));
1103 }
1104 
1105 /*
1106  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1107  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1108  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1109  * @entry_failure_code.
1110  */
1111 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
1112 			       enum vm_entry_failure_code *entry_failure_code)
1113 {
1114 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1115 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1116 		return -EINVAL;
1117 	}
1118 
1119 	/*
1120 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1121 	 * must not be dereferenced.
1122 	 */
1123 	if (!nested_ept && is_pae_paging(vcpu) &&
1124 	    (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
1125 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1126 			*entry_failure_code = ENTRY_FAIL_PDPTE;
1127 			return -EINVAL;
1128 		}
1129 	}
1130 
1131 	/*
1132 	 * Unconditionally skip the TLB flush on fast CR3 switch, all TLB
1133 	 * flushes are handled by nested_vmx_transition_tlb_flush().  See
1134 	 * nested_vmx_transition_mmu_sync for details on skipping the MMU sync.
1135 	 */
1136 	if (!nested_ept)
1137 		kvm_mmu_new_pgd(vcpu, cr3, true,
1138 				!nested_vmx_transition_mmu_sync(vcpu));
1139 
1140 	vcpu->arch.cr3 = cr3;
1141 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1142 
1143 	kvm_init_mmu(vcpu, false);
1144 
1145 	return 0;
1146 }
1147 
1148 /*
1149  * Returns if KVM is able to config CPU to tag TLB entries
1150  * populated by L2 differently than TLB entries populated
1151  * by L1.
1152  *
1153  * If L0 uses EPT, L1 and L2 run with different EPTP because
1154  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1155  * are tagged with different EPTP.
1156  *
1157  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1158  * with different VPID (L1 entries are tagged with vmx->vpid
1159  * while L2 entries are tagged with vmx->nested.vpid02).
1160  */
1161 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1162 {
1163 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1164 
1165 	return enable_ept ||
1166 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1167 }
1168 
1169 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1170 					    struct vmcs12 *vmcs12,
1171 					    bool is_vmenter)
1172 {
1173 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1174 
1175 	/*
1176 	 * If VPID is disabled, linear and combined mappings are flushed on
1177 	 * VM-Enter/VM-Exit, and guest-physical mappings are valid only for
1178 	 * their associated EPTP.
1179 	 */
1180 	if (!enable_vpid)
1181 		return;
1182 
1183 	/*
1184 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1185 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit.
1186 	 *
1187 	 * If VPID is enabled and used by vmc12, but L2 does not have a unique
1188 	 * TLB tag (ASID), i.e. EPT is disabled and KVM was unable to allocate
1189 	 * a VPID for L2, flush the current context as the effective ASID is
1190 	 * common to both L1 and L2.
1191 	 *
1192 	 * Defer the flush so that it runs after vmcs02.EPTP has been set by
1193 	 * KVM_REQ_LOAD_MMU_PGD (if nested EPT is enabled) and to avoid
1194 	 * redundant flushes further down the nested pipeline.
1195 	 *
1196 	 * If a TLB flush isn't required due to any of the above, and vpid12 is
1197 	 * changing then the new "virtual" VPID (vpid12) will reuse the same
1198 	 * "real" VPID (vpid02), and so needs to be sync'd.  There is no direct
1199 	 * mapping between vpid02 and vpid12, vpid02 is per-vCPU and reused for
1200 	 * all nested vCPUs.
1201 	 */
1202 	if (!nested_cpu_has_vpid(vmcs12)) {
1203 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1204 	} else if (!nested_has_guest_tlb_tag(vcpu)) {
1205 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1206 	} else if (is_vmenter &&
1207 		   vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1208 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1209 		vpid_sync_context(nested_get_vpid02(vcpu));
1210 	}
1211 }
1212 
1213 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1214 {
1215 	superset &= mask;
1216 	subset &= mask;
1217 
1218 	return (superset | subset) == superset;
1219 }
1220 
1221 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1222 {
1223 	const u64 feature_and_reserved =
1224 		/* feature (except bit 48; see below) */
1225 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1226 		/* reserved */
1227 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1228 	u64 vmx_basic = vmx->nested.msrs.basic;
1229 
1230 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1231 		return -EINVAL;
1232 
1233 	/*
1234 	 * KVM does not emulate a version of VMX that constrains physical
1235 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1236 	 */
1237 	if (data & BIT_ULL(48))
1238 		return -EINVAL;
1239 
1240 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1241 	    vmx_basic_vmcs_revision_id(data))
1242 		return -EINVAL;
1243 
1244 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1245 		return -EINVAL;
1246 
1247 	vmx->nested.msrs.basic = data;
1248 	return 0;
1249 }
1250 
1251 static int
1252 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1253 {
1254 	u64 supported;
1255 	u32 *lowp, *highp;
1256 
1257 	switch (msr_index) {
1258 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1259 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1260 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1261 		break;
1262 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1263 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1264 		highp = &vmx->nested.msrs.procbased_ctls_high;
1265 		break;
1266 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1267 		lowp = &vmx->nested.msrs.exit_ctls_low;
1268 		highp = &vmx->nested.msrs.exit_ctls_high;
1269 		break;
1270 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1271 		lowp = &vmx->nested.msrs.entry_ctls_low;
1272 		highp = &vmx->nested.msrs.entry_ctls_high;
1273 		break;
1274 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1275 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1276 		highp = &vmx->nested.msrs.secondary_ctls_high;
1277 		break;
1278 	default:
1279 		BUG();
1280 	}
1281 
1282 	supported = vmx_control_msr(*lowp, *highp);
1283 
1284 	/* Check must-be-1 bits are still 1. */
1285 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1286 		return -EINVAL;
1287 
1288 	/* Check must-be-0 bits are still 0. */
1289 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1290 		return -EINVAL;
1291 
1292 	*lowp = data;
1293 	*highp = data >> 32;
1294 	return 0;
1295 }
1296 
1297 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1298 {
1299 	const u64 feature_and_reserved_bits =
1300 		/* feature */
1301 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1302 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1303 		/* reserved */
1304 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1305 	u64 vmx_misc;
1306 
1307 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1308 				   vmx->nested.msrs.misc_high);
1309 
1310 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1311 		return -EINVAL;
1312 
1313 	if ((vmx->nested.msrs.pinbased_ctls_high &
1314 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1315 	    vmx_misc_preemption_timer_rate(data) !=
1316 	    vmx_misc_preemption_timer_rate(vmx_misc))
1317 		return -EINVAL;
1318 
1319 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1320 		return -EINVAL;
1321 
1322 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1323 		return -EINVAL;
1324 
1325 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1326 		return -EINVAL;
1327 
1328 	vmx->nested.msrs.misc_low = data;
1329 	vmx->nested.msrs.misc_high = data >> 32;
1330 
1331 	return 0;
1332 }
1333 
1334 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1335 {
1336 	u64 vmx_ept_vpid_cap;
1337 
1338 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1339 					   vmx->nested.msrs.vpid_caps);
1340 
1341 	/* Every bit is either reserved or a feature bit. */
1342 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1343 		return -EINVAL;
1344 
1345 	vmx->nested.msrs.ept_caps = data;
1346 	vmx->nested.msrs.vpid_caps = data >> 32;
1347 	return 0;
1348 }
1349 
1350 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1351 {
1352 	u64 *msr;
1353 
1354 	switch (msr_index) {
1355 	case MSR_IA32_VMX_CR0_FIXED0:
1356 		msr = &vmx->nested.msrs.cr0_fixed0;
1357 		break;
1358 	case MSR_IA32_VMX_CR4_FIXED0:
1359 		msr = &vmx->nested.msrs.cr4_fixed0;
1360 		break;
1361 	default:
1362 		BUG();
1363 	}
1364 
1365 	/*
1366 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1367 	 * must be 1 in the restored value.
1368 	 */
1369 	if (!is_bitwise_subset(data, *msr, -1ULL))
1370 		return -EINVAL;
1371 
1372 	*msr = data;
1373 	return 0;
1374 }
1375 
1376 /*
1377  * Called when userspace is restoring VMX MSRs.
1378  *
1379  * Returns 0 on success, non-0 otherwise.
1380  */
1381 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1382 {
1383 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1384 
1385 	/*
1386 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1387 	 * is in VMX operation.
1388 	 */
1389 	if (vmx->nested.vmxon)
1390 		return -EBUSY;
1391 
1392 	switch (msr_index) {
1393 	case MSR_IA32_VMX_BASIC:
1394 		return vmx_restore_vmx_basic(vmx, data);
1395 	case MSR_IA32_VMX_PINBASED_CTLS:
1396 	case MSR_IA32_VMX_PROCBASED_CTLS:
1397 	case MSR_IA32_VMX_EXIT_CTLS:
1398 	case MSR_IA32_VMX_ENTRY_CTLS:
1399 		/*
1400 		 * The "non-true" VMX capability MSRs are generated from the
1401 		 * "true" MSRs, so we do not support restoring them directly.
1402 		 *
1403 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1404 		 * should restore the "true" MSRs with the must-be-1 bits
1405 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1406 		 * DEFAULT SETTINGS".
1407 		 */
1408 		return -EINVAL;
1409 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1410 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1411 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1412 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1413 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1414 		return vmx_restore_control_msr(vmx, msr_index, data);
1415 	case MSR_IA32_VMX_MISC:
1416 		return vmx_restore_vmx_misc(vmx, data);
1417 	case MSR_IA32_VMX_CR0_FIXED0:
1418 	case MSR_IA32_VMX_CR4_FIXED0:
1419 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1420 	case MSR_IA32_VMX_CR0_FIXED1:
1421 	case MSR_IA32_VMX_CR4_FIXED1:
1422 		/*
1423 		 * These MSRs are generated based on the vCPU's CPUID, so we
1424 		 * do not support restoring them directly.
1425 		 */
1426 		return -EINVAL;
1427 	case MSR_IA32_VMX_EPT_VPID_CAP:
1428 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1429 	case MSR_IA32_VMX_VMCS_ENUM:
1430 		vmx->nested.msrs.vmcs_enum = data;
1431 		return 0;
1432 	case MSR_IA32_VMX_VMFUNC:
1433 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1434 			return -EINVAL;
1435 		vmx->nested.msrs.vmfunc_controls = data;
1436 		return 0;
1437 	default:
1438 		/*
1439 		 * The rest of the VMX capability MSRs do not support restore.
1440 		 */
1441 		return -EINVAL;
1442 	}
1443 }
1444 
1445 /* Returns 0 on success, non-0 otherwise. */
1446 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1447 {
1448 	switch (msr_index) {
1449 	case MSR_IA32_VMX_BASIC:
1450 		*pdata = msrs->basic;
1451 		break;
1452 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1453 	case MSR_IA32_VMX_PINBASED_CTLS:
1454 		*pdata = vmx_control_msr(
1455 			msrs->pinbased_ctls_low,
1456 			msrs->pinbased_ctls_high);
1457 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1458 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1459 		break;
1460 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1461 	case MSR_IA32_VMX_PROCBASED_CTLS:
1462 		*pdata = vmx_control_msr(
1463 			msrs->procbased_ctls_low,
1464 			msrs->procbased_ctls_high);
1465 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1466 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1467 		break;
1468 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1469 	case MSR_IA32_VMX_EXIT_CTLS:
1470 		*pdata = vmx_control_msr(
1471 			msrs->exit_ctls_low,
1472 			msrs->exit_ctls_high);
1473 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1474 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1475 		break;
1476 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1477 	case MSR_IA32_VMX_ENTRY_CTLS:
1478 		*pdata = vmx_control_msr(
1479 			msrs->entry_ctls_low,
1480 			msrs->entry_ctls_high);
1481 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1482 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1483 		break;
1484 	case MSR_IA32_VMX_MISC:
1485 		*pdata = vmx_control_msr(
1486 			msrs->misc_low,
1487 			msrs->misc_high);
1488 		break;
1489 	case MSR_IA32_VMX_CR0_FIXED0:
1490 		*pdata = msrs->cr0_fixed0;
1491 		break;
1492 	case MSR_IA32_VMX_CR0_FIXED1:
1493 		*pdata = msrs->cr0_fixed1;
1494 		break;
1495 	case MSR_IA32_VMX_CR4_FIXED0:
1496 		*pdata = msrs->cr4_fixed0;
1497 		break;
1498 	case MSR_IA32_VMX_CR4_FIXED1:
1499 		*pdata = msrs->cr4_fixed1;
1500 		break;
1501 	case MSR_IA32_VMX_VMCS_ENUM:
1502 		*pdata = msrs->vmcs_enum;
1503 		break;
1504 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1505 		*pdata = vmx_control_msr(
1506 			msrs->secondary_ctls_low,
1507 			msrs->secondary_ctls_high);
1508 		break;
1509 	case MSR_IA32_VMX_EPT_VPID_CAP:
1510 		*pdata = msrs->ept_caps |
1511 			((u64)msrs->vpid_caps << 32);
1512 		break;
1513 	case MSR_IA32_VMX_VMFUNC:
1514 		*pdata = msrs->vmfunc_controls;
1515 		break;
1516 	default:
1517 		return 1;
1518 	}
1519 
1520 	return 0;
1521 }
1522 
1523 /*
1524  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1525  * been modified by the L1 guest.  Note, "writable" in this context means
1526  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1527  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1528  * VM-exit information fields (which are actually writable if the vCPU is
1529  * configured to support "VMWRITE to any supported field in the VMCS").
1530  */
1531 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1532 {
1533 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1534 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1535 	struct shadow_vmcs_field field;
1536 	unsigned long val;
1537 	int i;
1538 
1539 	if (WARN_ON(!shadow_vmcs))
1540 		return;
1541 
1542 	preempt_disable();
1543 
1544 	vmcs_load(shadow_vmcs);
1545 
1546 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1547 		field = shadow_read_write_fields[i];
1548 		val = __vmcs_readl(field.encoding);
1549 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1550 	}
1551 
1552 	vmcs_clear(shadow_vmcs);
1553 	vmcs_load(vmx->loaded_vmcs->vmcs);
1554 
1555 	preempt_enable();
1556 }
1557 
1558 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1559 {
1560 	const struct shadow_vmcs_field *fields[] = {
1561 		shadow_read_write_fields,
1562 		shadow_read_only_fields
1563 	};
1564 	const int max_fields[] = {
1565 		max_shadow_read_write_fields,
1566 		max_shadow_read_only_fields
1567 	};
1568 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1569 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1570 	struct shadow_vmcs_field field;
1571 	unsigned long val;
1572 	int i, q;
1573 
1574 	if (WARN_ON(!shadow_vmcs))
1575 		return;
1576 
1577 	vmcs_load(shadow_vmcs);
1578 
1579 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1580 		for (i = 0; i < max_fields[q]; i++) {
1581 			field = fields[q][i];
1582 			val = vmcs12_read_any(vmcs12, field.encoding,
1583 					      field.offset);
1584 			__vmcs_writel(field.encoding, val);
1585 		}
1586 	}
1587 
1588 	vmcs_clear(shadow_vmcs);
1589 	vmcs_load(vmx->loaded_vmcs->vmcs);
1590 }
1591 
1592 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1593 {
1594 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1595 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1596 
1597 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1598 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1599 	vmcs12->guest_rip = evmcs->guest_rip;
1600 
1601 	if (unlikely(!(evmcs->hv_clean_fields &
1602 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1603 		vmcs12->guest_rsp = evmcs->guest_rsp;
1604 		vmcs12->guest_rflags = evmcs->guest_rflags;
1605 		vmcs12->guest_interruptibility_info =
1606 			evmcs->guest_interruptibility_info;
1607 	}
1608 
1609 	if (unlikely(!(evmcs->hv_clean_fields &
1610 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1611 		vmcs12->cpu_based_vm_exec_control =
1612 			evmcs->cpu_based_vm_exec_control;
1613 	}
1614 
1615 	if (unlikely(!(evmcs->hv_clean_fields &
1616 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1617 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1618 	}
1619 
1620 	if (unlikely(!(evmcs->hv_clean_fields &
1621 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1622 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1623 	}
1624 
1625 	if (unlikely(!(evmcs->hv_clean_fields &
1626 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1627 		vmcs12->vm_entry_intr_info_field =
1628 			evmcs->vm_entry_intr_info_field;
1629 		vmcs12->vm_entry_exception_error_code =
1630 			evmcs->vm_entry_exception_error_code;
1631 		vmcs12->vm_entry_instruction_len =
1632 			evmcs->vm_entry_instruction_len;
1633 	}
1634 
1635 	if (unlikely(!(evmcs->hv_clean_fields &
1636 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1637 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1638 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1639 		vmcs12->host_cr0 = evmcs->host_cr0;
1640 		vmcs12->host_cr3 = evmcs->host_cr3;
1641 		vmcs12->host_cr4 = evmcs->host_cr4;
1642 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1643 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1644 		vmcs12->host_rip = evmcs->host_rip;
1645 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1646 		vmcs12->host_es_selector = evmcs->host_es_selector;
1647 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1648 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1649 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1650 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1651 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1652 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1653 	}
1654 
1655 	if (unlikely(!(evmcs->hv_clean_fields &
1656 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1657 		vmcs12->pin_based_vm_exec_control =
1658 			evmcs->pin_based_vm_exec_control;
1659 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1660 		vmcs12->secondary_vm_exec_control =
1661 			evmcs->secondary_vm_exec_control;
1662 	}
1663 
1664 	if (unlikely(!(evmcs->hv_clean_fields &
1665 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1666 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1667 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1668 	}
1669 
1670 	if (unlikely(!(evmcs->hv_clean_fields &
1671 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1672 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1673 	}
1674 
1675 	if (unlikely(!(evmcs->hv_clean_fields &
1676 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1677 		vmcs12->guest_es_base = evmcs->guest_es_base;
1678 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1679 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1680 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1681 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1682 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1683 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1684 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1685 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1686 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1687 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1688 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1689 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1690 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1691 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1692 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1693 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1694 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1695 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1696 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1697 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1698 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1699 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1700 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1701 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1702 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1703 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1704 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1705 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1706 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1707 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1708 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1709 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1710 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1711 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1712 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1713 	}
1714 
1715 	if (unlikely(!(evmcs->hv_clean_fields &
1716 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1717 		vmcs12->tsc_offset = evmcs->tsc_offset;
1718 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1719 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1720 	}
1721 
1722 	if (unlikely(!(evmcs->hv_clean_fields &
1723 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1724 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1725 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1726 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1727 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1728 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1729 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1730 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1731 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1732 	}
1733 
1734 	if (unlikely(!(evmcs->hv_clean_fields &
1735 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1736 		vmcs12->host_fs_base = evmcs->host_fs_base;
1737 		vmcs12->host_gs_base = evmcs->host_gs_base;
1738 		vmcs12->host_tr_base = evmcs->host_tr_base;
1739 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1740 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1741 		vmcs12->host_rsp = evmcs->host_rsp;
1742 	}
1743 
1744 	if (unlikely(!(evmcs->hv_clean_fields &
1745 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1746 		vmcs12->ept_pointer = evmcs->ept_pointer;
1747 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1748 	}
1749 
1750 	if (unlikely(!(evmcs->hv_clean_fields &
1751 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1752 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1753 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1754 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1755 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1756 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1757 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1758 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1759 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1760 		vmcs12->guest_pending_dbg_exceptions =
1761 			evmcs->guest_pending_dbg_exceptions;
1762 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1763 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1764 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1765 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1766 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1767 	}
1768 
1769 	/*
1770 	 * Not used?
