xref: /linux/arch/x86/kvm/vmx/nested.c (revision 0c8a32eed1625a65798286fb73fea8710a908545)
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 "x86.h"
16 
17 static bool __read_mostly enable_shadow_vmcs = 1;
18 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
19 
20 static bool __read_mostly nested_early_check = 0;
21 module_param(nested_early_check, bool, S_IRUGO);
22 
23 #define CC(consistency_check)						\
24 ({									\
25 	bool failed = (consistency_check);				\
26 	if (failed)							\
27 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
28 	failed;								\
29 })
30 
31 /*
32  * Hyper-V requires all of these, so mark them as supported even though
33  * they are just treated the same as all-context.
34  */
35 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
36 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
37 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
38 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
39 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
40 
41 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
42 
43 enum {
44 	VMX_VMREAD_BITMAP,
45 	VMX_VMWRITE_BITMAP,
46 	VMX_BITMAP_NR
47 };
48 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
49 
50 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
51 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
52 
53 struct shadow_vmcs_field {
54 	u16	encoding;
55 	u16	offset;
56 };
57 static struct shadow_vmcs_field shadow_read_only_fields[] = {
58 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
59 #include "vmcs_shadow_fields.h"
60 };
61 static int max_shadow_read_only_fields =
62 	ARRAY_SIZE(shadow_read_only_fields);
63 
64 static struct shadow_vmcs_field shadow_read_write_fields[] = {
65 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
66 #include "vmcs_shadow_fields.h"
67 };
68 static int max_shadow_read_write_fields =
69 	ARRAY_SIZE(shadow_read_write_fields);
70 
71 static void init_vmcs_shadow_fields(void)
72 {
73 	int i, j;
74 
75 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
76 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
77 
78 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
79 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
80 		u16 field = entry.encoding;
81 
82 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
83 		    (i + 1 == max_shadow_read_only_fields ||
84 		     shadow_read_only_fields[i + 1].encoding != field + 1))
85 			pr_err("Missing field from shadow_read_only_field %x\n",
86 			       field + 1);
87 
88 		clear_bit(field, vmx_vmread_bitmap);
89 		if (field & 1)
90 #ifdef CONFIG_X86_64
91 			continue;
92 #else
93 			entry.offset += sizeof(u32);
94 #endif
95 		shadow_read_only_fields[j++] = entry;
96 	}
97 	max_shadow_read_only_fields = j;
98 
99 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
100 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
101 		u16 field = entry.encoding;
102 
103 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
104 		    (i + 1 == max_shadow_read_write_fields ||
105 		     shadow_read_write_fields[i + 1].encoding != field + 1))
106 			pr_err("Missing field from shadow_read_write_field %x\n",
107 			       field + 1);
108 
109 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
110 			  field <= GUEST_TR_AR_BYTES,
111 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
112 
113 		/*
114 		 * PML and the preemption timer can be emulated, but the
115 		 * processor cannot vmwrite to fields that don't exist
116 		 * on bare metal.
117 		 */
118 		switch (field) {
119 		case GUEST_PML_INDEX:
120 			if (!cpu_has_vmx_pml())
121 				continue;
122 			break;
123 		case VMX_PREEMPTION_TIMER_VALUE:
124 			if (!cpu_has_vmx_preemption_timer())
125 				continue;
126 			break;
127 		case GUEST_INTR_STATUS:
128 			if (!cpu_has_vmx_apicv())
129 				continue;
130 			break;
131 		default:
132 			break;
133 		}
134 
135 		clear_bit(field, vmx_vmwrite_bitmap);
136 		clear_bit(field, vmx_vmread_bitmap);
137 		if (field & 1)
138 #ifdef CONFIG_X86_64
139 			continue;
140 #else
141 			entry.offset += sizeof(u32);
142 #endif
143 		shadow_read_write_fields[j++] = entry;
144 	}
145 	max_shadow_read_write_fields = j;
146 }
147 
148 /*
149  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
150  * set the success or error code of an emulated VMX instruction (as specified
151  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
152  * instruction.
153  */
154 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
155 {
156 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
157 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
158 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
159 	return kvm_skip_emulated_instruction(vcpu);
160 }
161 
162 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
163 {
164 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
165 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
166 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
167 			| X86_EFLAGS_CF);
168 	return kvm_skip_emulated_instruction(vcpu);
169 }
170 
171 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
172 				u32 vm_instruction_error)
173 {
174 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
175 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
176 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
177 			| X86_EFLAGS_ZF);
178 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
179 	/*
180 	 * We don't need to force a shadow sync because
181 	 * VM_INSTRUCTION_ERROR is not shadowed
182 	 */
183 	return kvm_skip_emulated_instruction(vcpu);
184 }
185 
186 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
187 {
188 	struct vcpu_vmx *vmx = to_vmx(vcpu);
189 
190 	/*
191 	 * failValid writes the error number to the current VMCS, which
192 	 * can't be done if there isn't a current VMCS.
193 	 */
194 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
195 		return nested_vmx_failInvalid(vcpu);
196 
197 	return nested_vmx_failValid(vcpu, vm_instruction_error);
198 }
199 
200 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
201 {
202 	/* TODO: not to reset guest simply here. */
203 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
204 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
205 }
206 
207 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
208 {
209 	return fixed_bits_valid(control, low, high);
210 }
211 
212 static inline u64 vmx_control_msr(u32 low, u32 high)
213 {
214 	return low | ((u64)high << 32);
215 }
216 
217 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
218 {
219 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
220 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
221 	vmx->nested.need_vmcs12_to_shadow_sync = false;
222 }
223 
224 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
225 {
226 	struct vcpu_vmx *vmx = to_vmx(vcpu);
227 
228 	if (!vmx->nested.hv_evmcs)
229 		return;
230 
231 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
232 	vmx->nested.hv_evmcs_vmptr = 0;
233 	vmx->nested.hv_evmcs = NULL;
234 }
235 
236 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
237 				     struct loaded_vmcs *prev)
238 {
239 	struct vmcs_host_state *dest, *src;
240 
241 	if (unlikely(!vmx->guest_state_loaded))
242 		return;
243 
244 	src = &prev->host_state;
245 	dest = &vmx->loaded_vmcs->host_state;
246 
247 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
248 	dest->ldt_sel = src->ldt_sel;
249 #ifdef CONFIG_X86_64
250 	dest->ds_sel = src->ds_sel;
251 	dest->es_sel = src->es_sel;
252 #endif
253 }
254 
255 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
256 {
257 	struct vcpu_vmx *vmx = to_vmx(vcpu);
258 	struct loaded_vmcs *prev;
259 	int cpu;
260 
261 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
262 		return;
263 
264 	cpu = get_cpu();
265 	prev = vmx->loaded_vmcs;
266 	vmx->loaded_vmcs = vmcs;
267 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
268 	vmx_sync_vmcs_host_state(vmx, prev);
269 	put_cpu();
270 
271 	vmx_register_cache_reset(vcpu);
272 }
273 
274 /*
275  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
276  * just stops using VMX.
277  */
278 static void free_nested(struct kvm_vcpu *vcpu)
279 {
280 	struct vcpu_vmx *vmx = to_vmx(vcpu);
281 
282 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
283 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
284 
285 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
286 		return;
287 
288 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
289 
290 	vmx->nested.vmxon = false;
291 	vmx->nested.smm.vmxon = false;
292 	free_vpid(vmx->nested.vpid02);
293 	vmx->nested.posted_intr_nv = -1;
294 	vmx->nested.current_vmptr = -1ull;
295 	if (enable_shadow_vmcs) {
296 		vmx_disable_shadow_vmcs(vmx);
297 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
298 		free_vmcs(vmx->vmcs01.shadow_vmcs);
299 		vmx->vmcs01.shadow_vmcs = NULL;
300 	}
301 	kfree(vmx->nested.cached_vmcs12);
302 	vmx->nested.cached_vmcs12 = NULL;
303 	kfree(vmx->nested.cached_shadow_vmcs12);
304 	vmx->nested.cached_shadow_vmcs12 = NULL;
305 	/* Unpin physical memory we referred to in the vmcs02 */
306 	if (vmx->nested.apic_access_page) {
307 		kvm_release_page_clean(vmx->nested.apic_access_page);
308 		vmx->nested.apic_access_page = NULL;
309 	}
310 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
311 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
312 	vmx->nested.pi_desc = NULL;
313 
314 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
315 
316 	nested_release_evmcs(vcpu);
317 
318 	free_loaded_vmcs(&vmx->nested.vmcs02);
319 }
320 
321 /*
322  * Ensure that the current vmcs of the logical processor is the
323  * vmcs01 of the vcpu before calling free_nested().
324  */
325 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
326 {
327 	vcpu_load(vcpu);
328 	vmx_leave_nested(vcpu);
329 	vcpu_put(vcpu);
330 }
331 
332 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
333 		struct x86_exception *fault)
334 {
335 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
336 	struct vcpu_vmx *vmx = to_vmx(vcpu);
337 	u32 vm_exit_reason;
338 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
339 
340 	if (vmx->nested.pml_full) {
341 		vm_exit_reason = EXIT_REASON_PML_FULL;
342 		vmx->nested.pml_full = false;
343 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
344 	} else if (fault->error_code & PFERR_RSVD_MASK)
345 		vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
346 	else
347 		vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
348 
349 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
350 	vmcs12->guest_physical_address = fault->address;
351 }
352 
353 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
354 {
355 	WARN_ON(mmu_is_nested(vcpu));
356 
357 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
358 	kvm_init_shadow_ept_mmu(vcpu,
359 			to_vmx(vcpu)->nested.msrs.ept_caps &
360 			VMX_EPT_EXECUTE_ONLY_BIT,
361 			nested_ept_ad_enabled(vcpu),
362 			nested_ept_get_eptp(vcpu));
363 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
364 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
365 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
366 
367 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
368 }
369 
370 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
371 {
372 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
373 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
374 }
375 
376 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
377 					    u16 error_code)
378 {
379 	bool inequality, bit;
380 
381 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
382 	inequality =
383 		(error_code & vmcs12->page_fault_error_code_mask) !=
384 		 vmcs12->page_fault_error_code_match;
385 	return inequality ^ bit;
386 }
387 
388 
389 /*
390  * KVM wants to inject page-faults which it got to the guest. This function
391  * checks whether in a nested guest, we need to inject them to L1 or L2.
392  */
393 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
394 {
395 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
396 	unsigned int nr = vcpu->arch.exception.nr;
397 	bool has_payload = vcpu->arch.exception.has_payload;
398 	unsigned long payload = vcpu->arch.exception.payload;
399 
400 	if (nr == PF_VECTOR) {
401 		if (vcpu->arch.exception.nested_apf) {
402 			*exit_qual = vcpu->arch.apf.nested_apf_token;
403 			return 1;
404 		}
405 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
406 						    vcpu->arch.exception.error_code)) {
407 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
408 			return 1;
409 		}
410 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
411 		if (nr == DB_VECTOR) {
412 			if (!has_payload) {
413 				payload = vcpu->arch.dr6;
414 				payload &= ~(DR6_FIXED_1 | DR6_BT);
415 				payload ^= DR6_RTM;
416 			}
417 			*exit_qual = payload;
418 		} else
419 			*exit_qual = 0;
420 		return 1;
421 	}
422 
423 	return 0;
424 }
425 
426 
427 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
428 		struct x86_exception *fault)
429 {
430 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
431 
432 	WARN_ON(!is_guest_mode(vcpu));
433 
434 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
435 		!to_vmx(vcpu)->nested.nested_run_pending) {
436 		vmcs12->vm_exit_intr_error_code = fault->error_code;
437 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
438 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
439 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
440 				  fault->address);
441 	} else {
442 		kvm_inject_page_fault(vcpu, fault);
443 	}
444 }
445 
446 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
447 					       struct vmcs12 *vmcs12)
448 {
449 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
450 		return 0;
451 
452 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
453 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
454 		return -EINVAL;
455 
456 	return 0;
457 }
458 
459 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
460 						struct vmcs12 *vmcs12)
461 {
462 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
463 		return 0;
464 
465 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
466 		return -EINVAL;
467 
468 	return 0;
469 }
470 
471 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
472 						struct vmcs12 *vmcs12)
473 {
474 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
475 		return 0;
476 
477 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
478 		return -EINVAL;
479 
480 	return 0;
481 }
482 
483 /*
484  * Check if MSR is intercepted for L01 MSR bitmap.
485  */
486 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
487 {
488 	unsigned long *msr_bitmap;
489 	int f = sizeof(unsigned long);
490 
491 	if (!cpu_has_vmx_msr_bitmap())
492 		return true;
493 
494 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
495 
496 	if (msr <= 0x1fff) {
497 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
498 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
499 		msr &= 0x1fff;
500 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
501 	}
502 
503 	return true;
504 }
505 
506 /*
507  * If a msr is allowed by L0, we should check whether it is allowed by L1.
508  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
509  */
510 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
511 					       unsigned long *msr_bitmap_nested,
512 					       u32 msr, int type)
513 {
514 	int f = sizeof(unsigned long);
515 
516 	/*
517 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
518 	 * have the write-low and read-high bitmap offsets the wrong way round.
519 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
520 	 */
521 	if (msr <= 0x1fff) {
522 		if (type & MSR_TYPE_R &&
523 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
524 			/* read-low */
525 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
526 
527 		if (type & MSR_TYPE_W &&
528 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
529 			/* write-low */
530 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
531 
532 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
533 		msr &= 0x1fff;
534 		if (type & MSR_TYPE_R &&
535 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
536 			/* read-high */
537 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
538 
539 		if (type & MSR_TYPE_W &&
540 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
541 			/* write-high */
542 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
543 
544 	}
545 }
546 
547 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
548 {
549 	int msr;
550 
551 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
552 		unsigned word = msr / BITS_PER_LONG;
553 
554 		msr_bitmap[word] = ~0;
555 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
556 	}
557 }
558 
559 /*
560  * Merge L0's and L1's MSR bitmap, return false to indicate that
561  * we do not use the hardware.
562  */
563 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
564 						 struct vmcs12 *vmcs12)
565 {
566 	int msr;
567 	unsigned long *msr_bitmap_l1;
568 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
569 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
570 
571 	/* Nothing to do if the MSR bitmap is not in use.  */
572 	if (!cpu_has_vmx_msr_bitmap() ||
573 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
574 		return false;
575 
576 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
577 		return false;
578 
579 	msr_bitmap_l1 = (unsigned long *)map->hva;
580 
581 	/*
582 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
583 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
584 	 * the x2APIC MSR range and selectively disable them below.
585 	 */
586 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
587 
588 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
589 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
590 			/*
591 			 * L0 need not intercept reads for MSRs between 0x800
592 			 * and 0x8ff, it just lets the processor take the value
593 			 * from the virtual-APIC page; take those 256 bits
594 			 * directly from the L1 bitmap.
595 			 */
596 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
597 				unsigned word = msr / BITS_PER_LONG;
598 
599 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
600 			}
601 		}
602 
603 		nested_vmx_disable_intercept_for_msr(
604 			msr_bitmap_l1, msr_bitmap_l0,
605 			X2APIC_MSR(APIC_TASKPRI),
606 			MSR_TYPE_R | MSR_TYPE_W);
607 
608 		if (nested_cpu_has_vid(vmcs12)) {
609 			nested_vmx_disable_intercept_for_msr(
610 				msr_bitmap_l1, msr_bitmap_l0,
611 				X2APIC_MSR(APIC_EOI),
612 				MSR_TYPE_W);
613 			nested_vmx_disable_intercept_for_msr(
614 				msr_bitmap_l1, msr_bitmap_l0,
615 				X2APIC_MSR(APIC_SELF_IPI),
616 				MSR_TYPE_W);
617 		}
618 	}
619 
620 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
621 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
622 					     MSR_FS_BASE, MSR_TYPE_RW);
623 
624 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
625 					     MSR_GS_BASE, MSR_TYPE_RW);
626 
627 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
628 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
629 
630 	/*
631 	 * Checking the L0->L1 bitmap is trying to verify two things:
632 	 *
633 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
634 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
635 	 *    from the L12 MSR bitmap that is too permissive.
636 	 * 2. That L1 or L2s have actually used the MSR. This avoids
637 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
638 	 *    works properly because we only update the L01 MSR bitmap lazily.
639 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
640 	 *    updated to reflect this when L1 (or its L2s) actually write to
641 	 *    the MSR.
642 	 */
643 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
644 		nested_vmx_disable_intercept_for_msr(
645 					msr_bitmap_l1, msr_bitmap_l0,
646 					MSR_IA32_SPEC_CTRL,
647 					MSR_TYPE_R | MSR_TYPE_W);
648 
649 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
650 		nested_vmx_disable_intercept_for_msr(
651 					msr_bitmap_l1, msr_bitmap_l0,
652 					MSR_IA32_PRED_CMD,
653 					MSR_TYPE_W);
654 
655 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
656 
657 	return true;
658 }
659 
660 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
661 				       struct vmcs12 *vmcs12)
662 {
663 	struct kvm_host_map map;
664 	struct vmcs12 *shadow;
665 
666 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
667 	    vmcs12->vmcs_link_pointer == -1ull)
668 		return;
669 
670 	shadow = get_shadow_vmcs12(vcpu);
671 
672 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
673 		return;
674 
675 	memcpy(shadow, map.hva, VMCS12_SIZE);
676 	kvm_vcpu_unmap(vcpu, &map, false);
677 }
678 
679 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
680 					      struct vmcs12 *vmcs12)
681 {
682 	struct vcpu_vmx *vmx = to_vmx(vcpu);
683 
684 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
685 	    vmcs12->vmcs_link_pointer == -1ull)
686 		return;
687 
688 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
689 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
690 }
691 
692 /*
693  * In nested virtualization, check if L1 has set
694  * VM_EXIT_ACK_INTR_ON_EXIT
695  */
696 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
697 {
698 	return get_vmcs12(vcpu)->vm_exit_controls &
699 		VM_EXIT_ACK_INTR_ON_EXIT;
700 }
701 
702 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
703 					  struct vmcs12 *vmcs12)
704 {
705 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
706 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
707 		return -EINVAL;
708 	else
709 		return 0;
710 }
711 
712 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
713 					   struct vmcs12 *vmcs12)
714 {
715 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
716 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
717 	    !nested_cpu_has_vid(vmcs12) &&
718 	    !nested_cpu_has_posted_intr(vmcs12))
719 		return 0;
720 
721 	/*
722 	 * If virtualize x2apic mode is enabled,
723 	 * virtualize apic access must be disabled.
724 	 */
725 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
726 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
727 		return -EINVAL;
728 
729 	/*
730 	 * If virtual interrupt delivery is enabled,
731 	 * we must exit on external interrupts.
732 	 */
733 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
734 		return -EINVAL;
735 
736 	/*
737 	 * bits 15:8 should be zero in posted_intr_nv,
738 	 * the descriptor address has been already checked
739 	 * in nested_get_vmcs12_pages.
740 	 *
741 	 * bits 5:0 of posted_intr_desc_addr should be zero.
742 	 */
743 	if (nested_cpu_has_posted_intr(vmcs12) &&
744 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
745 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
746 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
747 	    CC((vmcs12->posted_intr_desc_addr & 0x3f)) ||
748 	    CC((vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu)))))
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 	int maxphyaddr;
762 
763 	if (count == 0)
764 		return 0;
765 	maxphyaddr = cpuid_maxphyaddr(vcpu);
766 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
767 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
768 		return -EINVAL;
769 
770 	return 0;
771 }
772 
773 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
774 						     struct vmcs12 *vmcs12)
775 {
776 	if (CC(nested_vmx_check_msr_switch(vcpu,
777 					   vmcs12->vm_exit_msr_load_count,
778 					   vmcs12->vm_exit_msr_load_addr)) ||
779 	    CC(nested_vmx_check_msr_switch(vcpu,
780 					   vmcs12->vm_exit_msr_store_count,
781 					   vmcs12->vm_exit_msr_store_addr)))
782 		return -EINVAL;
783 
784 	return 0;
785 }
786 
787 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
788                                                       struct vmcs12 *vmcs12)
789 {
790 	if (CC(nested_vmx_check_msr_switch(vcpu,
791 					   vmcs12->vm_entry_msr_load_count,
792 					   vmcs12->vm_entry_msr_load_addr)))
793                 return -EINVAL;
794 
795 	return 0;
796 }
797 
798 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
799 					 struct vmcs12 *vmcs12)
800 {
801 	if (!nested_cpu_has_pml(vmcs12))
802 		return 0;
803 
804 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
805 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
806 		return -EINVAL;
807 
808 	return 0;
809 }
810 
811 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
812 							struct vmcs12 *vmcs12)
813 {
814 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
815 	       !nested_cpu_has_ept(vmcs12)))
816 		return -EINVAL;
817 	return 0;
818 }
819 
820 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
821 							 struct vmcs12 *vmcs12)
822 {
823 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
824 	       !nested_cpu_has_ept(vmcs12)))
825 		return -EINVAL;
826 	return 0;
827 }
828 
829 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
830 						 struct vmcs12 *vmcs12)
831 {
832 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
833 		return 0;
834 
835 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
836 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
837 		return -EINVAL;
838 
839 	return 0;
840 }
841 
842 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
843 				       struct vmx_msr_entry *e)
844 {
845 	/* x2APIC MSR accesses are not allowed */
846 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
847 		return -EINVAL;
848 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
849 	    CC(e->index == MSR_IA32_UCODE_REV))
850 		return -EINVAL;
851 	if (CC(e->reserved != 0))
852 		return -EINVAL;
853 	return 0;
854 }
855 
856 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
857 				     struct vmx_msr_entry *e)
858 {
859 	if (CC(e->index == MSR_FS_BASE) ||
860 	    CC(e->index == MSR_GS_BASE) ||
861 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
862 	    nested_vmx_msr_check_common(vcpu, e))
863 		return -EINVAL;
864 	return 0;
865 }
866 
867 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
868 				      struct vmx_msr_entry *e)
869 {
870 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
871 	    nested_vmx_msr_check_common(vcpu, e))
872 		return -EINVAL;
873 	return 0;
874 }
875 
876 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
877 {
878 	struct vcpu_vmx *vmx = to_vmx(vcpu);
879 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
880 				       vmx->nested.msrs.misc_high);
881 
882 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
883 }
884 
885 /*
886  * Load guest's/host's msr at nested entry/exit.
