xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 82df5b73)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/frame.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 	struct vcpu_vmx *vmx = to_vmx(vcpu);
175 
176 	/*
177 	 * failValid writes the error number to the current VMCS, which
178 	 * can't be done if there isn't a current VMCS.
179 	 */
180 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
181 		return nested_vmx_failInvalid(vcpu);
182 
183 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
184 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
185 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
186 			| X86_EFLAGS_ZF);
187 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
188 	/*
189 	 * We don't need to force a shadow sync because
190 	 * VM_INSTRUCTION_ERROR is not shadowed
191 	 */
192 	return kvm_skip_emulated_instruction(vcpu);
193 }
194 
195 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
196 {
197 	/* TODO: not to reset guest simply here. */
198 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
199 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
200 }
201 
202 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
203 {
204 	return fixed_bits_valid(control, low, high);
205 }
206 
207 static inline u64 vmx_control_msr(u32 low, u32 high)
208 {
209 	return low | ((u64)high << 32);
210 }
211 
212 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
213 {
214 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
215 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
216 	vmx->nested.need_vmcs12_to_shadow_sync = false;
217 }
218 
219 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
220 {
221 	struct vcpu_vmx *vmx = to_vmx(vcpu);
222 
223 	if (!vmx->nested.hv_evmcs)
224 		return;
225 
226 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
227 	vmx->nested.hv_evmcs_vmptr = 0;
228 	vmx->nested.hv_evmcs = NULL;
229 }
230 
231 /*
232  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
233  * just stops using VMX.
234  */
235 static void free_nested(struct kvm_vcpu *vcpu)
236 {
237 	struct vcpu_vmx *vmx = to_vmx(vcpu);
238 
239 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
240 		return;
241 
242 	kvm_clear_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
243 
244 	vmx->nested.vmxon = false;
245 	vmx->nested.smm.vmxon = false;
246 	free_vpid(vmx->nested.vpid02);
247 	vmx->nested.posted_intr_nv = -1;
248 	vmx->nested.current_vmptr = -1ull;
249 	if (enable_shadow_vmcs) {
250 		vmx_disable_shadow_vmcs(vmx);
251 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
252 		free_vmcs(vmx->vmcs01.shadow_vmcs);
253 		vmx->vmcs01.shadow_vmcs = NULL;
254 	}
255 	kfree(vmx->nested.cached_vmcs12);
256 	vmx->nested.cached_vmcs12 = NULL;
257 	kfree(vmx->nested.cached_shadow_vmcs12);
258 	vmx->nested.cached_shadow_vmcs12 = NULL;
259 	/* Unpin physical memory we referred to in the vmcs02 */
260 	if (vmx->nested.apic_access_page) {
261 		kvm_release_page_clean(vmx->nested.apic_access_page);
262 		vmx->nested.apic_access_page = NULL;
263 	}
264 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
265 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
266 	vmx->nested.pi_desc = NULL;
267 
268 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
269 
270 	nested_release_evmcs(vcpu);
271 
272 	free_loaded_vmcs(&vmx->nested.vmcs02);
273 }
274 
275 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
276 				     struct loaded_vmcs *prev)
277 {
278 	struct vmcs_host_state *dest, *src;
279 
280 	if (unlikely(!vmx->guest_state_loaded))
281 		return;
282 
283 	src = &prev->host_state;
284 	dest = &vmx->loaded_vmcs->host_state;
285 
286 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
287 	dest->ldt_sel = src->ldt_sel;
288 #ifdef CONFIG_X86_64
289 	dest->ds_sel = src->ds_sel;
290 	dest->es_sel = src->es_sel;
291 #endif
292 }
293 
294 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
295 {
296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
297 	struct loaded_vmcs *prev;
298 	int cpu;
299 
300 	if (vmx->loaded_vmcs == vmcs)
301 		return;
302 
303 	cpu = get_cpu();
304 	prev = vmx->loaded_vmcs;
305 	vmx->loaded_vmcs = vmcs;
306 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
307 	vmx_sync_vmcs_host_state(vmx, prev);
308 	put_cpu();
309 
310 	vmx_register_cache_reset(vcpu);
311 }
312 
313 /*
314  * Ensure that the current vmcs of the logical processor is the
315  * vmcs01 of the vcpu before calling free_nested().
316  */
317 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
318 {
319 	vcpu_load(vcpu);
320 	vmx_leave_nested(vcpu);
321 	vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
322 	free_nested(vcpu);
323 	vcpu_put(vcpu);
324 }
325 
326 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
327 		struct x86_exception *fault)
328 {
329 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
331 	u32 vm_exit_reason;
332 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
333 
334 	if (vmx->nested.pml_full) {
335 		vm_exit_reason = EXIT_REASON_PML_FULL;
336 		vmx->nested.pml_full = false;
337 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
338 	} else if (fault->error_code & PFERR_RSVD_MASK)
339 		vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
340 	else
341 		vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
342 
343 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
344 	vmcs12->guest_physical_address = fault->address;
345 }
346 
347 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
348 {
349 	WARN_ON(mmu_is_nested(vcpu));
350 
351 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
352 	kvm_init_shadow_ept_mmu(vcpu,
353 			to_vmx(vcpu)->nested.msrs.ept_caps &
354 			VMX_EPT_EXECUTE_ONLY_BIT,
355 			nested_ept_ad_enabled(vcpu),
356 			nested_ept_get_eptp(vcpu));
357 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
358 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
359 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
360 
361 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
362 }
363 
364 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
365 {
366 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
367 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
368 }
369 
370 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
371 					    u16 error_code)
372 {
373 	bool inequality, bit;
374 
375 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
376 	inequality =
377 		(error_code & vmcs12->page_fault_error_code_mask) !=
378 		 vmcs12->page_fault_error_code_match;
379 	return inequality ^ bit;
380 }
381 
382 
383 /*
384  * KVM wants to inject page-faults which it got to the guest. This function
385  * checks whether in a nested guest, we need to inject them to L1 or L2.
386  */
387 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
388 {
389 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
390 	unsigned int nr = vcpu->arch.exception.nr;
391 	bool has_payload = vcpu->arch.exception.has_payload;
392 	unsigned long payload = vcpu->arch.exception.payload;
393 
394 	if (nr == PF_VECTOR) {
395 		if (vcpu->arch.exception.nested_apf) {
396 			*exit_qual = vcpu->arch.apf.nested_apf_token;
397 			return 1;
398 		}
399 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
400 						    vcpu->arch.exception.error_code)) {
401 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
402 			return 1;
403 		}
404 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
405 		if (nr == DB_VECTOR) {
406 			if (!has_payload) {
407 				payload = vcpu->arch.dr6;
408 				payload &= ~(DR6_FIXED_1 | DR6_BT);
409 				payload ^= DR6_RTM;
410 			}
411 			*exit_qual = payload;
412 		} else
413 			*exit_qual = 0;
414 		return 1;
415 	}
416 
417 	return 0;
418 }
419 
420 
421 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
422 		struct x86_exception *fault)
423 {
424 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
425 
426 	WARN_ON(!is_guest_mode(vcpu));
427 
428 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
429 		!to_vmx(vcpu)->nested.nested_run_pending) {
430 		vmcs12->vm_exit_intr_error_code = fault->error_code;
431 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
432 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
433 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
434 				  fault->address);
435 	} else {
436 		kvm_inject_page_fault(vcpu, fault);
437 	}
438 }
439 
440 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
441 					       struct vmcs12 *vmcs12)
442 {
443 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
444 		return 0;
445 
446 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
447 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
448 		return -EINVAL;
449 
450 	return 0;
451 }
452 
453 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
454 						struct vmcs12 *vmcs12)
455 {
456 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
457 		return 0;
458 
459 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
460 		return -EINVAL;
461 
462 	return 0;
463 }
464 
465 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
466 						struct vmcs12 *vmcs12)
467 {
468 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
469 		return 0;
470 
471 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
472 		return -EINVAL;
473 
474 	return 0;
475 }
476 
477 /*
478  * Check if MSR is intercepted for L01 MSR bitmap.
479  */
480 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
481 {
482 	unsigned long *msr_bitmap;
483 	int f = sizeof(unsigned long);
484 
485 	if (!cpu_has_vmx_msr_bitmap())
486 		return true;
487 
488 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
489 
490 	if (msr <= 0x1fff) {
491 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
492 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
493 		msr &= 0x1fff;
494 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
495 	}
496 
497 	return true;
498 }
499 
500 /*
501  * If a msr is allowed by L0, we should check whether it is allowed by L1.
502  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
503  */
504 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
505 					       unsigned long *msr_bitmap_nested,
506 					       u32 msr, int type)
507 {
508 	int f = sizeof(unsigned long);
509 
510 	/*
511 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
512 	 * have the write-low and read-high bitmap offsets the wrong way round.
513 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
514 	 */
515 	if (msr <= 0x1fff) {
516 		if (type & MSR_TYPE_R &&
517 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
518 			/* read-low */
519 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
520 
521 		if (type & MSR_TYPE_W &&
522 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
523 			/* write-low */
524 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
525 
526 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
527 		msr &= 0x1fff;
528 		if (type & MSR_TYPE_R &&
529 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
530 			/* read-high */
531 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
532 
533 		if (type & MSR_TYPE_W &&
534 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
535 			/* write-high */
536 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
537 
538 	}
539 }
540 
541 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
542 {
543 	int msr;
544 
545 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
546 		unsigned word = msr / BITS_PER_LONG;
547 
548 		msr_bitmap[word] = ~0;
549 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
550 	}
551 }
552 
553 /*
554  * Merge L0's and L1's MSR bitmap, return false to indicate that
555  * we do not use the hardware.
556  */
557 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
558 						 struct vmcs12 *vmcs12)
559 {
560 	int msr;
561 	unsigned long *msr_bitmap_l1;
562 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
563 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
564 
565 	/* Nothing to do if the MSR bitmap is not in use.  */
566 	if (!cpu_has_vmx_msr_bitmap() ||
567 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
568 		return false;
569 
570 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
571 		return false;
572 
573 	msr_bitmap_l1 = (unsigned long *)map->hva;
574 
575 	/*
576 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
577 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
578 	 * the x2APIC MSR range and selectively disable them below.
579 	 */
580 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
581 
582 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
583 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
584 			/*
585 			 * L0 need not intercept reads for MSRs between 0x800
586 			 * and 0x8ff, it just lets the processor take the value
587 			 * from the virtual-APIC page; take those 256 bits
588 			 * directly from the L1 bitmap.
589 			 */
590 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
591 				unsigned word = msr / BITS_PER_LONG;
592 
593 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
594 			}
595 		}
596 
597 		nested_vmx_disable_intercept_for_msr(
598 			msr_bitmap_l1, msr_bitmap_l0,
599 			X2APIC_MSR(APIC_TASKPRI),
600 			MSR_TYPE_R | MSR_TYPE_W);
601 
602 		if (nested_cpu_has_vid(vmcs12)) {
603 			nested_vmx_disable_intercept_for_msr(
604 				msr_bitmap_l1, msr_bitmap_l0,
605 				X2APIC_MSR(APIC_EOI),
606 				MSR_TYPE_W);
607 			nested_vmx_disable_intercept_for_msr(
608 				msr_bitmap_l1, msr_bitmap_l0,
609 				X2APIC_MSR(APIC_SELF_IPI),
610 				MSR_TYPE_W);
611 		}
612 	}
613 
614 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
615 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
616 					     MSR_FS_BASE, MSR_TYPE_RW);
617 
618 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
619 					     MSR_GS_BASE, MSR_TYPE_RW);
620 
621 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
622 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
623 
624 	/*
625 	 * Checking the L0->L1 bitmap is trying to verify two things:
626 	 *
627 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
628 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
629 	 *    from the L12 MSR bitmap that is too permissive.