1771 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1772 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1773 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1774 	 * vmcs12->page_fault_error_code_mask =
1775 	 *		evmcs->page_fault_error_code_mask;
1776 	 * vmcs12->page_fault_error_code_match =
1777 	 *		evmcs->page_fault_error_code_match;
1778 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1779 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1780 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1781 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1782 	 */
1783 
1784 	/*
1785 	 * Read only fields:
1786 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1787 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1788 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1789 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1790 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1791 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1792 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1793 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1794 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1795 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1796 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1797 	 *
1798 	 * Not present in struct vmcs12:
1799 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1800 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1801 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1802 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1803 	 */
1804 
1805 	return 0;
1806 }
1807 
1808 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1809 {
1810 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1811 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1812 
1813 	/*
1814 	 * Should not be changed by KVM:
1815 	 *
1816 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1817 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1818 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1819 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1820 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1821 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1822 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1823 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1824 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1825 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1826 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1827 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1828 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1829 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1830 	 * evmcs->host_rip = vmcs12->host_rip;
1831 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1832 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1833 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1834 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1835 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1836 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1837 	 * evmcs->host_rsp = vmcs12->host_rsp;
1838 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1839 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1840 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1841 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1842 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1843 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1844 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1845 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1846 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1847 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1848 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1849 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1850 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1851 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1852 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1853 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1854 	 * evmcs->page_fault_error_code_mask =
1855 	 *		vmcs12->page_fault_error_code_mask;
1856 	 * evmcs->page_fault_error_code_match =
1857 	 *		vmcs12->page_fault_error_code_match;
1858 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1859 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1860 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1861 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1862 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1863 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1864 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1865 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1866 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1867 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1868 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1869 	 *
1870 	 * Not present in struct vmcs12:
1871 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1872 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1873 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1874 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1875 	 */
1876 
1877 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1878 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1879 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1880 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1881 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1882 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1883 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1884 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1885 
1886 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1887 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1888 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1889 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1890 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1891 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1892 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1893 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1894 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1895 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1896 
1897 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1898 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1899 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1900 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1901 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1902 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1903 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1904 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1905 
1906 	evmcs->guest_es_base = vmcs12->guest_es_base;
1907 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1908 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1909 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1910 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1911 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1912 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1913 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1914 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1915 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1916 
1917 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1918 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1919 
1920 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1921 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1922 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1923 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1924 
1925 	evmcs->guest_pending_dbg_exceptions =
1926 		vmcs12->guest_pending_dbg_exceptions;
1927 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1928 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1929 
1930 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1931 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1932 
1933 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1934 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1935 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1936 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1937 
1938 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1939 
1940 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1941 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1942 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1943 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1944 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1945 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1946 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1947 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1948 
1949 	evmcs->exit_qualification = vmcs12->exit_qualification;
1950 
1951 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1952 	evmcs->guest_rsp = vmcs12->guest_rsp;
1953 	evmcs->guest_rflags = vmcs12->guest_rflags;
1954 
1955 	evmcs->guest_interruptibility_info =
1956 		vmcs12->guest_interruptibility_info;
1957 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1958 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1959 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1960 	evmcs->vm_entry_exception_error_code =
1961 		vmcs12->vm_entry_exception_error_code;
1962 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1963 
1964 	evmcs->guest_rip = vmcs12->guest_rip;
1965 
1966 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1967 
1968 	return 0;
1969 }
1970 
1971 /*
1972  * This is an equivalent of the nested hypervisor executing the vmptrld
1973  * instruction.
1974  */
1975 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1976 	struct kvm_vcpu *vcpu, bool from_launch)
1977 {
1978 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1979 	bool evmcs_gpa_changed = false;
1980 	u64 evmcs_gpa;
1981 
1982 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1983 		return EVMPTRLD_DISABLED;
1984 
1985 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa))
1986 		return EVMPTRLD_DISABLED;
1987 
1988 	if (unlikely(!vmx->nested.hv_evmcs ||
1989 		     evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1990 		if (!vmx->nested.hv_evmcs)
1991 			vmx->nested.current_vmptr = -1ull;
1992 
1993 		nested_release_evmcs(vcpu);
1994 
1995 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1996 				 &vmx->nested.hv_evmcs_map))
1997 			return EVMPTRLD_ERROR;
1998 
1999 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2000 
2001 		/*
2002 		 * Currently, KVM only supports eVMCS version 1
2003 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2004 		 * value to first u32 field of eVMCS which should specify eVMCS
2005 		 * VersionNumber.
2006 		 *
2007 		 * Guest should be aware of supported eVMCS versions by host by
2008 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2009 		 * expected to set this CPUID leaf according to the value
2010 		 * returned in vmcs_version from nested_enable_evmcs().
2011 		 *
2012 		 * However, it turns out that Microsoft Hyper-V fails to comply
2013 		 * to their own invented interface: When Hyper-V use eVMCS, it
2014 		 * just sets first u32 field of eVMCS to revision_id specified
2015 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2016 		 * which is one of the supported versions specified in
2017 		 * CPUID.0x4000000A.EAX[0:15].
2018 		 *
2019 		 * To overcome Hyper-V bug, we accept here either a supported
2020 		 * eVMCS version or VMCS12 revision_id as valid values for first
2021 		 * u32 field of eVMCS.
2022 		 */
2023 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2024 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2025 			nested_release_evmcs(vcpu);
2026 			return EVMPTRLD_VMFAIL;
2027 		}
2028 
2029 		vmx->nested.dirty_vmcs12 = true;
2030 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2031 
2032 		evmcs_gpa_changed = true;
2033 		/*
2034 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2035 		 * reloaded from guest's memory (read only fields, fields not
2036 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2037 		 * are no leftovers.
2038 		 */
2039 		if (from_launch) {
2040 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2041 			memset(vmcs12, 0, sizeof(*vmcs12));
2042 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2043 		}
2044 
2045 	}
2046 
2047 	/*
2048 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2049 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2050 	 */
2051 	if (from_launch || evmcs_gpa_changed)
2052 		vmx->nested.hv_evmcs->hv_clean_fields &=
2053 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2054 
2055 	return EVMPTRLD_SUCCEEDED;
2056 }
2057 
2058 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2059 {
2060 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2061 
2062 	if (vmx->nested.hv_evmcs) {
2063 		copy_vmcs12_to_enlightened(vmx);
2064 		/* All fields are clean */
2065 		vmx->nested.hv_evmcs->hv_clean_fields |=
2066 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2067 	} else {
2068 		copy_vmcs12_to_shadow(vmx);
2069 	}
2070 
2071 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2072 }
2073 
2074 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2075 {
2076 	struct vcpu_vmx *vmx =
2077 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2078 
2079 	vmx->nested.preemption_timer_expired = true;
2080 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2081 	kvm_vcpu_kick(&vmx->vcpu);
2082 
2083 	return HRTIMER_NORESTART;
2084 }
2085 
2086 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2087 {
2088 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2089 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2090 
2091 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2092 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2093 
2094 	if (!vmx->nested.has_preemption_timer_deadline) {
2095 		vmx->nested.preemption_timer_deadline =
2096 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2097 		vmx->nested.has_preemption_timer_deadline = true;
2098 	}
2099 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2100 }
2101 
2102 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2103 					u64 preemption_timeout)
2104 {
2105 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2106 
2107 	/*
2108 	 * A timer value of zero is architecturally guaranteed to cause
2109 	 * a VMExit prior to executing any instructions in the guest.
2110 	 */
2111 	if (preemption_timeout == 0) {
2112 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2113 		return;
2114 	}
2115 
2116 	if (vcpu->arch.virtual_tsc_khz == 0)
2117 		return;
2118 
2119 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2120 	preemption_timeout *= 1000000;
2121 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2122 	hrtimer_start(&vmx->nested.preemption_timer,
2123 		      ktime_add_ns(ktime_get(), preemption_timeout),
2124 		      HRTIMER_MODE_ABS_PINNED);
2125 }
2126 
2127 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2128 {
2129 	if (vmx->nested.nested_run_pending &&
2130 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2131 		return vmcs12->guest_ia32_efer;
2132 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2133 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2134 	else
2135 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2136 }
2137 
2138 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2139 {
2140 	/*
2141 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2142 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2143 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2144 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2145 	 */
2146 	if (vmx->nested.vmcs02_initialized)
2147 		return;
2148 	vmx->nested.vmcs02_initialized = true;
2149 
2150 	/*
2151 	 * We don't care what the EPTP value is we just need to guarantee
2152 	 * it's valid so we don't get a false positive when doing early
2153 	 * consistency checks.
2154 	 */
2155 	if (enable_ept && nested_early_check)
2156 		vmcs_write64(EPT_POINTER,
2157 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2158 
2159 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2160 	if (cpu_has_vmx_vmfunc())
2161 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2162 
2163 	if (cpu_has_vmx_posted_intr())
2164 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2165 
2166 	if (cpu_has_vmx_msr_bitmap())
2167 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2168 
2169 	/*
2170 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2171 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2172 	 * deal with filtering out L2 GPAs from the buffer.
2173 	 */
2174 	if (enable_pml) {
2175 		vmcs_write64(PML_ADDRESS, 0);
2176 		vmcs_write16(GUEST_PML_INDEX, -1);
2177 	}
2178 
2179 	if (cpu_has_vmx_encls_vmexit())
2180 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2181 
2182 	/*
2183 	 * Set the MSR load/store lists to match L0's settings.  Only the
2184 	 * addresses are constant (for vmcs02), the counts can change based
2185 	 * on L2's behavior, e.g. switching to/from long mode.
2186 	 */
2187 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2188 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2189 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2190 
2191 	vmx_set_constant_host_state(vmx);
2192 }
2193 
2194 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2195 				      struct vmcs12 *vmcs12)
2196 {
2197 	prepare_vmcs02_constant_state(vmx);
2198 
2199 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2200 
2201 	if (enable_vpid) {
2202 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2203 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2204 		else
2205 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2206 	}
2207 }
2208 
2209 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2210 {
2211 	u32 exec_control;
2212 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2213 
2214 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
2215 		prepare_vmcs02_early_rare(vmx, vmcs12);
2216 
2217 	/*
2218 	 * PIN CONTROLS
2219 	 */
2220 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2221 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2222 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2223 
2224 	/* Posted interrupts setting is only taken from vmcs12.  */
2225 	if (nested_cpu_has_posted_intr(vmcs12)) {
2226 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2227 		vmx->nested.pi_pending = false;
2228 	} else {
2229 		exec_control &= ~PIN_BASED_POSTED_INTR;
2230 	}
2231 	pin_controls_set(vmx, exec_control);
2232 
2233 	/*
2234 	 * EXEC CONTROLS
2235 	 */
2236 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2237 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2238 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2239 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2240 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2241 
2242 	vmx->nested.l1_tpr_threshold = -1;
2243 	if (exec_control & CPU_BASED_TPR_SHADOW)
2244 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2245 #ifdef CONFIG_X86_64
2246 	else
2247 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2248 				CPU_BASED_CR8_STORE_EXITING;
2249 #endif
2250 
2251 	/*
2252 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2253 	 * for I/O port accesses.
2254 	 */
2255 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2256 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2257 
2258 	/*
2259 	 * This bit will be computed in nested_get_vmcs12_pages, because
2260 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2261 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2262 	 * only set/clear this bit.
2263 	 */
2264 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2265 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2266 
2267 	exec_controls_set(vmx, exec_control);
2268 
2269 	/*
2270 	 * SECONDARY EXEC CONTROLS
2271 	 */
2272 	if (cpu_has_secondary_exec_ctrls()) {
2273 		exec_control = vmx->secondary_exec_control;
2274 
2275 		/* Take the following fields only from vmcs12 */
2276 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2277 				  SECONDARY_EXEC_ENABLE_INVPCID |
2278 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2279 				  SECONDARY_EXEC_XSAVES |
2280 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2281 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2282 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2283 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2284 		if (nested_cpu_has(vmcs12,
2285 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2286 			exec_control |= vmcs12->secondary_vm_exec_control;
2287 
2288 		/* PML is emulated and never enabled in hardware for L2. */
2289 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2290 
2291 		/* VMCS shadowing for L2 is emulated for now */
2292 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2293 
2294 		/*
2295 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2296 		 * will not have to rewrite the controls just for this bit.
2297 		 */
2298 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2299 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2300 			exec_control |= SECONDARY_EXEC_DESC;
2301 
2302 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2303 			vmcs_write16(GUEST_INTR_STATUS,
2304 				vmcs12->guest_intr_status);
2305 
2306 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2307 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2308 
2309 		secondary_exec_controls_set(vmx, exec_control);
2310 	}
2311 
2312 	/*
2313 	 * ENTRY CONTROLS
2314 	 *
2315 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2316 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2317 	 * on the related bits (if supported by the CPU) in the hope that
2318 	 * we can avoid VMWrites during vmx_set_efer().
2319 	 */
2320 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2321 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2322 	if (cpu_has_load_ia32_efer()) {
2323 		if (guest_efer & EFER_LMA)
2324 			exec_control |= VM_ENTRY_IA32E_MODE;
2325 		if (guest_efer != host_efer)
2326 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2327 	}
2328 	vm_entry_controls_set(vmx, exec_control);
2329 
2330 	/*
2331 	 * EXIT CONTROLS
2332 	 *
2333 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2334 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2335 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2336 	 */
2337 	exec_control = vmx_vmexit_ctrl();
2338 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2339 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2340 	vm_exit_controls_set(vmx, exec_control);
2341 
2342 	/*
2343 	 * Interrupt/Exception Fields
2344 	 */
2345 	if (vmx->nested.nested_run_pending) {
2346 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2347 			     vmcs12->vm_entry_intr_info_field);
2348 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2349 			     vmcs12->vm_entry_exception_error_code);
2350 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2351 			     vmcs12->vm_entry_instruction_len);
2352 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2353 			     vmcs12->guest_interruptibility_info);
2354 		vmx->loaded_vmcs->nmi_known_unmasked =
2355 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2356 	} else {
2357 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2358 	}
2359 }
2360 
2361 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2362 {
2363 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2364 
2365 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2366 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2367 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2368 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2369 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2370 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2371 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2372 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2373 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2374 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2375 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2376 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2377 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2378 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2379 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2380 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2381 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2382 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2383 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2384 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2385 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2386 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2387 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2388 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2389 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2390 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2391 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2392 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2393 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2394 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2395 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2396 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2397 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2398 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2399 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2400 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2401 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2402 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2403 
2404 		vmx->segment_cache.bitmask = 0;
2405 	}
2406 
2407 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2408 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2409 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2410 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2411 			    vmcs12->guest_pending_dbg_exceptions);
2412 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2413 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2414 
2415 		/*
2416 		 * L1 may access the L2's PDPTR, so save them to construct
2417 		 * vmcs12
2418 		 */
2419 		if (enable_ept) {
2420 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2421 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2422 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2423 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2424 		}
2425 
2426 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2427 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2428 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2429 	}
2430 
2431 	if (nested_cpu_has_xsaves(vmcs12))
2432 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2433 
2434 	/*
2435 	 * Whether page-faults are trapped is determined by a combination of
2436 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2437 	 * doesn't care about page faults then we should set all of these to
2438 	 * L1's desires. However, if L0 does care about (some) page faults, it
2439 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2440 	 * simply ask to exit on each and every L2 page fault. This is done by
2441 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2442 	 * Note that below we don't need special code to set EB.PF beyond the
2443 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2444 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2445 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2446 	 */
2447 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2448 		/*
2449 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2450 		 * go ahead and use it?
2451 		 */
2452 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2453 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2454 	} else {
2455 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2456 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2457 	}
2458 
2459 	if (cpu_has_vmx_apicv()) {
2460 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2461 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2462 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2463 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2464 	}
2465 
2466 	/*
2467 	 * Make sure the msr_autostore list is up to date before we set the
2468 	 * count in the vmcs02.
2469 	 */
2470 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2471 
2472 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2473 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2474 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2475 
2476 	set_cr4_guest_host_mask(vmx);
2477 }
2478 
2479 /*
2480  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2481  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2482  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2483  * guest in a way that will both be appropriate to L1's requests, and our
2484  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2485  * function also has additional necessary side-effects, like setting various
2486  * vcpu->arch fields.
2487  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2488  * is assigned to entry_failure_code on failure.
2489  */
2490 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2491 			  enum vm_entry_failure_code *entry_failure_code)
2492 {
2493 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2494 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2495 	bool load_guest_pdptrs_vmcs12 = false;
2496 
2497 	if (vmx->nested.dirty_vmcs12 || hv_evmcs) {
2498 		prepare_vmcs02_rare(vmx, vmcs12);
2499 		vmx->nested.dirty_vmcs12 = false;
2500 
2501 		load_guest_pdptrs_vmcs12 = !hv_evmcs ||
2502 			!(hv_evmcs->hv_clean_fields &
2503 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2504 	}
2505 
2506 	if (vmx->nested.nested_run_pending &&
2507 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2508 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2509 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2510 	} else {
2511 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2512 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2513 	}
2514 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2515 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2516 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2517 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2518 
2519 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2520 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2521 	 * trap. Note that CR0.TS also needs updating - we do this later.
2522 	 */
2523 	vmx_update_exception_bitmap(vcpu);
2524 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2525 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2526 
2527 	if (vmx->nested.nested_run_pending &&
2528 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2529 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2530 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2531 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2532 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2533 	}
2534 
2535 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2536 
2537 	if (kvm_has_tsc_control)
2538 		decache_tsc_multiplier(vmx);
2539 
2540 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2541 
2542 	if (nested_cpu_has_ept(vmcs12))
2543 		nested_ept_init_mmu_context(vcpu);
2544 
2545 	/*
2546 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2547 	 * bits which we consider mandatory enabled.