887  * return 0 for success, entry index for failure.
888  *
889  * One of the failure modes for MSR load/store is when a list exceeds the
890  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
891  * as possible, process all valid entries before failing rather than precheck
892  * for a capacity violation.
893  */
894 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
895 {
896 	u32 i;
897 	struct vmx_msr_entry e;
898 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
899 
900 	for (i = 0; i < count; i++) {
901 		if (unlikely(i >= max_msr_list_size))
902 			goto fail;
903 
904 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
905 					&e, sizeof(e))) {
906 			pr_debug_ratelimited(
907 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
908 				__func__, i, gpa + i * sizeof(e));
909 			goto fail;
910 		}
911 		if (nested_vmx_load_msr_check(vcpu, &e)) {
912 			pr_debug_ratelimited(
913 				"%s check failed (%u, 0x%x, 0x%x)\n",
914 				__func__, i, e.index, e.reserved);
915 			goto fail;
916 		}
917 		if (kvm_set_msr(vcpu, e.index, e.value)) {
918 			pr_debug_ratelimited(
919 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
920 				__func__, i, e.index, e.value);
921 			goto fail;
922 		}
923 	}
924 	return 0;
925 fail:
926 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
927 	return i + 1;
928 }
929 
930 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
931 					    u32 msr_index,
932 					    u64 *data)
933 {
934 	struct vcpu_vmx *vmx = to_vmx(vcpu);
935 
936 	/*
937 	 * If the L0 hypervisor stored a more accurate value for the TSC that
938 	 * does not include the time taken for emulation of the L2->L1
939 	 * VM-exit in L0, use the more accurate value.
940 	 */
941 	if (msr_index == MSR_IA32_TSC) {
942 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
943 						    MSR_IA32_TSC);
944 
945 		if (i >= 0) {
946 			u64 val = vmx->msr_autostore.guest.val[i].value;
947 
948 			*data = kvm_read_l1_tsc(vcpu, val);
949 			return true;
950 		}
951 	}
952 
953 	if (kvm_get_msr(vcpu, msr_index, data)) {
954 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
955 			msr_index);
956 		return false;
957 	}
958 	return true;
959 }
960 
961 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
962 				     struct vmx_msr_entry *e)
963 {
964 	if (kvm_vcpu_read_guest(vcpu,
965 				gpa + i * sizeof(*e),
966 				e, 2 * sizeof(u32))) {
967 		pr_debug_ratelimited(
968 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
969 			__func__, i, gpa + i * sizeof(*e));
970 		return false;
971 	}
972 	if (nested_vmx_store_msr_check(vcpu, e)) {
973 		pr_debug_ratelimited(
974 			"%s check failed (%u, 0x%x, 0x%x)\n",
975 			__func__, i, e->index, e->reserved);
976 		return false;
977 	}
978 	return true;
979 }
980 
981 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
982 {
983 	u64 data;
984 	u32 i;
985 	struct vmx_msr_entry e;
986 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
987 
988 	for (i = 0; i < count; i++) {
989 		if (unlikely(i >= max_msr_list_size))
990 			return -EINVAL;
991 
992 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
993 			return -EINVAL;
994 
995 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
996 			return -EINVAL;
997 
998 		if (kvm_vcpu_write_guest(vcpu,
999 					 gpa + i * sizeof(e) +
1000 					     offsetof(struct vmx_msr_entry, value),
1001 					 &data, sizeof(data))) {
1002 			pr_debug_ratelimited(
1003 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1004 				__func__, i, e.index, data);
1005 			return -EINVAL;
1006 		}
1007 	}
1008 	return 0;
1009 }
1010 
1011 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1012 {
1013 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1014 	u32 count = vmcs12->vm_exit_msr_store_count;
1015 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1016 	struct vmx_msr_entry e;
1017 	u32 i;
1018 
1019 	for (i = 0; i < count; i++) {
1020 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1021 			return false;
1022 
1023 		if (e.index == msr_index)
1024 			return true;
1025 	}
1026 	return false;
1027 }
1028 
1029 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1030 					   u32 msr_index)
1031 {
1032 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1033 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1034 	bool in_vmcs12_store_list;
1035 	int msr_autostore_slot;
1036 	bool in_autostore_list;
1037 	int last;
1038 
1039 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1040 	in_autostore_list = msr_autostore_slot >= 0;
1041 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1042 
1043 	if (in_vmcs12_store_list && !in_autostore_list) {
1044 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1045 			/*
1046 			 * Emulated VMEntry does not fail here.  Instead a less
1047 			 * accurate value will be returned by
1048 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1049 			 * instead of reading the value from the vmcs02 VMExit
1050 			 * MSR-store area.
1051 			 */
1052 			pr_warn_ratelimited(
1053 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1054 				msr_index);
1055 			return;
1056 		}
1057 		last = autostore->nr++;
1058 		autostore->val[last].index = msr_index;
1059 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1060 		last = --autostore->nr;
1061 		autostore->val[msr_autostore_slot] = autostore->val[last];
1062 	}
1063 }
1064 
1065 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
1066 {
1067 	unsigned long invalid_mask;
1068 
1069 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1070 	return (val & invalid_mask) == 0;
1071 }
1072 
1073 /*
1074  * Returns true if the MMU needs to be sync'd on nested VM-Enter/VM-Exit.
1075  * tl;dr: the MMU needs a sync if L0 is using shadow paging and L1 didn't
1076  * enable VPID for L2 (implying it expects a TLB flush on VMX transitions).
1077  * Here's why.
1078  *
1079  * If EPT is enabled by L0 a sync is never needed:
1080  * - if it is disabled by L1, then L0 is not shadowing L1 or L2 PTEs, there
1081  *   cannot be unsync'd SPTEs for either L1 or L2.
1082  *
1083  * - if it is also enabled by L1, then L0 doesn't need to sync on VM-Enter
1084  *   VM-Enter as VM-Enter isn't required to invalidate guest-physical mappings
1085  *   (irrespective of VPID), i.e. L1 can't rely on the (virtual) CPU to flush
1086  *   stale guest-physical mappings for L2 from the TLB.  And as above, L0 isn't
1087  *   shadowing L1 PTEs so there are no unsync'd SPTEs to sync on VM-Exit.
1088  *
1089  * If EPT is disabled by L0:
1090  * - if VPID is enabled by L1 (for L2), the situation is similar to when L1
1091  *   enables EPT: L0 doesn't need to sync as VM-Enter and VM-Exit aren't
1092  *   required to invalidate linear mappings (EPT is disabled so there are
1093  *   no combined or guest-physical mappings), i.e. L1 can't rely on the
1094  *   (virtual) CPU to flush stale linear mappings for either L2 or itself (L1).
1095  *
1096  * - however if VPID is disabled by L1, then a sync is needed as L1 expects all
1097  *   linear mappings (EPT is disabled so there are no combined or guest-physical
1098  *   mappings) to be invalidated on both VM-Enter and VM-Exit.
1099  *
1100  * Note, this logic is subtly different than nested_has_guest_tlb_tag(), which
1101  * additionally checks that L2 has been assigned a VPID (when EPT is disabled).
1102  * Whether or not L2 has been assigned a VPID by L0 is irrelevant with respect
1103  * to L1's expectations, e.g. L0 needs to invalidate hardware TLB entries if L2
1104  * doesn't have a unique VPID to prevent reusing L1's entries (assuming L1 has
1105  * been assigned a VPID), but L0 doesn't need to do a MMU sync because L1
1106  * doesn't expect stale (virtual) TLB entries to be flushed, i.e. L1 doesn't
1107  * know that L0 will flush the TLB and so L1 will do INVVPID as needed to flush
1108  * stale TLB entries, at which point L0 will sync L2's MMU.
1109  */
1110 static bool nested_vmx_transition_mmu_sync(struct kvm_vcpu *vcpu)
1111 {
1112 	return !enable_ept && !nested_cpu_has_vpid(get_vmcs12(vcpu));
1113 }
1114 
1115 /*
1116  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1117  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1118  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1119  * @entry_failure_code.
1120  */
1121 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
1122 			       enum vm_entry_failure_code *entry_failure_code)
1123 {
1124 	if (CC(!nested_cr3_valid(vcpu, cr3))) {
1125 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1126 		return -EINVAL;
1127 	}
1128 
1129 	/*
1130 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1131 	 * must not be dereferenced.
1132 	 */
1133 	if (!nested_ept && is_pae_paging(vcpu) &&
1134 	    (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
1135 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1136 			*entry_failure_code = ENTRY_FAIL_PDPTE;
1137 			return -EINVAL;
1138 		}
1139 	}
1140 
1141 	/*
1142 	 * Unconditionally skip the TLB flush on fast CR3 switch, all TLB
1143 	 * flushes are handled by nested_vmx_transition_tlb_flush().  See
1144 	 * nested_vmx_transition_mmu_sync for details on skipping the MMU sync.
1145 	 */
1146 	if (!nested_ept)
1147 		kvm_mmu_new_pgd(vcpu, cr3, true,
1148 				!nested_vmx_transition_mmu_sync(vcpu));
1149 
1150 	vcpu->arch.cr3 = cr3;
1151 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1152 
1153 	kvm_init_mmu(vcpu, false);
1154 
1155 	return 0;
1156 }
1157 
1158 /*
1159  * Returns if KVM is able to config CPU to tag TLB entries
1160  * populated by L2 differently than TLB entries populated
1161  * by L1.
1162  *
1163  * If L0 uses EPT, L1 and L2 run with different EPTP because
1164  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1165  * are tagged with different EPTP.
1166  *
1167  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1168  * with different VPID (L1 entries are tagged with vmx->vpid
1169  * while L2 entries are tagged with vmx->nested.vpid02).
1170  */
1171 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1172 {
1173 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1174 
1175 	return enable_ept ||
1176 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1177 }
1178 
1179 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1180 					    struct vmcs12 *vmcs12,
1181 					    bool is_vmenter)
1182 {
1183 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1184 
1185 	/*
1186 	 * If VPID is disabled, linear and combined mappings are flushed on
1187 	 * VM-Enter/VM-Exit, and guest-physical mappings are valid only for
1188 	 * their associated EPTP.
1189 	 */
1190 	if (!enable_vpid)
1191 		return;
1192 
1193 	/*
1194 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1195 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit.
1196 	 *
1197 	 * If VPID is enabled and used by vmc12, but L2 does not have a unique
1198 	 * TLB tag (ASID), i.e. EPT is disabled and KVM was unable to allocate
1199 	 * a VPID for L2, flush the current context as the effective ASID is
1200 	 * common to both L1 and L2.
1201 	 *
1202 	 * Defer the flush so that it runs after vmcs02.EPTP has been set by
1203 	 * KVM_REQ_LOAD_MMU_PGD (if nested EPT is enabled) and to avoid
1204 	 * redundant flushes further down the nested pipeline.
1205 	 *
1206 	 * If a TLB flush isn't required due to any of the above, and vpid12 is
1207 	 * changing then the new "virtual" VPID (vpid12) will reuse the same
1208 	 * "real" VPID (vpid02), and so needs to be sync'd.  There is no direct
1209 	 * mapping between vpid02 and vpid12, vpid02 is per-vCPU and reused for
1210 	 * all nested vCPUs.
1211 	 */
1212 	if (!nested_cpu_has_vpid(vmcs12)) {
1213 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1214 	} else if (!nested_has_guest_tlb_tag(vcpu)) {
1215 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1216 	} else if (is_vmenter &&
1217 		   vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1218 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1219 		vpid_sync_context(nested_get_vpid02(vcpu));
1220 	}
1221 }
1222 
1223 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1224 {
1225 	superset &= mask;
1226 	subset &= mask;
1227 
1228 	return (superset | subset) == superset;
1229 }
1230 
1231 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1232 {
1233 	const u64 feature_and_reserved =
1234 		/* feature (except bit 48; see below) */
1235 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1236 		/* reserved */
1237 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1238 	u64 vmx_basic = vmx->nested.msrs.basic;
1239 
1240 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1241 		return -EINVAL;
1242 
1243 	/*
1244 	 * KVM does not emulate a version of VMX that constrains physical
1245 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1246 	 */
1247 	if (data & BIT_ULL(48))
1248 		return -EINVAL;
1249 
1250 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1251 	    vmx_basic_vmcs_revision_id(data))
1252 		return -EINVAL;
1253 
1254 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1255 		return -EINVAL;
1256 
1257 	vmx->nested.msrs.basic = data;
1258 	return 0;
1259 }
1260 
1261 static int
1262 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1263 {
1264 	u64 supported;
1265 	u32 *lowp, *highp;
1266 
1267 	switch (msr_index) {
1268 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1269 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1270 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1271 		break;
1272 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1273 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1274 		highp = &vmx->nested.msrs.procbased_ctls_high;
1275 		break;
1276 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1277 		lowp = &vmx->nested.msrs.exit_ctls_low;
1278 		highp = &vmx->nested.msrs.exit_ctls_high;
1279 		break;
1280 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1281 		lowp = &vmx->nested.msrs.entry_ctls_low;
1282 		highp = &vmx->nested.msrs.entry_ctls_high;
1283 		break;
1284 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1285 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1286 		highp = &vmx->nested.msrs.secondary_ctls_high;
1287 		break;
1288 	default:
1289 		BUG();
1290 	}
1291 
1292 	supported = vmx_control_msr(*lowp, *highp);
1293 
1294 	/* Check must-be-1 bits are still 1. */
1295 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1296 		return -EINVAL;
1297 
1298 	/* Check must-be-0 bits are still 0. */
1299 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1300 		return -EINVAL;
1301 
1302 	*lowp = data;
1303 	*highp = data >> 32;
1304 	return 0;
1305 }
1306 
1307 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1308 {
1309 	const u64 feature_and_reserved_bits =
1310 		/* feature */
1311 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1312 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1313 		/* reserved */
1314 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1315 	u64 vmx_misc;
1316 
1317 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1318 				   vmx->nested.msrs.misc_high);
1319 
1320 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1321 		return -EINVAL;
1322 
1323 	if ((vmx->nested.msrs.pinbased_ctls_high &
1324 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1325 	    vmx_misc_preemption_timer_rate(data) !=
1326 	    vmx_misc_preemption_timer_rate(vmx_misc))
1327 		return -EINVAL;
1328 
1329 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1330 		return -EINVAL;
1331 
1332 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1333 		return -EINVAL;
1334 
1335 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1336 		return -EINVAL;
1337 
1338 	vmx->nested.msrs.misc_low = data;
1339 	vmx->nested.msrs.misc_high = data >> 32;
1340 
1341 	return 0;
1342 }
1343 
1344 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1345 {
1346 	u64 vmx_ept_vpid_cap;
1347 
1348 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1349 					   vmx->nested.msrs.vpid_caps);
1350 
1351 	/* Every bit is either reserved or a feature bit. */
1352 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1353 		return -EINVAL;
1354 
1355 	vmx->nested.msrs.ept_caps = data;
1356 	vmx->nested.msrs.vpid_caps = data >> 32;
1357 	return 0;
1358 }
1359 
1360 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1361 {
1362 	u64 *msr;
1363 
1364 	switch (msr_index) {
1365 	case MSR_IA32_VMX_CR0_FIXED0:
1366 		msr = &vmx->nested.msrs.cr0_fixed0;
1367 		break;
1368 	case MSR_IA32_VMX_CR4_FIXED0:
1369 		msr = &vmx->nested.msrs.cr4_fixed0;
1370 		break;
1371 	default:
1372 		BUG();
1373 	}
1374 
1375 	/*
1376 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1377 	 * must be 1 in the restored value.
1378 	 */
1379 	if (!is_bitwise_subset(data, *msr, -1ULL))
1380 		return -EINVAL;
1381 
1382 	*msr = data;
1383 	return 0;
1384 }
1385 
1386 /*
1387  * Called when userspace is restoring VMX MSRs.
1388  *
1389  * Returns 0 on success, non-0 otherwise.
1390  */
1391 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1392 {
1393 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1394 
1395 	/*
1396 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1397 	 * is in VMX operation.
1398 	 */
1399 	if (vmx->nested.vmxon)
1400 		return -EBUSY;
1401 
1402 	switch (msr_index) {
1403 	case MSR_IA32_VMX_BASIC:
1404 		return vmx_restore_vmx_basic(vmx, data);
1405 	case MSR_IA32_VMX_PINBASED_CTLS:
1406 	case MSR_IA32_VMX_PROCBASED_CTLS:
1407 	case MSR_IA32_VMX_EXIT_CTLS:
1408 	case MSR_IA32_VMX_ENTRY_CTLS:
1409 		/*
1410 		 * The "non-true" VMX capability MSRs are generated from the
1411 		 * "true" MSRs, so we do not support restoring them directly.
1412 		 *
1413 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1414 		 * should restore the "true" MSRs with the must-be-1 bits
1415 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1416 		 * DEFAULT SETTINGS".
1417 		 */
1418 		return -EINVAL;
1419 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1420 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1421 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1422 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1423 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1424 		return vmx_restore_control_msr(vmx, msr_index, data);
1425 	case MSR_IA32_VMX_MISC:
1426 		return vmx_restore_vmx_misc(vmx, data);
1427 	case MSR_IA32_VMX_CR0_FIXED0:
1428 	case MSR_IA32_VMX_CR4_FIXED0:
1429 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1430 	case MSR_IA32_VMX_CR0_FIXED1:
1431 	case MSR_IA32_VMX_CR4_FIXED1:
1432 		/*
1433 		 * These MSRs are generated based on the vCPU's CPUID, so we
1434 		 * do not support restoring them directly.
1435 		 */
1436 		return -EINVAL;
1437 	case MSR_IA32_VMX_EPT_VPID_CAP:
1438 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1439 	case MSR_IA32_VMX_VMCS_ENUM:
1440 		vmx->nested.msrs.vmcs_enum = data;
1441 		return 0;
1442 	case MSR_IA32_VMX_VMFUNC:
1443 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1444 			return -EINVAL;
1445 		vmx->nested.msrs.vmfunc_controls = data;
1446 		return 0;
1447 	default:
1448 		/*
1449 		 * The rest of the VMX capability MSRs do not support restore.
1450 		 */
1451 		return -EINVAL;
1452 	}
1453 }
1454 
1455 /* Returns 0 on success, non-0 otherwise. */
1456 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1457 {
1458 	switch (msr_index) {
1459 	case MSR_IA32_VMX_BASIC:
1460 		*pdata = msrs->basic;
1461 		break;
1462 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1463 	case MSR_IA32_VMX_PINBASED_CTLS:
1464 		*pdata = vmx_control_msr(
1465 			msrs->pinbased_ctls_low,
1466 			msrs->pinbased_ctls_high);
1467 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1468 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1469 		break;
1470 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1471 	case MSR_IA32_VMX_PROCBASED_CTLS:
1472 		*pdata = vmx_control_msr(
1473 			msrs->procbased_ctls_low,
1474 			msrs->procbased_ctls_high);
1475 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1476 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1477 		break;
1478 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1479 	case MSR_IA32_VMX_EXIT_CTLS:
1480 		*pdata = vmx_control_msr(
1481 			msrs->exit_ctls_low,
1482 			msrs->exit_ctls_high);
1483 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1484 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1485 		break;
1486 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1487 	case MSR_IA32_VMX_ENTRY_CTLS:
1488 		*pdata = vmx_control_msr(
1489 			msrs->entry_ctls_low,
1490 			msrs->entry_ctls_high);
1491 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1492 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1493 		break;
1494 	case MSR_IA32_VMX_MISC:
1495 		*pdata = vmx_control_msr(
1496 			msrs->misc_low,
1497 			msrs->misc_high);
1498 		break;
1499 	case MSR_IA32_VMX_CR0_FIXED0:
1500 		*pdata = msrs->cr0_fixed0;
1501 		break;
1502 	case MSR_IA32_VMX_CR0_FIXED1:
1503 		*pdata = msrs->cr0_fixed1;
1504 		break;
1505 	case MSR_IA32_VMX_CR4_FIXED0:
1506 		*pdata = msrs->cr4_fixed0;
1507 		break;
1508 	case MSR_IA32_VMX_CR4_FIXED1:
1509 		*pdata = msrs->cr4_fixed1;
1510 		break;
1511 	case MSR_IA32_VMX_VMCS_ENUM:
1512 		*pdata = msrs->vmcs_enum;
1513 		break;
1514 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1515 		*pdata = vmx_control_msr(
1516 			msrs->secondary_ctls_low,
1517 			msrs->secondary_ctls_high);
1518 		break;
1519 	case MSR_IA32_VMX_EPT_VPID_CAP:
1520 		*pdata = msrs->ept_caps |
1521 			((u64)msrs->vpid_caps << 32);
1522 		break;
1523 	case MSR_IA32_VMX_VMFUNC:
1524 		*pdata = msrs->vmfunc_controls;
1525 		break;
1526 	default:
1527 		return 1;
1528 	}
1529 
1530 	return 0;
1531 }
1532 
1533 /*
1534  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1535  * been modified by the L1 guest.  Note, "writable" in this context means
1536  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1537  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1538  * VM-exit information fields (which are actually writable if the vCPU is
1539  * configured to support "VMWRITE to any supported field in the VMCS").