630 	 * 2. That L1 or L2s have actually used the MSR. This avoids
631 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
632 	 *    works properly because we only update the L01 MSR bitmap lazily.
633 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
634 	 *    updated to reflect this when L1 (or its L2s) actually write to
635 	 *    the MSR.
636 	 */
637 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
638 		nested_vmx_disable_intercept_for_msr(
639 					msr_bitmap_l1, msr_bitmap_l0,
640 					MSR_IA32_SPEC_CTRL,
641 					MSR_TYPE_R | MSR_TYPE_W);
642 
643 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
644 		nested_vmx_disable_intercept_for_msr(
645 					msr_bitmap_l1, msr_bitmap_l0,
646 					MSR_IA32_PRED_CMD,
647 					MSR_TYPE_W);
648 
649 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
650 
651 	return true;
652 }
653 
654 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
655 				       struct vmcs12 *vmcs12)
656 {
657 	struct kvm_host_map map;
658 	struct vmcs12 *shadow;
659 
660 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
661 	    vmcs12->vmcs_link_pointer == -1ull)
662 		return;
663 
664 	shadow = get_shadow_vmcs12(vcpu);
665 
666 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
667 		return;
668 
669 	memcpy(shadow, map.hva, VMCS12_SIZE);
670 	kvm_vcpu_unmap(vcpu, &map, false);
671 }
672 
673 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
674 					      struct vmcs12 *vmcs12)
675 {
676 	struct vcpu_vmx *vmx = to_vmx(vcpu);
677 
678 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
679 	    vmcs12->vmcs_link_pointer == -1ull)
680 		return;
681 
682 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
683 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
684 }
685 
686 /*
687  * In nested virtualization, check if L1 has set
688  * VM_EXIT_ACK_INTR_ON_EXIT
689  */
690 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
691 {
692 	return get_vmcs12(vcpu)->vm_exit_controls &
693 		VM_EXIT_ACK_INTR_ON_EXIT;
694 }
695 
696 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
697 					  struct vmcs12 *vmcs12)
698 {
699 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
700 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
701 		return -EINVAL;
702 	else
703 		return 0;
704 }
705 
706 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
707 					   struct vmcs12 *vmcs12)
708 {
709 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
710 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
711 	    !nested_cpu_has_vid(vmcs12) &&
712 	    !nested_cpu_has_posted_intr(vmcs12))
713 		return 0;
714 
715 	/*
716 	 * If virtualize x2apic mode is enabled,
717 	 * virtualize apic access must be disabled.
718 	 */
719 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
720 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
721 		return -EINVAL;
722 
723 	/*
724 	 * If virtual interrupt delivery is enabled,
725 	 * we must exit on external interrupts.
726 	 */
727 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
728 		return -EINVAL;
729 
730 	/*
731 	 * bits 15:8 should be zero in posted_intr_nv,
732 	 * the descriptor address has been already checked
733 	 * in nested_get_vmcs12_pages.
734 	 *
735 	 * bits 5:0 of posted_intr_desc_addr should be zero.
736 	 */
737 	if (nested_cpu_has_posted_intr(vmcs12) &&
738 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
739 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
740 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
741 	    CC((vmcs12->posted_intr_desc_addr & 0x3f)) ||
742 	    CC((vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu)))))
743 		return -EINVAL;
744 
745 	/* tpr shadow is needed by all apicv features. */
746 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
747 		return -EINVAL;
748 
749 	return 0;
750 }
751 
752 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
753 				       u32 count, u64 addr)
754 {
755 	int maxphyaddr;
756 
757 	if (count == 0)
758 		return 0;
759 	maxphyaddr = cpuid_maxphyaddr(vcpu);
760 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
761 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
762 		return -EINVAL;
763 
764 	return 0;
765 }
766 
767 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
768 						     struct vmcs12 *vmcs12)
769 {
770 	if (CC(nested_vmx_check_msr_switch(vcpu,
771 					   vmcs12->vm_exit_msr_load_count,
772 					   vmcs12->vm_exit_msr_load_addr)) ||
773 	    CC(nested_vmx_check_msr_switch(vcpu,
774 					   vmcs12->vm_exit_msr_store_count,
775 					   vmcs12->vm_exit_msr_store_addr)))
776 		return -EINVAL;
777 
778 	return 0;
779 }
780 
781 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
782                                                       struct vmcs12 *vmcs12)
783 {
784 	if (CC(nested_vmx_check_msr_switch(vcpu,
785 					   vmcs12->vm_entry_msr_load_count,
786 					   vmcs12->vm_entry_msr_load_addr)))
787                 return -EINVAL;
788 
789 	return 0;
790 }
791 
792 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
793 					 struct vmcs12 *vmcs12)
794 {
795 	if (!nested_cpu_has_pml(vmcs12))
796 		return 0;
797 
798 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
799 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
800 		return -EINVAL;
801 
802 	return 0;
803 }
804 
805 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
806 							struct vmcs12 *vmcs12)
807 {
808 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
809 	       !nested_cpu_has_ept(vmcs12)))
810 		return -EINVAL;
811 	return 0;
812 }
813 
814 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
815 							 struct vmcs12 *vmcs12)
816 {
817 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
818 	       !nested_cpu_has_ept(vmcs12)))
819 		return -EINVAL;
820 	return 0;
821 }
822 
823 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
824 						 struct vmcs12 *vmcs12)
825 {
826 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
827 		return 0;
828 
829 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
830 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
831 		return -EINVAL;
832 
833 	return 0;
834 }
835 
836 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
837 				       struct vmx_msr_entry *e)
838 {
839 	/* x2APIC MSR accesses are not allowed */
840 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
841 		return -EINVAL;
842 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
843 	    CC(e->index == MSR_IA32_UCODE_REV))
844 		return -EINVAL;
845 	if (CC(e->reserved != 0))
846 		return -EINVAL;
847 	return 0;
848 }
849 
850 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
851 				     struct vmx_msr_entry *e)
852 {
853 	if (CC(e->index == MSR_FS_BASE) ||
854 	    CC(e->index == MSR_GS_BASE) ||
855 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
856 	    nested_vmx_msr_check_common(vcpu, e))
857 		return -EINVAL;
858 	return 0;
859 }
860 
861 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
862 				      struct vmx_msr_entry *e)
863 {
864 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
865 	    nested_vmx_msr_check_common(vcpu, e))
866 		return -EINVAL;
867 	return 0;
868 }
869 
870 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
871 {
872 	struct vcpu_vmx *vmx = to_vmx(vcpu);
873 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
874 				       vmx->nested.msrs.misc_high);
875 
876 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
877 }
878 
879 /*
880  * Load guest's/host's msr at nested entry/exit.
881  * return 0 for success, entry index for failure.
882  *
883  * One of the failure modes for MSR load/store is when a list exceeds the
884  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
885  * as possible, process all valid entries before failing rather than precheck
886  * for a capacity violation.
887  */
888 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
889 {
890 	u32 i;
891 	struct vmx_msr_entry e;
892 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
893 
894 	for (i = 0; i < count; i++) {
895 		if (unlikely(i >= max_msr_list_size))
896 			goto fail;
897 
898 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
899 					&e, sizeof(e))) {
900 			pr_debug_ratelimited(
901 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
902 				__func__, i, gpa + i * sizeof(e));
903 			goto fail;
904 		}
905 		if (nested_vmx_load_msr_check(vcpu, &e)) {
906 			pr_debug_ratelimited(
907 				"%s check failed (%u, 0x%x, 0x%x)\n",
908 				__func__, i, e.index, e.reserved);
909 			goto fail;
910 		}
911 		if (kvm_set_msr(vcpu, e.index, e.value)) {
912 			pr_debug_ratelimited(
913 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
914 				__func__, i, e.index, e.value);
915 			goto fail;
916 		}
917 	}
918 	return 0;
919 fail:
920 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
921 	return i + 1;
922 }
923 
924 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
925 					    u32 msr_index,
926 					    u64 *data)
927 {
928 	struct vcpu_vmx *vmx = to_vmx(vcpu);
929 
930 	/*
931 	 * If the L0 hypervisor stored a more accurate value for the TSC that
932 	 * does not include the time taken for emulation of the L2->L1
933 	 * VM-exit in L0, use the more accurate value.
934 	 */
935 	if (msr_index == MSR_IA32_TSC) {
936 		int index = vmx_find_msr_index(&vmx->msr_autostore.guest,
937 					       MSR_IA32_TSC);
938 
939 		if (index >= 0) {
940 			u64 val = vmx->msr_autostore.guest.val[index].value;
941 
942 			*data = kvm_read_l1_tsc(vcpu, val);
943 			return true;
944 		}
945 	}
946 
947 	if (kvm_get_msr(vcpu, msr_index, data)) {
948 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
949 			msr_index);
950 		return false;
951 	}
952 	return true;
953 }
954 
955 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
956 				     struct vmx_msr_entry *e)
957 {
958 	if (kvm_vcpu_read_guest(vcpu,
959 				gpa + i * sizeof(*e),
960 				e, 2 * sizeof(u32))) {
961 		pr_debug_ratelimited(
962 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
963 			__func__, i, gpa + i * sizeof(*e));
964 		return false;
965 	}
966 	if (nested_vmx_store_msr_check(vcpu, e)) {
967 		pr_debug_ratelimited(
968 			"%s check failed (%u, 0x%x, 0x%x)\n",
969 			__func__, i, e->index, e->reserved);
970 		return false;
971 	}
972 	return true;
973 }
974 
975 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
976 {
977 	u64 data;
978 	u32 i;
979 	struct vmx_msr_entry e;
980 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
981 
982 	for (i = 0; i < count; i++) {
983 		if (unlikely(i >= max_msr_list_size))
984 			return -EINVAL;
985 
986 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
987 			return -EINVAL;
988 
989 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
990 			return -EINVAL;
991 
992 		if (kvm_vcpu_write_guest(vcpu,
993 					 gpa + i * sizeof(e) +
994 					     offsetof(struct vmx_msr_entry, value),
995 					 &data, sizeof(data))) {
996 			pr_debug_ratelimited(
997 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
998 				__func__, i, e.index, data);
999 			return -EINVAL;
1000 		}
1001 	}
1002 	return 0;
1003 }
1004 
1005 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1006 {
1007 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1008 	u32 count = vmcs12->vm_exit_msr_store_count;
1009 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1010 	struct vmx_msr_entry e;
1011 	u32 i;
1012 
1013 	for (i = 0; i < count; i++) {
1014 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1015 			return false;
1016 
1017 		if (e.index == msr_index)
1018 			return true;
1019 	}
1020 	return false;
1021 }
1022 
1023 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1024 					   u32 msr_index)
1025 {
1026 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1027 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1028 	bool in_vmcs12_store_list;
1029 	int msr_autostore_index;
1030 	bool in_autostore_list;
1031 	int last;
1032 
1033 	msr_autostore_index = vmx_find_msr_index(autostore, msr_index);
1034 	in_autostore_list = msr_autostore_index >= 0;
1035 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1036 
1037 	if (in_vmcs12_store_list && !in_autostore_list) {
1038 		if (autostore->nr == NR_LOADSTORE_MSRS) {
1039 			/*
1040 			 * Emulated VMEntry does not fail here.  Instead a less
1041 			 * accurate value will be returned by
1042 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1043 			 * instead of reading the value from the vmcs02 VMExit
1044 			 * MSR-store area.