2548 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2549 	 * the specifications by L1; It's not enough to take
2550 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2551 	 * have more bits than L1 expected.
2552 	 */
2553 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2554 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2555 
2556 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2557 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2558 
2559 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2560 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2561 	vmx_set_efer(vcpu, vcpu->arch.efer);
2562 
2563 	/*
2564 	 * Guest state is invalid and unrestricted guest is disabled,
2565 	 * which means L1 attempted VMEntry to L2 with invalid state.
2566 	 * Fail the VMEntry.
2567 	 */
2568 	if (CC(!vmx_guest_state_valid(vcpu))) {
2569 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2570 		return -EINVAL;
2571 	}
2572 
2573 	/* Shadow page tables on either EPT or shadow page tables. */
2574 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2575 				entry_failure_code))
2576 		return -EINVAL;
2577 
2578 	/*
2579 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2580 	 * on nested VM-Exit, which can occur without actually running L2 and
2581 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2582 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2583 	 * transition to HLT instead of running L2.
2584 	 */
2585 	if (enable_ept)
2586 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2587 
2588 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2589 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2590 	    is_pae_paging(vcpu)) {
2591 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2592 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2593 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2594 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2595 	}
2596 
2597 	if (!enable_ept)
2598 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2599 
2600 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2601 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2602 				     vmcs12->guest_ia32_perf_global_ctrl)))
2603 		return -EINVAL;
2604 
2605 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2606 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2607 	return 0;
2608 }
2609 
2610 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2611 {
2612 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2613 	       nested_cpu_has_virtual_nmis(vmcs12)))
2614 		return -EINVAL;
2615 
2616 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2617 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2618 		return -EINVAL;
2619 
2620 	return 0;
2621 }
2622 
2623 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2624 {
2625 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2626 
2627 	/* Check for memory type validity */
2628 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2629 	case VMX_EPTP_MT_UC:
2630 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2631 			return false;
2632 		break;
2633 	case VMX_EPTP_MT_WB:
2634 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2635 			return false;
2636 		break;
2637 	default:
2638 		return false;
2639 	}
2640 
2641 	/* Page-walk levels validity. */
2642 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2643 	case VMX_EPTP_PWL_5:
2644 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2645 			return false;
2646 		break;
2647 	case VMX_EPTP_PWL_4:
2648 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2649 			return false;
2650 		break;
2651 	default:
2652 		return false;
2653 	}
2654 
2655 	/* Reserved bits should not be set */
2656 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2657 		return false;
2658 
2659 	/* AD, if set, should be supported */
2660 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2661 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2662 			return false;
2663 	}
2664 
2665 	return true;
2666 }
2667 
2668 /*
2669  * Checks related to VM-Execution Control Fields
2670  */
2671 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2672                                               struct vmcs12 *vmcs12)
2673 {
2674 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2675 
2676 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2677 				   vmx->nested.msrs.pinbased_ctls_low,
2678 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2679 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2680 				   vmx->nested.msrs.procbased_ctls_low,
2681 				   vmx->nested.msrs.procbased_ctls_high)))
2682 		return -EINVAL;
2683 
2684 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2685 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2686 				   vmx->nested.msrs.secondary_ctls_low,
2687 				   vmx->nested.msrs.secondary_ctls_high)))
2688 		return -EINVAL;
2689 
2690 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2691 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2692 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2693 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2694 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2695 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2696 	    nested_vmx_check_nmi_controls(vmcs12) ||
2697 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2698 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2699 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2700 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2701 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2702 		return -EINVAL;
2703 
2704 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2705 	    nested_cpu_has_save_preemption_timer(vmcs12))
2706 		return -EINVAL;
2707 
2708 	if (nested_cpu_has_ept(vmcs12) &&
2709 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2710 		return -EINVAL;
2711 
2712 	if (nested_cpu_has_vmfunc(vmcs12)) {
2713 		if (CC(vmcs12->vm_function_control &
2714 		       ~vmx->nested.msrs.vmfunc_controls))
2715 			return -EINVAL;
2716 
2717 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2718 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2719 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2720 				return -EINVAL;
2721 		}
2722 	}
2723 
2724 	return 0;
2725 }
2726 
2727 /*
2728  * Checks related to VM-Exit Control Fields
2729  */
2730 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2731                                          struct vmcs12 *vmcs12)
2732 {
2733 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2734 
2735 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2736 				    vmx->nested.msrs.exit_ctls_low,
2737 				    vmx->nested.msrs.exit_ctls_high)) ||
2738 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2739 		return -EINVAL;
2740 
2741 	return 0;
2742 }
2743 
2744 /*
2745  * Checks related to VM-Entry Control Fields
2746  */
2747 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2748 					  struct vmcs12 *vmcs12)
2749 {
2750 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2751 
2752 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2753 				    vmx->nested.msrs.entry_ctls_low,
2754 				    vmx->nested.msrs.entry_ctls_high)))
2755 		return -EINVAL;
2756 
2757 	/*
2758 	 * From the Intel SDM, volume 3:
2759 	 * Fields relevant to VM-entry event injection must be set properly.
2760 	 * These fields are the VM-entry interruption-information field, the
2761 	 * VM-entry exception error code, and the VM-entry instruction length.
2762 	 */
2763 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2764 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2765 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2766 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2767 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2768 		bool should_have_error_code;
2769 		bool urg = nested_cpu_has2(vmcs12,
2770 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2771 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2772 
2773 		/* VM-entry interruption-info field: interruption type */
2774 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2775 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2776 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2777 			return -EINVAL;
2778 
2779 		/* VM-entry interruption-info field: vector */
2780 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2781 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2782 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2783 			return -EINVAL;
2784 
2785 		/* VM-entry interruption-info field: deliver error code */
2786 		should_have_error_code =
2787 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2788 			x86_exception_has_error_code(vector);
2789 		if (CC(has_error_code != should_have_error_code))
2790 			return -EINVAL;
2791 
2792 		/* VM-entry exception error code */
2793 		if (CC(has_error_code &&
2794 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2795 			return -EINVAL;
2796 
2797 		/* VM-entry interruption-info field: reserved bits */
2798 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2799 			return -EINVAL;
2800 
2801 		/* VM-entry instruction length */
2802 		switch (intr_type) {
2803 		case INTR_TYPE_SOFT_EXCEPTION:
2804 		case INTR_TYPE_SOFT_INTR:
2805 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2806 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2807 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2808 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2809 				return -EINVAL;
2810 		}
2811 	}
2812 
2813 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2814 		return -EINVAL;
2815 
2816 	return 0;
2817 }
2818 
2819 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2820 				     struct vmcs12 *vmcs12)
2821 {
2822 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2823 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2824 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2825 		return -EINVAL;
2826 
2827 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2828 		return nested_evmcs_check_controls(vmcs12);
2829 
2830 	return 0;
2831 }
2832 
2833 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2834 				       struct vmcs12 *vmcs12)
2835 {
2836 	bool ia32e;
2837 
2838 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2839 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2840 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2841 		return -EINVAL;
2842 
2843 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2844 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2845 		return -EINVAL;
2846 
2847 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2848 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2849 		return -EINVAL;
2850 
2851 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2852 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2853 					   vmcs12->host_ia32_perf_global_ctrl)))
2854 		return -EINVAL;
2855 
2856 #ifdef CONFIG_X86_64
2857 	ia32e = !!(vcpu->arch.efer & EFER_LMA);
2858 #else
2859 	ia32e = false;
2860 #endif
2861 
2862 	if (ia32e) {
2863 		if (CC(!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)) ||
2864 		    CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2865 			return -EINVAL;
2866 	} else {
2867 		if (CC(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) ||
2868 		    CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2869 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2870 		    CC((vmcs12->host_rip) >> 32))
2871 			return -EINVAL;
2872 	}
2873 
2874 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2875 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2876 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2877 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2878 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2879 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2880 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2881 	    CC(vmcs12->host_cs_selector == 0) ||
2882 	    CC(vmcs12->host_tr_selector == 0) ||
2883 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2884 		return -EINVAL;
2885 
2886 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2887 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2888 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2889 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2890 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2891 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2892 		return -EINVAL;
2893 
2894 	/*
2895 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2896 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2897 	 * the values of the LMA and LME bits in the field must each be that of
2898 	 * the host address-space size VM-exit control.
2899 	 */
2900 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2901 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2902 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2903 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2904 			return -EINVAL;
2905 	}
2906 
2907 	return 0;
2908 }
2909 
2910 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2911 					  struct vmcs12 *vmcs12)
2912 {
2913 	int r = 0;
2914 	struct vmcs12 *shadow;
2915 	struct kvm_host_map map;
2916 
2917 	if (vmcs12->vmcs_link_pointer == -1ull)
2918 		return 0;
2919 
2920 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2921 		return -EINVAL;
2922 
2923 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2924 		return -EINVAL;
2925 
2926 	shadow = map.hva;
2927 
2928 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2929 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2930 		r = -EINVAL;
2931 
2932 	kvm_vcpu_unmap(vcpu, &map, false);
2933 	return r;
2934 }
2935 
2936 /*
2937  * Checks related to Guest Non-register State
2938  */
2939 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2940 {
2941 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2942 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2943 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2944 		return -EINVAL;
2945 
2946 	return 0;
2947 }
2948 
2949 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2950 					struct vmcs12 *vmcs12,
2951 					enum vm_entry_failure_code *entry_failure_code)
2952 {
2953 	bool ia32e;
2954 
2955 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
2956 
2957 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2958 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2959 		return -EINVAL;
2960 
2961 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2962 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2963 		return -EINVAL;
2964 
2965 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2966 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2967 		return -EINVAL;
2968 
2969 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2970 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
2971 		return -EINVAL;
2972 	}
2973 
2974 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2975 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2976 					   vmcs12->guest_ia32_perf_global_ctrl)))
2977 		return -EINVAL;
2978 
2979 	/*
2980 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2981 	 * are performed on the field for the IA32_EFER MSR:
2982 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2983 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2984 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2985 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2986 	 *   CR0.PG) is 1.
2987 	 */
2988 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2989 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2990 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2991 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
2992 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
2993 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
2994 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
2995 			return -EINVAL;
2996 	}
2997 
2998 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2999 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3000 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3001 		return -EINVAL;
3002 
3003 	if (nested_check_guest_non_reg_state(vmcs12))
3004 		return -EINVAL;
3005 
3006 	return 0;
3007 }
3008 
3009 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3010 {
3011 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3012 	unsigned long cr3, cr4;
3013 	bool vm_fail;
3014 
3015 	if (!nested_early_check)
3016 		return 0;
3017 
3018 	if (vmx->msr_autoload.host.nr)
3019 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3020 	if (vmx->msr_autoload.guest.nr)
3021 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3022 
3023 	preempt_disable();
3024 
3025 	vmx_prepare_switch_to_guest(vcpu);
3026 
3027 	/*
3028 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3029 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3030 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3031 	 * there is no need to preserve other bits or save/restore the field.
3032 	 */
3033 	vmcs_writel(GUEST_RFLAGS, 0);
3034 
3035 	cr3 = __get_current_cr3_fast();
3036 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3037 		vmcs_writel(HOST_CR3, cr3);
3038 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3039 	}
3040 
3041 	cr4 = cr4_read_shadow();
3042 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3043 		vmcs_writel(HOST_CR4, cr4);
3044 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3045 	}
3046 
3047 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3048 				 vmx->loaded_vmcs->launched);
3049 
3050 	if (vmx->msr_autoload.host.nr)
3051 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3052 	if (vmx->msr_autoload.guest.nr)
3053 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3054 
3055 	if (vm_fail) {
3056 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3057 
3058 		preempt_enable();
3059 
3060 		trace_kvm_nested_vmenter_failed(
3061 			"early hardware check VM-instruction error: ", error);
3062 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3063 		return 1;
3064 	}
3065 
3066 	/*
3067 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3068 	 */
3069 	if (hw_breakpoint_active())
3070 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3071 	local_irq_enable();
3072 	preempt_enable();
3073 
3074 	/*
3075 	 * A non-failing VMEntry means we somehow entered guest mode with
3076 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3077 	 * is no telling what memory has been modified or what state has
3078 	 * been exposed to unknown code.  Hitting this all but guarantees
3079 	 * a (very critical) hardware issue.
3080 	 */
3081 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3082 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3083 
3084 	return 0;
3085 }
3086 
3087 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3088 {
3089 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3090 
3091 	/*
3092 	 * hv_evmcs may end up being not mapped after migration (when
3093 	 * L2 was running), map it here to make sure vmcs12 changes are
3094 	 * properly reflected.
3095 	 */
3096 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs) {
3097 		enum nested_evmptrld_status evmptrld_status =
3098 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3099 
3100 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3101 		    evmptrld_status == EVMPTRLD_ERROR) {
3102 			pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3103 					     __func__);
3104 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3105 			vcpu->run->internal.suberror =
3106 				KVM_INTERNAL_ERROR_EMULATION;
3107 			vcpu->run->internal.ndata = 0;
3108 			return false;
3109 		}
3110 	}
3111 
3112 	return true;
3113 }
3114 
3115 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3116 {
3117 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3118 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3119 	struct kvm_host_map *map;
3120 	struct page *page;
3121 	u64 hpa;
3122 
3123 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3124 		/*
3125 		 * Translate L1 physical address to host physical
3126 		 * address for vmcs02. Keep the page pinned, so this
3127 		 * physical address remains valid. We keep a reference
3128 		 * to it so we can release it later.
3129 		 */
3130 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3131 			kvm_release_page_clean(vmx->nested.apic_access_page);
3132 			vmx->nested.apic_access_page = NULL;
3133 		}
3134 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3135 		if (!is_error_page(page)) {
3136 			vmx->nested.apic_access_page = page;
3137 			hpa = page_to_phys(vmx->nested.apic_access_page);
3138 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3139 		} else {
3140 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3141 					     __func__);
3142 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3143 			vcpu->run->internal.suberror =
3144 				KVM_INTERNAL_ERROR_EMULATION;
3145 			vcpu->run->internal.ndata = 0;
3146 			return false;
3147 		}
3148 	}
3149 
3150 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3151 		map = &vmx->nested.virtual_apic_map;
3152 
3153 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3154 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3155 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3156 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3157 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3158 			/*
3159 			 * The processor will never use the TPR shadow, simply
3160 			 * clear the bit from the execution control.  Such a
3161 			 * configuration is useless, but it happens in tests.
3162 			 * For any other configuration, failing the vm entry is
3163 			 * _not_ what the processor does but it's basically the
3164 			 * only possibility we have.
3165 			 */
3166 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3167 		} else {
3168 			/*
3169 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3170 			 * force VM-Entry to fail.
3171 			 */
3172 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
3173 		}
3174 	}
3175 
3176 	if (nested_cpu_has_posted_intr(vmcs12)) {
3177 		map = &vmx->nested.pi_desc_map;
3178 
3179 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3180 			vmx->nested.pi_desc =
3181 				(struct pi_desc *)(((void *)map->hva) +
3182 				offset_in_page(vmcs12->posted_intr_desc_addr));
3183 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3184 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3185 		}
3186 	}
3187 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3188 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3189 	else
3190 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3191 
3192 	return true;
3193 }
3194 
3195 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3196 {
3197 	if (!nested_get_evmcs_page(vcpu))
3198 		return false;
3199 
3200 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3201 		return false;
3202 
3203 	return true;
3204 }
3205 
3206 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3207 {
3208 	struct vmcs12 *vmcs12;
3209 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3210 	gpa_t dst;
3211 
3212 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3213 		return 0;
3214 
3215 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3216 		return 1;
3217 
3218 	/*
3219 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3220 	 * set is already checked as part of A/D emulation.
3221 	 */
3222 	vmcs12 = get_vmcs12(vcpu);
3223 	if (!nested_cpu_has_pml(vmcs12))
3224 		return 0;
3225 
3226 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3227 		vmx->nested.pml_full = true;
3228 		return 1;
3229 	}
3230 
3231 	gpa &= ~0xFFFull;
3232 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3233 
3234 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3235 				 offset_in_page(dst), sizeof(gpa)))
3236 		return 0;
3237 
3238 	vmcs12->guest_pml_index--;
3239 
3240 	return 0;
3241 }
3242 
3243 /*
3244  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3245  * for running VMX instructions (except VMXON, whose prerequisites are
3246  * slightly different). It also specifies what exception to inject otherwise.
3247  * Note that many of these exceptions have priority over VM exits, so they
3248  * don't have to be checked again here.
3249  */
3250 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3251 {
3252 	if (!to_vmx(vcpu)->nested.vmxon) {
3253 		kvm_queue_exception(vcpu, UD_VECTOR);
3254 		return 0;
3255 	}
3256 
3257 	if (vmx_get_cpl(vcpu)) {
3258 		kvm_inject_gp(vcpu, 0);
3259 		return 0;
3260 	}
3261 
3262 	return 1;
3263 }
3264 
3265 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3266 {
3267 	u8 rvi = vmx_get_rvi();
3268 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3269 
3270 	return ((rvi & 0xf0) > (vppr & 0xf0));
3271 }
3272 
3273 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3274 				   struct vmcs12 *vmcs12);
3275 
3276 /*
3277  * If from_vmentry is false, this is being called from state restore (either RSM
3278  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3279  *
3280  * Returns:
3281  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3282  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3283  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3284  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3285  */
3286 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3287 							bool from_vmentry)
3288 {
3289 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3290 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3291 	enum vm_entry_failure_code entry_failure_code;
3292 	bool evaluate_pending_interrupts;
3293 	union vmx_exit_reason exit_reason = {
3294 		.basic = EXIT_REASON_INVALID_STATE,
3295 		.failed_vmentry = 1,
3296 	};
3297 	u32 failed_index;
3298 
3299 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3300 		kvm_vcpu_flush_tlb_current(vcpu);
3301 
3302 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3303 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3304 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3305 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3306 
3307 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3308 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3309 	if (kvm_mpx_supported() &&
3310 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3311 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3312 
3313 	/*
3314 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3315 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3316 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3317 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3318 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3319 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3320 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3321 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3322 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3323 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3324 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3325 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3326 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3327 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3328 	 */
3329 	if (!enable_ept && !nested_early_check)
3330 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3331 
3332 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3333 
3334 	prepare_vmcs02_early(vmx, vmcs12);
3335 
3336 	if (from_vmentry) {
3337 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3338 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3339 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3340 		}
3341 
3342 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3343 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3344 			return NVMX_VMENTRY_VMFAIL;
3345 		}
3346 
3347 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3348 						 &entry_failure_code)) {
3349 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3350 			vmcs12->exit_qualification = entry_failure_code;
3351 			goto vmentry_fail_vmexit;
3352 		}
3353 	}
3354 
3355 	enter_guest_mode(vcpu);
3356 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3357 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3358 
3359 	if (prepare_vmcs02(vcpu, vmcs12, &entry_failure_code)) {
3360 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3361 		vmcs12->exit_qualification = entry_failure_code;
3362 		goto vmentry_fail_vmexit_guest_mode;
3363 	}
3364 
3365 	if (from_vmentry) {
3366 		failed_index = nested_vmx_load_msr(vcpu,
3367 						   vmcs12->vm_entry_msr_load_addr,
3368 						   vmcs12->vm_entry_msr_load_count);
3369 		if (failed_index) {
3370 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3371 			vmcs12->exit_qualification = failed_index;
3372 			goto vmentry_fail_vmexit_guest_mode;
3373 		}
3374 	} else {
3375 		/*
3376 		 * The MMU is not initialized to point at the right entities yet and
3377 		 * "get pages" would need to read data from the guest (i.e. we will
3378 		 * need to perform gpa to hpa translation). Request a call
3379 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3380 		 * have already been set at vmentry time and should not be reset.