1540  */
1541 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1542 {
1543 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1544 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1545 	struct shadow_vmcs_field field;
1546 	unsigned long val;
1547 	int i;
1548 
1549 	if (WARN_ON(!shadow_vmcs))
1550 		return;
1551 
1552 	preempt_disable();
1553 
1554 	vmcs_load(shadow_vmcs);
1555 
1556 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1557 		field = shadow_read_write_fields[i];
1558 		val = __vmcs_readl(field.encoding);
1559 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1560 	}
1561 
1562 	vmcs_clear(shadow_vmcs);
1563 	vmcs_load(vmx->loaded_vmcs->vmcs);
1564 
1565 	preempt_enable();
1566 }
1567 
1568 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1569 {
1570 	const struct shadow_vmcs_field *fields[] = {
1571 		shadow_read_write_fields,
1572 		shadow_read_only_fields
1573 	};
1574 	const int max_fields[] = {
1575 		max_shadow_read_write_fields,
1576 		max_shadow_read_only_fields
1577 	};
1578 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1579 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1580 	struct shadow_vmcs_field field;
1581 	unsigned long val;
1582 	int i, q;
1583 
1584 	if (WARN_ON(!shadow_vmcs))
1585 		return;
1586 
1587 	vmcs_load(shadow_vmcs);
1588 
1589 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1590 		for (i = 0; i < max_fields[q]; i++) {
1591 			field = fields[q][i];
1592 			val = vmcs12_read_any(vmcs12, field.encoding,
1593 					      field.offset);
1594 			__vmcs_writel(field.encoding, val);
1595 		}
1596 	}
1597 
1598 	vmcs_clear(shadow_vmcs);
1599 	vmcs_load(vmx->loaded_vmcs->vmcs);
1600 }
1601 
1602 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1603 {
1604 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1605 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1606 
1607 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1608 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1609 	vmcs12->guest_rip = evmcs->guest_rip;
1610 
1611 	if (unlikely(!(evmcs->hv_clean_fields &
1612 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1613 		vmcs12->guest_rsp = evmcs->guest_rsp;
1614 		vmcs12->guest_rflags = evmcs->guest_rflags;
1615 		vmcs12->guest_interruptibility_info =
1616 			evmcs->guest_interruptibility_info;
1617 	}
1618 
1619 	if (unlikely(!(evmcs->hv_clean_fields &
1620 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1621 		vmcs12->cpu_based_vm_exec_control =
1622 			evmcs->cpu_based_vm_exec_control;
1623 	}
1624 
1625 	if (unlikely(!(evmcs->hv_clean_fields &
1626 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1627 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1628 	}
1629 
1630 	if (unlikely(!(evmcs->hv_clean_fields &
1631 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1632 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1633 	}
1634 
1635 	if (unlikely(!(evmcs->hv_clean_fields &
1636 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1637 		vmcs12->vm_entry_intr_info_field =
1638 			evmcs->vm_entry_intr_info_field;
1639 		vmcs12->vm_entry_exception_error_code =
1640 			evmcs->vm_entry_exception_error_code;
1641 		vmcs12->vm_entry_instruction_len =
1642 			evmcs->vm_entry_instruction_len;
1643 	}
1644 
1645 	if (unlikely(!(evmcs->hv_clean_fields &
1646 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1647 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1648 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1649 		vmcs12->host_cr0 = evmcs->host_cr0;
1650 		vmcs12->host_cr3 = evmcs->host_cr3;
1651 		vmcs12->host_cr4 = evmcs->host_cr4;
1652 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1653 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1654 		vmcs12->host_rip = evmcs->host_rip;
1655 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1656 		vmcs12->host_es_selector = evmcs->host_es_selector;
1657 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1658 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1659 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1660 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1661 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1662 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1663 	}
1664 
1665 	if (unlikely(!(evmcs->hv_clean_fields &
1666 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1667 		vmcs12->pin_based_vm_exec_control =
1668 			evmcs->pin_based_vm_exec_control;
1669 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1670 		vmcs12->secondary_vm_exec_control =
1671 			evmcs->secondary_vm_exec_control;
1672 	}
1673 
1674 	if (unlikely(!(evmcs->hv_clean_fields &
1675 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1676 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1677 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1678 	}
1679 
1680 	if (unlikely(!(evmcs->hv_clean_fields &
1681 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1682 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1683 	}
1684 
1685 	if (unlikely(!(evmcs->hv_clean_fields &
1686 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1687 		vmcs12->guest_es_base = evmcs->guest_es_base;
1688 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1689 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1690 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1691 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1692 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1693 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1694 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1695 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1696 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1697 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1698 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1699 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1700 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1701 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1702 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1703 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1704 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1705 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1706 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1707 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1708 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1709 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1710 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1711 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1712 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1713 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1714 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1715 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1716 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1717 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1718 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1719 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1720 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1721 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1722 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1723 	}
1724 
1725 	if (unlikely(!(evmcs->hv_clean_fields &
1726 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1727 		vmcs12->tsc_offset = evmcs->tsc_offset;
1728 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1729 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1730 	}
1731 
1732 	if (unlikely(!(evmcs->hv_clean_fields &
1733 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1734 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1735 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1736 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1737 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1738 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1739 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1740 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1741 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1742 	}
1743 
1744 	if (unlikely(!(evmcs->hv_clean_fields &
1745 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1746 		vmcs12->host_fs_base = evmcs->host_fs_base;
1747 		vmcs12->host_gs_base = evmcs->host_gs_base;
1748 		vmcs12->host_tr_base = evmcs->host_tr_base;
1749 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1750 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1751 		vmcs12->host_rsp = evmcs->host_rsp;
1752 	}
1753 
1754 	if (unlikely(!(evmcs->hv_clean_fields &
1755 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1756 		vmcs12->ept_pointer = evmcs->ept_pointer;
1757 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1758 	}
1759 
1760 	if (unlikely(!(evmcs->hv_clean_fields &
1761 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1762 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1763 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1764 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1765 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1766 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1767 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1768 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1769 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1770 		vmcs12->guest_pending_dbg_exceptions =
1771 			evmcs->guest_pending_dbg_exceptions;
1772 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1773 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1774 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1775 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1776 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1777 	}
1778 
1779 	/*
1780 	 * Not used?
1781 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1782 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1783 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1784 	 * vmcs12->page_fault_error_code_mask =
1785 	 *		evmcs->page_fault_error_code_mask;
1786 	 * vmcs12->page_fault_error_code_match =
1787 	 *		evmcs->page_fault_error_code_match;
1788 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1789 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1790 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1791 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1792 	 */
1793 
1794 	/*
1795 	 * Read only fields:
1796 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1797 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1798 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1799 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1800 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1801 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1802 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1803 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1804 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1805 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1806 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1807 	 *
1808 	 * Not present in struct vmcs12:
1809 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1810 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1811 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1812 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1813 	 */
1814 
1815 	return 0;
1816 }
1817 
1818 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1819 {
1820 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1821 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1822 
1823 	/*
1824 	 * Should not be changed by KVM:
1825 	 *
1826 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1827 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1828 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1829 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1830 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1831 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1832 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1833 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1834 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1835 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1836 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1837 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1838 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1839 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1840 	 * evmcs->host_rip = vmcs12->host_rip;
1841 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1842 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1843 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1844 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1845 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1846 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1847 	 * evmcs->host_rsp = vmcs12->host_rsp;
1848 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1849 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1850 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1851 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1852 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1853 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1854 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1855 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1856 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1857 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1858 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1859 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1860 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1861 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1862 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1863 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1864 	 * evmcs->page_fault_error_code_mask =
1865 	 *		vmcs12->page_fault_error_code_mask;
1866 	 * evmcs->page_fault_error_code_match =
1867 	 *		vmcs12->page_fault_error_code_match;
1868 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1869 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1870 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1871 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1872 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1873 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1874 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1875 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1876 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1877 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1878 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1879 	 *
1880 	 * Not present in struct vmcs12:
1881 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1882 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1883 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1884 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1885 	 */
1886 
1887 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1888 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1889 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1890 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1891 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1892 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1893 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1894 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1895 
1896 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1897 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1898 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1899 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1900 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1901 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1902 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1903 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1904 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1905 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1906 
1907 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1908 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1909 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1910 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1911 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1912 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1913 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1914 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1915 
1916 	evmcs->guest_es_base = vmcs12->guest_es_base;
1917 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1918 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1919 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1920 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1921 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1922 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1923 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1924 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1925 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1926 
1927 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1928 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1929 
1930 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1931 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1932 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1933 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1934 
1935 	evmcs->guest_pending_dbg_exceptions =
1936 		vmcs12->guest_pending_dbg_exceptions;
1937 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1938 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1939 
1940 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1941 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1942 
1943 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1944 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1945 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1946 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1947 
1948 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1949 
1950 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1951 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1952 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1953 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1954 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1955 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1956 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1957 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1958 
1959 	evmcs->exit_qualification = vmcs12->exit_qualification;
1960 
1961 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1962 	evmcs->guest_rsp = vmcs12->guest_rsp;
1963 	evmcs->guest_rflags = vmcs12->guest_rflags;
1964 
1965 	evmcs->guest_interruptibility_info =
1966 		vmcs12->guest_interruptibility_info;
1967 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1968 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1969 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1970 	evmcs->vm_entry_exception_error_code =
1971 		vmcs12->vm_entry_exception_error_code;
1972 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1973 
1974 	evmcs->guest_rip = vmcs12->guest_rip;
1975 
1976 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1977 
1978 	return 0;
1979 }
1980 
1981 /*
1982  * This is an equivalent of the nested hypervisor executing the vmptrld
1983  * instruction.
1984  */
1985 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1986 	struct kvm_vcpu *vcpu, bool from_launch)
1987 {
1988 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1989 	bool evmcs_gpa_changed = false;
1990 	u64 evmcs_gpa;
1991 
1992 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1993 		return EVMPTRLD_DISABLED;
1994 
1995 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa))
1996 		return EVMPTRLD_DISABLED;
1997 
1998 	if (unlikely(!vmx->nested.hv_evmcs ||
1999 		     evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
2000 		if (!vmx->nested.hv_evmcs)
2001 			vmx->nested.current_vmptr = -1ull;
2002 
2003 		nested_release_evmcs(vcpu);
2004 
2005 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2006 				 &vmx->nested.hv_evmcs_map))
2007 			return EVMPTRLD_ERROR;
2008 
2009 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2010 
2011 		/*
2012 		 * Currently, KVM only supports eVMCS version 1
2013 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2014 		 * value to first u32 field of eVMCS which should specify eVMCS
2015 		 * VersionNumber.
2016 		 *
2017 		 * Guest should be aware of supported eVMCS versions by host by
2018 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2019 		 * expected to set this CPUID leaf according to the value
2020 		 * returned in vmcs_version from nested_enable_evmcs().
2021 		 *
2022 		 * However, it turns out that Microsoft Hyper-V fails to comply
2023 		 * to their own invented interface: When Hyper-V use eVMCS, it
2024 		 * just sets first u32 field of eVMCS to revision_id specified
2025 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2026 		 * which is one of the supported versions specified in
2027 		 * CPUID.0x4000000A.EAX[0:15].
2028 		 *
2029 		 * To overcome Hyper-V bug, we accept here either a supported
2030 		 * eVMCS version or VMCS12 revision_id as valid values for first
2031 		 * u32 field of eVMCS.
2032 		 */
2033 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2034 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2035 			nested_release_evmcs(vcpu);
2036 			return EVMPTRLD_VMFAIL;
2037 		}
2038 
2039 		vmx->nested.dirty_vmcs12 = true;
2040 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2041 
2042 		evmcs_gpa_changed = true;
2043 		/*
2044 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2045 		 * reloaded from guest's memory (read only fields, fields not
2046 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2047 		 * are no leftovers.
2048 		 */
2049 		if (from_launch) {
2050 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2051 			memset(vmcs12, 0, sizeof(*vmcs12));
2052 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2053 		}
2054 
2055 	}
2056 
2057 	/*
2058 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2059 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2060 	 */
2061 	if (from_launch || evmcs_gpa_changed)
2062 		vmx->nested.hv_evmcs->hv_clean_fields &=
2063 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2064 
2065 	return EVMPTRLD_SUCCEEDED;
2066 }
2067 
2068 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2069 {
2070 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2071 
2072 	if (vmx->nested.hv_evmcs) {
2073 		copy_vmcs12_to_enlightened(vmx);
2074 		/* All fields are clean */
2075 		vmx->nested.hv_evmcs->hv_clean_fields |=
2076 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2077 	} else {
2078 		copy_vmcs12_to_shadow(vmx);
2079 	}
2080 
2081 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2082 }
2083 
2084 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2085 {
2086 	struct vcpu_vmx *vmx =
2087 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2088 
2089 	vmx->nested.preemption_timer_expired = true;
2090 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2091 	kvm_vcpu_kick(&vmx->vcpu);
2092 
2093 	return HRTIMER_NORESTART;
2094 }
2095 
2096 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2097 {
2098 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2099 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2100 
2101 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2102 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2103 
2104 	if (!vmx->nested.has_preemption_timer_deadline) {
2105 		vmx->nested.preemption_timer_deadline =
2106 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2107 		vmx->nested.has_preemption_timer_deadline = true;
2108 	}
2109 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2110 }
2111 
2112 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2113 					u64 preemption_timeout)
2114 {
2115 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2116 
2117 	/*
2118 	 * A timer value of zero is architecturally guaranteed to cause
2119 	 * a VMExit prior to executing any instructions in the guest.
2120 	 */
2121 	if (preemption_timeout == 0) {
2122 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2123 		return;
2124 	}
2125 
2126 	if (vcpu->arch.virtual_tsc_khz == 0)
2127 		return;
2128 
2129 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2130 	preemption_timeout *= 1000000;
2131 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2132 	hrtimer_start(&vmx->nested.preemption_timer,
2133 		      ktime_add_ns(ktime_get(), preemption_timeout),
2134 		      HRTIMER_MODE_ABS_PINNED);
2135 }
2136 
2137 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2138 {
2139 	if (vmx->nested.nested_run_pending &&
2140 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2141 		return vmcs12->guest_ia32_efer;
2142 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2143 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2144 	else
2145 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2146 }
2147 
2148 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2149 {
2150 	/*
2151 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2152 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2153 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2154 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2155 	 */
2156 	if (vmx->nested.vmcs02_initialized)
2157 		return;
2158 	vmx->nested.vmcs02_initialized = true;
2159 
2160 	/*
2161 	 * We don't care what the EPTP value is we just need to guarantee
2162 	 * it's valid so we don't get a false positive when doing early
2163 	 * consistency checks.
2164 	 */
2165 	if (enable_ept && nested_early_check)
2166 		vmcs_write64(EPT_POINTER,
2167 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2168 
2169 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2170 	if (cpu_has_vmx_vmfunc())
2171 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2172 
2173 	if (cpu_has_vmx_posted_intr())
2174 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2175 
2176 	if (cpu_has_vmx_msr_bitmap())
2177 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2178 
2179 	/*
2180 	 * The PML address never changes, so it is constant in vmcs02.
2181 	 * Conceptually we want to copy the PML index from vmcs01 here,
2182 	 * and then back to vmcs01 on nested vmexit.  But since we flush
2183 	 * the log and reset GUEST_PML_INDEX on each vmexit, the PML
2184 	 * index is also effectively constant in vmcs02.
2185 	 */
2186 	if (enable_pml) {
2187 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
2188 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2189 	}
2190 
2191 	if (cpu_has_vmx_encls_vmexit())
2192 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2193 
2194 	/*
2195 	 * Set the MSR load/store lists to match L0's settings.  Only the
2196 	 * addresses are constant (for vmcs02), the counts can change based
2197 	 * on L2's behavior, e.g. switching to/from long mode.
2198 	 */
2199 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2200 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2201 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2202 
2203 	vmx_set_constant_host_state(vmx);
2204 }
2205 
2206 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2207 				      struct vmcs12 *vmcs12)
2208 {
2209 	prepare_vmcs02_constant_state(vmx);
2210 
2211 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2212 
2213 	if (enable_vpid) {
2214 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2215 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2216 		else
2217 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2218 	}
2219 }
2220 
2221 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2222 {
2223 	u32 exec_control, vmcs12_exec_ctrl;
2224 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2225 
2226 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
2227 		prepare_vmcs02_early_rare(vmx, vmcs12);
2228 
2229 	/*
2230 	 * PIN CONTROLS
2231 	 */
2232 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2233 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2234 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2235 
2236 	/* Posted interrupts setting is only taken from vmcs12.  */
2237 	if (nested_cpu_has_posted_intr(vmcs12)) {
2238 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2239 		vmx->nested.pi_pending = false;
2240 	} else {
2241 		exec_control &= ~PIN_BASED_POSTED_INTR;
2242 	}
2243 	pin_controls_set(vmx, exec_control);
2244 
2245 	/*
2246 	 * EXEC CONTROLS
2247 	 */
2248 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2249 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2250 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2251 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2252 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2253 
2254 	vmx->nested.l1_tpr_threshold = -1;
2255 	if (exec_control & CPU_BASED_TPR_SHADOW)
2256 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2257 #ifdef CONFIG_X86_64
2258 	else
2259 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2260 				CPU_BASED_CR8_STORE_EXITING;
2261 #endif
2262 
2263 	/*
2264 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2265 	 * for I/O port accesses.
2266 	 */
2267 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2268 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2269 
2270 	/*
2271 	 * This bit will be computed in nested_get_vmcs12_pages, because
2272 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2273 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2274 	 * only set/clear this bit.
2275 	 */
2276 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2277 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2278 
2279 	exec_controls_set(vmx, exec_control);
2280 
2281 	/*
2282 	 * SECONDARY EXEC CONTROLS
2283 	 */
2284 	if (cpu_has_secondary_exec_ctrls()) {
2285 		exec_control = vmx->secondary_exec_control;
2286 
2287 		/* Take the following fields only from vmcs12 */
2288 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2289 				  SECONDARY_EXEC_ENABLE_INVPCID |
2290 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2291 				  SECONDARY_EXEC_XSAVES |
2292 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2293 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2294 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2295 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2296 		if (nested_cpu_has(vmcs12,
2297 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2298 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2299 				~SECONDARY_EXEC_ENABLE_PML;
2300 			exec_control |= vmcs12_exec_ctrl;
2301 		}
2302 
2303 		/* VMCS shadowing for L2 is emulated for now */
2304 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2305 
2306 		/*
2307 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2308 		 * will not have to rewrite the controls just for this bit.
2309 		 */
2310 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2311 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2312 			exec_control |= SECONDARY_EXEC_DESC;
2313 
2314 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2315 			vmcs_write16(GUEST_INTR_STATUS,
2316 				vmcs12->guest_intr_status);
2317 
2318 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2319 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2320 
2321 		secondary_exec_controls_set(vmx, exec_control);
2322 	}
2323 
2324 	/*
2325 	 * ENTRY CONTROLS
2326 	 *
2327 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2328 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2329 	 * on the related bits (if supported by the CPU) in the hope that
2330 	 * we can avoid VMWrites during vmx_set_efer().
2331 	 */
2332 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2333 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2334 	if (cpu_has_load_ia32_efer()) {
2335 		if (guest_efer & EFER_LMA)
2336 			exec_control |= VM_ENTRY_IA32E_MODE;
2337 		if (guest_efer != host_efer)
2338 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2339 	}
2340 	vm_entry_controls_set(vmx, exec_control);
2341 
2342 	/*
2343 	 * EXIT CONTROLS
2344 	 *
2345 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2346 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2347 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2348 	 */
2349 	exec_control = vmx_vmexit_ctrl();
2350 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2351 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2352 	vm_exit_controls_set(vmx, exec_control);
2353 
2354 	/*
2355 	 * Interrupt/Exception Fields
2356 	 */
2357 	if (vmx->nested.nested_run_pending) {
2358 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2359 			     vmcs12->vm_entry_intr_info_field);
2360 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2361 			     vmcs12->vm_entry_exception_error_code);
2362 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2363 			     vmcs12->vm_entry_instruction_len);
2364 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2365 			     vmcs12->guest_interruptibility_info);
2366 		vmx->loaded_vmcs->nmi_known_unmasked =
2367 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2368 	} else {
2369 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2370 	}
2371 }
2372 
2373 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2374 {
2375 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2376 
2377 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2378 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2379 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2380 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2381 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2382 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2383 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2384 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2385 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2386 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2387 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2388 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2389 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2390 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2391 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2392 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2393 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2394 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2395 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2396 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2397 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2398 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2399 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2400 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2401 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2402 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2403 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2404 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2405 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2406 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2407 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2408 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2409 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2410 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2411 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2412 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2413 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2414 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2415 
2416 		vmx->segment_cache.bitmask = 0;
2417 	}
2418 
2419 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2420 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2421 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2422 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2423 			    vmcs12->guest_pending_dbg_exceptions);
2424 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2425 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2426 
2427 		/*
2428 		 * L1 may access the L2's PDPTR, so save them to construct
2429 		 * vmcs12
2430 		 */
2431 		if (enable_ept) {
2432 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2433 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2434 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2435 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2436 		}
2437 
2438 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2439 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2440 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2441 	}
2442 
2443 	if (nested_cpu_has_xsaves(vmcs12))
2444 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2445 
2446 	/*
2447 	 * Whether page-faults are trapped is determined by a combination of
2448 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2449 	 * doesn't care about page faults then we should set all of these to
2450 	 * L1's desires. However, if L0 does care about (some) page faults, it
2451 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2452 	 * simply ask to exit on each and every L2 page fault. This is done by
2453 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2454 	 * Note that below we don't need special code to set EB.PF beyond the
2455 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2456 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2457 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2458 	 */
2459 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2460 		/*
2461 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2462 		 * go ahead and use it?
2463 		 */
2464 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2465 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2466 	} else {
2467 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2468 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2469 	}
2470 
2471 	if (cpu_has_vmx_apicv()) {
2472 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2473 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2474 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2475 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2476 	}
2477 
2478 	/*
2479 	 * Make sure the msr_autostore list is up to date before we set the
2480 	 * count in the vmcs02.
2481 	 */
2482 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2483 
2484 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2485 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2486 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2487 
2488 	set_cr4_guest_host_mask(vmx);
2489 }
2490 
2491 /*
2492  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2493  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2494  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2495  * guest in a way that will both be appropriate to L1's requests, and our
2496  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2497  * function also has additional necessary side-effects, like setting various
2498  * vcpu->arch fields.
2499  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2500  * is assigned to entry_failure_code on failure.
2501  */
2502 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2503 			  enum vm_entry_failure_code *entry_failure_code)
2504 {
2505 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2506 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2507 	bool load_guest_pdptrs_vmcs12 = false;
2508 
2509 	if (vmx->nested.dirty_vmcs12 || hv_evmcs) {
2510 		prepare_vmcs02_rare(vmx, vmcs12);
2511 		vmx->nested.dirty_vmcs12 = false;
2512 
2513 		load_guest_pdptrs_vmcs12 = !hv_evmcs ||
2514 			!(hv_evmcs->hv_clean_fields &
2515 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2516 	}
2517 
2518 	if (vmx->nested.nested_run_pending &&
2519 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2520 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2521 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2522 	} else {
2523 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2524 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2525 	}
2526 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2527 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2528 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2529 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2530 
2531 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2532 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2533 	 * trap. Note that CR0.TS also needs updating - we do this later.