1045 			 */
1046 			pr_warn_ratelimited(
1047 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1048 				msr_index);
1049 			return;
1050 		}
1051 		last = autostore->nr++;
1052 		autostore->val[last].index = msr_index;
1053 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1054 		last = --autostore->nr;
1055 		autostore->val[msr_autostore_index] = autostore->val[last];
1056 	}
1057 }
1058 
1059 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
1060 {
1061 	unsigned long invalid_mask;
1062 
1063 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1064 	return (val & invalid_mask) == 0;
1065 }
1066 
1067 /*
1068  * Returns true if the MMU needs to be sync'd on nested VM-Enter/VM-Exit.
1069  * tl;dr: the MMU needs a sync if L0 is using shadow paging and L1 didn't
1070  * enable VPID for L2 (implying it expects a TLB flush on VMX transitions).
1071  * Here's why.
1072  *
1073  * If EPT is enabled by L0 a sync is never needed:
1074  * - if it is disabled by L1, then L0 is not shadowing L1 or L2 PTEs, there
1075  *   cannot be unsync'd SPTEs for either L1 or L2.
1076  *
1077  * - if it is also enabled by L1, then L0 doesn't need to sync on VM-Enter
1078  *   VM-Enter as VM-Enter isn't required to invalidate guest-physical mappings
1079  *   (irrespective of VPID), i.e. L1 can't rely on the (virtual) CPU to flush
1080  *   stale guest-physical mappings for L2 from the TLB.  And as above, L0 isn't
1081  *   shadowing L1 PTEs so there are no unsync'd SPTEs to sync on VM-Exit.
1082  *
1083  * If EPT is disabled by L0:
1084  * - if VPID is enabled by L1 (for L2), the situation is similar to when L1
1085  *   enables EPT: L0 doesn't need to sync as VM-Enter and VM-Exit aren't
1086  *   required to invalidate linear mappings (EPT is disabled so there are
1087  *   no combined or guest-physical mappings), i.e. L1 can't rely on the
1088  *   (virtual) CPU to flush stale linear mappings for either L2 or itself (L1).
1089  *
1090  * - however if VPID is disabled by L1, then a sync is needed as L1 expects all
1091  *   linear mappings (EPT is disabled so there are no combined or guest-physical
1092  *   mappings) to be invalidated on both VM-Enter and VM-Exit.
1093  *
1094  * Note, this logic is subtly different than nested_has_guest_tlb_tag(), which
1095  * additionally checks that L2 has been assigned a VPID (when EPT is disabled).
1096  * Whether or not L2 has been assigned a VPID by L0 is irrelevant with respect
1097  * to L1's expectations, e.g. L0 needs to invalidate hardware TLB entries if L2
1098  * doesn't have a unique VPID to prevent reusing L1's entries (assuming L1 has
1099  * been assigned a VPID), but L0 doesn't need to do a MMU sync because L1
1100  * doesn't expect stale (virtual) TLB entries to be flushed, i.e. L1 doesn't
1101  * know that L0 will flush the TLB and so L1 will do INVVPID as needed to flush
1102  * stale TLB entries, at which point L0 will sync L2's MMU.
1103  */
1104 static bool nested_vmx_transition_mmu_sync(struct kvm_vcpu *vcpu)
1105 {
1106 	return !enable_ept && !nested_cpu_has_vpid(get_vmcs12(vcpu));
1107 }
1108 
1109 /*
1110  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1111  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1112  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1113  * @entry_failure_code.
1114  */
1115 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
1116 			       enum vm_entry_failure_code *entry_failure_code)
1117 {
1118 	if (CC(!nested_cr3_valid(vcpu, cr3))) {
1119 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1120 		return -EINVAL;
1121 	}
1122 
1123 	/*
1124 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1125 	 * must not be dereferenced.
1126 	 */
1127 	if (!nested_ept && is_pae_paging(vcpu) &&
1128 	    (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
1129 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1130 			*entry_failure_code = ENTRY_FAIL_PDPTE;
1131 			return -EINVAL;
1132 		}
1133 	}
1134 
1135 	/*
1136 	 * Unconditionally skip the TLB flush on fast CR3 switch, all TLB
1137 	 * flushes are handled by nested_vmx_transition_tlb_flush().  See
1138 	 * nested_vmx_transition_mmu_sync for details on skipping the MMU sync.
1139 	 */
1140 	if (!nested_ept)
1141 		kvm_mmu_new_pgd(vcpu, cr3, true,
1142 				!nested_vmx_transition_mmu_sync(vcpu));
1143 
1144 	vcpu->arch.cr3 = cr3;
1145 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1146 
1147 	kvm_init_mmu(vcpu, false);
1148 
1149 	return 0;
1150 }
1151 
1152 /*
1153  * Returns if KVM is able to config CPU to tag TLB entries
1154  * populated by L2 differently than TLB entries populated
1155  * by L1.
1156  *
1157  * If L0 uses EPT, L1 and L2 run with different EPTP because
1158  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1159  * are tagged with different EPTP.
1160  *
1161  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1162  * with different VPID (L1 entries are tagged with vmx->vpid
1163  * while L2 entries are tagged with vmx->nested.vpid02).
1164  */
1165 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1166 {
1167 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1168 
1169 	return enable_ept ||
1170 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1171 }
1172 
1173 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1174 					    struct vmcs12 *vmcs12,
1175 					    bool is_vmenter)
1176 {
1177 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1178 
1179 	/*
1180 	 * If VPID is disabled, linear and combined mappings are flushed on
1181 	 * VM-Enter/VM-Exit, and guest-physical mappings are valid only for
1182 	 * their associated EPTP.
1183 	 */
1184 	if (!enable_vpid)
1185 		return;
1186 
1187 	/*
1188 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1189 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit.
1190 	 *
1191 	 * If VPID is enabled and used by vmc12, but L2 does not have a unique
1192 	 * TLB tag (ASID), i.e. EPT is disabled and KVM was unable to allocate
1193 	 * a VPID for L2, flush the current context as the effective ASID is
1194 	 * common to both L1 and L2.
1195 	 *
1196 	 * Defer the flush so that it runs after vmcs02.EPTP has been set by
1197 	 * KVM_REQ_LOAD_MMU_PGD (if nested EPT is enabled) and to avoid
1198 	 * redundant flushes further down the nested pipeline.
1199 	 *
1200 	 * If a TLB flush isn't required due to any of the above, and vpid12 is
1201 	 * changing then the new "virtual" VPID (vpid12) will reuse the same
1202 	 * "real" VPID (vpid02), and so needs to be sync'd.  There is no direct
1203 	 * mapping between vpid02 and vpid12, vpid02 is per-vCPU and reused for
1204 	 * all nested vCPUs.
1205 	 */
1206 	if (!nested_cpu_has_vpid(vmcs12)) {
1207 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1208 	} else if (!nested_has_guest_tlb_tag(vcpu)) {
1209 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1210 	} else if (is_vmenter &&
1211 		   vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1212 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1213 		vpid_sync_context(nested_get_vpid02(vcpu));
1214 	}
1215 }
1216 
1217 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1218 {
1219 	superset &= mask;
1220 	subset &= mask;
1221 
1222 	return (superset | subset) == superset;
1223 }
1224 
1225 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1226 {
1227 	const u64 feature_and_reserved =
1228 		/* feature (except bit 48; see below) */
1229 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1230 		/* reserved */
1231 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1232 	u64 vmx_basic = vmx->nested.msrs.basic;
1233 
1234 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1235 		return -EINVAL;
1236 
1237 	/*
1238 	 * KVM does not emulate a version of VMX that constrains physical
1239 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1240 	 */
1241 	if (data & BIT_ULL(48))
1242 		return -EINVAL;
1243 
1244 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1245 	    vmx_basic_vmcs_revision_id(data))
1246 		return -EINVAL;
1247 
1248 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1249 		return -EINVAL;
1250 
1251 	vmx->nested.msrs.basic = data;
1252 	return 0;
1253 }
1254 
1255 static int
1256 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1257 {
1258 	u64 supported;
1259 	u32 *lowp, *highp;
1260 
1261 	switch (msr_index) {
1262 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1263 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1264 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1265 		break;
1266 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1267 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1268 		highp = &vmx->nested.msrs.procbased_ctls_high;
1269 		break;
1270 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1271 		lowp = &vmx->nested.msrs.exit_ctls_low;
1272 		highp = &vmx->nested.msrs.exit_ctls_high;
1273 		break;
1274 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1275 		lowp = &vmx->nested.msrs.entry_ctls_low;
1276 		highp = &vmx->nested.msrs.entry_ctls_high;
1277 		break;
1278 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1279 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1280 		highp = &vmx->nested.msrs.secondary_ctls_high;
1281 		break;
1282 	default:
1283 		BUG();
1284 	}
1285 
1286 	supported = vmx_control_msr(*lowp, *highp);
1287 
1288 	/* Check must-be-1 bits are still 1. */
1289 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1290 		return -EINVAL;
1291 
1292 	/* Check must-be-0 bits are still 0. */
1293 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1294 		return -EINVAL;
1295 
1296 	*lowp = data;
1297 	*highp = data >> 32;
1298 	return 0;
1299 }
1300 
1301 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1302 {
1303 	const u64 feature_and_reserved_bits =
1304 		/* feature */
1305 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1306 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1307 		/* reserved */
1308 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1309 	u64 vmx_misc;
1310 
1311 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1312 				   vmx->nested.msrs.misc_high);
1313 
1314 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1315 		return -EINVAL;
1316 
1317 	if ((vmx->nested.msrs.pinbased_ctls_high &
1318 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1319 	    vmx_misc_preemption_timer_rate(data) !=
1320 	    vmx_misc_preemption_timer_rate(vmx_misc))
1321 		return -EINVAL;
1322 
1323 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1324 		return -EINVAL;
1325 
1326 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1327 		return -EINVAL;
1328 
1329 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1330 		return -EINVAL;
1331 
1332 	vmx->nested.msrs.misc_low = data;
1333 	vmx->nested.msrs.misc_high = data >> 32;
1334 
1335 	return 0;
1336 }
1337 
1338 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1339 {
1340 	u64 vmx_ept_vpid_cap;
1341 
1342 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1343 					   vmx->nested.msrs.vpid_caps);
1344 
1345 	/* Every bit is either reserved or a feature bit. */
1346 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1347 		return -EINVAL;
1348 
1349 	vmx->nested.msrs.ept_caps = data;
1350 	vmx->nested.msrs.vpid_caps = data >> 32;
1351 	return 0;
1352 }
1353 
1354 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1355 {
1356 	u64 *msr;
1357 
1358 	switch (msr_index) {
1359 	case MSR_IA32_VMX_CR0_FIXED0:
1360 		msr = &vmx->nested.msrs.cr0_fixed0;
1361 		break;
1362 	case MSR_IA32_VMX_CR4_FIXED0:
1363 		msr = &vmx->nested.msrs.cr4_fixed0;
1364 		break;
1365 	default:
1366 		BUG();
1367 	}
1368 
1369 	/*
1370 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1371 	 * must be 1 in the restored value.