3381 		 */
3382 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3383 	}
3384 
3385 	/*
3386 	 * If L1 had a pending IRQ/NMI until it executed
3387 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3388 	 * disallowed (e.g. interrupts disabled), L0 needs to
3389 	 * evaluate if this pending event should cause an exit from L2
3390 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3391 	 * intercept EXTERNAL_INTERRUPT).
3392 	 *
3393 	 * Usually this would be handled by the processor noticing an
3394 	 * IRQ/NMI window request, or checking RVI during evaluation of
3395 	 * pending virtual interrupts.  However, this setting was done
3396 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3397 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3398 	 */
3399 	if (unlikely(evaluate_pending_interrupts))
3400 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3401 
3402 	/*
3403 	 * Do not start the preemption timer hrtimer until after we know
3404 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3405 	 * the timer.
3406 	 */
3407 	vmx->nested.preemption_timer_expired = false;
3408 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3409 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3410 		vmx_start_preemption_timer(vcpu, timer_value);
3411 	}
3412 
3413 	/*
3414 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3415 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3416 	 * returned as far as L1 is concerned. It will only return (and set
3417 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3418 	 */
3419 	return NVMX_VMENTRY_SUCCESS;
3420 
3421 	/*
3422 	 * A failed consistency check that leads to a VMExit during L1's
3423 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3424 	 * 26.7 "VM-entry failures during or after loading guest state".
3425 	 */
3426 vmentry_fail_vmexit_guest_mode:
3427 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3428 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3429 	leave_guest_mode(vcpu);
3430 
3431 vmentry_fail_vmexit:
3432 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3433 
3434 	if (!from_vmentry)
3435 		return NVMX_VMENTRY_VMEXIT;
3436 
3437 	load_vmcs12_host_state(vcpu, vmcs12);
3438 	vmcs12->vm_exit_reason = exit_reason.full;
3439 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3440 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3441 	return NVMX_VMENTRY_VMEXIT;
3442 }
3443 
3444 /*
3445  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3446  * for running an L2 nested guest.
3447  */
3448 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3449 {
3450 	struct vmcs12 *vmcs12;
3451 	enum nvmx_vmentry_status status;
3452 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3453 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3454 	enum nested_evmptrld_status evmptrld_status;
3455 
3456 	if (!nested_vmx_check_permission(vcpu))
3457 		return 1;
3458 
3459 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3460 	if (evmptrld_status == EVMPTRLD_ERROR) {
3461 		kvm_queue_exception(vcpu, UD_VECTOR);
3462 		return 1;
3463 	} else if (CC(evmptrld_status == EVMPTRLD_VMFAIL)) {
3464 		return nested_vmx_failInvalid(vcpu);
3465 	}
3466 
3467 	if (CC(!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull))
3468 		return nested_vmx_failInvalid(vcpu);
3469 
3470 	vmcs12 = get_vmcs12(vcpu);
3471 
3472 	/*
3473 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3474 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3475 	 * rather than RFLAGS.ZF, and no error number is stored to the
3476 	 * VM-instruction error field.
3477 	 */
3478 	if (CC(vmcs12->hdr.shadow_vmcs))
3479 		return nested_vmx_failInvalid(vcpu);
3480 
3481 	if (vmx->nested.hv_evmcs) {
3482 		copy_enlightened_to_vmcs12(vmx);
3483 		/* Enlightened VMCS doesn't have launch state */
3484 		vmcs12->launch_state = !launch;
3485 	} else if (enable_shadow_vmcs) {
3486 		copy_shadow_to_vmcs12(vmx);
3487 	}
3488 
3489 	/*
3490 	 * The nested entry process starts with enforcing various prerequisites
3491 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3492 	 * they fail: As the SDM explains, some conditions should cause the
3493 	 * instruction to fail, while others will cause the instruction to seem
3494 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3495 	 * To speed up the normal (success) code path, we should avoid checking
3496 	 * for misconfigurations which will anyway be caught by the processor
3497 	 * when using the merged vmcs02.
3498 	 */
3499 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3500 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3501 
3502 	if (CC(vmcs12->launch_state == launch))
3503 		return nested_vmx_fail(vcpu,
3504 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3505 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3506 
3507 	if (nested_vmx_check_controls(vcpu, vmcs12))
3508 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3509 
3510 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3511 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3512 
3513 	/*
3514 	 * We're finally done with prerequisite checking, and can start with
3515 	 * the nested entry.
3516 	 */
3517 	vmx->nested.nested_run_pending = 1;
3518 	vmx->nested.has_preemption_timer_deadline = false;
3519 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3520 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3521 		goto vmentry_failed;
3522 
3523 	/* Emulate processing of posted interrupts on VM-Enter. */
3524 	if (nested_cpu_has_posted_intr(vmcs12) &&
3525 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3526 		vmx->nested.pi_pending = true;
3527 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3528 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3529 	}
3530 
3531 	/* Hide L1D cache contents from the nested guest.  */
3532 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3533 
3534 	/*
3535 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3536 	 * also be used as part of restoring nVMX state for
3537 	 * snapshot restore (migration).
3538 	 *
3539 	 * In this flow, it is assumed that vmcs12 cache was
3540 	 * trasferred as part of captured nVMX state and should
3541 	 * therefore not be read from guest memory (which may not
3542 	 * exist on destination host yet).
3543 	 */
3544 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3545 
3546 	switch (vmcs12->guest_activity_state) {
3547 	case GUEST_ACTIVITY_HLT:
3548 		/*
3549 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3550 		 * awakened by event injection or by an NMI-window VM-exit or
3551 		 * by an interrupt-window VM-exit, halt the vcpu.
3552 		 */
3553 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3554 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3555 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3556 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3557 			vmx->nested.nested_run_pending = 0;
3558 			return kvm_vcpu_halt(vcpu);
3559 		}
3560 		break;
3561 	case GUEST_ACTIVITY_WAIT_SIPI:
3562 		vmx->nested.nested_run_pending = 0;
3563 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3564 		break;
3565 	default:
3566 		break;
3567 	}
3568 
3569 	return 1;
3570 
3571 vmentry_failed:
3572 	vmx->nested.nested_run_pending = 0;
3573 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3574 		return 0;
3575 	if (status == NVMX_VMENTRY_VMEXIT)
3576 		return 1;
3577 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3578 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3579 }
3580 
3581 /*
3582  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3583  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3584  * This function returns the new value we should put in vmcs12.guest_cr0.
3585  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3586  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3587  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3588  *     didn't trap the bit, because if L1 did, so would L0).
3589  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3590  *     been modified by L2, and L1 knows it. So just leave the old value of
3591  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3592  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3593  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3594  *     changed these bits, and therefore they need to be updated, but L0
3595  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3596  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3597  */
3598 static inline unsigned long
3599 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3600 {
3601 	return
3602 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3603 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3604 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3605 			vcpu->arch.cr0_guest_owned_bits));
3606 }
3607 
3608 static inline unsigned long
3609 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3610 {
3611 	return
3612 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3613 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3614 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3615 			vcpu->arch.cr4_guest_owned_bits));
3616 }
3617 
3618 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3619 				      struct vmcs12 *vmcs12)
3620 {
3621 	u32 idt_vectoring;
3622 	unsigned int nr;
3623 
3624 	if (vcpu->arch.exception.injected) {
3625 		nr = vcpu->arch.exception.nr;
3626 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3627 
3628 		if (kvm_exception_is_soft(nr)) {
3629 			vmcs12->vm_exit_instruction_len =
3630 				vcpu->arch.event_exit_inst_len;
3631 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3632 		} else
3633 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3634 
3635 		if (vcpu->arch.exception.has_error_code) {
3636 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3637 			vmcs12->idt_vectoring_error_code =
3638 				vcpu->arch.exception.error_code;
3639 		}
3640 
3641 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3642 	} else if (vcpu->arch.nmi_injected) {
3643 		vmcs12->idt_vectoring_info_field =
3644 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3645 	} else if (vcpu->arch.interrupt.injected) {
3646 		nr = vcpu->arch.interrupt.nr;
3647 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3648 
3649 		if (vcpu->arch.interrupt.soft) {
3650 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3651 			vmcs12->vm_entry_instruction_len =
3652 				vcpu->arch.event_exit_inst_len;
3653 		} else
3654 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3655 
3656 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3657 	}
3658 }
3659 
3660 
3661 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3662 {
3663 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3664 	gfn_t gfn;
3665 
3666 	/*
3667 	 * Don't need to mark the APIC access page dirty; it is never
3668 	 * written to by the CPU during APIC virtualization.
3669 	 */
3670 
3671 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3672 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3673 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3674 	}
3675 
3676 	if (nested_cpu_has_posted_intr(vmcs12)) {
3677 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3678 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3679 	}
3680 }
3681 
3682 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3683 {
3684 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3685 	int max_irr;
3686 	void *vapic_page;
3687 	u16 status;
3688 
3689 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3690 		return;
3691 
3692 	vmx->nested.pi_pending = false;
3693 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3694 		return;
3695 
3696 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3697 	if (max_irr != 256) {
3698 		vapic_page = vmx->nested.virtual_apic_map.hva;
3699 		if (!vapic_page)
3700 			return;
3701 
3702 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3703 			vapic_page, &max_irr);
3704 		status = vmcs_read16(GUEST_INTR_STATUS);
3705 		if ((u8)max_irr > ((u8)status & 0xff)) {
3706 			status &= ~0xff;
3707 			status |= (u8)max_irr;
3708 			vmcs_write16(GUEST_INTR_STATUS, status);
3709 		}
3710 	}
3711 
3712 	nested_mark_vmcs12_pages_dirty(vcpu);
3713 }
3714 
3715 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3716 					       unsigned long exit_qual)
3717 {
3718 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3719 	unsigned int nr = vcpu->arch.exception.nr;
3720 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3721 
3722 	if (vcpu->arch.exception.has_error_code) {
3723 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3724 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3725 	}
3726 
3727 	if (kvm_exception_is_soft(nr))
3728 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3729 	else
3730 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3731 
3732 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3733 	    vmx_get_nmi_mask(vcpu))
3734 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3735 
3736 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3737 }
3738 
3739 /*
3740  * Returns true if a debug trap is pending delivery.
3741  *
3742  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3743  * exception may be inferred from the presence of an exception payload.
3744  */
3745 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3746 {
3747 	return vcpu->arch.exception.pending &&
3748 			vcpu->arch.exception.nr == DB_VECTOR &&
3749 			vcpu->arch.exception.payload;
3750 }
3751 
3752 /*
3753  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3754  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3755  * represents these debug traps with a payload that is said to be compatible
3756  * with the 'pending debug exceptions' field, write the payload to the VMCS
3757  * field if a VM-exit is delivered before the debug trap.
3758  */
3759 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3760 {
3761 	if (vmx_pending_dbg_trap(vcpu))
3762 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3763 			    vcpu->arch.exception.payload);
3764 }
3765 
3766 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3767 {
3768 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3769 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3770 }
3771 
3772 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3773 {
3774 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3775 	unsigned long exit_qual;
3776 	bool block_nested_events =
3777 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3778 	bool mtf_pending = vmx->nested.mtf_pending;
3779 	struct kvm_lapic *apic = vcpu->arch.apic;
3780 
3781 	/*
3782 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3783 	 * this state is discarded.
3784 	 */
3785 	if (!block_nested_events)
3786 		vmx->nested.mtf_pending = false;
3787 
3788 	if (lapic_in_kernel(vcpu) &&
3789 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3790 		if (block_nested_events)
3791 			return -EBUSY;
3792 		nested_vmx_update_pending_dbg(vcpu);
3793 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3794 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3795 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3796 		return 0;
3797 	}
3798 
3799 	if (lapic_in_kernel(vcpu) &&
3800 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3801 		if (block_nested_events)
3802 			return -EBUSY;
3803 
3804 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3805 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3806 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3807 						apic->sipi_vector & 0xFFUL);
3808 		return 0;
3809 	}
3810 
3811 	/*
3812 	 * Process any exceptions that are not debug traps before MTF.
3813 	 */
3814 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3815 		if (block_nested_events)
3816 			return -EBUSY;
3817 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3818 			goto no_vmexit;
3819 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3820 		return 0;
3821 	}
3822 
3823 	if (mtf_pending) {
3824 		if (block_nested_events)
3825 			return -EBUSY;
3826 		nested_vmx_update_pending_dbg(vcpu);
3827 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3828 		return 0;
3829 	}
3830 
3831 	if (vcpu->arch.exception.pending) {
3832 		if (block_nested_events)
3833 			return -EBUSY;
3834 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3835 			goto no_vmexit;
3836 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3837 		return 0;
3838 	}
3839 
3840 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3841 		if (block_nested_events)
3842 			return -EBUSY;
3843 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3844 		return 0;
3845 	}
3846 
3847 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3848 		if (block_nested_events)
3849 			return -EBUSY;
3850 		goto no_vmexit;
3851 	}
3852 
3853 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3854 		if (block_nested_events)
3855 			return -EBUSY;
3856 		if (!nested_exit_on_nmi(vcpu))
3857 			goto no_vmexit;
3858 
3859 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3860 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3861 				  INTR_INFO_VALID_MASK, 0);
3862 		/*
3863 		 * The NMI-triggered VM exit counts as injection:
3864 		 * clear this one and block further NMIs.
3865 		 */
3866 		vcpu->arch.nmi_pending = 0;
3867 		vmx_set_nmi_mask(vcpu, true);
3868 		return 0;
3869 	}
3870 
3871 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3872 		if (block_nested_events)
3873 			return -EBUSY;
3874 		if (!nested_exit_on_intr(vcpu))
3875 			goto no_vmexit;
3876 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3877 		return 0;
3878 	}
3879 
3880 no_vmexit:
3881 	vmx_complete_nested_posted_interrupt(vcpu);
3882 	return 0;
3883 }
3884 
3885 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3886 {
3887 	ktime_t remaining =
3888 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3889 	u64 value;
3890 
3891 	if (ktime_to_ns(remaining) <= 0)
3892 		return 0;
3893 
3894 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3895 	do_div(value, 1000000);
3896 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3897 }
3898 
3899 static bool is_vmcs12_ext_field(unsigned long field)
3900 {
3901 	switch (field) {
3902 	case GUEST_ES_SELECTOR:
3903 	case GUEST_CS_SELECTOR:
3904 	case GUEST_SS_SELECTOR:
3905 	case GUEST_DS_SELECTOR:
3906 	case GUEST_FS_SELECTOR:
3907 	case GUEST_GS_SELECTOR:
3908 	case GUEST_LDTR_SELECTOR:
3909 	case GUEST_TR_SELECTOR:
3910 	case GUEST_ES_LIMIT:
3911 	case GUEST_CS_LIMIT:
3912 	case GUEST_SS_LIMIT:
3913 	case GUEST_DS_LIMIT:
3914 	case GUEST_FS_LIMIT:
3915 	case GUEST_GS_LIMIT:
3916 	case GUEST_LDTR_LIMIT:
3917 	case GUEST_TR_LIMIT:
3918 	case GUEST_GDTR_LIMIT:
3919 	case GUEST_IDTR_LIMIT:
3920 	case GUEST_ES_AR_BYTES:
3921 	case GUEST_DS_AR_BYTES:
3922 	case GUEST_FS_AR_BYTES:
3923 	case GUEST_GS_AR_BYTES:
3924 	case GUEST_LDTR_AR_BYTES:
3925 	case GUEST_TR_AR_BYTES:
3926 	case GUEST_ES_BASE:
3927 	case GUEST_CS_BASE:
3928 	case GUEST_SS_BASE:
3929 	case GUEST_DS_BASE:
3930 	case GUEST_FS_BASE:
3931 	case GUEST_GS_BASE:
3932 	case GUEST_LDTR_BASE:
3933 	case GUEST_TR_BASE:
3934 	case GUEST_GDTR_BASE:
3935 	case GUEST_IDTR_BASE:
3936 	case GUEST_PENDING_DBG_EXCEPTIONS:
3937 	case GUEST_BNDCFGS:
3938 		return true;
3939 	default:
3940 		break;
3941 	}
3942 
3943 	return false;
3944 }
3945 
3946 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3947 				       struct vmcs12 *vmcs12)
3948 {
3949 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3950 
3951 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3952 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3953 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3954 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3955 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3956 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3957 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3958 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3959 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3960 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3961 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3962 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3963 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3964 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3965 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3966 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3967 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3968 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3969 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3970 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3971 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3972 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3973 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3974 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3975 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3976 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3977 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3978 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3979 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3980 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3981 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3982 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3983 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3984 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3985 	vmcs12->guest_pending_dbg_exceptions =
3986 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3987 	if (kvm_mpx_supported())
3988 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3989 
3990 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
3991 }
3992 
3993 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3994 				       struct vmcs12 *vmcs12)
3995 {
3996 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3997 	int cpu;
3998 
3999 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4000 		return;
4001 
4002 
4003 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4004 
4005 	cpu = get_cpu();
4006 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4007 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4008 
4009 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4010 
4011 	vmx->loaded_vmcs = &vmx->vmcs01;
4012 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4013 	put_cpu();
4014 }
4015 
4016 /*
4017  * Update the guest state fields of vmcs12 to reflect changes that
4018  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4019  * VM-entry controls is also updated, since this is really a guest
4020  * state bit.)