2534 	 */
2535 	update_exception_bitmap(vcpu);
2536 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2537 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2538 
2539 	if (vmx->nested.nested_run_pending &&
2540 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2541 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2542 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2543 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2544 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2545 	}
2546 
2547 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2548 
2549 	if (kvm_has_tsc_control)
2550 		decache_tsc_multiplier(vmx);
2551 
2552 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2553 
2554 	if (nested_cpu_has_ept(vmcs12))
2555 		nested_ept_init_mmu_context(vcpu);
2556 
2557 	/*
2558 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2559 	 * bits which we consider mandatory enabled.
2560 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2561 	 * the specifications by L1; It's not enough to take
2562 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2563 	 * have more bits than L1 expected.
2564 	 */
2565 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2566 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2567 
2568 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2569 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2570 
2571 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2572 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2573 	vmx_set_efer(vcpu, vcpu->arch.efer);
2574 
2575 	/*
2576 	 * Guest state is invalid and unrestricted guest is disabled,
2577 	 * which means L1 attempted VMEntry to L2 with invalid state.
2578 	 * Fail the VMEntry.
2579 	 */
2580 	if (CC(!vmx_guest_state_valid(vcpu))) {
2581 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2582 		return -EINVAL;
2583 	}
2584 
2585 	/* Shadow page tables on either EPT or shadow page tables. */
2586 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2587 				entry_failure_code))
2588 		return -EINVAL;
2589 
2590 	/*
2591 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2592 	 * on nested VM-Exit, which can occur without actually running L2 and
2593 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2594 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2595 	 * transition to HLT instead of running L2.
2596 	 */
2597 	if (enable_ept)
2598 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2599 
2600 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2601 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2602 	    is_pae_paging(vcpu)) {
2603 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2604 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2605 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2606 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2607 	}
2608 
2609 	if (!enable_ept)
2610 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2611 
2612 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2613 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2614 				     vmcs12->guest_ia32_perf_global_ctrl)))
2615 		return -EINVAL;
2616 
2617 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2618 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2619 	return 0;
2620 }
2621 
2622 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2623 {
2624 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2625 	       nested_cpu_has_virtual_nmis(vmcs12)))
2626 		return -EINVAL;
2627 
2628 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2629 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2630 		return -EINVAL;
2631 
2632 	return 0;
2633 }
2634 
2635 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2636 {
2637 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2638 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2639 
2640 	/* Check for memory type validity */
2641 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2642 	case VMX_EPTP_MT_UC:
2643 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2644 			return false;
2645 		break;
2646 	case VMX_EPTP_MT_WB:
2647 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2648 			return false;
2649 		break;
2650 	default:
2651 		return false;
2652 	}
2653 
2654 	/* Page-walk levels validity. */
2655 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2656 	case VMX_EPTP_PWL_5:
2657 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2658 			return false;
2659 		break;
2660 	case VMX_EPTP_PWL_4:
2661 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2662 			return false;
2663 		break;
2664 	default:
2665 		return false;
2666 	}
2667 
2668 	/* Reserved bits should not be set */
2669 	if (CC(new_eptp >> maxphyaddr || ((new_eptp >> 7) & 0x1f)))
2670 		return false;
2671 
2672 	/* AD, if set, should be supported */
2673 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2674 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2675 			return false;
2676 	}
2677 
2678 	return true;
2679 }
2680 
2681 /*
2682  * Checks related to VM-Execution Control Fields
2683  */
2684 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2685                                               struct vmcs12 *vmcs12)
2686 {
2687 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2688 
2689 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2690 				   vmx->nested.msrs.pinbased_ctls_low,
2691 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2692 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2693 				   vmx->nested.msrs.procbased_ctls_low,
2694 				   vmx->nested.msrs.procbased_ctls_high)))
2695 		return -EINVAL;
2696 
2697 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2698 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2699 				   vmx->nested.msrs.secondary_ctls_low,
2700 				   vmx->nested.msrs.secondary_ctls_high)))
2701 		return -EINVAL;
2702 
2703 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2704 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2705 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2706 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2707 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2708 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2709 	    nested_vmx_check_nmi_controls(vmcs12) ||
2710 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2711 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2712 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2713 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2714 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2715 		return -EINVAL;
2716 
2717 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2718 	    nested_cpu_has_save_preemption_timer(vmcs12))
2719 		return -EINVAL;
2720 
2721 	if (nested_cpu_has_ept(vmcs12) &&
2722 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2723 		return -EINVAL;
2724 
2725 	if (nested_cpu_has_vmfunc(vmcs12)) {
2726 		if (CC(vmcs12->vm_function_control &
2727 		       ~vmx->nested.msrs.vmfunc_controls))
2728 			return -EINVAL;
2729 
2730 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2731 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2732 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2733 				return -EINVAL;
2734 		}
2735 	}
2736 
2737 	return 0;
2738 }
2739 
2740 /*
2741  * Checks related to VM-Exit Control Fields
2742  */
2743 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2744                                          struct vmcs12 *vmcs12)
2745 {
2746 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2747 
2748 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2749 				    vmx->nested.msrs.exit_ctls_low,
2750 				    vmx->nested.msrs.exit_ctls_high)) ||
2751 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2752 		return -EINVAL;
2753 
2754 	return 0;
2755 }
2756 
2757 /*
2758  * Checks related to VM-Entry Control Fields
2759  */
2760 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2761 					  struct vmcs12 *vmcs12)
2762 {
2763 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2764 
2765 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2766 				    vmx->nested.msrs.entry_ctls_low,
2767 				    vmx->nested.msrs.entry_ctls_high)))
2768 		return -EINVAL;
2769 
2770 	/*
2771 	 * From the Intel SDM, volume 3:
2772 	 * Fields relevant to VM-entry event injection must be set properly.
2773 	 * These fields are the VM-entry interruption-information field, the
2774 	 * VM-entry exception error code, and the VM-entry instruction length.
2775 	 */
2776 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2777 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2778 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2779 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2780 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2781 		bool should_have_error_code;
2782 		bool urg = nested_cpu_has2(vmcs12,
2783 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2784 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2785 
2786 		/* VM-entry interruption-info field: interruption type */
2787 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2788 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2789 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2790 			return -EINVAL;
2791 
2792 		/* VM-entry interruption-info field: vector */
2793 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2794 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2795 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2796 			return -EINVAL;
2797 
2798 		/* VM-entry interruption-info field: deliver error code */
2799 		should_have_error_code =
2800 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2801 			x86_exception_has_error_code(vector);
2802 		if (CC(has_error_code != should_have_error_code))
2803 			return -EINVAL;
2804 
2805 		/* VM-entry exception error code */
2806 		if (CC(has_error_code &&
2807 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2808 			return -EINVAL;
2809 
2810 		/* VM-entry interruption-info field: reserved bits */
2811 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2812 			return -EINVAL;
2813 
2814 		/* VM-entry instruction length */
2815 		switch (intr_type) {
2816 		case INTR_TYPE_SOFT_EXCEPTION:
2817 		case INTR_TYPE_SOFT_INTR:
2818 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2819 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2820 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2821 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2822 				return -EINVAL;
2823 		}
2824 	}
2825 
2826 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2827 		return -EINVAL;
2828 
2829 	return 0;
2830 }
2831 
2832 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2833 				     struct vmcs12 *vmcs12)
2834 {
2835 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2836 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2837 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2838 		return -EINVAL;
2839 
2840 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2841 		return nested_evmcs_check_controls(vmcs12);
2842 
2843 	return 0;
2844 }
2845 
2846 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2847 				       struct vmcs12 *vmcs12)
2848 {
2849 	bool ia32e;
2850 
2851 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2852 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2853 	    CC(!nested_cr3_valid(vcpu, vmcs12->host_cr3)))
2854 		return -EINVAL;
2855 
2856 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2857 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2858 		return -EINVAL;
2859 
2860 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2861 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2862 		return -EINVAL;
2863 
2864 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2865 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2866 					   vmcs12->host_ia32_perf_global_ctrl)))
2867 		return -EINVAL;
2868 
2869 #ifdef CONFIG_X86_64
2870 	ia32e = !!(vcpu->arch.efer & EFER_LMA);
2871 #else
2872 	ia32e = false;
2873 #endif
2874 
2875 	if (ia32e) {
2876 		if (CC(!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)) ||
2877 		    CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2878 			return -EINVAL;
2879 	} else {
2880 		if (CC(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) ||
2881 		    CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2882 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2883 		    CC((vmcs12->host_rip) >> 32))
2884 			return -EINVAL;
2885 	}
2886 
2887 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2888 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2889 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2890 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2891 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2892 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2893 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2894 	    CC(vmcs12->host_cs_selector == 0) ||
2895 	    CC(vmcs12->host_tr_selector == 0) ||
2896 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2897 		return -EINVAL;
2898 
2899 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2900 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2901 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2902 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2903 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2904 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2905 		return -EINVAL;
2906 
2907 	/*
2908 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2909 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2910 	 * the values of the LMA and LME bits in the field must each be that of
2911 	 * the host address-space size VM-exit control.
2912 	 */
2913 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2914 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2915 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2916 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2917 			return -EINVAL;
2918 	}
2919 
2920 	return 0;
2921 }
2922 
2923 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2924 					  struct vmcs12 *vmcs12)
2925 {
2926 	int r = 0;
2927 	struct vmcs12 *shadow;
2928 	struct kvm_host_map map;
2929 
2930 	if (vmcs12->vmcs_link_pointer == -1ull)
2931 		return 0;
2932 
2933 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2934 		return -EINVAL;
2935 
2936 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2937 		return -EINVAL;
2938 
2939 	shadow = map.hva;
2940 
2941 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2942 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2943 		r = -EINVAL;
2944 
2945 	kvm_vcpu_unmap(vcpu, &map, false);
2946 	return r;
2947 }
2948 
2949 /*
2950  * Checks related to Guest Non-register State
2951  */
2952 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2953 {
2954 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2955 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2956 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2957 		return -EINVAL;
2958 
2959 	return 0;
2960 }
2961 
2962 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2963 					struct vmcs12 *vmcs12,
2964 					enum vm_entry_failure_code *entry_failure_code)
2965 {
2966 	bool ia32e;
2967 
2968 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
2969 
2970 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2971 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2972 		return -EINVAL;
2973 
2974 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2975 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2976 		return -EINVAL;
2977 
2978 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2979 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2980 		return -EINVAL;
2981 
2982 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2983 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
2984 		return -EINVAL;
2985 	}
2986 
2987 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2988 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2989 					   vmcs12->guest_ia32_perf_global_ctrl)))
2990 		return -EINVAL;
2991 
2992 	/*
2993 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2994 	 * are performed on the field for the IA32_EFER MSR:
2995 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2996 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2997 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2998 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2999 	 *   CR0.PG) is 1.
3000 	 */
3001 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3002 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3003 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
3004 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3005 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3006 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3007 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3008 			return -EINVAL;
3009 	}
3010 
3011 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3012 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3013 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3014 		return -EINVAL;
3015 
3016 	if (nested_check_guest_non_reg_state(vmcs12))
3017 		return -EINVAL;
3018 
3019 	return 0;
3020 }
3021 
3022 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3023 {
3024 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3025 	unsigned long cr3, cr4;
3026 	bool vm_fail;
3027 
3028 	if (!nested_early_check)
3029 		return 0;
3030 
3031 	if (vmx->msr_autoload.host.nr)
3032 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3033 	if (vmx->msr_autoload.guest.nr)
3034 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3035 
3036 	preempt_disable();
3037 
3038 	vmx_prepare_switch_to_guest(vcpu);
3039 
3040 	/*
3041 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3042 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3043 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3044 	 * there is no need to preserve other bits or save/restore the field.
3045 	 */
3046 	vmcs_writel(GUEST_RFLAGS, 0);
3047 
3048 	cr3 = __get_current_cr3_fast();
3049 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3050 		vmcs_writel(HOST_CR3, cr3);
3051 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3052 	}
3053 
3054 	cr4 = cr4_read_shadow();
3055 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3056 		vmcs_writel(HOST_CR4, cr4);
3057 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3058 	}
3059 
3060 	asm(
3061 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
3062 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
3063 		"je 1f \n\t"
3064 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
3065 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
3066 		"1: \n\t"
3067 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
3068 
3069 		/* Check if vmlaunch or vmresume is needed */
3070 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
3071 
3072 		/*
3073 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
3074 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
3075 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
3076 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
3077 		 */
3078 		"call vmx_vmenter\n\t"
3079 
3080 		CC_SET(be)
3081 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
3082 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
3083 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
3084 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
3085 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
3086 		[wordsize]"i"(sizeof(ulong))
3087 	      : "memory"
3088 	);
3089 
3090 	if (vmx->msr_autoload.host.nr)
3091 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3092 	if (vmx->msr_autoload.guest.nr)
3093 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3094 
3095 	if (vm_fail) {
3096 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3097 
3098 		preempt_enable();
3099 
3100 		trace_kvm_nested_vmenter_failed(
3101 			"early hardware check VM-instruction error: ", error);
3102 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3103 		return 1;
3104 	}
3105 
3106 	/*
3107 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3108 	 */
3109 	if (hw_breakpoint_active())
3110 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3111 	local_irq_enable();
3112 	preempt_enable();
3113 
3114 	/*
3115 	 * A non-failing VMEntry means we somehow entered guest mode with
3116 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3117 	 * is no telling what memory has been modified or what state has
3118 	 * been exposed to unknown code.  Hitting this all but guarantees
3119 	 * a (very critical) hardware issue.
3120 	 */
3121 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3122 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3123 
3124 	return 0;
3125 }
3126 
3127 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3128 {
3129 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3130 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3131 	struct kvm_host_map *map;
3132 	struct page *page;
3133 	u64 hpa;
3134 
3135 	/*
3136 	 * hv_evmcs may end up being not mapped after migration (when
3137 	 * L2 was running), map it here to make sure vmcs12 changes are
3138 	 * properly reflected.
3139 	 */
3140 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs) {
3141 		enum nested_evmptrld_status evmptrld_status =
3142 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3143 
3144 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3145 		    evmptrld_status == EVMPTRLD_ERROR) {
3146 			pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3147 					     __func__);
3148 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3149 			vcpu->run->internal.suberror =
3150 				KVM_INTERNAL_ERROR_EMULATION;
3151 			vcpu->run->internal.ndata = 0;
3152 			return false;
3153 		}
3154 	}
3155 
3156 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3157 		/*
3158 		 * Translate L1 physical address to host physical
3159 		 * address for vmcs02. Keep the page pinned, so this
3160 		 * physical address remains valid. We keep a reference
3161 		 * to it so we can release it later.
3162 		 */
3163 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3164 			kvm_release_page_clean(vmx->nested.apic_access_page);
3165 			vmx->nested.apic_access_page = NULL;
3166 		}
3167 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3168 		if (!is_error_page(page)) {
3169 			vmx->nested.apic_access_page = page;
3170 			hpa = page_to_phys(vmx->nested.apic_access_page);
3171 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3172 		} else {
3173 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3174 					     __func__);
3175 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3176 			vcpu->run->internal.suberror =
3177 				KVM_INTERNAL_ERROR_EMULATION;
3178 			vcpu->run->internal.ndata = 0;
3179 			return false;
3180 		}
3181 	}
3182 
3183 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3184 		map = &vmx->nested.virtual_apic_map;
3185 
3186 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3187 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3188 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3189 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3190 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3191 			/*
3192 			 * The processor will never use the TPR shadow, simply
3193 			 * clear the bit from the execution control.  Such a
3194 			 * configuration is useless, but it happens in tests.
3195 			 * For any other configuration, failing the vm entry is
3196 			 * _not_ what the processor does but it's basically the
3197 			 * only possibility we have.
3198 			 */
3199 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3200 		} else {
3201 			/*
3202 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3203 			 * force VM-Entry to fail.
3204 			 */
3205 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
3206 		}
3207 	}
3208 
3209 	if (nested_cpu_has_posted_intr(vmcs12)) {
3210 		map = &vmx->nested.pi_desc_map;
3211 
3212 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3213 			vmx->nested.pi_desc =
3214 				(struct pi_desc *)(((void *)map->hva) +
3215 				offset_in_page(vmcs12->posted_intr_desc_addr));
3216 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3217 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3218 		}
3219 	}
3220 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3221 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3222 	else
3223 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3224 	return true;
3225 }
3226 
3227 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3228 {
3229 	struct vmcs12 *vmcs12;
3230 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3231 	gpa_t dst;
3232 
3233 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3234 		return 0;
3235 
3236 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3237 		return 1;
3238 
3239 	/*
3240 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3241 	 * set is already checked as part of A/D emulation.
3242 	 */
3243 	vmcs12 = get_vmcs12(vcpu);
3244 	if (!nested_cpu_has_pml(vmcs12))
3245 		return 0;
3246 
3247 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3248 		vmx->nested.pml_full = true;
3249 		return 1;
3250 	}
3251 
3252 	gpa &= ~0xFFFull;
3253 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3254 
3255 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3256 				 offset_in_page(dst), sizeof(gpa)))
3257 		return 0;
3258 
3259 	vmcs12->guest_pml_index--;
3260 
3261 	return 0;
3262 }
3263 
3264 /*
3265  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3266  * for running VMX instructions (except VMXON, whose prerequisites are
3267  * slightly different). It also specifies what exception to inject otherwise.
3268  * Note that many of these exceptions have priority over VM exits, so they
3269  * don't have to be checked again here.
3270  */
3271 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3272 {
3273 	if (!to_vmx(vcpu)->nested.vmxon) {
3274 		kvm_queue_exception(vcpu, UD_VECTOR);
3275 		return 0;
3276 	}
3277 
3278 	if (vmx_get_cpl(vcpu)) {
3279 		kvm_inject_gp(vcpu, 0);
3280 		return 0;
3281 	}
3282 
3283 	return 1;
3284 }
3285 
3286 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3287 {
3288 	u8 rvi = vmx_get_rvi();
3289 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3290 
3291 	return ((rvi & 0xf0) > (vppr & 0xf0));
3292 }
3293 
3294 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3295 				   struct vmcs12 *vmcs12);
3296 
3297 /*
3298  * If from_vmentry is false, this is being called from state restore (either RSM
3299  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3300  *
3301  * Returns:
3302  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3303  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3304  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3305  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3306  */
3307 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3308 							bool from_vmentry)
3309 {
3310 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3311 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3312 	enum vm_entry_failure_code entry_failure_code;
3313 	bool evaluate_pending_interrupts;
3314 	u32 exit_reason, failed_index;
3315 
3316 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3317 		kvm_vcpu_flush_tlb_current(vcpu);
3318 
3319 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3320 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3321 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3322 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3323 
3324 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3325 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3326 	if (kvm_mpx_supported() &&
3327 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3328 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3329 
3330 	/*
3331 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3332 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3333 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3334 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3335 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3336 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3337 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3338 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3339 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3340 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3341 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3342 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3343 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3344 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3345 	 */
3346 	if (!enable_ept && !nested_early_check)
3347 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3348 
3349 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3350 
3351 	prepare_vmcs02_early(vmx, vmcs12);
3352 
3353 	if (from_vmentry) {
3354 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3355 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3356 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3357 		}
3358 
3359 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3360 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3361 			return NVMX_VMENTRY_VMFAIL;
3362 		}
3363 
3364 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3365 						 &entry_failure_code)) {
3366 			exit_reason = EXIT_REASON_INVALID_STATE;
3367 			vmcs12->exit_qualification = entry_failure_code;
3368 			goto vmentry_fail_vmexit;
3369 		}
3370 	}
3371 
3372 	enter_guest_mode(vcpu);
3373 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3374 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3375 
3376 	if (prepare_vmcs02(vcpu, vmcs12, &entry_failure_code)) {
3377 		exit_reason = EXIT_REASON_INVALID_STATE;
3378 		vmcs12->exit_qualification = entry_failure_code;
3379 		goto vmentry_fail_vmexit_guest_mode;
3380 	}
3381 
3382 	if (from_vmentry) {
3383 		failed_index = nested_vmx_load_msr(vcpu,
3384 						   vmcs12->vm_entry_msr_load_addr,
3385 						   vmcs12->vm_entry_msr_load_count);
3386 		if (failed_index) {
3387 			exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3388 			vmcs12->exit_qualification = failed_index;
3389 			goto vmentry_fail_vmexit_guest_mode;
3390 		}
3391 	} else {
3392 		/*
3393 		 * The MMU is not initialized to point at the right entities yet and
3394 		 * "get pages" would need to read data from the guest (i.e. we will
3395 		 * need to perform gpa to hpa translation). Request a call
3396 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3397 		 * have already been set at vmentry time and should not be reset.
3398 		 */
3399 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3400 	}
3401 
3402 	/*
3403 	 * If L1 had a pending IRQ/NMI until it executed
3404 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3405 	 * disallowed (e.g. interrupts disabled), L0 needs to
3406 	 * evaluate if this pending event should cause an exit from L2
3407 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3408 	 * intercept EXTERNAL_INTERRUPT).
3409 	 *
3410 	 * Usually this would be handled by the processor noticing an
3411 	 * IRQ/NMI window request, or checking RVI during evaluation of
3412 	 * pending virtual interrupts.  However, this setting was done
3413 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3414 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3415 	 */
3416 	if (unlikely(evaluate_pending_interrupts))
3417 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3418 
3419 	/*
3420 	 * Do not start the preemption timer hrtimer until after we know
3421 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3422 	 * the timer.
3423 	 */
3424 	vmx->nested.preemption_timer_expired = false;
3425 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3426 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3427 		vmx_start_preemption_timer(vcpu, timer_value);
3428 	}
3429 
3430 	/*
3431 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3432 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3433 	 * returned as far as L1 is concerned. It will only return (and set
3434 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3435 	 */
3436 	return NVMX_VMENTRY_SUCCESS;
3437 
3438 	/*
3439 	 * A failed consistency check that leads to a VMExit during L1's
3440 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3441 	 * 26.7 "VM-entry failures during or after loading guest state".
3442 	 */
3443 vmentry_fail_vmexit_guest_mode:
3444 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3445 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3446 	leave_guest_mode(vcpu);
3447 
3448 vmentry_fail_vmexit:
3449 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3450 
3451 	if (!from_vmentry)
3452 		return NVMX_VMENTRY_VMEXIT;
3453 
3454 	load_vmcs12_host_state(vcpu, vmcs12);
3455 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3456 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3457 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3458 	return NVMX_VMENTRY_VMEXIT;
3459 }
3460 
3461 /*
3462  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3463  * for running an L2 nested guest.