1372 	 */
1373 	if (!is_bitwise_subset(data, *msr, -1ULL))
1374 		return -EINVAL;
1375 
1376 	*msr = data;
1377 	return 0;
1378 }
1379 
1380 /*
1381  * Called when userspace is restoring VMX MSRs.
1382  *
1383  * Returns 0 on success, non-0 otherwise.
1384  */
1385 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1386 {
1387 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1388 
1389 	/*
1390 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1391 	 * is in VMX operation.
1392 	 */
1393 	if (vmx->nested.vmxon)
1394 		return -EBUSY;
1395 
1396 	switch (msr_index) {
1397 	case MSR_IA32_VMX_BASIC:
1398 		return vmx_restore_vmx_basic(vmx, data);
1399 	case MSR_IA32_VMX_PINBASED_CTLS:
1400 	case MSR_IA32_VMX_PROCBASED_CTLS:
1401 	case MSR_IA32_VMX_EXIT_CTLS:
1402 	case MSR_IA32_VMX_ENTRY_CTLS:
1403 		/*
1404 		 * The "non-true" VMX capability MSRs are generated from the
1405 		 * "true" MSRs, so we do not support restoring them directly.
1406 		 *
1407 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1408 		 * should restore the "true" MSRs with the must-be-1 bits
1409 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1410 		 * DEFAULT SETTINGS".
1411 		 */
1412 		return -EINVAL;
1413 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1414 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1415 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1416 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1417 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1418 		return vmx_restore_control_msr(vmx, msr_index, data);
1419 	case MSR_IA32_VMX_MISC:
1420 		return vmx_restore_vmx_misc(vmx, data);
1421 	case MSR_IA32_VMX_CR0_FIXED0:
1422 	case MSR_IA32_VMX_CR4_FIXED0:
1423 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1424 	case MSR_IA32_VMX_CR0_FIXED1:
1425 	case MSR_IA32_VMX_CR4_FIXED1:
1426 		/*
1427 		 * These MSRs are generated based on the vCPU's CPUID, so we
1428 		 * do not support restoring them directly.
1429 		 */
1430 		return -EINVAL;
1431 	case MSR_IA32_VMX_EPT_VPID_CAP:
1432 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1433 	case MSR_IA32_VMX_VMCS_ENUM:
1434 		vmx->nested.msrs.vmcs_enum = data;
1435 		return 0;
1436 	case MSR_IA32_VMX_VMFUNC:
1437 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1438 			return -EINVAL;
1439 		vmx->nested.msrs.vmfunc_controls = data;
1440 		return 0;
1441 	default:
1442 		/*
1443 		 * The rest of the VMX capability MSRs do not support restore.
1444 		 */
1445 		return -EINVAL;
1446 	}
1447 }
1448 
1449 /* Returns 0 on success, non-0 otherwise. */
1450 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1451 {
1452 	switch (msr_index) {
1453 	case MSR_IA32_VMX_BASIC:
1454 		*pdata = msrs->basic;
1455 		break;
1456 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1457 	case MSR_IA32_VMX_PINBASED_CTLS:
1458 		*pdata = vmx_control_msr(
1459 			msrs->pinbased_ctls_low,
1460 			msrs->pinbased_ctls_high);
1461 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1462 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1463 		break;
1464 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1465 	case MSR_IA32_VMX_PROCBASED_CTLS:
1466 		*pdata = vmx_control_msr(
1467 			msrs->procbased_ctls_low,
1468 			msrs->procbased_ctls_high);
1469 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1470 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1471 		break;
1472 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1473 	case MSR_IA32_VMX_EXIT_CTLS:
1474 		*pdata = vmx_control_msr(
1475 			msrs->exit_ctls_low,
1476 			msrs->exit_ctls_high);
1477 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1478 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1479 		break;
1480 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1481 	case MSR_IA32_VMX_ENTRY_CTLS:
1482 		*pdata = vmx_control_msr(
1483 			msrs->entry_ctls_low,
1484 			msrs->entry_ctls_high);
1485 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1486 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1487 		break;
1488 	case MSR_IA32_VMX_MISC:
1489 		*pdata = vmx_control_msr(
1490 			msrs->misc_low,
1491 			msrs->misc_high);
1492 		break;
1493 	case MSR_IA32_VMX_CR0_FIXED0:
1494 		*pdata = msrs->cr0_fixed0;
1495 		break;
1496 	case MSR_IA32_VMX_CR0_FIXED1:
1497 		*pdata = msrs->cr0_fixed1;
1498 		break;
1499 	case MSR_IA32_VMX_CR4_FIXED0:
1500 		*pdata = msrs->cr4_fixed0;
1501 		break;
1502 	case MSR_IA32_VMX_CR4_FIXED1:
1503 		*pdata = msrs->cr4_fixed1;
1504 		break;
1505 	case MSR_IA32_VMX_VMCS_ENUM:
1506 		*pdata = msrs->vmcs_enum;
1507 		break;
1508 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1509 		*pdata = vmx_control_msr(
1510 			msrs->secondary_ctls_low,
1511 			msrs->secondary_ctls_high);
1512 		break;
1513 	case MSR_IA32_VMX_EPT_VPID_CAP:
1514 		*pdata = msrs->ept_caps |
1515 			((u64)msrs->vpid_caps << 32);
1516 		break;
1517 	case MSR_IA32_VMX_VMFUNC:
1518 		*pdata = msrs->vmfunc_controls;
1519 		break;
1520 	default:
1521 		return 1;
1522 	}
1523 
1524 	return 0;
1525 }
1526 
1527 /*
1528  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1529  * been modified by the L1 guest.  Note, "writable" in this context means
1530  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1531  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1532  * VM-exit information fields (which are actually writable if the vCPU is
1533  * configured to support "VMWRITE to any supported field in the VMCS").
1534  */
1535 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1536 {
1537 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1538 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1539 	struct shadow_vmcs_field field;
1540 	unsigned long val;
1541 	int i;
1542 
1543 	if (WARN_ON(!shadow_vmcs))
1544 		return;
1545 
1546 	preempt_disable();
1547 
1548 	vmcs_load(shadow_vmcs);
1549 
1550 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1551 		field = shadow_read_write_fields[i];
1552 		val = __vmcs_readl(field.encoding);
1553 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1554 	}
1555 
1556 	vmcs_clear(shadow_vmcs);
1557 	vmcs_load(vmx->loaded_vmcs->vmcs);
1558 
1559 	preempt_enable();
1560 }
1561 
1562 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1563 {
1564 	const struct shadow_vmcs_field *fields[] = {
1565 		shadow_read_write_fields,
1566 		shadow_read_only_fields
1567 	};
1568 	const int max_fields[] = {
1569 		max_shadow_read_write_fields,
1570 		max_shadow_read_only_fields
1571 	};
1572 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1573 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1574 	struct shadow_vmcs_field field;
1575 	unsigned long val;
1576 	int i, q;
1577 
1578 	if (WARN_ON(!shadow_vmcs))
1579 		return;
1580 
1581 	vmcs_load(shadow_vmcs);
1582 
1583 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1584 		for (i = 0; i < max_fields[q]; i++) {
1585 			field = fields[q][i];
1586 			val = vmcs12_read_any(vmcs12, field.encoding,
1587 					      field.offset);
1588 			__vmcs_writel(field.encoding, val);
1589 		}
1590 	}
1591 
1592 	vmcs_clear(shadow_vmcs);
1593 	vmcs_load(vmx->loaded_vmcs->vmcs);
1594 }
1595 
1596 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1597 {
1598 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1599 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1600 
1601 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1602 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1603 	vmcs12->guest_rip = evmcs->guest_rip;
1604 
1605 	if (unlikely(!(evmcs->hv_clean_fields &
1606 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1607 		vmcs12->guest_rsp = evmcs->guest_rsp;
1608 		vmcs12->guest_rflags = evmcs->guest_rflags;
1609 		vmcs12->guest_interruptibility_info =
1610 			evmcs->guest_interruptibility_info;
1611 	}
1612 
1613 	if (unlikely(!(evmcs->hv_clean_fields &
1614 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1615 		vmcs12->cpu_based_vm_exec_control =
1616 			evmcs->cpu_based_vm_exec_control;
1617 	}
1618 
1619 	if (unlikely(!(evmcs->hv_clean_fields &
1620 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1621 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1622 	}
1623 
1624 	if (unlikely(!(evmcs->hv_clean_fields &
1625 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1626 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1627 	}
1628 
1629 	if (unlikely(!(evmcs->hv_clean_fields &
1630 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1631 		vmcs12->vm_entry_intr_info_field =
1632 			evmcs->vm_entry_intr_info_field;
1633 		vmcs12->vm_entry_exception_error_code =
1634 			evmcs->vm_entry_exception_error_code;
1635 		vmcs12->vm_entry_instruction_len =
1636 			evmcs->vm_entry_instruction_len;
1637 	}
1638 
1639 	if (unlikely(!(evmcs->hv_clean_fields &
1640 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1641 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1642 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1643 		vmcs12->host_cr0 = evmcs->host_cr0;
1644 		vmcs12->host_cr3 = evmcs->host_cr3;
1645 		vmcs12->host_cr4 = evmcs->host_cr4;
1646 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1647 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1648 		vmcs12->host_rip = evmcs->host_rip;
1649 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1650 		vmcs12->host_es_selector = evmcs->host_es_selector;
1651 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1652 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1653 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1654 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1655 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1656 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1657 	}
1658 
1659 	if (unlikely(!(evmcs->hv_clean_fields &
1660 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1661 		vmcs12->pin_based_vm_exec_control =
1662 			evmcs->pin_based_vm_exec_control;
1663 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1664 		vmcs12->secondary_vm_exec_control =
1665 			evmcs->secondary_vm_exec_control;
1666 	}
1667 
1668 	if (unlikely(!(evmcs->hv_clean_fields &
1669 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1670 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1671 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1672 	}
1673 
1674 	if (unlikely(!(evmcs->hv_clean_fields &
1675 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1676 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1677 	}
1678 
1679 	if (unlikely(!(evmcs->hv_clean_fields &
1680 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1681 		vmcs12->guest_es_base = evmcs->guest_es_base;
1682 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1683 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1684 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1685 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1686 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1687 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1688 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1689 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1690 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1691 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1692 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1693 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1694 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1695 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1696 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1697 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1698 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1699 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1700 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1701 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1702 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1703 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1704 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1705 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1706 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1707 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1708 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1709 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1710 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1711 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1712 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1713 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1714 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1715 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1716 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1717 	}
1718 
1719 	if (unlikely(!(evmcs->hv_clean_fields &
1720 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1721 		vmcs12->tsc_offset = evmcs->tsc_offset;
1722 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1723 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1724 	}
1725 
1726 	if (unlikely(!(evmcs->hv_clean_fields &
1727 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1728 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1729 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1730 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1731 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1732 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1733 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1734 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1735 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1736 	}
1737 
1738 	if (unlikely(!(evmcs->hv_clean_fields &
1739 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1740 		vmcs12->host_fs_base = evmcs->host_fs_base;
1741 		vmcs12->host_gs_base = evmcs->host_gs_base;
1742 		vmcs12->host_tr_base = evmcs->host_tr_base;
1743 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1744 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1745 		vmcs12->host_rsp = evmcs->host_rsp;
1746 	}
1747 
1748 	if (unlikely(!(evmcs->hv_clean_fields &
1749 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1750 		vmcs12->ept_pointer = evmcs->ept_pointer;
1751 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1752 	}
1753 
1754 	if (unlikely(!(evmcs->hv_clean_fields &
1755 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1756 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1757 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1758 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1759 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1760 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1761 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1762 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1763 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1764 		vmcs12->guest_pending_dbg_exceptions =
1765 			evmcs->guest_pending_dbg_exceptions;
1766 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1767 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1768 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1769 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1770 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1771 	}
1772 
1773 	/*
1774 	 * Not used?