4021  */
4022 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4023 {
4024 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4025 
4026 	if (vmx->nested.hv_evmcs)
4027 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4028 
4029 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = !vmx->nested.hv_evmcs;
4030 
4031 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4032 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4033 
4034 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4035 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4036 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4037 
4038 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4039 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4040 
4041 	vmcs12->guest_interruptibility_info =
4042 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4043 
4044 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4045 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4046 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4047 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4048 	else
4049 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4050 
4051 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4052 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4053 	    !vmx->nested.nested_run_pending)
4054 		vmcs12->vmx_preemption_timer_value =
4055 			vmx_get_preemption_timer_value(vcpu);
4056 
4057 	/*
4058 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4059 	 * own CR3 without exiting. If it has changed it, we must keep it.
4060 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4061 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4062 	 *
4063 	 * Additionally, restore L2's PDPTR to vmcs12.
4064 	 */
4065 	if (enable_ept) {
4066 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4067 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4068 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4069 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4070 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4071 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4072 		}
4073 	}
4074 
4075 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4076 
4077 	if (nested_cpu_has_vid(vmcs12))
4078 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4079 
4080 	vmcs12->vm_entry_controls =
4081 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4082 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4083 
4084 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4085 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4086 
4087 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4088 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4089 }
4090 
4091 /*
4092  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4093  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4094  * and this function updates it to reflect the changes to the guest state while
4095  * L2 was running (and perhaps made some exits which were handled directly by L0
4096  * without going back to L1), and to reflect the exit reason.
4097  * Note that we do not have to copy here all VMCS fields, just those that
4098  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4099  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4100  * which already writes to vmcs12 directly.
4101  */
4102 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4103 			   u32 vm_exit_reason, u32 exit_intr_info,
4104 			   unsigned long exit_qualification)
4105 {
4106 	/* update exit information fields: */
4107 	vmcs12->vm_exit_reason = vm_exit_reason;
4108 	vmcs12->exit_qualification = exit_qualification;
4109 	vmcs12->vm_exit_intr_info = exit_intr_info;
4110 
4111 	vmcs12->idt_vectoring_info_field = 0;
4112 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4113 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4114 
4115 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4116 		vmcs12->launch_state = 1;
4117 
4118 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4119 		 * instead of reading the real value. */
4120 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4121 
4122 		/*
4123 		 * Transfer the event that L0 or L1 may wanted to inject into
4124 		 * L2 to IDT_VECTORING_INFO_FIELD.
4125 		 */
4126 		vmcs12_save_pending_event(vcpu, vmcs12);
4127 
4128 		/*
4129 		 * According to spec, there's no need to store the guest's
4130 		 * MSRs if the exit is due to a VM-entry failure that occurs
4131 		 * during or after loading the guest state. Since this exit
4132 		 * does not fall in that category, we need to save the MSRs.
4133 		 */
4134 		if (nested_vmx_store_msr(vcpu,
4135 					 vmcs12->vm_exit_msr_store_addr,
4136 					 vmcs12->vm_exit_msr_store_count))
4137 			nested_vmx_abort(vcpu,
4138 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4139 	}
4140 
4141 	/*
4142 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4143 	 * preserved above and would only end up incorrectly in L1.
4144 	 */
4145 	vcpu->arch.nmi_injected = false;
4146 	kvm_clear_exception_queue(vcpu);
4147 	kvm_clear_interrupt_queue(vcpu);
4148 }
4149 
4150 /*
4151  * A part of what we need to when the nested L2 guest exits and we want to
4152  * run its L1 parent, is to reset L1's guest state to the host state specified
4153  * in vmcs12.
4154  * This function is to be called not only on normal nested exit, but also on
4155  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4156  * Failures During or After Loading Guest State").
4157  * This function should be called when the active VMCS is L1's (vmcs01).
4158  */
4159 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4160 				   struct vmcs12 *vmcs12)
4161 {
4162 	enum vm_entry_failure_code ignored;
4163 	struct kvm_segment seg;
4164 
4165 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4166 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4167 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4168 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4169 	else
4170 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4171 	vmx_set_efer(vcpu, vcpu->arch.efer);
4172 
4173 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4174 	kvm_rip_write(vcpu, vmcs12->host_rip);
4175 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4176 	vmx_set_interrupt_shadow(vcpu, 0);
4177 
4178 	/*
4179 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4180 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4181 	 *
4182 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4183 	 * (KVM doesn't change it);
4184 	 */
4185 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4186 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4187 
4188 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4189 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4190 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4191 
4192 	nested_ept_uninit_mmu_context(vcpu);
4193 
4194 	/*
4195 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4196 	 * couldn't have changed.
4197 	 */
4198 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &ignored))
4199 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4200 
4201 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4202 
4203 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4204 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4205 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4206 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4207 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4208 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4209 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4210 
4211 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4212 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4213 		vmcs_write64(GUEST_BNDCFGS, 0);
4214 
4215 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4216 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4217 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4218 	}
4219 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4220 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4221 					 vmcs12->host_ia32_perf_global_ctrl));
4222 
4223 	/* Set L1 segment info according to Intel SDM
4224 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4225 	seg = (struct kvm_segment) {
4226 		.base = 0,
4227 		.limit = 0xFFFFFFFF,
4228 		.selector = vmcs12->host_cs_selector,
4229 		.type = 11,
4230 		.present = 1,
4231 		.s = 1,
4232 		.g = 1
4233 	};
4234 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4235 		seg.l = 1;
4236 	else
4237 		seg.db = 1;
4238 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4239 	seg = (struct kvm_segment) {
4240 		.base = 0,
4241 		.limit = 0xFFFFFFFF,
4242 		.type = 3,
4243 		.present = 1,
4244 		.s = 1,
4245 		.db = 1,
4246 		.g = 1
4247 	};
4248 	seg.selector = vmcs12->host_ds_selector;
4249 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4250 	seg.selector = vmcs12->host_es_selector;
4251 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4252 	seg.selector = vmcs12->host_ss_selector;
4253 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4254 	seg.selector = vmcs12->host_fs_selector;
4255 	seg.base = vmcs12->host_fs_base;
4256 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4257 	seg.selector = vmcs12->host_gs_selector;
4258 	seg.base = vmcs12->host_gs_base;
4259 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4260 	seg = (struct kvm_segment) {
4261 		.base = vmcs12->host_tr_base,
4262 		.limit = 0x67,
4263 		.selector = vmcs12->host_tr_selector,
4264 		.type = 11,
4265 		.present = 1
4266 	};
4267 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4268 
4269 	kvm_set_dr(vcpu, 7, 0x400);
4270 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4271 
4272 	if (cpu_has_vmx_msr_bitmap())
4273 		vmx_update_msr_bitmap(vcpu);
4274 
4275 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4276 				vmcs12->vm_exit_msr_load_count))
4277 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4278 }
4279 
4280 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4281 {
4282 	struct vmx_uret_msr *efer_msr;
4283 	unsigned int i;
4284 
4285 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4286 		return vmcs_read64(GUEST_IA32_EFER);
4287 
4288 	if (cpu_has_load_ia32_efer())
4289 		return host_efer;
4290 
4291 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4292 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4293 			return vmx->msr_autoload.guest.val[i].value;
4294 	}
4295 
4296 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4297 	if (efer_msr)
4298 		return efer_msr->data;
4299 
4300 	return host_efer;
4301 }
4302 
4303 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4304 {
4305 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4306 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4307 	struct vmx_msr_entry g, h;
4308 	gpa_t gpa;
4309 	u32 i, j;
4310 
4311 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4312 
4313 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4314 		/*
4315 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4316 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4317 		 * and vcpu->arch.dr7 is not squirreled away before the
4318 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4319 		 */
4320 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4321 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4322 		else
4323 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4324 	}
4325 
4326 	/*
4327 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4328 	 * handle a variety of side effects to KVM's software model.
4329 	 */
4330 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4331 
4332 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4333 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4334 
4335 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4336 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4337 
4338 	nested_ept_uninit_mmu_context(vcpu);
4339 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4340 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4341 
4342 	/*
4343 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4344 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4345 	 * VMFail, like everything else we just need to ensure our
4346 	 * software model is up-to-date.
4347 	 */
4348 	if (enable_ept && is_pae_paging(vcpu))
4349 		ept_save_pdptrs(vcpu);
4350 
4351 	kvm_mmu_reset_context(vcpu);
4352 
4353 	if (cpu_has_vmx_msr_bitmap())
4354 		vmx_update_msr_bitmap(vcpu);
4355 
4356 	/*
4357 	 * This nasty bit of open coding is a compromise between blindly
4358 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4359 	 * of VMFail), leaving the nested VM's MSRs in the software model
4360 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4361 	 * expensive since the lists are unbound by hardware).  For each
4362 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4363 	 * list, reload it from the exit load list if it exists and differs
4364 	 * from the guest value.  The intent is to stuff host state as
4365 	 * silently as possible, not to fully process the exit load list.
4366 	 */
4367 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4368 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4369 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4370 			pr_debug_ratelimited(
4371 				"%s read MSR index failed (%u, 0x%08llx)\n",
4372 				__func__, i, gpa);
4373 			goto vmabort;
4374 		}
4375 
4376 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4377 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4378 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4379 				pr_debug_ratelimited(
4380 					"%s read MSR failed (%u, 0x%08llx)\n",
4381 					__func__, j, gpa);
4382 				goto vmabort;
4383 			}
4384 			if (h.index != g.index)
4385 				continue;
4386 			if (h.value == g.value)
4387 				break;
4388 
4389 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4390 				pr_debug_ratelimited(
4391 					"%s check failed (%u, 0x%x, 0x%x)\n",
4392 					__func__, j, h.index, h.reserved);
4393 				goto vmabort;
4394 			}
4395 
4396 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4397 				pr_debug_ratelimited(
4398 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4399 					__func__, j, h.index, h.value);
4400 				goto vmabort;
4401 			}
4402 		}
4403 	}
4404 
4405 	return;
4406 
4407 vmabort:
4408 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4409 }
4410 
4411 /*
4412  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4413  * and modify vmcs12 to make it see what it would expect to see there if
4414  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4415  */
4416 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4417 		       u32 exit_intr_info, unsigned long exit_qualification)
4418 {
4419 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4420 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4421 
4422 	/* trying to cancel vmlaunch/vmresume is a bug */
4423 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4424 
4425 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
4426 
4427 	/* Service the TLB flush request for L2 before switching to L1. */
4428 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
4429 		kvm_vcpu_flush_tlb_current(vcpu);
4430 
4431 	/*
4432 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4433 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4434 	 * up-to-date before switching to L1.
4435 	 */
4436 	if (enable_ept && is_pae_paging(vcpu))
4437 		vmx_ept_load_pdptrs(vcpu);
4438 
4439 	leave_guest_mode(vcpu);
4440 
4441 	if (nested_cpu_has_preemption_timer(vmcs12))
4442 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4443 
4444 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
4445 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
4446 
4447 	if (likely(!vmx->fail)) {
4448 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4449 
4450 		if (vm_exit_reason != -1)
4451 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4452 				       exit_intr_info, exit_qualification);
4453 
4454 		/*
4455 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4456 		 * also be used to capture vmcs12 cache as part of
4457 		 * capturing nVMX state for snapshot (migration).
4458 		 *
4459 		 * Otherwise, this flush will dirty guest memory at a
4460 		 * point it is already assumed by user-space to be
4461 		 * immutable.
4462 		 */
4463 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4464 	} else {
4465 		/*
4466 		 * The only expected VM-instruction error is "VM entry with
4467 		 * invalid control field(s)." Anything else indicates a
4468 		 * problem with L0.  And we should never get here with a
4469 		 * VMFail of any type if early consistency checks are enabled.
4470 		 */
4471 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4472 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4473 		WARN_ON_ONCE(nested_early_check);
4474 	}
4475 
4476 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4477 
4478 	/* Update any VMCS fields that might have changed while L2 ran */
4479 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4480 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4481 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4482 	if (vmx->nested.l1_tpr_threshold != -1)
4483 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4484 
4485 	if (kvm_has_tsc_control)
4486 		decache_tsc_multiplier(vmx);
4487 
4488 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4489 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4490 		vmx_set_virtual_apic_mode(vcpu);
4491 	}
4492 
4493 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4494 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4495 		vmx_update_cpu_dirty_logging(vcpu);
4496 	}
4497 
4498 	/* Unpin physical memory we referred to in vmcs02 */
4499 	if (vmx->nested.apic_access_page) {
4500 		kvm_release_page_clean(vmx->nested.apic_access_page);
4501 		vmx->nested.apic_access_page = NULL;
4502 	}
4503 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4504 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4505 	vmx->nested.pi_desc = NULL;
4506 
4507 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4508 		vmx->nested.reload_vmcs01_apic_access_page = false;
4509 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4510 	}
4511 
4512 	if ((vm_exit_reason != -1) &&
4513 	    (enable_shadow_vmcs || vmx->nested.hv_evmcs))
4514 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4515 
4516 	/* in case we halted in L2 */
4517 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4518 
4519 	if (likely(!vmx->fail)) {
4520 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4521 		    nested_exit_intr_ack_set(vcpu)) {
4522 			int irq = kvm_cpu_get_interrupt(vcpu);
4523 			WARN_ON(irq < 0);
4524 			vmcs12->vm_exit_intr_info = irq |
4525 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4526 		}
4527 
4528 		if (vm_exit_reason != -1)
4529 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4530 						       vmcs12->exit_qualification,
4531 						       vmcs12->idt_vectoring_info_field,
4532 						       vmcs12->vm_exit_intr_info,
4533 						       vmcs12->vm_exit_intr_error_code,
4534 						       KVM_ISA_VMX);
4535 
4536 		load_vmcs12_host_state(vcpu, vmcs12);
4537 
4538 		return;
4539 	}
4540 
4541 	/*
4542 	 * After an early L2 VM-entry failure, we're now back
4543 	 * in L1 which thinks it just finished a VMLAUNCH or
4544 	 * VMRESUME instruction, so we need to set the failure
4545 	 * flag and the VM-instruction error field of the VMCS
4546 	 * accordingly, and skip the emulated instruction.
4547 	 */
4548 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4549 
4550 	/*
4551 	 * Restore L1's host state to KVM's software model.  We're here
4552 	 * because a consistency check was caught by hardware, which
4553 	 * means some amount of guest state has been propagated to KVM's
4554 	 * model and needs to be unwound to the host's state.
4555 	 */
4556 	nested_vmx_restore_host_state(vcpu);
4557 
4558 	vmx->fail = 0;
4559 }
4560 
4561 /*
4562  * Decode the memory-address operand of a vmx instruction, as recorded on an
4563  * exit caused by such an instruction (run by a guest hypervisor).
4564  * On success, returns 0. When the operand is invalid, returns 1 and throws
4565  * #UD, #GP, or #SS.
4566  */
4567 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4568 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4569 {
4570 	gva_t off;
4571 	bool exn;
4572 	struct kvm_segment s;
4573 
4574 	/*
4575 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4576 	 * Execution", on an exit, vmx_instruction_info holds most of the
4577 	 * addressing components of the operand. Only the displacement part
4578 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4579 	 * For how an actual address is calculated from all these components,
4580 	 * refer to Vol. 1, "Operand Addressing".
4581 	 */
4582 	int  scaling = vmx_instruction_info & 3;
4583 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4584 	bool is_reg = vmx_instruction_info & (1u << 10);
4585 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4586 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4587 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4588 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4589 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4590 
4591 	if (is_reg) {
4592 		kvm_queue_exception(vcpu, UD_VECTOR);
4593 		return 1;
4594 	}
4595 
4596 	/* Addr = segment_base + offset */
4597 	/* offset = base + [index * scale] + displacement */
4598 	off = exit_qualification; /* holds the displacement */
4599 	if (addr_size == 1)
4600 		off = (gva_t)sign_extend64(off, 31);
4601 	else if (addr_size == 0)
4602 		off = (gva_t)sign_extend64(off, 15);
4603 	if (base_is_valid)
4604 		off += kvm_register_read(vcpu, base_reg);
4605 	if (index_is_valid)
4606 		off += kvm_register_read(vcpu, index_reg) << scaling;
4607 	vmx_get_segment(vcpu, &s, seg_reg);
4608 
4609 	/*
4610 	 * The effective address, i.e. @off, of a memory operand is truncated
4611 	 * based on the address size of the instruction.  Note that this is
4612 	 * the *effective address*, i.e. the address prior to accounting for
4613 	 * the segment's base.
4614 	 */
4615 	if (addr_size == 1) /* 32 bit */
4616 		off &= 0xffffffff;
4617 	else if (addr_size == 0) /* 16 bit */
4618 		off &= 0xffff;
4619 
4620 	/* Checks for #GP/#SS exceptions. */
4621 	exn = false;
4622 	if (is_long_mode(vcpu)) {
4623 		/*
4624 		 * The virtual/linear address is never truncated in 64-bit
4625 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4626 		 * address when using FS/GS with a non-zero base.
4627 		 */
4628 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4629 			*ret = s.base + off;
4630 		else
4631 			*ret = off;
4632 
4633 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4634 		 * non-canonical form. This is the only check on the memory
4635 		 * destination for long mode!
4636 		 */
4637 		exn = is_noncanonical_address(*ret, vcpu);
4638 	} else {
4639 		/*
4640 		 * When not in long mode, the virtual/linear address is
4641 		 * unconditionally truncated to 32 bits regardless of the
4642 		 * address size.
4643 		 */
4644 		*ret = (s.base + off) & 0xffffffff;
4645 
4646 		/* Protected mode: apply checks for segment validity in the
4647 		 * following order:
4648 		 * - segment type check (#GP(0) may be thrown)
4649 		 * - usability check (#GP(0)/#SS(0))
4650 		 * - limit check (#GP(0)/#SS(0))
4651 		 */
4652 		if (wr)
4653 			/* #GP(0) if the destination operand is located in a
4654 			 * read-only data segment or any code segment.
4655 			 */
4656 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4657 		else
4658 			/* #GP(0) if the source operand is located in an
4659 			 * execute-only code segment
4660 			 */
4661 			exn = ((s.type & 0xa) == 8);
4662 		if (exn) {
4663 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4664 			return 1;
4665 		}
4666 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4667 		 */
4668 		exn = (s.unusable != 0);
4669 
4670 		/*
4671 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4672 		 * outside the segment limit.  All CPUs that support VMX ignore
4673 		 * limit checks for flat segments, i.e. segments with base==0,
4674 		 * limit==0xffffffff and of type expand-up data or code.
4675 		 */
4676 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4677 		     ((s.type & 8) || !(s.type & 4))))
4678 			exn = exn || ((u64)off + len - 1 > s.limit);
4679 	}
4680 	if (exn) {
4681 		kvm_queue_exception_e(vcpu,
4682 				      seg_reg == VCPU_SREG_SS ?