3464  */
3465 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3466 {
3467 	struct vmcs12 *vmcs12;
3468 	enum nvmx_vmentry_status status;
3469 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3470 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3471 	enum nested_evmptrld_status evmptrld_status;
3472 
3473 	if (!nested_vmx_check_permission(vcpu))
3474 		return 1;
3475 
3476 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3477 	if (evmptrld_status == EVMPTRLD_ERROR) {
3478 		kvm_queue_exception(vcpu, UD_VECTOR);
3479 		return 1;
3480 	} else if (CC(evmptrld_status == EVMPTRLD_VMFAIL)) {
3481 		return nested_vmx_failInvalid(vcpu);
3482 	}
3483 
3484 	if (CC(!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull))
3485 		return nested_vmx_failInvalid(vcpu);
3486 
3487 	vmcs12 = get_vmcs12(vcpu);
3488 
3489 	/*
3490 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3491 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3492 	 * rather than RFLAGS.ZF, and no error number is stored to the
3493 	 * VM-instruction error field.
3494 	 */
3495 	if (CC(vmcs12->hdr.shadow_vmcs))
3496 		return nested_vmx_failInvalid(vcpu);
3497 
3498 	if (vmx->nested.hv_evmcs) {
3499 		copy_enlightened_to_vmcs12(vmx);
3500 		/* Enlightened VMCS doesn't have launch state */
3501 		vmcs12->launch_state = !launch;
3502 	} else if (enable_shadow_vmcs) {
3503 		copy_shadow_to_vmcs12(vmx);
3504 	}
3505 
3506 	/*
3507 	 * The nested entry process starts with enforcing various prerequisites
3508 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3509 	 * they fail: As the SDM explains, some conditions should cause the
3510 	 * instruction to fail, while others will cause the instruction to seem
3511 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3512 	 * To speed up the normal (success) code path, we should avoid checking
3513 	 * for misconfigurations which will anyway be caught by the processor
3514 	 * when using the merged vmcs02.
3515 	 */
3516 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3517 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3518 
3519 	if (CC(vmcs12->launch_state == launch))
3520 		return nested_vmx_fail(vcpu,
3521 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3522 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3523 
3524 	if (nested_vmx_check_controls(vcpu, vmcs12))
3525 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3526 
3527 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3528 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3529 
3530 	/*
3531 	 * We're finally done with prerequisite checking, and can start with
3532 	 * the nested entry.
3533 	 */
3534 	vmx->nested.nested_run_pending = 1;
3535 	vmx->nested.has_preemption_timer_deadline = false;
3536 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3537 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3538 		goto vmentry_failed;
3539 
3540 	/* Emulate processing of posted interrupts on VM-Enter. */
3541 	if (nested_cpu_has_posted_intr(vmcs12) &&
3542 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3543 		vmx->nested.pi_pending = true;
3544 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3545 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3546 	}
3547 
3548 	/* Hide L1D cache contents from the nested guest.  */
3549 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3550 
3551 	/*
3552 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3553 	 * also be used as part of restoring nVMX state for
3554 	 * snapshot restore (migration).
3555 	 *
3556 	 * In this flow, it is assumed that vmcs12 cache was
3557 	 * trasferred as part of captured nVMX state and should
3558 	 * therefore not be read from guest memory (which may not
3559 	 * exist on destination host yet).
3560 	 */
3561 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3562 
3563 	switch (vmcs12->guest_activity_state) {
3564 	case GUEST_ACTIVITY_HLT:
3565 		/*
3566 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3567 		 * awakened by event injection or by an NMI-window VM-exit or
3568 		 * by an interrupt-window VM-exit, halt the vcpu.
3569 		 */
3570 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3571 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3572 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3573 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3574 			vmx->nested.nested_run_pending = 0;
3575 			return kvm_vcpu_halt(vcpu);
3576 		}
3577 		break;
3578 	case GUEST_ACTIVITY_WAIT_SIPI:
3579 		vmx->nested.nested_run_pending = 0;
3580 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3581 		break;
3582 	default:
3583 		break;
3584 	}
3585 
3586 	return 1;
3587 
3588 vmentry_failed:
3589 	vmx->nested.nested_run_pending = 0;
3590 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3591 		return 0;
3592 	if (status == NVMX_VMENTRY_VMEXIT)
3593 		return 1;
3594 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3595 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3596 }
3597 
3598 /*
3599  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3600  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3601  * This function returns the new value we should put in vmcs12.guest_cr0.
3602  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3603  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3604  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3605  *     didn't trap the bit, because if L1 did, so would L0).
3606  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3607  *     been modified by L2, and L1 knows it. So just leave the old value of
3608  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3609  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3610  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3611  *     changed these bits, and therefore they need to be updated, but L0
3612  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3613  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3614  */
3615 static inline unsigned long
3616 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3617 {
3618 	return
3619 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3620 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3621 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3622 			vcpu->arch.cr0_guest_owned_bits));
3623 }
3624 
3625 static inline unsigned long
3626 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3627 {
3628 	return
3629 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3630 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3631 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3632 			vcpu->arch.cr4_guest_owned_bits));
3633 }
3634 
3635 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3636 				      struct vmcs12 *vmcs12)
3637 {
3638 	u32 idt_vectoring;
3639 	unsigned int nr;
3640 
3641 	if (vcpu->arch.exception.injected) {
3642 		nr = vcpu->arch.exception.nr;
3643 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3644 
3645 		if (kvm_exception_is_soft(nr)) {
3646 			vmcs12->vm_exit_instruction_len =
3647 				vcpu->arch.event_exit_inst_len;
3648 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3649 		} else
3650 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3651 
3652 		if (vcpu->arch.exception.has_error_code) {
3653 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3654 			vmcs12->idt_vectoring_error_code =
3655 				vcpu->arch.exception.error_code;
3656 		}
3657 
3658 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3659 	} else if (vcpu->arch.nmi_injected) {
3660 		vmcs12->idt_vectoring_info_field =
3661 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3662 	} else if (vcpu->arch.interrupt.injected) {
3663 		nr = vcpu->arch.interrupt.nr;
3664 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3665 
3666 		if (vcpu->arch.interrupt.soft) {
3667 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3668 			vmcs12->vm_entry_instruction_len =
3669 				vcpu->arch.event_exit_inst_len;
3670 		} else
3671 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3672 
3673 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3674 	}
3675 }
3676 
3677 
3678 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3679 {
3680 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3681 	gfn_t gfn;
3682 
3683 	/*
3684 	 * Don't need to mark the APIC access page dirty; it is never
3685 	 * written to by the CPU during APIC virtualization.
3686 	 */
3687 
3688 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3689 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3690 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3691 	}
3692 
3693 	if (nested_cpu_has_posted_intr(vmcs12)) {
3694 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3695 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3696 	}
3697 }
3698 
3699 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3700 {
3701 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3702 	int max_irr;
3703 	void *vapic_page;
3704 	u16 status;
3705 
3706 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3707 		return;
3708 
3709 	vmx->nested.pi_pending = false;
3710 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3711 		return;
3712 
3713 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3714 	if (max_irr != 256) {
3715 		vapic_page = vmx->nested.virtual_apic_map.hva;
3716 		if (!vapic_page)
3717 			return;
3718 
3719 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3720 			vapic_page, &max_irr);
3721 		status = vmcs_read16(GUEST_INTR_STATUS);
3722 		if ((u8)max_irr > ((u8)status & 0xff)) {
3723 			status &= ~0xff;
3724 			status |= (u8)max_irr;
3725 			vmcs_write16(GUEST_INTR_STATUS, status);
3726 		}
3727 	}
3728 
3729 	nested_mark_vmcs12_pages_dirty(vcpu);
3730 }
3731 
3732 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3733 					       unsigned long exit_qual)
3734 {
3735 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3736 	unsigned int nr = vcpu->arch.exception.nr;
3737 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3738 
3739 	if (vcpu->arch.exception.has_error_code) {
3740 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3741 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3742 	}
3743 
3744 	if (kvm_exception_is_soft(nr))
3745 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3746 	else
3747 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3748 
3749 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3750 	    vmx_get_nmi_mask(vcpu))
3751 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3752 
3753 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3754 }
3755 
3756 /*
3757  * Returns true if a debug trap is pending delivery.
3758  *
3759  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3760  * exception may be inferred from the presence of an exception payload.
3761  */
3762 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3763 {
3764 	return vcpu->arch.exception.pending &&
3765 			vcpu->arch.exception.nr == DB_VECTOR &&
3766 			vcpu->arch.exception.payload;
3767 }
3768 
3769 /*
3770  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3771  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3772  * represents these debug traps with a payload that is said to be compatible
3773  * with the 'pending debug exceptions' field, write the payload to the VMCS
3774  * field if a VM-exit is delivered before the debug trap.
3775  */
3776 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3777 {
3778 	if (vmx_pending_dbg_trap(vcpu))
3779 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3780 			    vcpu->arch.exception.payload);
3781 }
3782 
3783 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3784 {
3785 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3786 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3787 }
3788 
3789 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3790 {
3791 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3792 	unsigned long exit_qual;
3793 	bool block_nested_events =
3794 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3795 	bool mtf_pending = vmx->nested.mtf_pending;
3796 	struct kvm_lapic *apic = vcpu->arch.apic;
3797 
3798 	/*
3799 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3800 	 * this state is discarded.
3801 	 */
3802 	if (!block_nested_events)
3803 		vmx->nested.mtf_pending = false;
3804 
3805 	if (lapic_in_kernel(vcpu) &&
3806 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3807 		if (block_nested_events)
3808 			return -EBUSY;
3809 		nested_vmx_update_pending_dbg(vcpu);
3810 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3811 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3812 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3813 		return 0;
3814 	}
3815 
3816 	if (lapic_in_kernel(vcpu) &&
3817 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3818 		if (block_nested_events)
3819 			return -EBUSY;
3820 
3821 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3822 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3823 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3824 						apic->sipi_vector & 0xFFUL);
3825 		return 0;
3826 	}
3827 
3828 	/*
3829 	 * Process any exceptions that are not debug traps before MTF.
3830 	 */
3831 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
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 (mtf_pending) {
3841 		if (block_nested_events)
3842 			return -EBUSY;
3843 		nested_vmx_update_pending_dbg(vcpu);
3844 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3845 		return 0;
3846 	}
3847 
3848 	if (vcpu->arch.exception.pending) {
3849 		if (block_nested_events)
3850 			return -EBUSY;
3851 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3852 			goto no_vmexit;
3853 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3854 		return 0;
3855 	}
3856 
3857 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3858 		if (block_nested_events)
3859 			return -EBUSY;
3860 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3861 		return 0;
3862 	}
3863 
3864 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3865 		if (block_nested_events)
3866 			return -EBUSY;
3867 		goto no_vmexit;
3868 	}
3869 
3870 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3871 		if (block_nested_events)
3872 			return -EBUSY;
3873 		if (!nested_exit_on_nmi(vcpu))
3874 			goto no_vmexit;
3875 
3876 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3877 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3878 				  INTR_INFO_VALID_MASK, 0);
3879 		/*
3880 		 * The NMI-triggered VM exit counts as injection:
3881 		 * clear this one and block further NMIs.
3882 		 */
3883 		vcpu->arch.nmi_pending = 0;
3884 		vmx_set_nmi_mask(vcpu, true);
3885 		return 0;
3886 	}
3887 
3888 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3889 		if (block_nested_events)
3890 			return -EBUSY;
3891 		if (!nested_exit_on_intr(vcpu))
3892 			goto no_vmexit;
3893 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3894 		return 0;
3895 	}
3896 
3897 no_vmexit:
3898 	vmx_complete_nested_posted_interrupt(vcpu);
3899 	return 0;
3900 }
3901 
3902 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3903 {
3904 	ktime_t remaining =
3905 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3906 	u64 value;
3907 
3908 	if (ktime_to_ns(remaining) <= 0)
3909 		return 0;
3910 
3911 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3912 	do_div(value, 1000000);
3913 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3914 }
3915 
3916 static bool is_vmcs12_ext_field(unsigned long field)
3917 {
3918 	switch (field) {
3919 	case GUEST_ES_SELECTOR:
3920 	case GUEST_CS_SELECTOR:
3921 	case GUEST_SS_SELECTOR:
3922 	case GUEST_DS_SELECTOR:
3923 	case GUEST_FS_SELECTOR:
3924 	case GUEST_GS_SELECTOR:
3925 	case GUEST_LDTR_SELECTOR:
3926 	case GUEST_TR_SELECTOR:
3927 	case GUEST_ES_LIMIT:
3928 	case GUEST_CS_LIMIT:
3929 	case GUEST_SS_LIMIT:
3930 	case GUEST_DS_LIMIT:
3931 	case GUEST_FS_LIMIT:
3932 	case GUEST_GS_LIMIT:
3933 	case GUEST_LDTR_LIMIT:
3934 	case GUEST_TR_LIMIT:
3935 	case GUEST_GDTR_LIMIT:
3936 	case GUEST_IDTR_LIMIT:
3937 	case GUEST_ES_AR_BYTES:
3938 	case GUEST_DS_AR_BYTES:
3939 	case GUEST_FS_AR_BYTES:
3940 	case GUEST_GS_AR_BYTES:
3941 	case GUEST_LDTR_AR_BYTES:
3942 	case GUEST_TR_AR_BYTES:
3943 	case GUEST_ES_BASE:
3944 	case GUEST_CS_BASE:
3945 	case GUEST_SS_BASE:
3946 	case GUEST_DS_BASE:
3947 	case GUEST_FS_BASE:
3948 	case GUEST_GS_BASE:
3949 	case GUEST_LDTR_BASE:
3950 	case GUEST_TR_BASE:
3951 	case GUEST_GDTR_BASE:
3952 	case GUEST_IDTR_BASE:
3953 	case GUEST_PENDING_DBG_EXCEPTIONS:
3954 	case GUEST_BNDCFGS:
3955 		return true;
3956 	default:
3957 		break;
3958 	}
3959 
3960 	return false;
3961 }
3962 
3963 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3964 				       struct vmcs12 *vmcs12)
3965 {
3966 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3967 
3968 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3969 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3970 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3971 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3972 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3973 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3974 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3975 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3976 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3977 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3978 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3979 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3980 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3981 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3982 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3983 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3984 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3985 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3986 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3987 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3988 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3989 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3990 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3991 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3992 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3993 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3994 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3995 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3996 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3997 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3998 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3999 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4000 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4001 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4002 	vmcs12->guest_pending_dbg_exceptions =
4003 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4004 	if (kvm_mpx_supported())
4005 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
4006 
4007 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4008 }
4009 
4010 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4011 				       struct vmcs12 *vmcs12)
4012 {
4013 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4014 	int cpu;
4015 
4016 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4017 		return;
4018 
4019 
4020 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4021 
4022 	cpu = get_cpu();
4023 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4024 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4025 
4026 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4027 
4028 	vmx->loaded_vmcs = &vmx->vmcs01;
4029 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4030 	put_cpu();
4031 }
4032 
4033 /*
4034  * Update the guest state fields of vmcs12 to reflect changes that
4035  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4036  * VM-entry controls is also updated, since this is really a guest
4037  * state bit.)
4038  */
4039 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4040 {
4041 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4042 
4043 	if (vmx->nested.hv_evmcs)
4044 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4045 
4046 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = !vmx->nested.hv_evmcs;
4047 
4048 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4049 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4050 
4051 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4052 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4053 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4054 
4055 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4056 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4057 
4058 	vmcs12->guest_interruptibility_info =
4059 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4060 
4061 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4062 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4063 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4064 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4065 	else
4066 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4067 
4068 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4069 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4070 	    !vmx->nested.nested_run_pending)
4071 		vmcs12->vmx_preemption_timer_value =
4072 			vmx_get_preemption_timer_value(vcpu);
4073 
4074 	/*
4075 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4076 	 * own CR3 without exiting. If it has changed it, we must keep it.
4077 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4078 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4079 	 *
4080 	 * Additionally, restore L2's PDPTR to vmcs12.
4081 	 */
4082 	if (enable_ept) {
4083 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4084 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4085 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4086 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4087 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4088 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4089 		}
4090 	}
4091 
4092 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4093 
4094 	if (nested_cpu_has_vid(vmcs12))
4095 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4096 
4097 	vmcs12->vm_entry_controls =
4098 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4099 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4100 
4101 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4102 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4103 
4104 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4105 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4106 }
4107 
4108 /*
4109  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4110  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4111  * and this function updates it to reflect the changes to the guest state while
4112  * L2 was running (and perhaps made some exits which were handled directly by L0
4113  * without going back to L1), and to reflect the exit reason.
4114  * Note that we do not have to copy here all VMCS fields, just those that
4115  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4116  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4117  * which already writes to vmcs12 directly.
4118  */
4119 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4120 			   u32 vm_exit_reason, u32 exit_intr_info,
4121 			   unsigned long exit_qualification)
4122 {
4123 	/* update exit information fields: */
4124 	vmcs12->vm_exit_reason = vm_exit_reason;
4125 	vmcs12->exit_qualification = exit_qualification;
4126 	vmcs12->vm_exit_intr_info = exit_intr_info;
4127 
4128 	vmcs12->idt_vectoring_info_field = 0;
4129 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4130 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4131 
4132 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4133 		vmcs12->launch_state = 1;
4134 
4135 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4136 		 * instead of reading the real value. */
4137 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4138 
4139 		/*
4140 		 * Transfer the event that L0 or L1 may wanted to inject into
4141 		 * L2 to IDT_VECTORING_INFO_FIELD.
4142 		 */
4143 		vmcs12_save_pending_event(vcpu, vmcs12);
4144 
4145 		/*
4146 		 * According to spec, there's no need to store the guest's
4147 		 * MSRs if the exit is due to a VM-entry failure that occurs
4148 		 * during or after loading the guest state. Since this exit
4149 		 * does not fall in that category, we need to save the MSRs.
4150 		 */
4151 		if (nested_vmx_store_msr(vcpu,
4152 					 vmcs12->vm_exit_msr_store_addr,
4153 					 vmcs12->vm_exit_msr_store_count))
4154 			nested_vmx_abort(vcpu,
4155 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4156 	}
4157 
4158 	/*
4159 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4160 	 * preserved above and would only end up incorrectly in L1.
4161 	 */
4162 	vcpu->arch.nmi_injected = false;
4163 	kvm_clear_exception_queue(vcpu);
4164 	kvm_clear_interrupt_queue(vcpu);
4165 }
4166 
4167 /*
4168  * A part of what we need to when the nested L2 guest exits and we want to
4169  * run its L1 parent, is to reset L1's guest state to the host state specified
4170  * in vmcs12.
4171  * This function is to be called not only on normal nested exit, but also on
4172  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4173  * Failures During or After Loading Guest State").
4174  * This function should be called when the active VMCS is L1's (vmcs01).
4175  */
4176 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4177 				   struct vmcs12 *vmcs12)
4178 {
4179 	enum vm_entry_failure_code ignored;
4180 	struct kvm_segment seg;
4181 
4182 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4183 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4184 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4185 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4186 	else
4187 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4188 	vmx_set_efer(vcpu, vcpu->arch.efer);
4189 
4190 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4191 	kvm_rip_write(vcpu, vmcs12->host_rip);
4192 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4193 	vmx_set_interrupt_shadow(vcpu, 0);
4194 
4195 	/*
4196 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4197 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4198 	 *
4199 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4200 	 * (KVM doesn't change it);
4201 	 */
4202 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4203 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4204 
4205 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4206 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4207 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4208 
4209 	nested_ept_uninit_mmu_context(vcpu);
4210 
4211 	/*
4212 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4213 	 * couldn't have changed.
4214 	 */
4215 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &ignored))
4216 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4217 
4218 	if (!enable_ept)
4219 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
4220 
4221 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4222 
4223 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4224 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4225 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4226 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4227 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4228 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4229 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4230 
4231 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4232 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4233 		vmcs_write64(GUEST_BNDCFGS, 0);
4234 
4235 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4236 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4237 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4238 	}
4239 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4240 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4241 					 vmcs12->host_ia32_perf_global_ctrl));
4242 
4243 	/* Set L1 segment info according to Intel SDM
4244 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4245 	seg = (struct kvm_segment) {
4246 		.base = 0,
4247 		.limit = 0xFFFFFFFF,
4248 		.selector = vmcs12->host_cs_selector,
4249 		.type = 11,
4250 		.present = 1,
4251 		.s = 1,
4252 		.g = 1
4253 	};
4254 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4255 		seg.l = 1;
4256 	else
4257 		seg.db = 1;
4258 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4259 	seg = (struct kvm_segment) {
4260 		.base = 0,
4261 		.limit = 0xFFFFFFFF,
4262 		.type = 3,
4263 		.present = 1,
4264 		.s = 1,
4265 		.db = 1,
4266 		.g = 1
4267 	};
4268 	seg.selector = vmcs12->host_ds_selector;
4269 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4270 	seg.selector = vmcs12->host_es_selector;
4271 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4272 	seg.selector = vmcs12->host_ss_selector;
4273 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4274 	seg.selector = vmcs12->host_fs_selector;
4275 	seg.base = vmcs12->host_fs_base;
4276 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4277 	seg.selector = vmcs12->host_gs_selector;
4278 	seg.base = vmcs12->host_gs_base;
4279 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4280 	seg = (struct kvm_segment) {
4281 		.base = vmcs12->host_tr_base,
4282 		.limit = 0x67,
4283 		.selector = vmcs12->host_tr_selector,
4284 		.type = 11,
4285 		.present = 1
4286 	};
4287 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4288 
4289 	kvm_set_dr(vcpu, 7, 0x400);
4290 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4291 
4292 	if (cpu_has_vmx_msr_bitmap())
4293 		vmx_update_msr_bitmap(vcpu);
4294 
4295 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4296 				vmcs12->vm_exit_msr_load_count))
4297 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4298 }
4299 
4300 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4301 {
4302 	struct vmx_uret_msr *efer_msr;
4303 	unsigned int i;
4304 
4305 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4306 		return vmcs_read64(GUEST_IA32_EFER);
4307 
4308 	if (cpu_has_load_ia32_efer())
4309 		return host_efer;
4310 
4311 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4312 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4313 			return vmx->msr_autoload.guest.val[i].value;
4314 	}
4315 
4316 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4317 	if (efer_msr)
4318 		return efer_msr->data;
4319 
4320 	return host_efer;
4321 }
4322 
4323 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4324 {
4325 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4326 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4327 	struct vmx_msr_entry g, h;
4328 	gpa_t gpa;
4329 	u32 i, j;
4330 
4331 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4332 
4333 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4334 		/*
4335 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4336 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4337 		 * and vcpu->arch.dr7 is not squirreled away before the
4338 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4339 		 */
4340 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4341 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4342 		else
4343 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4344 	}
4345 
4346 	/*
4347 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4348 	 * handle a variety of side effects to KVM's software model.