1775 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1776 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1777 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1778 	 * vmcs12->page_fault_error_code_mask =
1779 	 *		evmcs->page_fault_error_code_mask;
1780 	 * vmcs12->page_fault_error_code_match =
1781 	 *		evmcs->page_fault_error_code_match;
1782 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1783 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1784 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1785 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1786 	 */
1787 
1788 	/*
1789 	 * Read only fields:
1790 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1791 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1792 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1793 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1794 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1795 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1796 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1797 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1798 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1799 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1800 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1801 	 *
1802 	 * Not present in struct vmcs12:
1803 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1804 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1805 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1806 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1807 	 */
1808 
1809 	return 0;
1810 }
1811 
1812 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1813 {
1814 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1815 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1816 
1817 	/*
1818 	 * Should not be changed by KVM:
1819 	 *
1820 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1821 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1822 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1823 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1824 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1825 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1826 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1827 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1828 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1829 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1830 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1831 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1832 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1833 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1834 	 * evmcs->host_rip = vmcs12->host_rip;
1835 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1836 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1837 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1838 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1839 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1840 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1841 	 * evmcs->host_rsp = vmcs12->host_rsp;
1842 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1843 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1844 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1845 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1846 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1847 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1848 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1849 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1850 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1851 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1852 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1853 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1854 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1855 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1856 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1857 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1858 	 * evmcs->page_fault_error_code_mask =
1859 	 *		vmcs12->page_fault_error_code_mask;
1860 	 * evmcs->page_fault_error_code_match =
1861 	 *		vmcs12->page_fault_error_code_match;
1862 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1863 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1864 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1865 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1866 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1867 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1868 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1869 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1870 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1871 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1872 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1873 	 *
1874 	 * Not present in struct vmcs12:
1875 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1876 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1877 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1878 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1879 	 */
1880 
1881 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1882 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1883 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1884 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1885 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1886 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1887 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1888 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1889 
1890 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1891 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1892 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1893 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1894 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1895 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1896 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1897 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1898 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1899 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1900 
1901 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1902 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1903 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1904 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1905 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1906 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1907 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1908 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1909 
1910 	evmcs->guest_es_base = vmcs12->guest_es_base;
1911 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1912 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1913 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1914 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1915 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1916 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1917 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1918 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1919 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1920 
1921 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1922 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1923 
1924 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1925 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1926 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1927 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1928 
1929 	evmcs->guest_pending_dbg_exceptions =
1930 		vmcs12->guest_pending_dbg_exceptions;
1931 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1932 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1933 
1934 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1935 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1936 
1937 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1938 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1939 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1940 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1941 
1942 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1943 
1944 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1945 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1946 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1947 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1948 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1949 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1950 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1951 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1952 
1953 	evmcs->exit_qualification = vmcs12->exit_qualification;
1954 
1955 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1956 	evmcs->guest_rsp = vmcs12->guest_rsp;
1957 	evmcs->guest_rflags = vmcs12->guest_rflags;
1958 
1959 	evmcs->guest_interruptibility_info =
1960 		vmcs12->guest_interruptibility_info;
1961 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1962 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1963 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1964 	evmcs->vm_entry_exception_error_code =
1965 		vmcs12->vm_entry_exception_error_code;
1966 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1967 
1968 	evmcs->guest_rip = vmcs12->guest_rip;
1969 
1970 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1971 
1972 	return 0;
1973 }
1974 
1975 /*
1976  * This is an equivalent of the nested hypervisor executing the vmptrld
1977  * instruction.
1978  */
1979 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1980 	struct kvm_vcpu *vcpu, bool from_launch)
1981 {
1982 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1983 	bool evmcs_gpa_changed = false;
1984 	u64 evmcs_gpa;
1985 
1986 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1987 		return EVMPTRLD_DISABLED;
1988 
1989 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa))
1990 		return EVMPTRLD_DISABLED;
1991 
1992 	if (unlikely(!vmx->nested.hv_evmcs ||
1993 		     evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1994 		if (!vmx->nested.hv_evmcs)
1995 			vmx->nested.current_vmptr = -1ull;
1996 
1997 		nested_release_evmcs(vcpu);
1998 
1999 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2000 				 &vmx->nested.hv_evmcs_map))
2001 			return EVMPTRLD_ERROR;
2002 
2003 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2004 
2005 		/*
2006 		 * Currently, KVM only supports eVMCS version 1
2007 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2008 		 * value to first u32 field of eVMCS which should specify eVMCS
2009 		 * VersionNumber.
2010 		 *
2011 		 * Guest should be aware of supported eVMCS versions by host by
2012 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2013 		 * expected to set this CPUID leaf according to the value
2014 		 * returned in vmcs_version from nested_enable_evmcs().
2015 		 *
2016 		 * However, it turns out that Microsoft Hyper-V fails to comply
2017 		 * to their own invented interface: When Hyper-V use eVMCS, it
2018 		 * just sets first u32 field of eVMCS to revision_id specified
2019 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2020 		 * which is one of the supported versions specified in
2021 		 * CPUID.0x4000000A.EAX[0:15].
2022 		 *
2023 		 * To overcome Hyper-V bug, we accept here either a supported
2024 		 * eVMCS version or VMCS12 revision_id as valid values for first
2025 		 * u32 field of eVMCS.
2026 		 */
2027 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2028 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2029 			nested_release_evmcs(vcpu);
2030 			return EVMPTRLD_VMFAIL;
2031 		}
2032 
2033 		vmx->nested.dirty_vmcs12 = true;
2034 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2035 
2036 		evmcs_gpa_changed = true;
2037 		/*
2038 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2039 		 * reloaded from guest's memory (read only fields, fields not
2040 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2041 		 * are no leftovers.
2042 		 */
2043 		if (from_launch) {
2044 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2045 			memset(vmcs12, 0, sizeof(*vmcs12));
2046 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2047 		}
2048 
2049 	}
2050 
2051 	/*
2052 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2053 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2054 	 */
2055 	if (from_launch || evmcs_gpa_changed)
2056 		vmx->nested.hv_evmcs->hv_clean_fields &=
2057 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2058 
2059 	return EVMPTRLD_SUCCEEDED;
2060 }
2061 
2062 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2063 {
2064 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2065 
2066 	if (vmx->nested.hv_evmcs) {
2067 		copy_vmcs12_to_enlightened(vmx);
2068 		/* All fields are clean */
2069 		vmx->nested.hv_evmcs->hv_clean_fields |=
2070 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2071 	} else {
2072 		copy_vmcs12_to_shadow(vmx);
2073 	}
2074 
2075 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2076 }
2077 
2078 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2079 {
2080 	struct vcpu_vmx *vmx =
2081 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2082 
2083 	vmx->nested.preemption_timer_expired = true;
2084 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2085 	kvm_vcpu_kick(&vmx->vcpu);
2086 
2087 	return HRTIMER_NORESTART;
2088 }
2089 
2090 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2091 {
2092 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2093 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2094 
2095 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2096 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2097 
2098 	if (!vmx->nested.has_preemption_timer_deadline) {
2099 		vmx->nested.preemption_timer_deadline =
2100 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2101 		vmx->nested.has_preemption_timer_deadline = true;
2102 	}
2103 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2104 }
2105 
2106 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2107 					u64 preemption_timeout)
2108 {
2109 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2110 
2111 	/*
2112 	 * A timer value of zero is architecturally guaranteed to cause
2113 	 * a VMExit prior to executing any instructions in the guest.
2114 	 */
2115 	if (preemption_timeout == 0) {
2116 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2117 		return;
2118 	}
2119 
2120 	if (vcpu->arch.virtual_tsc_khz == 0)
2121 		return;
2122 
2123 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2124 	preemption_timeout *= 1000000;
2125 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2126 	hrtimer_start(&vmx->nested.preemption_timer,
2127 		      ktime_add_ns(ktime_get(), preemption_timeout),
2128 		      HRTIMER_MODE_ABS_PINNED);
2129 }
2130 
2131 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2132 {
2133 	if (vmx->nested.nested_run_pending &&
2134 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2135 		return vmcs12->guest_ia32_efer;
2136 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2137 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2138 	else
2139 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2140 }
2141 
2142 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2143 {
2144 	/*
2145 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2146 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2147 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2148 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2149 	 */
2150 	if (vmx->nested.vmcs02_initialized)
2151 		return;
2152 	vmx->nested.vmcs02_initialized = true;
2153 
2154 	/*
2155 	 * We don't care what the EPTP value is we just need to guarantee
2156 	 * it's valid so we don't get a false positive when doing early
2157 	 * consistency checks.
2158 	 */
2159 	if (enable_ept && nested_early_check)
2160 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
2161 
2162 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2163 	if (cpu_has_vmx_vmfunc())
2164 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2165 
2166 	if (cpu_has_vmx_posted_intr())
2167 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2168 
2169 	if (cpu_has_vmx_msr_bitmap())
2170 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2171 
2172 	/*
2173 	 * The PML address never changes, so it is constant in vmcs02.
2174 	 * Conceptually we want to copy the PML index from vmcs01 here,
2175 	 * and then back to vmcs01 on nested vmexit.  But since we flush
2176 	 * the log and reset GUEST_PML_INDEX on each vmexit, the PML
2177 	 * index is also effectively constant in vmcs02.
2178 	 */
2179 	if (enable_pml) {
2180 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
2181 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2182 	}
2183 
2184 	if (cpu_has_vmx_encls_vmexit())
2185 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2186 
2187 	/*
2188 	 * Set the MSR load/store lists to match L0's settings.  Only the
2189 	 * addresses are constant (for vmcs02), the counts can change based
2190 	 * on L2's behavior, e.g. switching to/from long mode.