4683 						SS_VECTOR : GP_VECTOR,
4684 				      0);
4685 		return 1;
4686 	}
4687 
4688 	return 0;
4689 }
4690 
4691 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4692 {
4693 	struct vcpu_vmx *vmx;
4694 
4695 	if (!nested_vmx_allowed(vcpu))
4696 		return;
4697 
4698 	vmx = to_vmx(vcpu);
4699 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4700 		vmx->nested.msrs.entry_ctls_high |=
4701 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4702 		vmx->nested.msrs.exit_ctls_high |=
4703 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4704 	} else {
4705 		vmx->nested.msrs.entry_ctls_high &=
4706 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4707 		vmx->nested.msrs.exit_ctls_high &=
4708 				~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4709 	}
4710 }
4711 
4712 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4713 				int *ret)
4714 {
4715 	gva_t gva;
4716 	struct x86_exception e;
4717 	int r;
4718 
4719 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4720 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4721 				sizeof(*vmpointer), &gva)) {
4722 		*ret = 1;
4723 		return -EINVAL;
4724 	}
4725 
4726 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4727 	if (r != X86EMUL_CONTINUE) {
4728 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4729 		return -EINVAL;
4730 	}
4731 
4732 	return 0;
4733 }
4734 
4735 /*
4736  * Allocate a shadow VMCS and associate it with the currently loaded
4737  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4738  * VMCS is also VMCLEARed, so that it is ready for use.
4739  */
4740 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4741 {
4742 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4743 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4744 
4745 	/*
4746 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4747 	 * executes VMXON and free it when L1 executes VMXOFF.
4748 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4749 	 * here when vmcs01 already have an allocated shadow vmcs.
4750 	 */
4751 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4752 
4753 	if (!loaded_vmcs->shadow_vmcs) {
4754 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4755 		if (loaded_vmcs->shadow_vmcs)
4756 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4757 	}
4758 	return loaded_vmcs->shadow_vmcs;
4759 }
4760 
4761 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4762 {
4763 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4764 	int r;
4765 
4766 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4767 	if (r < 0)
4768 		goto out_vmcs02;
4769 
4770 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4771 	if (!vmx->nested.cached_vmcs12)
4772 		goto out_cached_vmcs12;
4773 
4774 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4775 	if (!vmx->nested.cached_shadow_vmcs12)
4776 		goto out_cached_shadow_vmcs12;
4777 
4778 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4779 		goto out_shadow_vmcs;
4780 
4781 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4782 		     HRTIMER_MODE_ABS_PINNED);
4783 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4784 
4785 	vmx->nested.vpid02 = allocate_vpid();
4786 
4787 	vmx->nested.vmcs02_initialized = false;
4788 	vmx->nested.vmxon = true;
4789 
4790 	if (vmx_pt_mode_is_host_guest()) {
4791 		vmx->pt_desc.guest.ctl = 0;
4792 		pt_update_intercept_for_msr(vcpu);
4793 	}
4794 
4795 	return 0;
4796 
4797 out_shadow_vmcs:
4798 	kfree(vmx->nested.cached_shadow_vmcs12);
4799 
4800 out_cached_shadow_vmcs12:
4801 	kfree(vmx->nested.cached_vmcs12);
4802 
4803 out_cached_vmcs12:
4804 	free_loaded_vmcs(&vmx->nested.vmcs02);
4805 
4806 out_vmcs02:
4807 	return -ENOMEM;
4808 }
4809 
4810 /*
4811  * Emulate the VMXON instruction.
4812  * Currently, we just remember that VMX is active, and do not save or even
4813  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4814  * do not currently need to store anything in that guest-allocated memory
4815  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4816  * argument is different from the VMXON pointer (which the spec says they do).
4817  */
4818 static int handle_vmon(struct kvm_vcpu *vcpu)
4819 {
4820 	int ret;
4821 	gpa_t vmptr;
4822 	uint32_t revision;
4823 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4824 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4825 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4826 
4827 	/*
4828 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4829 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4830 	 * 1 (see vmx_is_valid_cr4() for when we allow the guest to set this).
4831 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4832 	 * have already been checked by hardware, prior to the VM-exit for
4833 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4834 	 * that bit set to 1 in non-root mode.
4835 	 */
4836 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4837 		kvm_queue_exception(vcpu, UD_VECTOR);
4838 		return 1;
4839 	}
4840 
4841 	/* CPL=0 must be checked manually. */
4842 	if (vmx_get_cpl(vcpu)) {
4843 		kvm_inject_gp(vcpu, 0);
4844 		return 1;
4845 	}
4846 
4847 	if (vmx->nested.vmxon)
4848 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4849 
4850 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4851 			!= VMXON_NEEDED_FEATURES) {
4852 		kvm_inject_gp(vcpu, 0);
4853 		return 1;
4854 	}
4855 
4856 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4857 		return ret;
4858 
4859 	/*
4860 	 * SDM 3: 24.11.5
4861 	 * The first 4 bytes of VMXON region contain the supported
4862 	 * VMCS revision identifier
4863 	 *
4864 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4865 	 * which replaces physical address width with 32
4866 	 */
4867 	if (!page_address_valid(vcpu, vmptr))
4868 		return nested_vmx_failInvalid(vcpu);
4869 
4870 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4871 	    revision != VMCS12_REVISION)
4872 		return nested_vmx_failInvalid(vcpu);
4873 
4874 	vmx->nested.vmxon_ptr = vmptr;
4875 	ret = enter_vmx_operation(vcpu);
4876 	if (ret)
4877 		return ret;
4878 
4879 	return nested_vmx_succeed(vcpu);
4880 }
4881 
4882 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4883 {
4884 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4885 
4886 	if (vmx->nested.current_vmptr == -1ull)
4887 		return;
4888 
4889 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4890 
4891 	if (enable_shadow_vmcs) {
4892 		/* copy to memory all shadowed fields in case
4893 		   they were modified */
4894 		copy_shadow_to_vmcs12(vmx);
4895 		vmx_disable_shadow_vmcs(vmx);
4896 	}
4897 	vmx->nested.posted_intr_nv = -1;
4898 
4899 	/* Flush VMCS12 to guest memory */
4900 	kvm_vcpu_write_guest_page(vcpu,
4901 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4902 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4903 
4904 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4905 
4906 	vmx->nested.current_vmptr = -1ull;
4907 }
4908 
4909 /* Emulate the VMXOFF instruction */
4910 static int handle_vmoff(struct kvm_vcpu *vcpu)
4911 {
4912 	if (!nested_vmx_check_permission(vcpu))
4913 		return 1;
4914 
4915 	free_nested(vcpu);
4916 
4917 	/* Process a latched INIT during time CPU was in VMX operation */
4918 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4919 
4920 	return nested_vmx_succeed(vcpu);
4921 }
4922 
4923 /* Emulate the VMCLEAR instruction */
4924 static int handle_vmclear(struct kvm_vcpu *vcpu)
4925 {
4926 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4927 	u32 zero = 0;
4928 	gpa_t vmptr;
4929 	u64 evmcs_gpa;
4930 	int r;
4931 
4932 	if (!nested_vmx_check_permission(vcpu))
4933 		return 1;
4934 
4935 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
4936 		return r;
4937 
4938 	if (!page_address_valid(vcpu, vmptr))
4939 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
4940 
4941 	if (vmptr == vmx->nested.vmxon_ptr)
4942 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
4943 
4944 	/*
4945 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4946 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4947 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4948 	 * writing to the non-existent 'launch_state' field. The area doesn't
4949 	 * have to be the currently active EVMCS on the calling CPU and there's
4950 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
4951 	 * state. It is possible that the area will stay mapped as
4952 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
4953 	 */
4954 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
4955 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
4956 		if (vmptr == vmx->nested.current_vmptr)
4957 			nested_release_vmcs12(vcpu);
4958 
4959 		kvm_vcpu_write_guest(vcpu,
4960 				     vmptr + offsetof(struct vmcs12,
4961 						      launch_state),
4962 				     &zero, sizeof(zero));
4963 	}
4964 
4965 	return nested_vmx_succeed(vcpu);
4966 }
4967 
4968 /* Emulate the VMLAUNCH instruction */
4969 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4970 {
4971 	return nested_vmx_run(vcpu, true);
4972 }
4973 
4974 /* Emulate the VMRESUME instruction */
4975 static int handle_vmresume(struct kvm_vcpu *vcpu)
4976 {
4977 
4978 	return nested_vmx_run(vcpu, false);
4979 }
4980 
4981 static int handle_vmread(struct kvm_vcpu *vcpu)
4982 {
4983 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4984 						    : get_vmcs12(vcpu);
4985 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
4986 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4987 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4988 	struct x86_exception e;
4989 	unsigned long field;
4990 	u64 value;
4991 	gva_t gva = 0;
4992 	short offset;
4993 	int len, r;
4994 
4995 	if (!nested_vmx_check_permission(vcpu))
4996 		return 1;
4997 
4998 	/*
4999 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
5000 	 * any VMREAD sets the ALU flags for VMfailInvalid.
5001 	 */
5002 	if (vmx->nested.current_vmptr == -1ull ||
5003 	    (is_guest_mode(vcpu) &&
5004 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
5005 		return nested_vmx_failInvalid(vcpu);
5006 
5007 	/* Decode instruction info and find the field to read */
5008 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
5009 
5010 	offset = vmcs_field_to_offset(field);
5011 	if (offset < 0)
5012 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5013 
5014 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5015 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5016 
5017 	/* Read the field, zero-extended to a u64 value */
5018 	value = vmcs12_read_any(vmcs12, field, offset);
5019 
5020 	/*
5021 	 * Now copy part of this value to register or memory, as requested.
5022 	 * Note that the number of bits actually copied is 32 or 64 depending
5023 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5024 	 */
5025 	if (instr_info & BIT(10)) {
5026 		kvm_register_writel(vcpu, (((instr_info) >> 3) & 0xf), value);
5027 	} else {
5028 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5029 		if (get_vmx_mem_address(vcpu, exit_qualification,
5030 					instr_info, true, len, &gva))
5031 			return 1;
5032 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5033 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5034 		if (r != X86EMUL_CONTINUE)
5035 			return kvm_handle_memory_failure(vcpu, r, &e);
5036 	}
5037 
5038 	return nested_vmx_succeed(vcpu);
5039 }
5040 
5041 static bool is_shadow_field_rw(unsigned long field)
5042 {
5043 	switch (field) {
5044 #define SHADOW_FIELD_RW(x, y) case x:
5045 #include "vmcs_shadow_fields.h"
5046 		return true;
5047 	default:
5048 		break;
5049 	}
5050 	return false;
5051 }
5052 
5053 static bool is_shadow_field_ro(unsigned long field)
5054 {
5055 	switch (field) {
5056 #define SHADOW_FIELD_RO(x, y) case x:
5057 #include "vmcs_shadow_fields.h"
5058 		return true;
5059 	default:
5060 		break;
5061 	}
5062 	return false;
5063 }
5064 
5065 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5066 {
5067 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5068 						    : get_vmcs12(vcpu);
5069 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5070 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5071 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5072 	struct x86_exception e;
5073 	unsigned long field;
5074 	short offset;
5075 	gva_t gva;
5076 	int len, r;
5077 
5078 	/*
5079 	 * The value to write might be 32 or 64 bits, depending on L1's long
5080 	 * mode, and eventually we need to write that into a field of several
5081 	 * possible lengths. The code below first zero-extends the value to 64
5082 	 * bit (value), and then copies only the appropriate number of
5083 	 * bits into the vmcs12 field.
5084 	 */
5085 	u64 value = 0;
5086 
5087 	if (!nested_vmx_check_permission(vcpu))
5088 		return 1;
5089 
5090 	/*
5091 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
5092 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5093 	 */
5094 	if (vmx->nested.current_vmptr == -1ull ||
5095 	    (is_guest_mode(vcpu) &&
5096 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
5097 		return nested_vmx_failInvalid(vcpu);
5098 
5099 	if (instr_info & BIT(10))
5100 		value = kvm_register_readl(vcpu, (((instr_info) >> 3) & 0xf));
5101 	else {
5102 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5103 		if (get_vmx_mem_address(vcpu, exit_qualification,
5104 					instr_info, false, len, &gva))
5105 			return 1;
5106 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5107 		if (r != X86EMUL_CONTINUE)
5108 			return kvm_handle_memory_failure(vcpu, r, &e);
5109 	}
5110 
5111 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
5112 
5113 	offset = vmcs_field_to_offset(field);
5114 	if (offset < 0)
5115 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5116 
5117 	/*
5118 	 * If the vCPU supports "VMWRITE to any supported field in the
5119 	 * VMCS," then the "read-only" fields are actually read/write.
5120 	 */
5121 	if (vmcs_field_readonly(field) &&
5122 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5123 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5124 
5125 	/*
5126 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5127 	 * vmcs12, else we may crush a field or consume a stale value.
5128 	 */
5129 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5130 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5131 
5132 	/*
5133 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5134 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5135 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5136 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5137 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5138 	 * the stripped down value, L2 sees the full value as stored by KVM).
5139 	 */
5140 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5141 		value &= 0x1f0ff;
5142 
5143 	vmcs12_write_any(vmcs12, field, offset, value);
5144 
5145 	/*
5146 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5147 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5148 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5149 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5150 	 */
5151 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5152 		/*
5153 		 * L1 can read these fields without exiting, ensure the
5154 		 * shadow VMCS is up-to-date.
5155 		 */
5156 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5157 			preempt_disable();
5158 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5159 
5160 			__vmcs_writel(field, value);
5161 
5162 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5163 			vmcs_load(vmx->loaded_vmcs->vmcs);
5164 			preempt_enable();
5165 		}
5166 		vmx->nested.dirty_vmcs12 = true;
5167 	}
5168 
5169 	return nested_vmx_succeed(vcpu);
5170 }
5171 
5172 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5173 {
5174 	vmx->nested.current_vmptr = vmptr;
5175 	if (enable_shadow_vmcs) {
5176 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5177 		vmcs_write64(VMCS_LINK_POINTER,
5178 			     __pa(vmx->vmcs01.shadow_vmcs));
5179 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5180 	}
5181 	vmx->nested.dirty_vmcs12 = true;
5182 }
5183 
5184 /* Emulate the VMPTRLD instruction */
5185 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5186 {
5187 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5188 	gpa_t vmptr;
5189 	int r;
5190 
5191 	if (!nested_vmx_check_permission(vcpu))
5192 		return 1;
5193 
5194 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5195 		return r;
5196 
5197 	if (!page_address_valid(vcpu, vmptr))
5198 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5199 
5200 	if (vmptr == vmx->nested.vmxon_ptr)
5201 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5202 
5203 	/* Forbid normal VMPTRLD if Enlightened version was used */
5204 	if (vmx->nested.hv_evmcs)
5205 		return 1;
5206 
5207 	if (vmx->nested.current_vmptr != vmptr) {
5208 		struct kvm_host_map map;
5209 		struct vmcs12 *new_vmcs12;
5210 
5211 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
5212 			/*
5213 			 * Reads from an unbacked page return all 1s,
5214 			 * which means that the 32 bits located at the
5215 			 * given physical address won't match the required
5216 			 * VMCS12_REVISION identifier.
5217 			 */
5218 			return nested_vmx_fail(vcpu,
5219 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5220 		}
5221 
5222 		new_vmcs12 = map.hva;
5223 
5224 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5225 		    (new_vmcs12->hdr.shadow_vmcs &&
5226 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5227 			kvm_vcpu_unmap(vcpu, &map, false);
5228 			return nested_vmx_fail(vcpu,
5229 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5230 		}
5231 
5232 		nested_release_vmcs12(vcpu);
5233 
5234 		/*
5235 		 * Load VMCS12 from guest memory since it is not already
5236 		 * cached.
5237 		 */
5238 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
5239 		kvm_vcpu_unmap(vcpu, &map, false);
5240 
5241 		set_current_vmptr(vmx, vmptr);
5242 	}
5243 
5244 	return nested_vmx_succeed(vcpu);
5245 }
5246 
5247 /* Emulate the VMPTRST instruction */
5248 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5249 {
5250 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5251 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5252 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5253 	struct x86_exception e;
5254 	gva_t gva;
5255 	int r;
5256 
5257 	if (!nested_vmx_check_permission(vcpu))
5258 		return 1;
5259 
5260 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
5261 		return 1;
5262 
5263 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5264 				true, sizeof(gpa_t), &gva))
5265 		return 1;
5266 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5267 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5268 					sizeof(gpa_t), &e);
5269 	if (r != X86EMUL_CONTINUE)
5270 		return kvm_handle_memory_failure(vcpu, r, &e);
5271 
5272 	return nested_vmx_succeed(vcpu);
5273 }
5274 
5275 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
5276 
5277 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
5278 {
5279 	return VALID_PAGE(root_hpa) &&
5280 		((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
5281 }
5282 
5283 /* Emulate the INVEPT instruction */
5284 static int handle_invept(struct kvm_vcpu *vcpu)
5285 {
5286 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5287 	u32 vmx_instruction_info, types;
5288 	unsigned long type, roots_to_free;
5289 	struct kvm_mmu *mmu;
5290 	gva_t gva;
5291 	struct x86_exception e;
5292 	struct {
5293 		u64 eptp, gpa;
5294 	} operand;
5295 	int i, r;
5296 
5297 	if (!(vmx->nested.msrs.secondary_ctls_high &
5298 	      SECONDARY_EXEC_ENABLE_EPT) ||
5299 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5300 		kvm_queue_exception(vcpu, UD_VECTOR);
5301 		return 1;
5302 	}
5303 
5304 	if (!nested_vmx_check_permission(vcpu))
5305 		return 1;
5306 
5307 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5308 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5309 
5310 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5311 
5312 	if (type >= 32 || !(types & (1 << type)))
5313 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5314 
5315 	/* According to the Intel VMX instruction reference, the memory
5316 	 * operand is read even if it isn't needed (e.g., for type==global)
5317 	 */
5318 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5319 			vmx_instruction_info, false, sizeof(operand), &gva))
5320 		return 1;
5321 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5322 	if (r != X86EMUL_CONTINUE)
5323 		return kvm_handle_memory_failure(vcpu, r, &e);
5324 
5325 	/*
5326 	 * Nested EPT roots are always held through guest_mmu,
5327 	 * not root_mmu.