4349 	 */
4350 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4351 
4352 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4353 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4354 
4355 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4356 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4357 
4358 	nested_ept_uninit_mmu_context(vcpu);
4359 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4360 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4361 
4362 	/*
4363 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4364 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4365 	 * VMFail, like everything else we just need to ensure our
4366 	 * software model is up-to-date.
4367 	 */
4368 	if (enable_ept && is_pae_paging(vcpu))
4369 		ept_save_pdptrs(vcpu);
4370 
4371 	kvm_mmu_reset_context(vcpu);
4372 
4373 	if (cpu_has_vmx_msr_bitmap())
4374 		vmx_update_msr_bitmap(vcpu);
4375 
4376 	/*
4377 	 * This nasty bit of open coding is a compromise between blindly
4378 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4379 	 * of VMFail), leaving the nested VM's MSRs in the software model
4380 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4381 	 * expensive since the lists are unbound by hardware).  For each
4382 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4383 	 * list, reload it from the exit load list if it exists and differs
4384 	 * from the guest value.  The intent is to stuff host state as
4385 	 * silently as possible, not to fully process the exit load list.
4386 	 */
4387 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4388 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4389 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4390 			pr_debug_ratelimited(
4391 				"%s read MSR index failed (%u, 0x%08llx)\n",
4392 				__func__, i, gpa);
4393 			goto vmabort;
4394 		}
4395 
4396 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4397 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4398 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4399 				pr_debug_ratelimited(
4400 					"%s read MSR failed (%u, 0x%08llx)\n",
4401 					__func__, j, gpa);
4402 				goto vmabort;
4403 			}
4404 			if (h.index != g.index)
4405 				continue;
4406 			if (h.value == g.value)
4407 				break;
4408 
4409 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4410 				pr_debug_ratelimited(
4411 					"%s check failed (%u, 0x%x, 0x%x)\n",
4412 					__func__, j, h.index, h.reserved);
4413 				goto vmabort;
4414 			}
4415 
4416 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4417 				pr_debug_ratelimited(
4418 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4419 					__func__, j, h.index, h.value);
4420 				goto vmabort;
4421 			}
4422 		}
4423 	}
4424 
4425 	return;
4426 
4427 vmabort:
4428 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4429 }
4430 
4431 /*
4432  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4433  * and modify vmcs12 to make it see what it would expect to see there if
4434  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4435  */
4436 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4437 		       u32 exit_intr_info, unsigned long exit_qualification)
4438 {
4439 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4440 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4441 
4442 	/* trying to cancel vmlaunch/vmresume is a bug */
4443 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4444 
4445 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
4446 
4447 	/* Service the TLB flush request for L2 before switching to L1. */
4448 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
4449 		kvm_vcpu_flush_tlb_current(vcpu);
4450 
4451 	/*
4452 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4453 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4454 	 * up-to-date before switching to L1.
4455 	 */
4456 	if (enable_ept && is_pae_paging(vcpu))
4457 		vmx_ept_load_pdptrs(vcpu);
4458 
4459 	leave_guest_mode(vcpu);
4460 
4461 	if (nested_cpu_has_preemption_timer(vmcs12))
4462 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4463 
4464 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
4465 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
4466 
4467 	if (likely(!vmx->fail)) {
4468 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4469 
4470 		if (vm_exit_reason != -1)
4471 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4472 				       exit_intr_info, exit_qualification);
4473 
4474 		/*
4475 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4476 		 * also be used to capture vmcs12 cache as part of
4477 		 * capturing nVMX state for snapshot (migration).
4478 		 *
4479 		 * Otherwise, this flush will dirty guest memory at a
4480 		 * point it is already assumed by user-space to be
4481 		 * immutable.
4482 		 */
4483 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4484 	} else {
4485 		/*
4486 		 * The only expected VM-instruction error is "VM entry with
4487 		 * invalid control field(s)." Anything else indicates a
4488 		 * problem with L0.  And we should never get here with a
4489 		 * VMFail of any type if early consistency checks are enabled.
4490 		 */
4491 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4492 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4493 		WARN_ON_ONCE(nested_early_check);
4494 	}
4495 
4496 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4497 
4498 	/* Update any VMCS fields that might have changed while L2 ran */
4499 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4500 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4501 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4502 	if (vmx->nested.l1_tpr_threshold != -1)
4503 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4504 
4505 	if (kvm_has_tsc_control)
4506 		decache_tsc_multiplier(vmx);
4507 
4508 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4509 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4510 		vmx_set_virtual_apic_mode(vcpu);
4511 	}
4512 
4513 	/* Unpin physical memory we referred to in vmcs02 */
4514 	if (vmx->nested.apic_access_page) {
4515 		kvm_release_page_clean(vmx->nested.apic_access_page);
4516 		vmx->nested.apic_access_page = NULL;
4517 	}
4518 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4519 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4520 	vmx->nested.pi_desc = NULL;
4521 
4522 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4523 		vmx->nested.reload_vmcs01_apic_access_page = false;
4524 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4525 	}
4526 
4527 	if ((vm_exit_reason != -1) &&
4528 	    (enable_shadow_vmcs || vmx->nested.hv_evmcs))
4529 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4530 
4531 	/* in case we halted in L2 */
4532 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4533 
4534 	if (likely(!vmx->fail)) {
4535 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4536 		    nested_exit_intr_ack_set(vcpu)) {
4537 			int irq = kvm_cpu_get_interrupt(vcpu);
4538 			WARN_ON(irq < 0);
4539 			vmcs12->vm_exit_intr_info = irq |
4540 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4541 		}
4542 
4543 		if (vm_exit_reason != -1)
4544 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4545 						       vmcs12->exit_qualification,
4546 						       vmcs12->idt_vectoring_info_field,
4547 						       vmcs12->vm_exit_intr_info,
4548 						       vmcs12->vm_exit_intr_error_code,
4549 						       KVM_ISA_VMX);
4550 
4551 		load_vmcs12_host_state(vcpu, vmcs12);
4552 
4553 		return;
4554 	}
4555 
4556 	/*
4557 	 * After an early L2 VM-entry failure, we're now back
4558 	 * in L1 which thinks it just finished a VMLAUNCH or
4559 	 * VMRESUME instruction, so we need to set the failure
4560 	 * flag and the VM-instruction error field of the VMCS
4561 	 * accordingly, and skip the emulated instruction.
4562 	 */
4563 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4564 
4565 	/*
4566 	 * Restore L1's host state to KVM's software model.  We're here
4567 	 * because a consistency check was caught by hardware, which
4568 	 * means some amount of guest state has been propagated to KVM's
4569 	 * model and needs to be unwound to the host's state.
4570 	 */
4571 	nested_vmx_restore_host_state(vcpu);
4572 
4573 	vmx->fail = 0;
4574 }
4575 
4576 /*
4577  * Decode the memory-address operand of a vmx instruction, as recorded on an
4578  * exit caused by such an instruction (run by a guest hypervisor).
4579  * On success, returns 0. When the operand is invalid, returns 1 and throws
4580  * #UD, #GP, or #SS.
4581  */
4582 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4583 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4584 {
4585 	gva_t off;
4586 	bool exn;
4587 	struct kvm_segment s;
4588 
4589 	/*
4590 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4591 	 * Execution", on an exit, vmx_instruction_info holds most of the
4592 	 * addressing components of the operand. Only the displacement part
4593 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4594 	 * For how an actual address is calculated from all these components,
4595 	 * refer to Vol. 1, "Operand Addressing".
4596 	 */
4597 	int  scaling = vmx_instruction_info & 3;
4598 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4599 	bool is_reg = vmx_instruction_info & (1u << 10);
4600 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4601 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4602 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4603 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4604 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4605 
4606 	if (is_reg) {
4607 		kvm_queue_exception(vcpu, UD_VECTOR);
4608 		return 1;
4609 	}
4610 
4611 	/* Addr = segment_base + offset */
4612 	/* offset = base + [index * scale] + displacement */
4613 	off = exit_qualification; /* holds the displacement */
4614 	if (addr_size == 1)
4615 		off = (gva_t)sign_extend64(off, 31);
4616 	else if (addr_size == 0)
4617 		off = (gva_t)sign_extend64(off, 15);
4618 	if (base_is_valid)
4619 		off += kvm_register_read(vcpu, base_reg);
4620 	if (index_is_valid)
4621 		off += kvm_register_read(vcpu, index_reg) << scaling;
4622 	vmx_get_segment(vcpu, &s, seg_reg);
4623 
4624 	/*
4625 	 * The effective address, i.e. @off, of a memory operand is truncated
4626 	 * based on the address size of the instruction.  Note that this is
4627 	 * the *effective address*, i.e. the address prior to accounting for
4628 	 * the segment's base.
4629 	 */
4630 	if (addr_size == 1) /* 32 bit */
4631 		off &= 0xffffffff;
4632 	else if (addr_size == 0) /* 16 bit */
4633 		off &= 0xffff;
4634 
4635 	/* Checks for #GP/#SS exceptions. */
4636 	exn = false;
4637 	if (is_long_mode(vcpu)) {
4638 		/*
4639 		 * The virtual/linear address is never truncated in 64-bit
4640 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4641 		 * address when using FS/GS with a non-zero base.
4642 		 */
4643 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4644 			*ret = s.base + off;
4645 		else
4646 			*ret = off;
4647 
4648 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4649 		 * non-canonical form. This is the only check on the memory
4650 		 * destination for long mode!
4651 		 */
4652 		exn = is_noncanonical_address(*ret, vcpu);
4653 	} else {
4654 		/*
4655 		 * When not in long mode, the virtual/linear address is
4656 		 * unconditionally truncated to 32 bits regardless of the
4657 		 * address size.
4658 		 */
4659 		*ret = (s.base + off) & 0xffffffff;
4660 
4661 		/* Protected mode: apply checks for segment validity in the
4662 		 * following order:
4663 		 * - segment type check (#GP(0) may be thrown)
4664 		 * - usability check (#GP(0)/#SS(0))
4665 		 * - limit check (#GP(0)/#SS(0))
4666 		 */
4667 		if (wr)
4668 			/* #GP(0) if the destination operand is located in a
4669 			 * read-only data segment or any code segment.
4670 			 */
4671 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4672 		else
4673 			/* #GP(0) if the source operand is located in an
4674 			 * execute-only code segment
4675 			 */
4676 			exn = ((s.type & 0xa) == 8);
4677 		if (exn) {
4678 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4679 			return 1;
4680 		}
4681 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4682 		 */
4683 		exn = (s.unusable != 0);
4684 
4685 		/*
4686 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4687 		 * outside the segment limit.  All CPUs that support VMX ignore
4688 		 * limit checks for flat segments, i.e. segments with base==0,
4689 		 * limit==0xffffffff and of type expand-up data or code.
4690 		 */
4691 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4692 		     ((s.type & 8) || !(s.type & 4))))
4693 			exn = exn || ((u64)off + len - 1 > s.limit);
4694 	}
4695 	if (exn) {
4696 		kvm_queue_exception_e(vcpu,
4697 				      seg_reg == VCPU_SREG_SS ?
4698 						SS_VECTOR : GP_VECTOR,
4699 				      0);
4700 		return 1;
4701 	}
4702 
4703 	return 0;
4704 }
4705 
4706 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4707 {
4708 	struct vcpu_vmx *vmx;
4709 
4710 	if (!nested_vmx_allowed(vcpu))
4711 		return;
4712 
4713 	vmx = to_vmx(vcpu);
4714 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4715 		vmx->nested.msrs.entry_ctls_high |=
4716 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4717 		vmx->nested.msrs.exit_ctls_high |=
4718 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4719 	} else {
4720 		vmx->nested.msrs.entry_ctls_high &=
4721 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4722 		vmx->nested.msrs.exit_ctls_high &=
4723 				~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4724 	}
4725 }
4726 
4727 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4728 				int *ret)
4729 {
4730 	gva_t gva;
4731 	struct x86_exception e;
4732 	int r;
4733 
4734 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4735 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4736 				sizeof(*vmpointer), &gva)) {
4737 		*ret = 1;
4738 		return -EINVAL;
4739 	}
4740 
4741 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4742 	if (r != X86EMUL_CONTINUE) {
4743 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4744 		return -EINVAL;
4745 	}
4746 
4747 	return 0;
4748 }
4749 
4750 /*
4751  * Allocate a shadow VMCS and associate it with the currently loaded
4752  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4753  * VMCS is also VMCLEARed, so that it is ready for use.
4754  */
4755 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4756 {
4757 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4758 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4759 
4760 	/*
4761 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4762 	 * executes VMXON and free it when L1 executes VMXOFF.
4763 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4764 	 * here when vmcs01 already have an allocated shadow vmcs.
4765 	 */
4766 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4767 
4768 	if (!loaded_vmcs->shadow_vmcs) {
4769 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4770 		if (loaded_vmcs->shadow_vmcs)
4771 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4772 	}
4773 	return loaded_vmcs->shadow_vmcs;
4774 }
4775 
4776 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4777 {
4778 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4779 	int r;
4780 
4781 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4782 	if (r < 0)
4783 		goto out_vmcs02;
4784 
4785 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4786 	if (!vmx->nested.cached_vmcs12)
4787 		goto out_cached_vmcs12;
4788 
4789 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4790 	if (!vmx->nested.cached_shadow_vmcs12)
4791 		goto out_cached_shadow_vmcs12;
4792 
4793 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4794 		goto out_shadow_vmcs;
4795 
4796 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4797 		     HRTIMER_MODE_ABS_PINNED);
4798 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4799 
4800 	vmx->nested.vpid02 = allocate_vpid();
4801 
4802 	vmx->nested.vmcs02_initialized = false;
4803 	vmx->nested.vmxon = true;
4804 
4805 	if (vmx_pt_mode_is_host_guest()) {
4806 		vmx->pt_desc.guest.ctl = 0;
4807 		pt_update_intercept_for_msr(vcpu);
4808 	}
4809 
4810 	return 0;
4811 
4812 out_shadow_vmcs:
4813 	kfree(vmx->nested.cached_shadow_vmcs12);
4814 
4815 out_cached_shadow_vmcs12:
4816 	kfree(vmx->nested.cached_vmcs12);
4817 
4818 out_cached_vmcs12:
4819 	free_loaded_vmcs(&vmx->nested.vmcs02);
4820 
4821 out_vmcs02:
4822 	return -ENOMEM;
4823 }
4824 
4825 /*
4826  * Emulate the VMXON instruction.
4827  * Currently, we just remember that VMX is active, and do not save or even
4828  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4829  * do not currently need to store anything in that guest-allocated memory
4830  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4831  * argument is different from the VMXON pointer (which the spec says they do).
4832  */
4833 static int handle_vmon(struct kvm_vcpu *vcpu)
4834 {
4835 	int ret;
4836 	gpa_t vmptr;
4837 	uint32_t revision;
4838 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4839 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4840 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4841 
4842 	/*
4843 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4844 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4845 	 * 1 (see vmx_is_valid_cr4() for when we allow the guest to set this).
4846 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4847 	 * have already been checked by hardware, prior to the VM-exit for
4848 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4849 	 * that bit set to 1 in non-root mode.
4850 	 */
4851 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4852 		kvm_queue_exception(vcpu, UD_VECTOR);
4853 		return 1;
4854 	}
4855 
4856 	/* CPL=0 must be checked manually. */
4857 	if (vmx_get_cpl(vcpu)) {
4858 		kvm_inject_gp(vcpu, 0);
4859 		return 1;
4860 	}
4861 
4862 	if (vmx->nested.vmxon)
4863 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4864 
4865 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4866 			!= VMXON_NEEDED_FEATURES) {
4867 		kvm_inject_gp(vcpu, 0);
4868 		return 1;
4869 	}
4870 
4871 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4872 		return ret;
4873 
4874 	/*
4875 	 * SDM 3: 24.11.5
4876 	 * The first 4 bytes of VMXON region contain the supported
4877 	 * VMCS revision identifier
4878 	 *
4879 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4880 	 * which replaces physical address width with 32
4881 	 */
4882 	if (!page_address_valid(vcpu, vmptr))
4883 		return nested_vmx_failInvalid(vcpu);
4884 
4885 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4886 	    revision != VMCS12_REVISION)
4887 		return nested_vmx_failInvalid(vcpu);
4888 
4889 	vmx->nested.vmxon_ptr = vmptr;
4890 	ret = enter_vmx_operation(vcpu);
4891 	if (ret)
4892 		return ret;
4893 
4894 	return nested_vmx_succeed(vcpu);
4895 }
4896 
4897 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4898 {
4899 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4900 
4901 	if (vmx->nested.current_vmptr == -1ull)
4902 		return;
4903 
4904 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4905 
4906 	if (enable_shadow_vmcs) {
4907 		/* copy to memory all shadowed fields in case
4908 		   they were modified */
4909 		copy_shadow_to_vmcs12(vmx);
4910 		vmx_disable_shadow_vmcs(vmx);
4911 	}
4912 	vmx->nested.posted_intr_nv = -1;
4913 
4914 	/* Flush VMCS12 to guest memory */
4915 	kvm_vcpu_write_guest_page(vcpu,
4916 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4917 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4918 
4919 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4920 
4921 	vmx->nested.current_vmptr = -1ull;
4922 }
4923 
4924 /* Emulate the VMXOFF instruction */
4925 static int handle_vmoff(struct kvm_vcpu *vcpu)
4926 {
4927 	if (!nested_vmx_check_permission(vcpu))
4928 		return 1;
4929 
4930 	free_nested(vcpu);
4931 
4932 	/* Process a latched INIT during time CPU was in VMX operation */
4933 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4934 
4935 	return nested_vmx_succeed(vcpu);
4936 }
4937 
4938 /* Emulate the VMCLEAR instruction */
4939 static int handle_vmclear(struct kvm_vcpu *vcpu)
4940 {
4941 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4942 	u32 zero = 0;
4943 	gpa_t vmptr;
4944 	u64 evmcs_gpa;
4945 	int r;
4946 
4947 	if (!nested_vmx_check_permission(vcpu))
4948 		return 1;
4949 
4950 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
4951 		return r;
4952 
4953 	if (!page_address_valid(vcpu, vmptr))
4954 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
4955 
4956 	if (vmptr == vmx->nested.vmxon_ptr)
4957 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
4958 
4959 	/*
4960 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4961 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4962 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4963 	 * writing to the non-existent 'launch_state' field. The area doesn't
4964 	 * have to be the currently active EVMCS on the calling CPU and there's
4965 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
4966 	 * state. It is possible that the area will stay mapped as
4967 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
4968 	 */
4969 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
4970 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
4971 		if (vmptr == vmx->nested.current_vmptr)
4972 			nested_release_vmcs12(vcpu);
4973 
4974 		kvm_vcpu_write_guest(vcpu,
4975 				     vmptr + offsetof(struct vmcs12,
4976 						      launch_state),
4977 				     &zero, sizeof(zero));
4978 	}
4979 
4980 	return nested_vmx_succeed(vcpu);
4981 }
4982 
4983 /* Emulate the VMLAUNCH instruction */
4984 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4985 {
4986 	return nested_vmx_run(vcpu, true);
4987 }
4988 
4989 /* Emulate the VMRESUME instruction */
4990 static int handle_vmresume(struct kvm_vcpu *vcpu)
4991 {
4992 
4993 	return nested_vmx_run(vcpu, false);
4994 }
4995 
4996 static int handle_vmread(struct kvm_vcpu *vcpu)
4997 {
4998 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4999 						    : get_vmcs12(vcpu);
5000 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5001 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5002 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5003 	struct x86_exception e;
5004 	unsigned long field;
5005 	u64 value;
5006 	gva_t gva = 0;
5007 	short offset;
5008 	int len, r;
5009 
5010 	if (!nested_vmx_check_permission(vcpu))
5011 		return 1;
5012 
5013 	/*
5014 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
5015 	 * any VMREAD sets the ALU flags for VMfailInvalid.
5016 	 */
5017 	if (vmx->nested.current_vmptr == -1ull ||
5018 	    (is_guest_mode(vcpu) &&
5019 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
5020 		return nested_vmx_failInvalid(vcpu);
5021 
5022 	/* Decode instruction info and find the field to read */
5023 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
5024 
5025 	offset = vmcs_field_to_offset(field);
5026 	if (offset < 0)
5027 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5028 
5029 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5030 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5031 
5032 	/* Read the field, zero-extended to a u64 value */
5033 	value = vmcs12_read_any(vmcs12, field, offset);
5034 
5035 	/*
5036 	 * Now copy part of this value to register or memory, as requested.
5037 	 * Note that the number of bits actually copied is 32 or 64 depending
5038 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5039 	 */
5040 	if (instr_info & BIT(10)) {
5041 		kvm_register_writel(vcpu, (((instr_info) >> 3) & 0xf), value);
5042 	} else {
5043 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5044 		if (get_vmx_mem_address(vcpu, exit_qualification,
5045 					instr_info, true, len, &gva))
5046 			return 1;
5047 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5048 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5049 		if (r != X86EMUL_CONTINUE)
5050 			return kvm_handle_memory_failure(vcpu, r, &e);
5051 	}
5052 
5053 	return nested_vmx_succeed(vcpu);
5054 }
5055 
5056 static bool is_shadow_field_rw(unsigned long field)
5057 {
5058 	switch (field) {
5059 #define SHADOW_FIELD_RW(x, y) case x:
5060 #include "vmcs_shadow_fields.h"
5061 		return true;
5062 	default:
5063 		break;
5064 	}
5065 	return false;
5066 }
5067 
5068 static bool is_shadow_field_ro(unsigned long field)
5069 {
5070 	switch (field) {
5071 #define SHADOW_FIELD_RO(x, y) case x:
5072 #include "vmcs_shadow_fields.h"
5073 		return true;
5074 	default:
5075 		break;
5076 	}
5077 	return false;
5078 }
5079 
5080 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5081 {
5082 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5083 						    : get_vmcs12(vcpu);
5084 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5085 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5086 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5087 	struct x86_exception e;
5088 	unsigned long field;
5089 	short offset;
5090 	gva_t gva;
5091 	int len, r;
5092 
5093 	/*
5094 	 * The value to write might be 32 or 64 bits, depending on L1's long
5095 	 * mode, and eventually we need to write that into a field of several
5096 	 * possible lengths. The code below first zero-extends the value to 64
5097 	 * bit (value), and then copies only the appropriate number of
5098 	 * bits into the vmcs12 field.