2191 	 */
2192 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2193 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2194 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2195 
2196 	vmx_set_constant_host_state(vmx);
2197 }
2198 
2199 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2200 				      struct vmcs12 *vmcs12)
2201 {
2202 	prepare_vmcs02_constant_state(vmx);
2203 
2204 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2205 
2206 	if (enable_vpid) {
2207 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2208 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2209 		else
2210 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2211 	}
2212 }
2213 
2214 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2215 {
2216 	u32 exec_control, vmcs12_exec_ctrl;
2217 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2218 
2219 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
2220 		prepare_vmcs02_early_rare(vmx, vmcs12);
2221 
2222 	/*
2223 	 * PIN CONTROLS
2224 	 */
2225 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2226 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2227 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2228 
2229 	/* Posted interrupts setting is only taken from vmcs12.  */
2230 	if (nested_cpu_has_posted_intr(vmcs12)) {
2231 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2232 		vmx->nested.pi_pending = false;
2233 	} else {
2234 		exec_control &= ~PIN_BASED_POSTED_INTR;
2235 	}
2236 	pin_controls_set(vmx, exec_control);
2237 
2238 	/*
2239 	 * EXEC CONTROLS
2240 	 */
2241 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2242 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2243 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2244 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2245 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2246 
2247 	vmx->nested.l1_tpr_threshold = -1;
2248 	if (exec_control & CPU_BASED_TPR_SHADOW)
2249 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2250 #ifdef CONFIG_X86_64
2251 	else
2252 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2253 				CPU_BASED_CR8_STORE_EXITING;
2254 #endif
2255 
2256 	/*
2257 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2258 	 * for I/O port accesses.
2259 	 */
2260 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2261 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2262 
2263 	/*
2264 	 * This bit will be computed in nested_get_vmcs12_pages, because
2265 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2266 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2267 	 * only set/clear this bit.
2268 	 */
2269 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2270 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2271 
2272 	exec_controls_set(vmx, exec_control);
2273 
2274 	/*
2275 	 * SECONDARY EXEC CONTROLS
2276 	 */
2277 	if (cpu_has_secondary_exec_ctrls()) {
2278 		exec_control = vmx->secondary_exec_control;
2279 
2280 		/* Take the following fields only from vmcs12 */
2281 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2282 				  SECONDARY_EXEC_ENABLE_INVPCID |
2283 				  SECONDARY_EXEC_RDTSCP |
2284 				  SECONDARY_EXEC_XSAVES |
2285 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2286 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2287 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2288 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2289 		if (nested_cpu_has(vmcs12,
2290 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2291 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2292 				~SECONDARY_EXEC_ENABLE_PML;
2293 			exec_control |= vmcs12_exec_ctrl;
2294 		}
2295 
2296 		/* VMCS shadowing for L2 is emulated for now */
2297 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2298 
2299 		/*
2300 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2301 		 * will not have to rewrite the controls just for this bit.
2302 		 */
2303 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2304 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2305 			exec_control |= SECONDARY_EXEC_DESC;
2306 
2307 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2308 			vmcs_write16(GUEST_INTR_STATUS,
2309 				vmcs12->guest_intr_status);
2310 
2311 		secondary_exec_controls_set(vmx, exec_control);
2312 	}
2313 
2314 	/*
2315 	 * ENTRY CONTROLS
2316 	 *
2317 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2318 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2319 	 * on the related bits (if supported by the CPU) in the hope that
2320 	 * we can avoid VMWrites during vmx_set_efer().
2321 	 */
2322 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2323 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2324 	if (cpu_has_load_ia32_efer()) {
2325 		if (guest_efer & EFER_LMA)
2326 			exec_control |= VM_ENTRY_IA32E_MODE;
2327 		if (guest_efer != host_efer)
2328 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2329 	}
2330 	vm_entry_controls_set(vmx, exec_control);
2331 
2332 	/*
2333 	 * EXIT CONTROLS
2334 	 *
2335 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2336 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2337 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2338 	 */
2339 	exec_control = vmx_vmexit_ctrl();
2340 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2341 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2342 	vm_exit_controls_set(vmx, exec_control);
2343 
2344 	/*
2345 	 * Interrupt/Exception Fields
2346 	 */
2347 	if (vmx->nested.nested_run_pending) {
2348 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2349 			     vmcs12->vm_entry_intr_info_field);
2350 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2351 			     vmcs12->vm_entry_exception_error_code);
2352 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2353 			     vmcs12->vm_entry_instruction_len);
2354 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2355 			     vmcs12->guest_interruptibility_info);
2356 		vmx->loaded_vmcs->nmi_known_unmasked =
2357 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2358 	} else {
2359 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2360 	}
2361 }
2362 
2363 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2364 {
2365 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2366 
2367 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2368 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2369 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2370 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2371 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2372 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2373 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2374 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2375 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2376 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2377 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2378 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2379 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2380 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2381 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2382 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2383 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2384 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2385 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2386 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2387 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2388 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2389 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2390 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2391 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2392 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2393 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2394 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2395 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2396 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2397 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2398 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2399 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2400 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2401 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2402 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2403 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2404 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2405 	}
2406 
2407 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2408 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2409 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2410 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2411 			    vmcs12->guest_pending_dbg_exceptions);
2412 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2413 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2414 
2415 		/*
2416 		 * L1 may access the L2's PDPTR, so save them to construct
2417 		 * vmcs12
2418 		 */
2419 		if (enable_ept) {
2420 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2421 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2422 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2423 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2424 		}
2425 
2426 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2427 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2428 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2429 	}
2430 
2431 	if (nested_cpu_has_xsaves(vmcs12))
2432 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2433 
2434 	/*
2435 	 * Whether page-faults are trapped is determined by a combination of
2436 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2437 	 * If enable_ept, L0 doesn't care about page faults and we should
2438 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2439 	 * care about (at least some) page faults, and because it is not easy
2440 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2441 	 * to exit on each and every L2 page fault. This is done by setting
2442 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2443 	 * Note that below we don't need special code to set EB.PF beyond the
2444 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2445 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2446 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2447 	 */
2448 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2449 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2450 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2451 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2452 
2453 	if (cpu_has_vmx_apicv()) {
2454 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2455 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2456 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2457 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2458 	}
2459 
2460 	/*
2461 	 * Make sure the msr_autostore list is up to date before we set the
2462 	 * count in the vmcs02.
2463 	 */
2464 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2465 
2466 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2467 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2468 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2469 
2470 	set_cr4_guest_host_mask(vmx);
2471 }
2472 
2473 /*
2474  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2475  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2476  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2477  * guest in a way that will both be appropriate to L1's requests, and our
2478  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2479  * function also has additional necessary side-effects, like setting various
2480  * vcpu->arch fields.
2481  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2482  * is assigned to entry_failure_code on failure.
2483  */
2484 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2485 			  enum vm_entry_failure_code *entry_failure_code)
2486 {
2487 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2488 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2489 	bool load_guest_pdptrs_vmcs12 = false;
2490 
2491 	if (vmx->nested.dirty_vmcs12 || hv_evmcs) {
2492 		prepare_vmcs02_rare(vmx, vmcs12);
2493 		vmx->nested.dirty_vmcs12 = false;
2494 
2495 		load_guest_pdptrs_vmcs12 = !hv_evmcs ||
2496 			!(hv_evmcs->hv_clean_fields &
2497 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2498 	}
2499 
2500 	if (vmx->nested.nested_run_pending &&
2501 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2502 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2503 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2504 	} else {
2505 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2506 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2507 	}
2508 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2509 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2510 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2511 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2512 
2513 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2514 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2515 	 * trap. Note that CR0.TS also needs updating - we do this later.
2516 	 */
2517 	update_exception_bitmap(vcpu);
2518 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2519 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2520 
2521 	if (vmx->nested.nested_run_pending &&
2522 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2523 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2524 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2525 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2526 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2527 	}
2528 
2529 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2530 
2531 	if (kvm_has_tsc_control)
2532 		decache_tsc_multiplier(vmx);
2533 
2534 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2535 
2536 	if (nested_cpu_has_ept(vmcs12))
2537 		nested_ept_init_mmu_context(vcpu);
2538 
2539 	/*
2540 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2541 	 * bits which we consider mandatory enabled.
2542 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2543 	 * the specifications by L1; It's not enough to take
2544 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2545 	 * have more bits than L1 expected.
2546 	 */
2547 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2548 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2549 
2550 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2551 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2552 
2553 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2554 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2555 	vmx_set_efer(vcpu, vcpu->arch.efer);
2556 
2557 	/*
2558 	 * Guest state is invalid and unrestricted guest is disabled,
2559 	 * which means L1 attempted VMEntry to L2 with invalid state.
2560 	 * Fail the VMEntry.
2561 	 */
2562 	if (vmx->emulation_required) {
2563 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2564 		return -EINVAL;
2565 	}
2566 
2567 	/* Shadow page tables on either EPT or shadow page tables. */
2568 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2569 				entry_failure_code))
2570 		return -EINVAL;
2571 
2572 	/*
2573 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2574 	 * on nested VM-Exit, which can occur without actually running L2 and
2575 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2576 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2577 	 * transition to HLT instead of running L2.
2578 	 */
2579 	if (enable_ept)
2580 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2581 
2582 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2583 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2584 	    is_pae_paging(vcpu)) {
2585 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2586 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2587 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2588 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2589 	}
2590 
2591 	if (!enable_ept)
2592 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2593 
2594 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2595 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2596 				     vmcs12->guest_ia32_perf_global_ctrl)))
2597 		return -EINVAL;
2598 
2599 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2600 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2601 	return 0;
2602 }
2603 
2604 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2605 {
2606 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2607 	       nested_cpu_has_virtual_nmis(vmcs12)))
2608 		return -EINVAL;
2609 
2610 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2611 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2612 		return -EINVAL;
2613 
2614 	return 0;
2615 }
2616 
2617 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2618 {
2619 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2620 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2621 
2622 	/* Check for memory type validity */
2623 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2624 	case VMX_EPTP_MT_UC:
2625 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2626 			return false;
2627 		break;
2628 	case VMX_EPTP_MT_WB:
2629 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2630 			return false;
2631 		break;
2632 	default:
2633 		return false;
2634 	}
2635 
2636 	/* Page-walk levels validity. */
2637 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2638 	case VMX_EPTP_PWL_5:
2639 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2640 			return false;
2641 		break;
2642 	case VMX_EPTP_PWL_4:
2643 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2644 			return false;
2645 		break;
2646 	default:
2647 		return false;
2648 	}
2649 
2650 	/* Reserved bits should not be set */
2651 	if (CC(new_eptp >> maxphyaddr || ((new_eptp >> 7) & 0x1f)))
2652 		return false;
2653 
2654 	/* AD, if set, should be supported */
2655 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2656 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2657 			return false;
2658 	}
2659 
2660 	return true;
2661 }
2662 
2663 /*
2664  * Checks related to VM-Execution Control Fields
2665  */
2666 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2667                                               struct vmcs12 *vmcs12)
2668 {
2669 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2670 
2671 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2672 				   vmx->nested.msrs.pinbased_ctls_low,
2673 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2674 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2675 				   vmx->nested.msrs.procbased_ctls_low,
2676 				   vmx->nested.msrs.procbased_ctls_high)))
2677 		return -EINVAL;
2678 
2679 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2680 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2681 				   vmx->nested.msrs.secondary_ctls_low,
2682 				   vmx->nested.msrs.secondary_ctls_high)))
2683 		return -EINVAL;
2684 
2685 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2686 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2687 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2688 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2689 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2690 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2691 	    nested_vmx_check_nmi_controls(vmcs12) ||
2692 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2693 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2694 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2695 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2696 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2697 		return -EINVAL;
2698 
2699 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2700 	    nested_cpu_has_save_preemption_timer(vmcs12))
2701 		return -EINVAL;
2702 
2703 	if (nested_cpu_has_ept(vmcs12) &&
2704 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2705 		return -EINVAL;
2706 
2707 	if (nested_cpu_has_vmfunc(vmcs12)) {
2708 		if (CC(vmcs12->vm_function_control &
2709 		       ~vmx->nested.msrs.vmfunc_controls))
2710 			return -EINVAL;
2711 
2712 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2713 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2714 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2715 				return -EINVAL;
2716 		}
2717 	}
2718 
2719 	return 0;
2720 }
2721 
2722 /*
2723  * Checks related to VM-Exit Control Fields
2724  */
2725 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2726                                          struct vmcs12 *vmcs12)
2727 {
2728 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2729 
2730 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2731 				    vmx->nested.msrs.exit_ctls_low,
2732 				    vmx->nested.msrs.exit_ctls_high)) ||
2733 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2734 		return -EINVAL;
2735 
2736 	return 0;
2737 }
2738 
2739 /*
2740  * Checks related to VM-Entry Control Fields
2741  */
2742 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2743 					  struct vmcs12 *vmcs12)
2744 {
2745 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2746 
2747 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2748 				    vmx->nested.msrs.entry_ctls_low,
2749 				    vmx->nested.msrs.entry_ctls_high)))
2750 		return -EINVAL;
2751 
2752 	/*
2753 	 * From the Intel SDM, volume 3:
2754 	 * Fields relevant to VM-entry event injection must be set properly.