5328 	 */
5329 	mmu = &vcpu->arch.guest_mmu;
5330 
5331 	switch (type) {
5332 	case VMX_EPT_EXTENT_CONTEXT:
5333 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5334 			return nested_vmx_fail(vcpu,
5335 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5336 
5337 		roots_to_free = 0;
5338 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
5339 					    operand.eptp))
5340 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5341 
5342 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5343 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5344 						    mmu->prev_roots[i].pgd,
5345 						    operand.eptp))
5346 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5347 		}
5348 		break;
5349 	case VMX_EPT_EXTENT_GLOBAL:
5350 		roots_to_free = KVM_MMU_ROOTS_ALL;
5351 		break;
5352 	default:
5353 		BUG();
5354 		break;
5355 	}
5356 
5357 	if (roots_to_free)
5358 		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
5359 
5360 	return nested_vmx_succeed(vcpu);
5361 }
5362 
5363 static int handle_invvpid(struct kvm_vcpu *vcpu)
5364 {
5365 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5366 	u32 vmx_instruction_info;
5367 	unsigned long type, types;
5368 	gva_t gva;
5369 	struct x86_exception e;
5370 	struct {
5371 		u64 vpid;
5372 		u64 gla;
5373 	} operand;
5374 	u16 vpid02;
5375 	int r;
5376 
5377 	if (!(vmx->nested.msrs.secondary_ctls_high &
5378 	      SECONDARY_EXEC_ENABLE_VPID) ||
5379 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5380 		kvm_queue_exception(vcpu, UD_VECTOR);
5381 		return 1;
5382 	}
5383 
5384 	if (!nested_vmx_check_permission(vcpu))
5385 		return 1;
5386 
5387 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5388 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5389 
5390 	types = (vmx->nested.msrs.vpid_caps &
5391 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5392 
5393 	if (type >= 32 || !(types & (1 << type)))
5394 		return nested_vmx_fail(vcpu,
5395 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5396 
5397 	/* according to the intel vmx instruction reference, the memory
5398 	 * operand is read even if it isn't needed (e.g., for type==global)
5399 	 */
5400 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5401 			vmx_instruction_info, false, sizeof(operand), &gva))
5402 		return 1;
5403 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5404 	if (r != X86EMUL_CONTINUE)
5405 		return kvm_handle_memory_failure(vcpu, r, &e);
5406 
5407 	if (operand.vpid >> 16)
5408 		return nested_vmx_fail(vcpu,
5409 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5410 
5411 	vpid02 = nested_get_vpid02(vcpu);
5412 	switch (type) {
5413 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5414 		if (!operand.vpid ||
5415 		    is_noncanonical_address(operand.gla, vcpu))
5416 			return nested_vmx_fail(vcpu,
5417 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5418 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5419 		break;
5420 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5421 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5422 		if (!operand.vpid)
5423 			return nested_vmx_fail(vcpu,
5424 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5425 		vpid_sync_context(vpid02);
5426 		break;
5427 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5428 		vpid_sync_context(vpid02);
5429 		break;
5430 	default:
5431 		WARN_ON_ONCE(1);
5432 		return kvm_skip_emulated_instruction(vcpu);
5433 	}
5434 
5435 	/*
5436 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5437 	 * linear mappings for L2 (tagged with L2's VPID).  Free all roots as
5438 	 * VPIDs are not tracked in the MMU role.
5439 	 *
5440 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5441 	 * an MMU when EPT is disabled.
5442 	 *
5443 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5444 	 */
5445 	if (!enable_ept)
5446 		kvm_mmu_free_roots(vcpu, &vcpu->arch.root_mmu,
5447 				   KVM_MMU_ROOTS_ALL);
5448 
5449 	return nested_vmx_succeed(vcpu);
5450 }
5451 
5452 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5453 				     struct vmcs12 *vmcs12)
5454 {
5455 	u32 index = kvm_rcx_read(vcpu);
5456 	u64 new_eptp;
5457 	bool accessed_dirty;
5458 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
5459 
5460 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
5461 	    !nested_cpu_has_ept(vmcs12))
5462 		return 1;
5463 
5464 	if (index >= VMFUNC_EPTP_ENTRIES)
5465 		return 1;
5466 
5467 
5468 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5469 				     &new_eptp, index * 8, 8))
5470 		return 1;
5471 
5472 	accessed_dirty = !!(new_eptp & VMX_EPTP_AD_ENABLE_BIT);
5473 
5474 	/*
5475 	 * If the (L2) guest does a vmfunc to the currently
5476 	 * active ept pointer, we don't have to do anything else
5477 	 */
5478 	if (vmcs12->ept_pointer != new_eptp) {
5479 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5480 			return 1;
5481 
5482 		kvm_mmu_unload(vcpu);
5483 		mmu->ept_ad = accessed_dirty;
5484 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
5485 		vmcs12->ept_pointer = new_eptp;
5486 		/*
5487 		 * TODO: Check what's the correct approach in case
5488 		 * mmu reload fails. Currently, we just let the next
5489 		 * reload potentially fail
5490 		 */
5491 		kvm_mmu_reload(vcpu);
5492 	}
5493 
5494 	return 0;
5495 }
5496 
5497 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5498 {
5499 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5500 	struct vmcs12 *vmcs12;
5501 	u32 function = kvm_rax_read(vcpu);
5502 
5503 	/*
5504 	 * VMFUNC is only supported for nested guests, but we always enable the
5505 	 * secondary control for simplicity; for non-nested mode, fake that we
5506 	 * didn't by injecting #UD.
5507 	 */
5508 	if (!is_guest_mode(vcpu)) {
5509 		kvm_queue_exception(vcpu, UD_VECTOR);
5510 		return 1;
5511 	}
5512 
5513 	vmcs12 = get_vmcs12(vcpu);
5514 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
5515 		goto fail;
5516 
5517 	switch (function) {
5518 	case 0:
5519 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5520 			goto fail;
5521 		break;
5522 	default:
5523 		goto fail;
5524 	}
5525 	return kvm_skip_emulated_instruction(vcpu);
5526 
5527 fail:
5528 	/*
5529 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5530 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5531 	 * EXIT_REASON_VMFUNC as the exit reason.
5532 	 */
5533 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5534 			  vmx_get_intr_info(vcpu),
5535 			  vmx_get_exit_qual(vcpu));
5536 	return 1;
5537 }
5538 
5539 /*
5540  * Return true if an IO instruction with the specified port and size should cause
5541  * a VM-exit into L1.
5542  */
5543 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5544 				 int size)
5545 {
5546 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5547 	gpa_t bitmap, last_bitmap;
5548 	u8 b;
5549 
5550 	last_bitmap = (gpa_t)-1;
5551 	b = -1;
5552 
5553 	while (size > 0) {
5554 		if (port < 0x8000)
5555 			bitmap = vmcs12->io_bitmap_a;
5556 		else if (port < 0x10000)
5557 			bitmap = vmcs12->io_bitmap_b;
5558 		else
5559 			return true;
5560 		bitmap += (port & 0x7fff) / 8;
5561 
5562 		if (last_bitmap != bitmap)
5563 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5564 				return true;
5565 		if (b & (1 << (port & 7)))
5566 			return true;
5567 
5568 		port++;
5569 		size--;
5570 		last_bitmap = bitmap;
5571 	}
5572 
5573 	return false;
5574 }
5575 
5576 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5577 				       struct vmcs12 *vmcs12)
5578 {
5579 	unsigned long exit_qualification;
5580 	unsigned short port;
5581 	int size;
5582 
5583 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5584 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5585 
5586 	exit_qualification = vmx_get_exit_qual(vcpu);
5587 
5588 	port = exit_qualification >> 16;
5589 	size = (exit_qualification & 7) + 1;
5590 
5591 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5592 }
5593 
5594 /*
5595  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5596  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5597  * disinterest in the current event (read or write a specific MSR) by using an
5598  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5599  */
5600 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5601 					struct vmcs12 *vmcs12,
5602 					union vmx_exit_reason exit_reason)
5603 {
5604 	u32 msr_index = kvm_rcx_read(vcpu);
5605 	gpa_t bitmap;
5606 
5607 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5608 		return true;
5609 
5610 	/*
5611 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5612 	 * for the four combinations of read/write and low/high MSR numbers.
5613 	 * First we need to figure out which of the four to use:
5614 	 */
5615 	bitmap = vmcs12->msr_bitmap;
5616 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5617 		bitmap += 2048;
5618 	if (msr_index >= 0xc0000000) {
5619 		msr_index -= 0xc0000000;
5620 		bitmap += 1024;
5621 	}
5622 
5623 	/* Then read the msr_index'th bit from this bitmap: */
5624 	if (msr_index < 1024*8) {
5625 		unsigned char b;
5626 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5627 			return true;
5628 		return 1 & (b >> (msr_index & 7));
5629 	} else
5630 		return true; /* let L1 handle the wrong parameter */
5631 }
5632 
5633 /*
5634  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5635  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5636  * intercept (via guest_host_mask etc.) the current event.
5637  */
5638 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5639 	struct vmcs12 *vmcs12)
5640 {
5641 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5642 	int cr = exit_qualification & 15;
5643 	int reg;
5644 	unsigned long val;
5645 
5646 	switch ((exit_qualification >> 4) & 3) {
5647 	case 0: /* mov to cr */
5648 		reg = (exit_qualification >> 8) & 15;
5649 		val = kvm_register_readl(vcpu, reg);
5650 		switch (cr) {
5651 		case 0:
5652 			if (vmcs12->cr0_guest_host_mask &
5653 			    (val ^ vmcs12->cr0_read_shadow))
5654 				return true;
5655 			break;
5656 		case 3:
5657 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5658 				return true;
5659 			break;
5660 		case 4:
5661 			if (vmcs12->cr4_guest_host_mask &
5662 			    (vmcs12->cr4_read_shadow ^ val))
5663 				return true;
5664 			break;
5665 		case 8:
5666 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5667 				return true;
5668 			break;
5669 		}
5670 		break;
5671 	case 2: /* clts */
5672 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5673 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5674 			return true;
5675 		break;
5676 	case 1: /* mov from cr */
5677 		switch (cr) {
5678 		case 3:
5679 			if (vmcs12->cpu_based_vm_exec_control &
5680 			    CPU_BASED_CR3_STORE_EXITING)
5681 				return true;
5682 			break;
5683 		case 8:
5684 			if (vmcs12->cpu_based_vm_exec_control &
5685 			    CPU_BASED_CR8_STORE_EXITING)
5686 				return true;
5687 			break;
5688 		}
5689 		break;
5690 	case 3: /* lmsw */
5691 		/*
5692 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5693 		 * cr0. Other attempted changes are ignored, with no exit.
5694 		 */
5695 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5696 		if (vmcs12->cr0_guest_host_mask & 0xe &
5697 		    (val ^ vmcs12->cr0_read_shadow))
5698 			return true;
5699 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5700 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5701 		    (val & 0x1))
5702 			return true;
5703 		break;
5704 	}
5705 	return false;
5706 }
5707 
5708 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5709 	struct vmcs12 *vmcs12, gpa_t bitmap)
5710 {
5711 	u32 vmx_instruction_info;
5712 	unsigned long field;
5713 	u8 b;
5714 
5715 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5716 		return true;
5717 
5718 	/* Decode instruction info and find the field to access */
5719 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5720 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5721 
5722 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5723 	if (field >> 15)
5724 		return true;
5725 
5726 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5727 		return true;
5728 
5729 	return 1 & (b >> (field & 7));
5730 }
5731 
5732 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5733 {
5734 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5735 
5736 	if (nested_cpu_has_mtf(vmcs12))
5737 		return true;
5738 
5739 	/*
5740 	 * An MTF VM-exit may be injected into the guest by setting the
5741 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5742 	 * is the case regardless of the 'monitor trap flag' VM-execution
5743 	 * control.
5744 	 */
5745 	return entry_intr_info == (INTR_INFO_VALID_MASK
5746 				   | INTR_TYPE_OTHER_EVENT);
5747 }
5748 
5749 /*
5750  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5751  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5752  */
5753 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
5754 				     union vmx_exit_reason exit_reason)
5755 {
5756 	u32 intr_info;
5757 
5758 	switch ((u16)exit_reason.basic) {
5759 	case EXIT_REASON_EXCEPTION_NMI:
5760 		intr_info = vmx_get_intr_info(vcpu);
5761 		if (is_nmi(intr_info))
5762 			return true;
5763 		else if (is_page_fault(intr_info))
5764 			return vcpu->arch.apf.host_apf_flags || !enable_ept;
5765 		else if (is_debug(intr_info) &&
5766 			 vcpu->guest_debug &
5767 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5768 			return true;
5769 		else if (is_breakpoint(intr_info) &&
5770 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5771 			return true;
5772 		return false;
5773 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5774 		return true;
5775 	case EXIT_REASON_MCE_DURING_VMENTRY:
5776 		return true;
5777 	case EXIT_REASON_EPT_VIOLATION:
5778 		/*
5779 		 * L0 always deals with the EPT violation. If nested EPT is
5780 		 * used, and the nested mmu code discovers that the address is
5781 		 * missing in the guest EPT table (EPT12), the EPT violation
5782 		 * will be injected with nested_ept_inject_page_fault()
5783 		 */
5784 		return true;
5785 	case EXIT_REASON_EPT_MISCONFIG:
5786 		/*
5787 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5788 		 * table (shadow on EPT) or a merged EPT table that L0 built
5789 		 * (EPT on EPT). So any problems with the structure of the
5790 		 * table is L0's fault.
5791 		 */
5792 		return true;
5793 	case EXIT_REASON_PREEMPTION_TIMER:
5794 		return true;
5795 	case EXIT_REASON_PML_FULL:
5796 		/*
5797 		 * PML is emulated for an L1 VMM and should never be enabled in
5798 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
5799 		 */
5800 		return true;
5801 	case EXIT_REASON_VMFUNC:
5802 		/* VM functions are emulated through L2->L0 vmexits. */
5803 		return true;
5804 	case EXIT_REASON_ENCLS:
5805 		/* SGX is never exposed to L1 */
5806 		return true;
5807 	default:
5808 		break;
5809 	}
5810 	return false;
5811 }
5812 
5813 /*
5814  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5815  * is_guest_mode (L2).
5816  */
5817 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
5818 				     union vmx_exit_reason exit_reason)
5819 {
5820 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5821 	u32 intr_info;
5822 
5823 	switch ((u16)exit_reason.basic) {
5824 	case EXIT_REASON_EXCEPTION_NMI:
5825 		intr_info = vmx_get_intr_info(vcpu);
5826 		if (is_nmi(intr_info))
5827 			return true;
5828 		else if (is_page_fault(intr_info))
5829 			return true;
5830 		return vmcs12->exception_bitmap &
5831 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5832 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5833 		return nested_exit_on_intr(vcpu);
5834 	case EXIT_REASON_TRIPLE_FAULT:
5835 		return true;
5836 	case EXIT_REASON_INTERRUPT_WINDOW:
5837 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5838 	case EXIT_REASON_NMI_WINDOW:
5839 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5840 	case EXIT_REASON_TASK_SWITCH:
5841 		return true;
5842 	case EXIT_REASON_CPUID:
5843 		return true;
5844 	case EXIT_REASON_HLT:
5845 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5846 	case EXIT_REASON_INVD:
5847 		return true;
5848 	case EXIT_REASON_INVLPG:
5849 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5850 	case EXIT_REASON_RDPMC:
5851 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5852 	case EXIT_REASON_RDRAND:
5853 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5854 	case EXIT_REASON_RDSEED:
5855 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5856 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5857 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5858 	case EXIT_REASON_VMREAD:
5859 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5860 			vmcs12->vmread_bitmap);
5861 	case EXIT_REASON_VMWRITE:
5862 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5863 			vmcs12->vmwrite_bitmap);
5864 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5865 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5866 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5867 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5868 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5869 		/*
5870 		 * VMX instructions trap unconditionally. This allows L1 to
5871 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5872 		 */
5873 		return true;
5874 	case EXIT_REASON_CR_ACCESS:
5875 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5876 	case EXIT_REASON_DR_ACCESS:
5877 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5878 	case EXIT_REASON_IO_INSTRUCTION:
5879 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5880 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5881 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5882 	case EXIT_REASON_MSR_READ:
5883 	case EXIT_REASON_MSR_WRITE:
5884 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5885 	case EXIT_REASON_INVALID_STATE:
5886 		return true;
5887 	case EXIT_REASON_MWAIT_INSTRUCTION:
5888 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5889 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5890 		return nested_vmx_exit_handled_mtf(vmcs12);
5891 	case EXIT_REASON_MONITOR_INSTRUCTION:
5892 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5893 	case EXIT_REASON_PAUSE_INSTRUCTION:
5894 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5895 			nested_cpu_has2(vmcs12,
5896 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5897 	case EXIT_REASON_MCE_DURING_VMENTRY:
5898 		return true;
5899 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5900 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5901 	case EXIT_REASON_APIC_ACCESS:
5902 	case EXIT_REASON_APIC_WRITE:
5903 	case EXIT_REASON_EOI_INDUCED:
5904 		/*
5905 		 * The controls for "virtualize APIC accesses," "APIC-
5906 		 * register virtualization," and "virtual-interrupt
5907 		 * delivery" only come from vmcs12.
5908 		 */
5909 		return true;
5910 	case EXIT_REASON_INVPCID:
5911 		return
5912 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5913 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5914 	case EXIT_REASON_WBINVD:
5915 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5916 	case EXIT_REASON_XSETBV:
5917 		return true;
5918 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5919 		/*
5920 		 * This should never happen, since it is not possible to
5921 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5922 		 * If if it were, XSS would have to be checked against
5923 		 * the XSS exit bitmap in vmcs12.
5924 		 */
5925 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5926 	case EXIT_REASON_UMWAIT:
5927 	case EXIT_REASON_TPAUSE:
5928 		return nested_cpu_has2(vmcs12,
5929 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
5930 	default:
5931 		return true;
5932 	}
5933 }
5934 
5935 /*
5936  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
5937  * reflected into L1.
5938  */
5939 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
5940 {
5941 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5942 	union vmx_exit_reason exit_reason = vmx->exit_reason;
5943 	unsigned long exit_qual;
5944 	u32 exit_intr_info;
5945 
5946 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
5947 
5948 	/*
5949 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
5950 	 * has already loaded L2's state.
5951 	 */
5952 	if (unlikely(vmx->fail)) {
5953 		trace_kvm_nested_vmenter_failed(
5954 			"hardware VM-instruction error: ",
5955 			vmcs_read32(VM_INSTRUCTION_ERROR));
5956 		exit_intr_info = 0;
5957 		exit_qual = 0;
5958 		goto reflect_vmexit;
5959 	}
5960 
5961 	trace_kvm_nested_vmexit(exit_reason.full, vcpu, KVM_ISA_VMX);
5962 
5963 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
5964 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
5965 		return false;
5966 
5967 	/* If L1 doesn't want the exit, handle it in L0. */
5968 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
5969 		return false;
5970 
5971 	/*
5972 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
5973 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
5974 	 * need to be synthesized by querying the in-kernel LAPIC, but external
5975 	 * interrupts are never reflected to L1 so it's a non-issue.