5099 	 */
5100 	u64 value = 0;
5101 
5102 	if (!nested_vmx_check_permission(vcpu))
5103 		return 1;
5104 
5105 	/*
5106 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
5107 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5108 	 */
5109 	if (vmx->nested.current_vmptr == -1ull ||
5110 	    (is_guest_mode(vcpu) &&
5111 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
5112 		return nested_vmx_failInvalid(vcpu);
5113 
5114 	if (instr_info & BIT(10))
5115 		value = kvm_register_readl(vcpu, (((instr_info) >> 3) & 0xf));
5116 	else {
5117 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5118 		if (get_vmx_mem_address(vcpu, exit_qualification,
5119 					instr_info, false, len, &gva))
5120 			return 1;
5121 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5122 		if (r != X86EMUL_CONTINUE)
5123 			return kvm_handle_memory_failure(vcpu, r, &e);
5124 	}
5125 
5126 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
5127 
5128 	offset = vmcs_field_to_offset(field);
5129 	if (offset < 0)
5130 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5131 
5132 	/*
5133 	 * If the vCPU supports "VMWRITE to any supported field in the
5134 	 * VMCS," then the "read-only" fields are actually read/write.
5135 	 */
5136 	if (vmcs_field_readonly(field) &&
5137 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5138 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5139 
5140 	/*
5141 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5142 	 * vmcs12, else we may crush a field or consume a stale value.
5143 	 */
5144 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5145 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5146 
5147 	/*
5148 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5149 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5150 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5151 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5152 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5153 	 * the stripped down value, L2 sees the full value as stored by KVM).
5154 	 */
5155 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5156 		value &= 0x1f0ff;
5157 
5158 	vmcs12_write_any(vmcs12, field, offset, value);
5159 
5160 	/*
5161 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5162 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5163 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5164 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5165 	 */
5166 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5167 		/*
5168 		 * L1 can read these fields without exiting, ensure the
5169 		 * shadow VMCS is up-to-date.
5170 		 */
5171 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5172 			preempt_disable();
5173 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5174 
5175 			__vmcs_writel(field, value);
5176 
5177 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5178 			vmcs_load(vmx->loaded_vmcs->vmcs);
5179 			preempt_enable();
5180 		}
5181 		vmx->nested.dirty_vmcs12 = true;
5182 	}
5183 
5184 	return nested_vmx_succeed(vcpu);
5185 }
5186 
5187 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5188 {
5189 	vmx->nested.current_vmptr = vmptr;
5190 	if (enable_shadow_vmcs) {
5191 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5192 		vmcs_write64(VMCS_LINK_POINTER,
5193 			     __pa(vmx->vmcs01.shadow_vmcs));
5194 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5195 	}
5196 	vmx->nested.dirty_vmcs12 = true;
5197 }
5198 
5199 /* Emulate the VMPTRLD instruction */
5200 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5201 {
5202 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5203 	gpa_t vmptr;
5204 	int r;
5205 
5206 	if (!nested_vmx_check_permission(vcpu))
5207 		return 1;
5208 
5209 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5210 		return r;
5211 
5212 	if (!page_address_valid(vcpu, vmptr))
5213 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5214 
5215 	if (vmptr == vmx->nested.vmxon_ptr)
5216 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5217 
5218 	/* Forbid normal VMPTRLD if Enlightened version was used */
5219 	if (vmx->nested.hv_evmcs)
5220 		return 1;
5221 
5222 	if (vmx->nested.current_vmptr != vmptr) {
5223 		struct kvm_host_map map;
5224 		struct vmcs12 *new_vmcs12;
5225 
5226 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
5227 			/*
5228 			 * Reads from an unbacked page return all 1s,
5229 			 * which means that the 32 bits located at the
5230 			 * given physical address won't match the required
5231 			 * VMCS12_REVISION identifier.
5232 			 */
5233 			return nested_vmx_fail(vcpu,
5234 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5235 		}
5236 
5237 		new_vmcs12 = map.hva;
5238 
5239 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5240 		    (new_vmcs12->hdr.shadow_vmcs &&
5241 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5242 			kvm_vcpu_unmap(vcpu, &map, false);
5243 			return nested_vmx_fail(vcpu,
5244 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5245 		}
5246 
5247 		nested_release_vmcs12(vcpu);
5248 
5249 		/*
5250 		 * Load VMCS12 from guest memory since it is not already
5251 		 * cached.
5252 		 */
5253 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
5254 		kvm_vcpu_unmap(vcpu, &map, false);
5255 
5256 		set_current_vmptr(vmx, vmptr);
5257 	}
5258 
5259 	return nested_vmx_succeed(vcpu);
5260 }
5261 
5262 /* Emulate the VMPTRST instruction */
5263 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5264 {
5265 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5266 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5267 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5268 	struct x86_exception e;
5269 	gva_t gva;
5270 	int r;
5271 
5272 	if (!nested_vmx_check_permission(vcpu))
5273 		return 1;
5274 
5275 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
5276 		return 1;
5277 
5278 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5279 				true, sizeof(gpa_t), &gva))
5280 		return 1;
5281 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5282 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5283 					sizeof(gpa_t), &e);
5284 	if (r != X86EMUL_CONTINUE)
5285 		return kvm_handle_memory_failure(vcpu, r, &e);
5286 
5287 	return nested_vmx_succeed(vcpu);
5288 }
5289 
5290 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
5291 
5292 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
5293 {
5294 	return VALID_PAGE(root_hpa) &&
5295 		((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
5296 }
5297 
5298 /* Emulate the INVEPT instruction */
5299 static int handle_invept(struct kvm_vcpu *vcpu)
5300 {
5301 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5302 	u32 vmx_instruction_info, types;
5303 	unsigned long type, roots_to_free;
5304 	struct kvm_mmu *mmu;
5305 	gva_t gva;
5306 	struct x86_exception e;
5307 	struct {
5308 		u64 eptp, gpa;
5309 	} operand;
5310 	int i, r;
5311 
5312 	if (!(vmx->nested.msrs.secondary_ctls_high &
5313 	      SECONDARY_EXEC_ENABLE_EPT) ||
5314 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5315 		kvm_queue_exception(vcpu, UD_VECTOR);
5316 		return 1;
5317 	}
5318 
5319 	if (!nested_vmx_check_permission(vcpu))
5320 		return 1;
5321 
5322 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5323 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5324 
5325 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5326 
5327 	if (type >= 32 || !(types & (1 << type)))
5328 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5329 
5330 	/* According to the Intel VMX instruction reference, the memory
5331 	 * operand is read even if it isn't needed (e.g., for type==global)
5332 	 */
5333 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5334 			vmx_instruction_info, false, sizeof(operand), &gva))
5335 		return 1;
5336 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5337 	if (r != X86EMUL_CONTINUE)
5338 		return kvm_handle_memory_failure(vcpu, r, &e);
5339 
5340 	/*
5341 	 * Nested EPT roots are always held through guest_mmu,
5342 	 * not root_mmu.
5343 	 */
5344 	mmu = &vcpu->arch.guest_mmu;
5345 
5346 	switch (type) {
5347 	case VMX_EPT_EXTENT_CONTEXT:
5348 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5349 			return nested_vmx_fail(vcpu,
5350 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5351 
5352 		roots_to_free = 0;
5353 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
5354 					    operand.eptp))
5355 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5356 
5357 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5358 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5359 						    mmu->prev_roots[i].pgd,
5360 						    operand.eptp))
5361 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5362 		}
5363 		break;
5364 	case VMX_EPT_EXTENT_GLOBAL:
5365 		roots_to_free = KVM_MMU_ROOTS_ALL;
5366 		break;
5367 	default:
5368 		BUG();
5369 		break;
5370 	}
5371 
5372 	if (roots_to_free)
5373 		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
5374 
5375 	return nested_vmx_succeed(vcpu);
5376 }
5377 
5378 static int handle_invvpid(struct kvm_vcpu *vcpu)
5379 {
5380 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5381 	u32 vmx_instruction_info;
5382 	unsigned long type, types;
5383 	gva_t gva;
5384 	struct x86_exception e;
5385 	struct {
5386 		u64 vpid;
5387 		u64 gla;
5388 	} operand;
5389 	u16 vpid02;
5390 	int r;
5391 
5392 	if (!(vmx->nested.msrs.secondary_ctls_high &
5393 	      SECONDARY_EXEC_ENABLE_VPID) ||
5394 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5395 		kvm_queue_exception(vcpu, UD_VECTOR);
5396 		return 1;
5397 	}
5398 
5399 	if (!nested_vmx_check_permission(vcpu))
5400 		return 1;
5401 
5402 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5403 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5404 
5405 	types = (vmx->nested.msrs.vpid_caps &
5406 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5407 
5408 	if (type >= 32 || !(types & (1 << type)))
5409 		return nested_vmx_fail(vcpu,
5410 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5411 
5412 	/* according to the intel vmx instruction reference, the memory
5413 	 * operand is read even if it isn't needed (e.g., for type==global)
5414 	 */
5415 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5416 			vmx_instruction_info, false, sizeof(operand), &gva))
5417 		return 1;
5418 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5419 	if (r != X86EMUL_CONTINUE)
5420 		return kvm_handle_memory_failure(vcpu, r, &e);
5421 
5422 	if (operand.vpid >> 16)
5423 		return nested_vmx_fail(vcpu,
5424 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5425 
5426 	vpid02 = nested_get_vpid02(vcpu);
5427 	switch (type) {
5428 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5429 		if (!operand.vpid ||
5430 		    is_noncanonical_address(operand.gla, vcpu))
5431 			return nested_vmx_fail(vcpu,
5432 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5433 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5434 		break;
5435 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5436 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5437 		if (!operand.vpid)
5438 			return nested_vmx_fail(vcpu,
5439 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5440 		vpid_sync_context(vpid02);
5441 		break;
5442 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5443 		vpid_sync_context(vpid02);
5444 		break;
5445 	default:
5446 		WARN_ON_ONCE(1);
5447 		return kvm_skip_emulated_instruction(vcpu);
5448 	}
5449 
5450 	/*
5451 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5452 	 * linear mappings for L2 (tagged with L2's VPID).  Free all roots as
5453 	 * VPIDs are not tracked in the MMU role.
5454 	 *
5455 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5456 	 * an MMU when EPT is disabled.
5457 	 *
5458 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5459 	 */
5460 	if (!enable_ept)
5461 		kvm_mmu_free_roots(vcpu, &vcpu->arch.root_mmu,
5462 				   KVM_MMU_ROOTS_ALL);
5463 
5464 	return nested_vmx_succeed(vcpu);
5465 }
5466 
5467 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5468 				     struct vmcs12 *vmcs12)
5469 {
5470 	u32 index = kvm_rcx_read(vcpu);
5471 	u64 new_eptp;
5472 	bool accessed_dirty;
5473 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
5474 
5475 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
5476 	    !nested_cpu_has_ept(vmcs12))
5477 		return 1;
5478 
5479 	if (index >= VMFUNC_EPTP_ENTRIES)
5480 		return 1;
5481 
5482 
5483 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5484 				     &new_eptp, index * 8, 8))
5485 		return 1;
5486 
5487 	accessed_dirty = !!(new_eptp & VMX_EPTP_AD_ENABLE_BIT);
5488 
5489 	/*
5490 	 * If the (L2) guest does a vmfunc to the currently
5491 	 * active ept pointer, we don't have to do anything else
5492 	 */
5493 	if (vmcs12->ept_pointer != new_eptp) {
5494 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5495 			return 1;
5496 
5497 		kvm_mmu_unload(vcpu);
5498 		mmu->ept_ad = accessed_dirty;
5499 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
5500 		vmcs12->ept_pointer = new_eptp;
5501 		/*
5502 		 * TODO: Check what's the correct approach in case
5503 		 * mmu reload fails. Currently, we just let the next
5504 		 * reload potentially fail
5505 		 */
5506 		kvm_mmu_reload(vcpu);
5507 	}
5508 
5509 	return 0;
5510 }
5511 
5512 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5513 {
5514 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5515 	struct vmcs12 *vmcs12;
5516 	u32 function = kvm_rax_read(vcpu);
5517 
5518 	/*
5519 	 * VMFUNC is only supported for nested guests, but we always enable the
5520 	 * secondary control for simplicity; for non-nested mode, fake that we
5521 	 * didn't by injecting #UD.
5522 	 */
5523 	if (!is_guest_mode(vcpu)) {
5524 		kvm_queue_exception(vcpu, UD_VECTOR);
5525 		return 1;
5526 	}
5527 
5528 	vmcs12 = get_vmcs12(vcpu);
5529 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
5530 		goto fail;
5531 
5532 	switch (function) {
5533 	case 0:
5534 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5535 			goto fail;
5536 		break;
5537 	default:
5538 		goto fail;
5539 	}
5540 	return kvm_skip_emulated_instruction(vcpu);
5541 
5542 fail:
5543 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
5544 			  vmx_get_intr_info(vcpu),
5545 			  vmx_get_exit_qual(vcpu));
5546 	return 1;
5547 }
5548 
5549 /*
5550  * Return true if an IO instruction with the specified port and size should cause
5551  * a VM-exit into L1.
5552  */
5553 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5554 				 int size)
5555 {
5556 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5557 	gpa_t bitmap, last_bitmap;
5558 	u8 b;
5559 
5560 	last_bitmap = (gpa_t)-1;
5561 	b = -1;
5562 
5563 	while (size > 0) {
5564 		if (port < 0x8000)
5565 			bitmap = vmcs12->io_bitmap_a;
5566 		else if (port < 0x10000)
5567 			bitmap = vmcs12->io_bitmap_b;
5568 		else
5569 			return true;
5570 		bitmap += (port & 0x7fff) / 8;
5571 
5572 		if (last_bitmap != bitmap)
5573 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5574 				return true;
5575 		if (b & (1 << (port & 7)))
5576 			return true;
5577 
5578 		port++;
5579 		size--;
5580 		last_bitmap = bitmap;
5581 	}
5582 
5583 	return false;
5584 }
5585 
5586 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5587 				       struct vmcs12 *vmcs12)
5588 {
5589 	unsigned long exit_qualification;
5590 	unsigned short port;
5591 	int size;
5592 
5593 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5594 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5595 
5596 	exit_qualification = vmx_get_exit_qual(vcpu);
5597 
5598 	port = exit_qualification >> 16;
5599 	size = (exit_qualification & 7) + 1;
5600 
5601 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5602 }
5603 
5604 /*
5605  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5606  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5607  * disinterest in the current event (read or write a specific MSR) by using an
5608  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5609  */
5610 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5611 	struct vmcs12 *vmcs12, u32 exit_reason)
5612 {
5613 	u32 msr_index = kvm_rcx_read(vcpu);
5614 	gpa_t bitmap;
5615 
5616 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5617 		return true;
5618 
5619 	/*
5620 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5621 	 * for the four combinations of read/write and low/high MSR numbers.
5622 	 * First we need to figure out which of the four to use:
5623 	 */
5624 	bitmap = vmcs12->msr_bitmap;
5625 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5626 		bitmap += 2048;
5627 	if (msr_index >= 0xc0000000) {
5628 		msr_index -= 0xc0000000;
5629 		bitmap += 1024;
5630 	}
5631 
5632 	/* Then read the msr_index'th bit from this bitmap: */
5633 	if (msr_index < 1024*8) {
5634 		unsigned char b;
5635 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5636 			return true;
5637 		return 1 & (b >> (msr_index & 7));
5638 	} else
5639 		return true; /* let L1 handle the wrong parameter */
5640 }
5641 
5642 /*
5643  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5644  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5645  * intercept (via guest_host_mask etc.) the current event.
5646  */
5647 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5648 	struct vmcs12 *vmcs12)
5649 {
5650 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5651 	int cr = exit_qualification & 15;
5652 	int reg;
5653 	unsigned long val;
5654 
5655 	switch ((exit_qualification >> 4) & 3) {
5656 	case 0: /* mov to cr */
5657 		reg = (exit_qualification >> 8) & 15;
5658 		val = kvm_register_readl(vcpu, reg);
5659 		switch (cr) {
5660 		case 0:
5661 			if (vmcs12->cr0_guest_host_mask &
5662 			    (val ^ vmcs12->cr0_read_shadow))
5663 				return true;
5664 			break;
5665 		case 3:
5666 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5667 				return true;
5668 			break;
5669 		case 4:
5670 			if (vmcs12->cr4_guest_host_mask &
5671 			    (vmcs12->cr4_read_shadow ^ val))
5672 				return true;
5673 			break;
5674 		case 8:
5675 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5676 				return true;
5677 			break;
5678 		}
5679 		break;
5680 	case 2: /* clts */
5681 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5682 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5683 			return true;
5684 		break;
5685 	case 1: /* mov from cr */
5686 		switch (cr) {
5687 		case 3:
5688 			if (vmcs12->cpu_based_vm_exec_control &
5689 			    CPU_BASED_CR3_STORE_EXITING)
5690 				return true;
5691 			break;
5692 		case 8:
5693 			if (vmcs12->cpu_based_vm_exec_control &
5694 			    CPU_BASED_CR8_STORE_EXITING)
5695 				return true;
5696 			break;
5697 		}
5698 		break;
5699 	case 3: /* lmsw */
5700 		/*
5701 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5702 		 * cr0. Other attempted changes are ignored, with no exit.
5703 		 */
5704 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5705 		if (vmcs12->cr0_guest_host_mask & 0xe &
5706 		    (val ^ vmcs12->cr0_read_shadow))
5707 			return true;
5708 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5709 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5710 		    (val & 0x1))
5711 			return true;
5712 		break;
5713 	}
5714 	return false;
5715 }
5716 
5717 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5718 	struct vmcs12 *vmcs12, gpa_t bitmap)
5719 {
5720 	u32 vmx_instruction_info;
5721 	unsigned long field;
5722 	u8 b;
5723 
5724 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5725 		return true;
5726 
5727 	/* Decode instruction info and find the field to access */
5728 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5729 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5730 
5731 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5732 	if (field >> 15)
5733 		return true;
5734 
5735 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5736 		return true;
5737 
5738 	return 1 & (b >> (field & 7));
5739 }
5740 
5741 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5742 {
5743 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5744 
5745 	if (nested_cpu_has_mtf(vmcs12))
5746 		return true;
5747 
5748 	/*
5749 	 * An MTF VM-exit may be injected into the guest by setting the
5750 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5751 	 * is the case regardless of the 'monitor trap flag' VM-execution
5752 	 * control.
5753 	 */
5754 	return entry_intr_info == (INTR_INFO_VALID_MASK
5755 				   | INTR_TYPE_OTHER_EVENT);
5756 }
5757 
5758 /*
5759  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5760  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5761  */
5762 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu, u32 exit_reason)
5763 {
5764 	u32 intr_info;
5765 
5766 	switch ((u16)exit_reason) {
5767 	case EXIT_REASON_EXCEPTION_NMI:
5768 		intr_info = vmx_get_intr_info(vcpu);
5769 		if (is_nmi(intr_info))
5770 			return true;
5771 		else if (is_page_fault(intr_info))
5772 			return vcpu->arch.apf.host_apf_flags || !enable_ept;
5773 		else if (is_debug(intr_info) &&
5774 			 vcpu->guest_debug &
5775 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5776 			return true;
5777 		else if (is_breakpoint(intr_info) &&
5778 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5779 			return true;
5780 		return false;
5781 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5782 		return true;
5783 	case EXIT_REASON_MCE_DURING_VMENTRY:
5784 		return true;
5785 	case EXIT_REASON_EPT_VIOLATION:
5786 		/*
5787 		 * L0 always deals with the EPT violation. If nested EPT is
5788 		 * used, and the nested mmu code discovers that the address is
5789 		 * missing in the guest EPT table (EPT12), the EPT violation
5790 		 * will be injected with nested_ept_inject_page_fault()
5791 		 */
5792 		return true;
5793 	case EXIT_REASON_EPT_MISCONFIG:
5794 		/*
5795 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5796 		 * table (shadow on EPT) or a merged EPT table that L0 built
5797 		 * (EPT on EPT). So any problems with the structure of the
5798 		 * table is L0's fault.
5799 		 */
5800 		return true;
5801 	case EXIT_REASON_PREEMPTION_TIMER:
5802 		return true;
5803 	case EXIT_REASON_PML_FULL:
5804 		/* We emulate PML support to L1. */
5805 		return true;
5806 	case EXIT_REASON_VMFUNC:
5807 		/* VM functions are emulated through L2->L0 vmexits. */
5808 		return true;
5809 	case EXIT_REASON_ENCLS:
5810 		/* SGX is never exposed to L1 */
5811 		return true;
5812 	default:
5813 		break;
5814 	}
5815 	return false;
5816 }
5817 
5818 /*
5819  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5820  * is_guest_mode (L2).