2755 	 * These fields are the VM-entry interruption-information field, the
2756 	 * VM-entry exception error code, and the VM-entry instruction length.
2757 	 */
2758 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2759 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2760 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2761 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2762 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2763 		bool should_have_error_code;
2764 		bool urg = nested_cpu_has2(vmcs12,
2765 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2766 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2767 
2768 		/* VM-entry interruption-info field: interruption type */
2769 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2770 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2771 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2772 			return -EINVAL;
2773 
2774 		/* VM-entry interruption-info field: vector */
2775 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2776 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2777 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2778 			return -EINVAL;
2779 
2780 		/* VM-entry interruption-info field: deliver error code */
2781 		should_have_error_code =
2782 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2783 			x86_exception_has_error_code(vector);
2784 		if (CC(has_error_code != should_have_error_code))
2785 			return -EINVAL;
2786 
2787 		/* VM-entry exception error code */
2788 		if (CC(has_error_code &&
2789 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2790 			return -EINVAL;
2791 
2792 		/* VM-entry interruption-info field: reserved bits */
2793 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2794 			return -EINVAL;
2795 
2796 		/* VM-entry instruction length */
2797 		switch (intr_type) {
2798 		case INTR_TYPE_SOFT_EXCEPTION:
2799 		case INTR_TYPE_SOFT_INTR:
2800 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2801 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2802 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2803 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2804 				return -EINVAL;
2805 		}
2806 	}
2807 
2808 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2809 		return -EINVAL;
2810 
2811 	return 0;
2812 }
2813 
2814 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2815 				     struct vmcs12 *vmcs12)
2816 {
2817 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2818 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2819 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2820 		return -EINVAL;
2821 
2822 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2823 		return nested_evmcs_check_controls(vmcs12);
2824 
2825 	return 0;
2826 }
2827 
2828 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2829 				       struct vmcs12 *vmcs12)
2830 {
2831 	bool ia32e;
2832 
2833 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2834 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2835 	    CC(!nested_cr3_valid(vcpu, vmcs12->host_cr3)))
2836 		return -EINVAL;
2837 
2838 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2839 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2840 		return -EINVAL;
2841 
2842 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2843 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2844 		return -EINVAL;
2845 
2846 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2847 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2848 					   vmcs12->host_ia32_perf_global_ctrl)))
2849 		return -EINVAL;
2850 
2851 #ifdef CONFIG_X86_64
2852 	ia32e = !!(vcpu->arch.efer & EFER_LMA);
2853 #else
2854 	ia32e = false;
2855 #endif
2856 
2857 	if (ia32e) {
2858 		if (CC(!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)) ||
2859 		    CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2860 			return -EINVAL;
2861 	} else {
2862 		if (CC(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) ||
2863 		    CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2864 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2865 		    CC((vmcs12->host_rip) >> 32))
2866 			return -EINVAL;
2867 	}
2868 
2869 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2870 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2871 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2872 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2873 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2874 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2875 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2876 	    CC(vmcs12->host_cs_selector == 0) ||
2877 	    CC(vmcs12->host_tr_selector == 0) ||
2878 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2879 		return -EINVAL;
2880 
2881 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2882 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2883 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2884 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2885 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2886 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2887 		return -EINVAL;
2888 
2889 	/*
2890 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2891 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2892 	 * the values of the LMA and LME bits in the field must each be that of
2893 	 * the host address-space size VM-exit control.
2894 	 */
2895 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2896 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2897 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2898 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2899 			return -EINVAL;
2900 	}
2901 
2902 	return 0;
2903 }
2904 
2905 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2906 					  struct vmcs12 *vmcs12)
2907 {
2908 	int r = 0;
2909 	struct vmcs12 *shadow;
2910 	struct kvm_host_map map;
2911 
2912 	if (vmcs12->vmcs_link_pointer == -1ull)
2913 		return 0;
2914 
2915 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2916 		return -EINVAL;
2917 
2918 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2919 		return -EINVAL;
2920 
2921 	shadow = map.hva;
2922 
2923 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2924 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2925 		r = -EINVAL;
2926 
2927 	kvm_vcpu_unmap(vcpu, &map, false);
2928 	return r;
2929 }
2930 
2931 /*
2932  * Checks related to Guest Non-register State
2933  */
2934 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2935 {
2936 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2937 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT))
2938 		return -EINVAL;
2939 
2940 	return 0;
2941 }
2942 
2943 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2944 					struct vmcs12 *vmcs12,
2945 					enum vm_entry_failure_code *entry_failure_code)
2946 {
2947 	bool ia32e;
2948 
2949 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
2950 
2951 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2952 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2953 		return -EINVAL;
2954 
2955 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2956 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2957 		return -EINVAL;
2958 
2959 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2960 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2961 		return -EINVAL;
2962 
2963 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2964 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
2965 		return -EINVAL;
2966 	}
2967 
2968 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2969 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2970 					   vmcs12->guest_ia32_perf_global_ctrl)))
2971 		return -EINVAL;
2972 
2973 	/*
2974 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2975 	 * are performed on the field for the IA32_EFER MSR:
2976 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2977 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2978 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2979 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2980 	 *   CR0.PG) is 1.
2981 	 */
2982 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2983 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2984 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2985 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
2986 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
2987 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
2988 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
2989 			return -EINVAL;
2990 	}
2991 
2992 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2993 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
2994 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
2995 		return -EINVAL;
2996 
2997 	if (nested_check_guest_non_reg_state(vmcs12))
2998 		return -EINVAL;
2999 
3000 	return 0;
3001 }
3002 
3003 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3004 {
3005 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3006 	unsigned long cr3, cr4;
3007 	bool vm_fail;
3008 
3009 	if (!nested_early_check)
3010 		return 0;
3011 
3012 	if (vmx->msr_autoload.host.nr)
3013 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3014 	if (vmx->msr_autoload.guest.nr)
3015 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3016 
3017 	preempt_disable();
3018 
3019 	vmx_prepare_switch_to_guest(vcpu);
3020 
3021 	/*
3022 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3023 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3024 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3025 	 * there is no need to preserve other bits or save/restore the field.
3026 	 */
3027 	vmcs_writel(GUEST_RFLAGS, 0);
3028 
3029 	cr3 = __get_current_cr3_fast();
3030 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3031 		vmcs_writel(HOST_CR3, cr3);
3032 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3033 	}
3034 
3035 	cr4 = cr4_read_shadow();
3036 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3037 		vmcs_writel(HOST_CR4, cr4);
3038 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3039 	}
3040 
3041 	asm(
3042 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
3043 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
3044 		"je 1f \n\t"
3045 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
3046 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
3047 		"1: \n\t"
3048 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
3049 
3050 		/* Check if vmlaunch or vmresume is needed */
3051 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
3052 
3053 		/*
3054 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
3055 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
3056 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
3057 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
3058 		 */
3059 		"call vmx_vmenter\n\t"
3060 
3061 		CC_SET(be)
3062 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
3063 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
3064 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
3065 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
3066 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
3067 		[wordsize]"i"(sizeof(ulong))
3068 	      : "memory"
3069 	);
3070 
3071 	if (vmx->msr_autoload.host.nr)
3072 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3073 	if (vmx->msr_autoload.guest.nr)
3074 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3075 
3076 	if (vm_fail) {
3077 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3078 
3079 		preempt_enable();
3080 
3081 		trace_kvm_nested_vmenter_failed(
3082 			"early hardware check VM-instruction error: ", error);
3083 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3084 		return 1;
3085 	}
3086 
3087 	/*
3088 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3089 	 */
3090 	if (hw_breakpoint_active())
3091 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3092 	local_irq_enable();
3093 	preempt_enable();
3094 
3095 	/*
3096 	 * A non-failing VMEntry means we somehow entered guest mode with
3097 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3098 	 * is no telling what memory has been modified or what state has
3099 	 * been exposed to unknown code.  Hitting this all but guarantees
3100 	 * a (very critical) hardware issue.
3101 	 */
3102 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3103 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3104 
3105 	return 0;
3106 }
3107 
3108 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3109 {
3110 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3111 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3112 	struct kvm_host_map *map;
3113 	struct page *page;
3114 	u64 hpa;
3115 
3116 	/*
3117 	 * hv_evmcs may end up being not mapped after migration (when
3118 	 * L2 was running), map it here to make sure vmcs12 changes are
3119 	 * properly reflected.
3120 	 */
3121 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs) {
3122 		enum nested_evmptrld_status evmptrld_status =
3123 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3124 
3125 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3126 		    evmptrld_status == EVMPTRLD_ERROR) {
3127 			pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3128 					     __func__);
3129 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3130 			vcpu->run->internal.suberror =
3131 				KVM_INTERNAL_ERROR_EMULATION;
3132 			vcpu->run->internal.ndata = 0;
3133 			return false;
3134 		}
3135 	}
3136 
3137 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3138 		/*
3139 		 * Translate L1 physical address to host physical
3140 		 * address for vmcs02. Keep the page pinned, so this
3141 		 * physical address remains valid. We keep a reference
3142 		 * to it so we can release it later.
3143 		 */
3144 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3145 			kvm_release_page_clean(vmx->nested.apic_access_page);
3146 			vmx->nested.apic_access_page = NULL;
3147 		}
3148 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3149 		if (!is_error_page(page)) {
3150 			vmx->nested.apic_access_page = page;
3151 			hpa = page_to_phys(vmx->nested.apic_access_page);
3152 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3153 		} else {
3154 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3155 					     __func__);
3156 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3157 			vcpu->run->internal.suberror =
3158 				KVM_INTERNAL_ERROR_EMULATION;
3159 			vcpu->run->internal.ndata = 0;
3160 			return false;
3161 		}
3162 	}
3163 
3164 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3165 		map = &vmx->nested.virtual_apic_map;
3166 
3167 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3168 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3169 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3170 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3171 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3172 			/*
3173 			 * The processor will never use the TPR shadow, simply
3174 			 * clear the bit from the execution control.  Such a
3175 			 * configuration is useless, but it happens in tests.