5976 	 */
5977 	exit_intr_info = vmx_get_intr_info(vcpu);
5978 	if (is_exception_with_error_code(exit_intr_info)) {
5979 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5980 
5981 		vmcs12->vm_exit_intr_error_code =
5982 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5983 	}
5984 	exit_qual = vmx_get_exit_qual(vcpu);
5985 
5986 reflect_vmexit:
5987 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
5988 	return true;
5989 }
5990 
5991 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5992 				struct kvm_nested_state __user *user_kvm_nested_state,
5993 				u32 user_data_size)
5994 {
5995 	struct vcpu_vmx *vmx;
5996 	struct vmcs12 *vmcs12;
5997 	struct kvm_nested_state kvm_state = {
5998 		.flags = 0,
5999 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6000 		.size = sizeof(kvm_state),
6001 		.hdr.vmx.flags = 0,
6002 		.hdr.vmx.vmxon_pa = -1ull,
6003 		.hdr.vmx.vmcs12_pa = -1ull,
6004 		.hdr.vmx.preemption_timer_deadline = 0,
6005 	};
6006 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6007 		&user_kvm_nested_state->data.vmx[0];
6008 
6009 	if (!vcpu)
6010 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6011 
6012 	vmx = to_vmx(vcpu);
6013 	vmcs12 = get_vmcs12(vcpu);
6014 
6015 	if (nested_vmx_allowed(vcpu) &&
6016 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6017 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6018 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6019 
6020 		if (vmx_has_valid_vmcs12(vcpu)) {
6021 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6022 
6023 			if (vmx->nested.hv_evmcs)
6024 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6025 
6026 			if (is_guest_mode(vcpu) &&
6027 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6028 			    vmcs12->vmcs_link_pointer != -1ull)
6029 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6030 		}
6031 
6032 		if (vmx->nested.smm.vmxon)
6033 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6034 
6035 		if (vmx->nested.smm.guest_mode)
6036 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6037 
6038 		if (is_guest_mode(vcpu)) {
6039 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6040 
6041 			if (vmx->nested.nested_run_pending)
6042 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6043 
6044 			if (vmx->nested.mtf_pending)
6045 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6046 
6047 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6048 			    vmx->nested.has_preemption_timer_deadline) {
6049 				kvm_state.hdr.vmx.flags |=
6050 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6051 				kvm_state.hdr.vmx.preemption_timer_deadline =
6052 					vmx->nested.preemption_timer_deadline;
6053 			}
6054 		}
6055 	}
6056 
6057 	if (user_data_size < kvm_state.size)
6058 		goto out;
6059 
6060 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6061 		return -EFAULT;
6062 
6063 	if (!vmx_has_valid_vmcs12(vcpu))
6064 		goto out;
6065 
6066 	/*
6067 	 * When running L2, the authoritative vmcs12 state is in the
6068 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6069 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6070 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6071 	 * vmcs12 state is in the vmcs12 already.
6072 	 */
6073 	if (is_guest_mode(vcpu)) {
6074 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6075 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6076 	} else  {
6077 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6078 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6079 			if (vmx->nested.hv_evmcs)
6080 				copy_enlightened_to_vmcs12(vmx);
6081 			else if (enable_shadow_vmcs)
6082 				copy_shadow_to_vmcs12(vmx);
6083 		}
6084 	}
6085 
6086 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6087 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6088 
6089 	/*
6090 	 * Copy over the full allocated size of vmcs12 rather than just the size
6091 	 * of the struct.
6092 	 */
6093 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6094 		return -EFAULT;
6095 
6096 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6097 	    vmcs12->vmcs_link_pointer != -1ull) {
6098 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6099 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6100 			return -EFAULT;
6101 	}
6102 out:
6103 	return kvm_state.size;
6104 }
6105 
6106 /*
6107  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6108  */
6109 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6110 {
6111 	if (is_guest_mode(vcpu)) {
6112 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6113 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6114 	}
6115 	free_nested(vcpu);
6116 }
6117 
6118 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6119 				struct kvm_nested_state __user *user_kvm_nested_state,
6120 				struct kvm_nested_state *kvm_state)
6121 {
6122 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6123 	struct vmcs12 *vmcs12;
6124 	enum vm_entry_failure_code ignored;
6125 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6126 		&user_kvm_nested_state->data.vmx[0];
6127 	int ret;
6128 
6129 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6130 		return -EINVAL;
6131 
6132 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
6133 		if (kvm_state->hdr.vmx.smm.flags)
6134 			return -EINVAL;
6135 
6136 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
6137 			return -EINVAL;
6138 
6139 		/*
6140 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6141 		 * enable eVMCS capability on vCPU. However, since then
6142 		 * code was changed such that flag signals vmcs12 should
6143 		 * be copied into eVMCS in guest memory.
6144 		 *
6145 		 * To preserve backwards compatability, allow user
6146 		 * to set this flag even when there is no VMXON region.
6147 		 */
6148 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6149 			return -EINVAL;
6150 	} else {
6151 		if (!nested_vmx_allowed(vcpu))
6152 			return -EINVAL;
6153 
6154 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6155 			return -EINVAL;
6156 	}
6157 
6158 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6159 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6160 		return -EINVAL;
6161 
6162 	if (kvm_state->hdr.vmx.smm.flags &
6163 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6164 		return -EINVAL;
6165 
6166 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6167 		return -EINVAL;
6168 
6169 	/*
6170 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6171 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6172 	 * must be zero.
6173 	 */
6174 	if (is_smm(vcpu) ?
6175 		(kvm_state->flags &
6176 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6177 		: kvm_state->hdr.vmx.smm.flags)
6178 		return -EINVAL;
6179 
6180 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6181 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6182 		return -EINVAL;
6183 
6184 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6185 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6186 			return -EINVAL;
6187 
6188 	vmx_leave_nested(vcpu);
6189 
6190 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
6191 		return 0;
6192 
6193 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6194 	ret = enter_vmx_operation(vcpu);
6195 	if (ret)
6196 		return ret;
6197 
6198 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6199 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6200 		/* See vmx_has_valid_vmcs12.  */
6201 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6202 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6203 		    (kvm_state->hdr.vmx.vmcs12_pa != -1ull))
6204 			return -EINVAL;
6205 		else
6206 			return 0;
6207 	}
6208 
6209 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
6210 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6211 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6212 			return -EINVAL;
6213 
6214 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6215 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6216 		/*
6217 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6218 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6219 		 * restored yet. EVMCS will be mapped from
6220 		 * nested_get_vmcs12_pages().
6221 		 */
6222 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6223 	} else {
6224 		return -EINVAL;
6225 	}
6226 
6227 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6228 		vmx->nested.smm.vmxon = true;
6229 		vmx->nested.vmxon = false;
6230 
6231 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6232 			vmx->nested.smm.guest_mode = true;
6233 	}
6234 
6235 	vmcs12 = get_vmcs12(vcpu);
6236 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6237 		return -EFAULT;
6238 
6239 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6240 		return -EINVAL;
6241 
6242 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6243 		return 0;
6244 
6245 	vmx->nested.nested_run_pending =
6246 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6247 
6248 	vmx->nested.mtf_pending =
6249 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6250 
6251 	ret = -EINVAL;
6252 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6253 	    vmcs12->vmcs_link_pointer != -1ull) {
6254 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6255 
6256 		if (kvm_state->size <
6257 		    sizeof(*kvm_state) +
6258 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6259 			goto error_guest_mode;
6260 
6261 		if (copy_from_user(shadow_vmcs12,
6262 				   user_vmx_nested_state->shadow_vmcs12,
6263 				   sizeof(*shadow_vmcs12))) {
6264 			ret = -EFAULT;
6265 			goto error_guest_mode;
6266 		}
6267 
6268 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6269 		    !shadow_vmcs12->hdr.shadow_vmcs)
6270 			goto error_guest_mode;
6271 	}
6272 
6273 	vmx->nested.has_preemption_timer_deadline = false;
6274 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6275 		vmx->nested.has_preemption_timer_deadline = true;
6276 		vmx->nested.preemption_timer_deadline =
6277 			kvm_state->hdr.vmx.preemption_timer_deadline;
6278 	}
6279 
6280 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6281 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6282 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6283 		goto error_guest_mode;
6284 
6285 	vmx->nested.dirty_vmcs12 = true;
6286 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6287 	if (ret)
6288 		goto error_guest_mode;
6289 
6290 	return 0;
6291 
6292 error_guest_mode:
6293 	vmx->nested.nested_run_pending = 0;
6294 	return ret;
6295 }
6296 
6297 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6298 {
6299 	if (enable_shadow_vmcs) {
6300 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6301 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6302 	}
6303 }
6304 
6305 /*
6306  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6307  * returned for the various VMX controls MSRs when nested VMX is enabled.
6308  * The same values should also be used to verify that vmcs12 control fields are
6309  * valid during nested entry from L1 to L2.
6310  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6311  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6312  * bit in the high half is on if the corresponding bit in the control field
6313  * may be on. See also vmx_control_verify().
6314  */
6315 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6316 {
6317 	/*
6318 	 * Note that as a general rule, the high half of the MSRs (bits in
6319 	 * the control fields which may be 1) should be initialized by the
6320 	 * intersection of the underlying hardware's MSR (i.e., features which
6321 	 * can be supported) and the list of features we want to expose -
6322 	 * because they are known to be properly supported in our code.
6323 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6324 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6325 	 * reason is that if one of these bits is necessary, it will appear
6326 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6327 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6328 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6329 	 * These rules have exceptions below.
6330 	 */
6331 
6332 	/* pin-based controls */
6333 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6334 		msrs->pinbased_ctls_low,
6335 		msrs->pinbased_ctls_high);
6336 	msrs->pinbased_ctls_low |=
6337 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6338 	msrs->pinbased_ctls_high &=
6339 		PIN_BASED_EXT_INTR_MASK |
6340 		PIN_BASED_NMI_EXITING |
6341 		PIN_BASED_VIRTUAL_NMIS |
6342 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6343 	msrs->pinbased_ctls_high |=
6344 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6345 		PIN_BASED_VMX_PREEMPTION_TIMER;
6346 
6347 	/* exit controls */
6348 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6349 		msrs->exit_ctls_low,
6350 		msrs->exit_ctls_high);
6351 	msrs->exit_ctls_low =
6352 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6353 
6354 	msrs->exit_ctls_high &=
6355 #ifdef CONFIG_X86_64
6356 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6357 #endif
6358 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6359 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6360 	msrs->exit_ctls_high |=
6361 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6362 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6363 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6364 
6365 	/* We support free control of debug control saving. */
6366 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6367 
6368 	/* entry controls */
6369 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6370 		msrs->entry_ctls_low,
6371 		msrs->entry_ctls_high);
6372 	msrs->entry_ctls_low =
6373 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6374 	msrs->entry_ctls_high &=
6375 #ifdef CONFIG_X86_64
6376 		VM_ENTRY_IA32E_MODE |
6377 #endif
6378 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6379 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6380 	msrs->entry_ctls_high |=
6381 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6382 
6383 	/* We support free control of debug control loading. */
6384 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6385 
6386 	/* cpu-based controls */
6387 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6388 		msrs->procbased_ctls_low,
6389 		msrs->procbased_ctls_high);
6390 	msrs->procbased_ctls_low =
6391 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6392 	msrs->procbased_ctls_high &=
6393 		CPU_BASED_INTR_WINDOW_EXITING |
6394 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6395 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6396 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6397 		CPU_BASED_CR3_STORE_EXITING |
6398 #ifdef CONFIG_X86_64
6399 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6400 #endif
6401 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6402 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6403 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6404 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6405 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6406 	/*
6407 	 * We can allow some features even when not supported by the
6408 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6409 	 * can use it to avoid exits to L1 - even when L0 runs L2
6410 	 * without MSR bitmaps.
6411 	 */
6412 	msrs->procbased_ctls_high |=
6413 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6414 		CPU_BASED_USE_MSR_BITMAPS;
6415 
6416 	/* We support free control of CR3 access interception. */
6417 	msrs->procbased_ctls_low &=
6418 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6419 
6420 	/*
6421 	 * secondary cpu-based controls.  Do not include those that
6422 	 * depend on CPUID bits, they are added later by
6423 	 * vmx_vcpu_after_set_cpuid.
6424 	 */
6425 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6426 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6427 		      msrs->secondary_ctls_low,
6428 		      msrs->secondary_ctls_high);
6429 
6430 	msrs->secondary_ctls_low = 0;
6431 	msrs->secondary_ctls_high &=
6432 		SECONDARY_EXEC_DESC |
6433 		SECONDARY_EXEC_ENABLE_RDTSCP |
6434 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6435 		SECONDARY_EXEC_WBINVD_EXITING |
6436 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6437 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6438 		SECONDARY_EXEC_RDRAND_EXITING |
6439 		SECONDARY_EXEC_ENABLE_INVPCID |
6440 		SECONDARY_EXEC_RDSEED_EXITING |
6441 		SECONDARY_EXEC_XSAVES;
6442 
6443 	/*
6444 	 * We can emulate "VMCS shadowing," even if the hardware
6445 	 * doesn't support it.
6446 	 */
6447 	msrs->secondary_ctls_high |=
6448 		SECONDARY_EXEC_SHADOW_VMCS;
6449 
6450 	if (enable_ept) {
6451 		/* nested EPT: emulate EPT also to L1 */
6452 		msrs->secondary_ctls_high |=
6453 			SECONDARY_EXEC_ENABLE_EPT;
6454 		msrs->ept_caps =
6455 			VMX_EPT_PAGE_WALK_4_BIT |
6456 			VMX_EPT_PAGE_WALK_5_BIT |
6457 			VMX_EPTP_WB_BIT |
6458 			VMX_EPT_INVEPT_BIT |
6459 			VMX_EPT_EXECUTE_ONLY_BIT;
6460 
6461 		msrs->ept_caps &= ept_caps;
6462 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6463 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6464 			VMX_EPT_1GB_PAGE_BIT;
6465 		if (enable_ept_ad_bits) {
6466 			msrs->secondary_ctls_high |=
6467 				SECONDARY_EXEC_ENABLE_PML;
6468 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6469 		}
6470 	}
6471 
6472 	if (cpu_has_vmx_vmfunc()) {
6473 		msrs->secondary_ctls_high |=
6474 			SECONDARY_EXEC_ENABLE_VMFUNC;
6475 		/*
6476 		 * Advertise EPTP switching unconditionally
6477 		 * since we emulate it
6478 		 */
6479 		if (enable_ept)
6480 			msrs->vmfunc_controls =
6481 				VMX_VMFUNC_EPTP_SWITCHING;
6482 	}
6483 
6484 	/*
6485 	 * Old versions of KVM use the single-context version without
6486 	 * checking for support, so declare that it is supported even
6487 	 * though it is treated as global context.  The alternative is
6488 	 * not failing the single-context invvpid, and it is worse.
6489 	 */
6490 	if (enable_vpid) {
6491 		msrs->secondary_ctls_high |=
6492 			SECONDARY_EXEC_ENABLE_VPID;
6493 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6494 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6495 	}
6496 
6497 	if (enable_unrestricted_guest)
6498 		msrs->secondary_ctls_high |=
6499 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6500 
6501 	if (flexpriority_enabled)
6502 		msrs->secondary_ctls_high |=
6503 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6504 
6505 	/* miscellaneous data */
6506 	rdmsr(MSR_IA32_VMX_MISC,
6507 		msrs->misc_low,
6508 		msrs->misc_high);
6509 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6510 	msrs->misc_low |=
6511 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6512 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6513 		VMX_MISC_ACTIVITY_HLT |
6514 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6515 	msrs->misc_high = 0;
6516 
6517 	/*
6518 	 * This MSR reports some information about VMX support. We
6519 	 * should return information about the VMX we emulate for the
6520 	 * guest, and the VMCS structure we give it - not about the
6521 	 * VMX support of the underlying hardware.
6522 	 */
6523 	msrs->basic =
6524 		VMCS12_REVISION |
6525 		VMX_BASIC_TRUE_CTLS |
6526 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6527 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6528 
6529 	if (cpu_has_vmx_basic_inout())
6530 		msrs->basic |= VMX_BASIC_INOUT;
6531 
6532 	/*
6533 	 * These MSRs specify bits which the guest must keep fixed on
6534 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6535 	 * We picked the standard core2 setting.
6536 	 */
6537 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6538 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6539 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6540 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6541 
6542 	/* These MSRs specify bits which the guest must keep fixed off. */
6543 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6544 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6545 
6546 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
6547 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
6548 }
6549 
6550 void nested_vmx_hardware_unsetup(void)
6551 {
6552 	int i;
6553 
6554 	if (enable_shadow_vmcs) {
6555 		for (i = 0; i < VMX_BITMAP_NR; i++)
6556 			free_page((unsigned long)vmx_bitmap[i]);
6557 	}
6558 }
6559 
6560 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6561 {
6562 	int i;
6563 
6564 	if (!cpu_has_vmx_shadow_vmcs())
6565 		enable_shadow_vmcs = 0;
6566 	if (enable_shadow_vmcs) {
6567 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6568 			/*
6569 			 * The vmx_bitmap is not tied to a VM and so should
6570 			 * not be charged to a memcg.
6571 			 */
6572 			vmx_bitmap[i] = (unsigned long *)
6573 				__get_free_page(GFP_KERNEL);
6574 			if (!vmx_bitmap[i]) {
6575 				nested_vmx_hardware_unsetup();
6576 				return -ENOMEM;
6577 			}
6578 		}
6579 
6580 		init_vmcs_shadow_fields();
6581 	}
6582 
6583 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6584 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6585 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6586 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6587 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6588 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6589 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6590 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6591 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6592 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6593 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6594 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6595 
6596 	return 0;
6597 }
6598 
6599 struct kvm_x86_nested_ops vmx_nested_ops = {
6600 	.check_events = vmx_check_nested_events,
6601 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6602 	.get_state = vmx_get_nested_state,
6603 	.set_state = vmx_set_nested_state,
6604 	.get_nested_state_pages = vmx_get_nested_state_pages,
6605 	.write_log_dirty = nested_vmx_write_pml_buffer,
6606 	.enable_evmcs = nested_enable_evmcs,
6607 	.get_evmcs_version = nested_get_evmcs_version,
6608 };
6609