5821  */
5822 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu, u32 exit_reason)
5823 {
5824 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5825 	u32 intr_info;
5826 
5827 	switch ((u16)exit_reason) {
5828 	case EXIT_REASON_EXCEPTION_NMI:
5829 		intr_info = vmx_get_intr_info(vcpu);
5830 		if (is_nmi(intr_info))
5831 			return true;
5832 		else if (is_page_fault(intr_info))
5833 			return true;
5834 		return vmcs12->exception_bitmap &
5835 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5836 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5837 		return nested_exit_on_intr(vcpu);
5838 	case EXIT_REASON_TRIPLE_FAULT:
5839 		return true;
5840 	case EXIT_REASON_INTERRUPT_WINDOW:
5841 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5842 	case EXIT_REASON_NMI_WINDOW:
5843 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5844 	case EXIT_REASON_TASK_SWITCH:
5845 		return true;
5846 	case EXIT_REASON_CPUID:
5847 		return true;
5848 	case EXIT_REASON_HLT:
5849 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5850 	case EXIT_REASON_INVD:
5851 		return true;
5852 	case EXIT_REASON_INVLPG:
5853 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5854 	case EXIT_REASON_RDPMC:
5855 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5856 	case EXIT_REASON_RDRAND:
5857 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5858 	case EXIT_REASON_RDSEED:
5859 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5860 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5861 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5862 	case EXIT_REASON_VMREAD:
5863 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5864 			vmcs12->vmread_bitmap);
5865 	case EXIT_REASON_VMWRITE:
5866 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5867 			vmcs12->vmwrite_bitmap);
5868 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5869 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5870 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5871 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5872 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5873 		/*
5874 		 * VMX instructions trap unconditionally. This allows L1 to
5875 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5876 		 */
5877 		return true;
5878 	case EXIT_REASON_CR_ACCESS:
5879 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5880 	case EXIT_REASON_DR_ACCESS:
5881 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5882 	case EXIT_REASON_IO_INSTRUCTION:
5883 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5884 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5885 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5886 	case EXIT_REASON_MSR_READ:
5887 	case EXIT_REASON_MSR_WRITE:
5888 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5889 	case EXIT_REASON_INVALID_STATE:
5890 		return true;
5891 	case EXIT_REASON_MWAIT_INSTRUCTION:
5892 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5893 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5894 		return nested_vmx_exit_handled_mtf(vmcs12);
5895 	case EXIT_REASON_MONITOR_INSTRUCTION:
5896 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5897 	case EXIT_REASON_PAUSE_INSTRUCTION:
5898 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5899 			nested_cpu_has2(vmcs12,
5900 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5901 	case EXIT_REASON_MCE_DURING_VMENTRY:
5902 		return true;
5903 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5904 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5905 	case EXIT_REASON_APIC_ACCESS:
5906 	case EXIT_REASON_APIC_WRITE:
5907 	case EXIT_REASON_EOI_INDUCED:
5908 		/*
5909 		 * The controls for "virtualize APIC accesses," "APIC-
5910 		 * register virtualization," and "virtual-interrupt
5911 		 * delivery" only come from vmcs12.
5912 		 */
5913 		return true;
5914 	case EXIT_REASON_INVPCID:
5915 		return
5916 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5917 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5918 	case EXIT_REASON_WBINVD:
5919 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5920 	case EXIT_REASON_XSETBV:
5921 		return true;
5922 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5923 		/*
5924 		 * This should never happen, since it is not possible to
5925 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5926 		 * If if it were, XSS would have to be checked against
5927 		 * the XSS exit bitmap in vmcs12.
5928 		 */
5929 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5930 	case EXIT_REASON_UMWAIT:
5931 	case EXIT_REASON_TPAUSE:
5932 		return nested_cpu_has2(vmcs12,
5933 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
5934 	default:
5935 		return true;
5936 	}
5937 }
5938 
5939 /*
5940  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
5941  * reflected into L1.
5942  */
5943 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
5944 {
5945 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5946 	u32 exit_reason = vmx->exit_reason;
5947 	unsigned long exit_qual;
5948 	u32 exit_intr_info;
5949 
5950 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
5951 
5952 	/*
5953 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
5954 	 * has already loaded L2's state.
5955 	 */
5956 	if (unlikely(vmx->fail)) {
5957 		trace_kvm_nested_vmenter_failed(
5958 			"hardware VM-instruction error: ",
5959 			vmcs_read32(VM_INSTRUCTION_ERROR));
5960 		exit_intr_info = 0;
5961 		exit_qual = 0;
5962 		goto reflect_vmexit;
5963 	}
5964 
5965 	trace_kvm_nested_vmexit(exit_reason, vcpu, KVM_ISA_VMX);
5966 
5967 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
5968 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
5969 		return false;
5970 
5971 	/* If L1 doesn't want the exit, handle it in L0. */
5972 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
5973 		return false;
5974 
5975 	/*
5976 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
5977 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
5978 	 * need to be synthesized by querying the in-kernel LAPIC, but external
5979 	 * interrupts are never reflected to L1 so it's a non-issue.
5980 	 */
5981 	exit_intr_info = vmx_get_intr_info(vcpu);
5982 	if (is_exception_with_error_code(exit_intr_info)) {
5983 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5984 
5985 		vmcs12->vm_exit_intr_error_code =
5986 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5987 	}
5988 	exit_qual = vmx_get_exit_qual(vcpu);
5989 
5990 reflect_vmexit:
5991 	nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info, exit_qual);
5992 	return true;
5993 }
5994 
5995 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5996 				struct kvm_nested_state __user *user_kvm_nested_state,
5997 				u32 user_data_size)
5998 {
5999 	struct vcpu_vmx *vmx;
6000 	struct vmcs12 *vmcs12;
6001 	struct kvm_nested_state kvm_state = {
6002 		.flags = 0,
6003 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6004 		.size = sizeof(kvm_state),
6005 		.hdr.vmx.flags = 0,
6006 		.hdr.vmx.vmxon_pa = -1ull,
6007 		.hdr.vmx.vmcs12_pa = -1ull,
6008 		.hdr.vmx.preemption_timer_deadline = 0,
6009 	};
6010 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6011 		&user_kvm_nested_state->data.vmx[0];
6012 
6013 	if (!vcpu)
6014 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6015 
6016 	vmx = to_vmx(vcpu);
6017 	vmcs12 = get_vmcs12(vcpu);
6018 
6019 	if (nested_vmx_allowed(vcpu) &&
6020 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6021 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6022 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6023 
6024 		if (vmx_has_valid_vmcs12(vcpu)) {
6025 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6026 
6027 			if (vmx->nested.hv_evmcs)
6028 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6029 
6030 			if (is_guest_mode(vcpu) &&
6031 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6032 			    vmcs12->vmcs_link_pointer != -1ull)
6033 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6034 		}
6035 
6036 		if (vmx->nested.smm.vmxon)
6037 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6038 
6039 		if (vmx->nested.smm.guest_mode)
6040 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6041 
6042 		if (is_guest_mode(vcpu)) {
6043 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6044 
6045 			if (vmx->nested.nested_run_pending)
6046 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6047 
6048 			if (vmx->nested.mtf_pending)
6049 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6050 
6051 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6052 			    vmx->nested.has_preemption_timer_deadline) {
6053 				kvm_state.hdr.vmx.flags |=
6054 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6055 				kvm_state.hdr.vmx.preemption_timer_deadline =
6056 					vmx->nested.preemption_timer_deadline;
6057 			}
6058 		}
6059 	}
6060 
6061 	if (user_data_size < kvm_state.size)
6062 		goto out;
6063 
6064 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6065 		return -EFAULT;
6066 
6067 	if (!vmx_has_valid_vmcs12(vcpu))
6068 		goto out;
6069 
6070 	/*
6071 	 * When running L2, the authoritative vmcs12 state is in the
6072 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6073 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6074 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6075 	 * vmcs12 state is in the vmcs12 already.
6076 	 */
6077 	if (is_guest_mode(vcpu)) {
6078 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6079 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6080 	} else if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6081 		if (vmx->nested.hv_evmcs)
6082 			copy_enlightened_to_vmcs12(vmx);
6083 		else if (enable_shadow_vmcs)
6084 			copy_shadow_to_vmcs12(vmx);
6085 	}
6086 
6087 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6088 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6089 
6090 	/*
6091 	 * Copy over the full allocated size of vmcs12 rather than just the size
6092 	 * of the struct.
6093 	 */
6094 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6095 		return -EFAULT;
6096 
6097 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6098 	    vmcs12->vmcs_link_pointer != -1ull) {
6099 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6100 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6101 			return -EFAULT;
6102 	}
6103 out:
6104 	return kvm_state.size;
6105 }
6106 
6107 /*
6108  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6109  */
6110 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6111 {
6112 	if (is_guest_mode(vcpu)) {
6113 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6114 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6115 	}
6116 	free_nested(vcpu);
6117 }
6118 
6119 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6120 				struct kvm_nested_state __user *user_kvm_nested_state,
6121 				struct kvm_nested_state *kvm_state)
6122 {
6123 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6124 	struct vmcs12 *vmcs12;
6125 	enum vm_entry_failure_code ignored;
6126 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6127 		&user_kvm_nested_state->data.vmx[0];
6128 	int ret;
6129 
6130 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6131 		return -EINVAL;
6132 
6133 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
6134 		if (kvm_state->hdr.vmx.smm.flags)
6135 			return -EINVAL;
6136 
6137 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
6138 			return -EINVAL;
6139 
6140 		/*
6141 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6142 		 * enable eVMCS capability on vCPU. However, since then
6143 		 * code was changed such that flag signals vmcs12 should
6144 		 * be copied into eVMCS in guest memory.
6145 		 *
6146 		 * To preserve backwards compatability, allow user
6147 		 * to set this flag even when there is no VMXON region.
6148 		 */
6149 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6150 			return -EINVAL;
6151 	} else {
6152 		if (!nested_vmx_allowed(vcpu))
6153 			return -EINVAL;
6154 
6155 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6156 			return -EINVAL;
6157 	}
6158 
6159 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6160 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6161 		return -EINVAL;
6162 
6163 	if (kvm_state->hdr.vmx.smm.flags &
6164 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6165 		return -EINVAL;
6166 
6167 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6168 		return -EINVAL;
6169 
6170 	/*
6171 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6172 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6173 	 * must be zero.
6174 	 */
6175 	if (is_smm(vcpu) ?
6176 		(kvm_state->flags &
6177 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6178 		: kvm_state->hdr.vmx.smm.flags)
6179 		return -EINVAL;
6180 
6181 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6182 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6183 		return -EINVAL;
6184 
6185 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6186 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6187 			return -EINVAL;
6188 
6189 	vmx_leave_nested(vcpu);
6190 
6191 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
6192 		return 0;
6193 
6194 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6195 	ret = enter_vmx_operation(vcpu);
6196 	if (ret)
6197 		return ret;
6198 
6199 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6200 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6201 		/* See vmx_has_valid_vmcs12.  */
6202 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6203 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6204 		    (kvm_state->hdr.vmx.vmcs12_pa != -1ull))
6205 			return -EINVAL;
6206 		else
6207 			return 0;
6208 	}
6209 
6210 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
6211 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6212 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6213 			return -EINVAL;
6214 
6215 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6216 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6217 		/*
6218 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6219 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6220 		 * restored yet. EVMCS will be mapped from
6221 		 * nested_get_vmcs12_pages().
6222 		 */
6223 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6224 	} else {
6225 		return -EINVAL;
6226 	}
6227 
6228 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6229 		vmx->nested.smm.vmxon = true;
6230 		vmx->nested.vmxon = false;
6231 
6232 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6233 			vmx->nested.smm.guest_mode = true;
6234 	}
6235 
6236 	vmcs12 = get_vmcs12(vcpu);
6237 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6238 		return -EFAULT;
6239 
6240 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6241 		return -EINVAL;
6242 
6243 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6244 		return 0;
6245 
6246 	vmx->nested.nested_run_pending =
6247 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6248 
6249 	vmx->nested.mtf_pending =
6250 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6251 
6252 	ret = -EINVAL;
6253 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6254 	    vmcs12->vmcs_link_pointer != -1ull) {
6255 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6256 
6257 		if (kvm_state->size <
6258 		    sizeof(*kvm_state) +
6259 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6260 			goto error_guest_mode;
6261 
6262 		if (copy_from_user(shadow_vmcs12,
6263 				   user_vmx_nested_state->shadow_vmcs12,
6264 				   sizeof(*shadow_vmcs12))) {
6265 			ret = -EFAULT;
6266 			goto error_guest_mode;
6267 		}
6268 
6269 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6270 		    !shadow_vmcs12->hdr.shadow_vmcs)
6271 			goto error_guest_mode;
6272 	}
6273 
6274 	vmx->nested.has_preemption_timer_deadline = false;
6275 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6276 		vmx->nested.has_preemption_timer_deadline = true;
6277 		vmx->nested.preemption_timer_deadline =
6278 			kvm_state->hdr.vmx.preemption_timer_deadline;
6279 	}
6280 
6281 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6282 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6283 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6284 		goto error_guest_mode;
6285 
6286 	vmx->nested.dirty_vmcs12 = true;
6287 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6288 	if (ret)
6289 		goto error_guest_mode;
6290 
6291 	return 0;
6292 
6293 error_guest_mode:
6294 	vmx->nested.nested_run_pending = 0;
6295 	return ret;
6296 }
6297 
6298 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6299 {
6300 	if (enable_shadow_vmcs) {
6301 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6302 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6303 	}
6304 }
6305 
6306 /*
6307  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6308  * returned for the various VMX controls MSRs when nested VMX is enabled.
6309  * The same values should also be used to verify that vmcs12 control fields are
6310  * valid during nested entry from L1 to L2.
6311  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6312  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6313  * bit in the high half is on if the corresponding bit in the control field
6314  * may be on. See also vmx_control_verify().
6315  */
6316 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6317 {
6318 	/*
6319 	 * Note that as a general rule, the high half of the MSRs (bits in
6320 	 * the control fields which may be 1) should be initialized by the
6321 	 * intersection of the underlying hardware's MSR (i.e., features which
6322 	 * can be supported) and the list of features we want to expose -
6323 	 * because they are known to be properly supported in our code.
6324 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6325 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6326 	 * reason is that if one of these bits is necessary, it will appear
6327 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6328 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6329 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6330 	 * These rules have exceptions below.
6331 	 */
6332 
6333 	/* pin-based controls */
6334 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6335 		msrs->pinbased_ctls_low,
6336 		msrs->pinbased_ctls_high);
6337 	msrs->pinbased_ctls_low |=
6338 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6339 	msrs->pinbased_ctls_high &=
6340 		PIN_BASED_EXT_INTR_MASK |
6341 		PIN_BASED_NMI_EXITING |
6342 		PIN_BASED_VIRTUAL_NMIS |
6343 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6344 	msrs->pinbased_ctls_high |=
6345 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6346 		PIN_BASED_VMX_PREEMPTION_TIMER;
6347 
6348 	/* exit controls */
6349 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6350 		msrs->exit_ctls_low,
6351 		msrs->exit_ctls_high);
6352 	msrs->exit_ctls_low =
6353 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6354 
6355 	msrs->exit_ctls_high &=
6356 #ifdef CONFIG_X86_64
6357 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6358 #endif
6359 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6360 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6361 	msrs->exit_ctls_high |=
6362 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6363 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6364 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6365 
6366 	/* We support free control of debug control saving. */
6367 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6368 
6369 	/* entry controls */
6370 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6371 		msrs->entry_ctls_low,
6372 		msrs->entry_ctls_high);
6373 	msrs->entry_ctls_low =
6374 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6375 	msrs->entry_ctls_high &=
6376 #ifdef CONFIG_X86_64
6377 		VM_ENTRY_IA32E_MODE |
6378 #endif
6379 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6380 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6381 	msrs->entry_ctls_high |=
6382 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6383 
6384 	/* We support free control of debug control loading. */
6385 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6386 
6387 	/* cpu-based controls */
6388 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6389 		msrs->procbased_ctls_low,
6390 		msrs->procbased_ctls_high);
6391 	msrs->procbased_ctls_low =
6392 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6393 	msrs->procbased_ctls_high &=
6394 		CPU_BASED_INTR_WINDOW_EXITING |
6395 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6396 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6397 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6398 		CPU_BASED_CR3_STORE_EXITING |
6399 #ifdef CONFIG_X86_64
6400 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6401 #endif
6402 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6403 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6404 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6405 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6406 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6407 	/*
6408 	 * We can allow some features even when not supported by the
6409 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6410 	 * can use it to avoid exits to L1 - even when L0 runs L2
6411 	 * without MSR bitmaps.
6412 	 */
6413 	msrs->procbased_ctls_high |=
6414 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6415 		CPU_BASED_USE_MSR_BITMAPS;
6416 
6417 	/* We support free control of CR3 access interception. */
6418 	msrs->procbased_ctls_low &=
6419 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6420 
6421 	/*
6422 	 * secondary cpu-based controls.  Do not include those that
6423 	 * depend on CPUID bits, they are added later by
6424 	 * vmx_vcpu_after_set_cpuid.
6425 	 */
6426 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6427 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6428 		      msrs->secondary_ctls_low,
6429 		      msrs->secondary_ctls_high);
6430 
6431 	msrs->secondary_ctls_low = 0;
6432 	msrs->secondary_ctls_high &=
6433 		SECONDARY_EXEC_DESC |
6434 		SECONDARY_EXEC_ENABLE_RDTSCP |
6435 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6436 		SECONDARY_EXEC_WBINVD_EXITING |
6437 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6438 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6439 		SECONDARY_EXEC_RDRAND_EXITING |
6440 		SECONDARY_EXEC_ENABLE_INVPCID |
6441 		SECONDARY_EXEC_RDSEED_EXITING |
6442 		SECONDARY_EXEC_XSAVES;
6443 
6444 	/*
6445 	 * We can emulate "VMCS shadowing," even if the hardware
6446 	 * doesn't support it.
6447 	 */
6448 	msrs->secondary_ctls_high |=
6449 		SECONDARY_EXEC_SHADOW_VMCS;
6450 
6451 	if (enable_ept) {
6452 		/* nested EPT: emulate EPT also to L1 */
6453 		msrs->secondary_ctls_high |=
6454 			SECONDARY_EXEC_ENABLE_EPT;
6455 		msrs->ept_caps =
6456 			VMX_EPT_PAGE_WALK_4_BIT |
6457 			VMX_EPT_PAGE_WALK_5_BIT |
6458 			VMX_EPTP_WB_BIT |
6459 			VMX_EPT_INVEPT_BIT |
6460 			VMX_EPT_EXECUTE_ONLY_BIT;
6461 
6462 		msrs->ept_caps &= ept_caps;
6463 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6464 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6465 			VMX_EPT_1GB_PAGE_BIT;
6466 		if (enable_ept_ad_bits) {
6467 			msrs->secondary_ctls_high |=
6468 				SECONDARY_EXEC_ENABLE_PML;
6469 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6470 		}
6471 	}
6472 
6473 	if (cpu_has_vmx_vmfunc()) {
6474 		msrs->secondary_ctls_high |=
6475 			SECONDARY_EXEC_ENABLE_VMFUNC;
6476 		/*
6477 		 * Advertise EPTP switching unconditionally
6478 		 * since we emulate it
6479 		 */
6480 		if (enable_ept)
6481 			msrs->vmfunc_controls =
6482 				VMX_VMFUNC_EPTP_SWITCHING;
6483 	}
6484 
6485 	/*
6486 	 * Old versions of KVM use the single-context version without
6487 	 * checking for support, so declare that it is supported even
6488 	 * though it is treated as global context.  The alternative is
6489 	 * not failing the single-context invvpid, and it is worse.
6490 	 */
6491 	if (enable_vpid) {
6492 		msrs->secondary_ctls_high |=
6493 			SECONDARY_EXEC_ENABLE_VPID;
6494 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6495 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6496 	}
6497 
6498 	if (enable_unrestricted_guest)
6499 		msrs->secondary_ctls_high |=
6500 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6501 
6502 	if (flexpriority_enabled)
6503 		msrs->secondary_ctls_high |=
6504 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6505 
6506 	/* miscellaneous data */
6507 	rdmsr(MSR_IA32_VMX_MISC,
6508 		msrs->misc_low,
6509 		msrs->misc_high);
6510 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6511 	msrs->misc_low |=
6512 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6513 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6514 		VMX_MISC_ACTIVITY_HLT |
6515 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6516 	msrs->misc_high = 0;
6517 
6518 	/*
6519 	 * This MSR reports some information about VMX support. We
6520 	 * should return information about the VMX we emulate for the
6521 	 * guest, and the VMCS structure we give it - not about the
6522 	 * VMX support of the underlying hardware.
6523 	 */
6524 	msrs->basic =
6525 		VMCS12_REVISION |
6526 		VMX_BASIC_TRUE_CTLS |
6527 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6528 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6529 
6530 	if (cpu_has_vmx_basic_inout())
6531 		msrs->basic |= VMX_BASIC_INOUT;
6532 
6533 	/*
6534 	 * These MSRs specify bits which the guest must keep fixed on
6535 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6536 	 * We picked the standard core2 setting.
6537 	 */
6538 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6539 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6540 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6541 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6542 
6543 	/* These MSRs specify bits which the guest must keep fixed off. */
6544 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6545 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6546 
6547 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
6548 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
6549 }
6550 
6551 void nested_vmx_hardware_unsetup(void)
6552 {
6553 	int i;
6554 
6555 	if (enable_shadow_vmcs) {
6556 		for (i = 0; i < VMX_BITMAP_NR; i++)
6557 			free_page((unsigned long)vmx_bitmap[i]);
6558 	}
6559 }
6560 
6561 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6562 {
6563 	int i;
6564 
6565 	if (!cpu_has_vmx_shadow_vmcs())
6566 		enable_shadow_vmcs = 0;
6567 	if (enable_shadow_vmcs) {
6568 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6569 			/*
6570 			 * The vmx_bitmap is not tied to a VM and so should
6571 			 * not be charged to a memcg.
6572 			 */
6573 			vmx_bitmap[i] = (unsigned long *)
6574 				__get_free_page(GFP_KERNEL);
6575 			if (!vmx_bitmap[i]) {
6576 				nested_vmx_hardware_unsetup();
6577 				return -ENOMEM;
6578 			}
6579 		}
6580 
6581 		init_vmcs_shadow_fields();
6582 	}
6583 
6584 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6585 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6586 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6587 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6588 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6589 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6590 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6591 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6592 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6593 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6594 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6595 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6596 
6597 	return 0;
6598 }
6599 
6600 struct kvm_x86_nested_ops vmx_nested_ops = {
6601 	.check_events = vmx_check_nested_events,
6602 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6603 	.get_state = vmx_get_nested_state,
6604 	.set_state = vmx_set_nested_state,
6605 	.get_nested_state_pages = nested_get_vmcs12_pages,
6606 	.write_log_dirty = nested_vmx_write_pml_buffer,
6607 	.enable_evmcs = nested_enable_evmcs,
6608 	.get_evmcs_version = nested_get_evmcs_version,
6609 };
6610