3176 			 * For any other configuration, failing the vm entry is
3177 			 * _not_ what the processor does but it's basically the
3178 			 * only possibility we have.
3179 			 */
3180 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3181 		} else {
3182 			/*
3183 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3184 			 * force VM-Entry to fail.
3185 			 */
3186 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
3187 		}
3188 	}
3189 
3190 	if (nested_cpu_has_posted_intr(vmcs12)) {
3191 		map = &vmx->nested.pi_desc_map;
3192 
3193 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3194 			vmx->nested.pi_desc =
3195 				(struct pi_desc *)(((void *)map->hva) +
3196 				offset_in_page(vmcs12->posted_intr_desc_addr));
3197 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3198 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3199 		}
3200 	}
3201 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3202 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3203 	else
3204 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3205 	return true;
3206 }
3207 
3208 /*
3209  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3210  * for running VMX instructions (except VMXON, whose prerequisites are
3211  * slightly different). It also specifies what exception to inject otherwise.
3212  * Note that many of these exceptions have priority over VM exits, so they
3213  * don't have to be checked again here.
3214  */
3215 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3216 {
3217 	if (!to_vmx(vcpu)->nested.vmxon) {
3218 		kvm_queue_exception(vcpu, UD_VECTOR);
3219 		return 0;
3220 	}
3221 
3222 	if (vmx_get_cpl(vcpu)) {
3223 		kvm_inject_gp(vcpu, 0);
3224 		return 0;
3225 	}
3226 
3227 	return 1;
3228 }
3229 
3230 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3231 {
3232 	u8 rvi = vmx_get_rvi();
3233 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3234 
3235 	return ((rvi & 0xf0) > (vppr & 0xf0));
3236 }
3237 
3238 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3239 				   struct vmcs12 *vmcs12);
3240 
3241 /*
3242  * If from_vmentry is false, this is being called from state restore (either RSM
3243  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3244  *
3245  * Returns:
3246  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3247  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3248  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3249  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3250  */
3251 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3252 							bool from_vmentry)
3253 {
3254 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3255 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3256 	enum vm_entry_failure_code entry_failure_code;
3257 	bool evaluate_pending_interrupts;
3258 	u32 exit_reason, failed_index;
3259 
3260 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3261 		kvm_vcpu_flush_tlb_current(vcpu);
3262 
3263 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3264 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3265 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3266 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3267 
3268 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3269 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3270 	if (kvm_mpx_supported() &&
3271 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3272 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3273 
3274 	/*
3275 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3276 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3277 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3278 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3279 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3280 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3281 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3282 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3283 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3284 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3285 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3286 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3287 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3288 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3289 	 */
3290 	if (!enable_ept && !nested_early_check)
3291 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3292 
3293 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3294 
3295 	prepare_vmcs02_early(vmx, vmcs12);
3296 
3297 	if (from_vmentry) {
3298 		if (unlikely(!nested_get_vmcs12_pages(vcpu)))
3299 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3300 
3301 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3302 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3303 			return NVMX_VMENTRY_VMFAIL;
3304 		}
3305 
3306 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3307 						 &entry_failure_code)) {
3308 			exit_reason = EXIT_REASON_INVALID_STATE;
3309 			vmcs12->exit_qualification = entry_failure_code;
3310 			goto vmentry_fail_vmexit;
3311 		}
3312 	}
3313 
3314 	enter_guest_mode(vcpu);
3315 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3316 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3317 
3318 	if (prepare_vmcs02(vcpu, vmcs12, &entry_failure_code)) {
3319 		exit_reason = EXIT_REASON_INVALID_STATE;
3320 		vmcs12->exit_qualification = entry_failure_code;
3321 		goto vmentry_fail_vmexit_guest_mode;
3322 	}
3323 
3324 	if (from_vmentry) {
3325 		failed_index = nested_vmx_load_msr(vcpu,
3326 						   vmcs12->vm_entry_msr_load_addr,
3327 						   vmcs12->vm_entry_msr_load_count);
3328 		if (failed_index) {
3329 			exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3330 			vmcs12->exit_qualification = failed_index;
3331 			goto vmentry_fail_vmexit_guest_mode;
3332 		}
3333 	} else {
3334 		/*
3335 		 * The MMU is not initialized to point at the right entities yet and
3336 		 * "get pages" would need to read data from the guest (i.e. we will
3337 		 * need to perform gpa to hpa translation). Request a call
3338 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3339 		 * have already been set at vmentry time and should not be reset.
3340 		 */
3341 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3342 	}
3343 
3344 	/*
3345 	 * If L1 had a pending IRQ/NMI until it executed
3346 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3347 	 * disallowed (e.g. interrupts disabled), L0 needs to
3348 	 * evaluate if this pending event should cause an exit from L2
3349 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3350 	 * intercept EXTERNAL_INTERRUPT).
3351 	 *
3352 	 * Usually this would be handled by the processor noticing an
3353 	 * IRQ/NMI window request, or checking RVI during evaluation of
3354 	 * pending virtual interrupts.  However, this setting was done
3355 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3356 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3357 	 */
3358 	if (unlikely(evaluate_pending_interrupts))
3359 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3360 
3361 	/*
3362 	 * Do not start the preemption timer hrtimer until after we know
3363 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3364 	 * the timer.
3365 	 */
3366 	vmx->nested.preemption_timer_expired = false;
3367 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3368 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3369 		vmx_start_preemption_timer(vcpu, timer_value);
3370 	}
3371 
3372 	/*
3373 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3374 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3375 	 * returned as far as L1 is concerned. It will only return (and set
3376 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3377 	 */
3378 	return NVMX_VMENTRY_SUCCESS;
3379 
3380 	/*
3381 	 * A failed consistency check that leads to a VMExit during L1's
3382 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3383 	 * 26.7 "VM-entry failures during or after loading guest state".
3384 	 */
3385 vmentry_fail_vmexit_guest_mode:
3386 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3387 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3388 	leave_guest_mode(vcpu);
3389 
3390 vmentry_fail_vmexit:
3391 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3392 
3393 	if (!from_vmentry)
3394 		return NVMX_VMENTRY_VMEXIT;
3395 
3396 	load_vmcs12_host_state(vcpu, vmcs12);
3397 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3398 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3399 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3400 	return NVMX_VMENTRY_VMEXIT;
3401 }
3402 
3403 /*
3404  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3405  * for running an L2 nested guest.
3406  */
3407 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3408 {
3409 	struct vmcs12 *vmcs12;
3410 	enum nvmx_vmentry_status status;
3411 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3412 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3413 	enum nested_evmptrld_status evmptrld_status;
3414 
3415 	if (!nested_vmx_check_permission(vcpu))
3416 		return 1;
3417 
3418 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3419 	if (evmptrld_status == EVMPTRLD_ERROR) {
3420 		kvm_queue_exception(vcpu, UD_VECTOR);
3421 		return 1;
3422 	} else if (evmptrld_status == EVMPTRLD_VMFAIL) {
3423 		return nested_vmx_failInvalid(vcpu);
3424 	}
3425 
3426 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3427 		return nested_vmx_failInvalid(vcpu);
3428 
3429 	vmcs12 = get_vmcs12(vcpu);
3430 
3431 	/*
3432 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3433 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3434 	 * rather than RFLAGS.ZF, and no error number is stored to the
3435 	 * VM-instruction error field.
3436 	 */
3437 	if (vmcs12->hdr.shadow_vmcs)
3438 		return nested_vmx_failInvalid(vcpu);
3439 
3440 	if (vmx->nested.hv_evmcs) {
3441 		copy_enlightened_to_vmcs12(vmx);
3442 		/* Enlightened VMCS doesn't have launch state */
3443 		vmcs12->launch_state = !launch;
3444 	} else if (enable_shadow_vmcs) {
3445 		copy_shadow_to_vmcs12(vmx);
3446 	}
3447 
3448 	/*
3449 	 * The nested entry process starts with enforcing various prerequisites
3450 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3451 	 * they fail: As the SDM explains, some conditions should cause the
3452 	 * instruction to fail, while others will cause the instruction to seem
3453 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3454 	 * To speed up the normal (success) code path, we should avoid checking
3455 	 * for misconfigurations which will anyway be caught by the processor
3456 	 * when using the merged vmcs02.
3457 	 */
3458 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3459 		return nested_vmx_failValid(vcpu,
3460 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3461 
3462 	if (vmcs12->launch_state == launch)
3463 		return nested_vmx_failValid(vcpu,
3464 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3465 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3466 
3467 	if (nested_vmx_check_controls(vcpu, vmcs12))
3468 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3469 
3470 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3471 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3472 
3473 	/*
3474 	 * We're finally done with prerequisite checking, and can start with
3475 	 * the nested entry.
3476 	 */
3477 	vmx->nested.nested_run_pending = 1;
3478 	vmx->nested.has_preemption_timer_deadline = false;
3479 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3480 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3481 		goto vmentry_failed;
3482 
3483 	/* Hide L1D cache contents from the nested guest.  */
3484 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3485 
3486 	/*
3487 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3488 	 * also be used as part of restoring nVMX state for
3489 	 * snapshot restore (migration).
3490 	 *
3491 	 * In this flow, it is assumed that vmcs12 cache was
3492 	 * trasferred as part of captured nVMX state and should
3493 	 * therefore not be read from guest memory (which may not
3494 	 * exist on destination host yet).
3495 	 */
3496 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3497 
3498 	/*
3499 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3500 	 * awakened by event injection or by an NMI-window VM-exit or
3501 	 * by an interrupt-window VM-exit, halt the vcpu.
3502 	 */
3503 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3504 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3505 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_NMI_WINDOW_EXITING) &&
3506 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_INTR_WINDOW_EXITING) &&
3507 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3508 		vmx->nested.nested_run_pending = 0;
3509 		return kvm_vcpu_halt(vcpu);
3510 	}
3511 	return 1;
3512 
3513 vmentry_failed:
3514 	vmx->nested.nested_run_pending = 0;
3515 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3516 		return 0;
3517 	if (status == NVMX_VMENTRY_VMEXIT)
3518 		return 1;
3519 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3520 	return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3521 }
3522 
3523 /*
3524  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3525  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3526  * This function returns the new value we should put in vmcs12.guest_cr0.
3527  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3528  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3529  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3530  *     didn't trap the bit, because if L1 did, so would L0).
3531  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3532  *     been modified by L2, and L1 knows it. So just leave the old value of
3533  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3534  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3535  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3536  *     changed these bits, and therefore they need to be updated, but L0
3537  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3538  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3539  */
3540 static inline unsigned long
3541 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3542 {
3543 	return
3544 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3545 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3546 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3547 			vcpu->arch.cr0_guest_owned_bits));
3548 }
3549 
3550 static inline unsigned long
3551 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3552 {
3553 	return
3554 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3555 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3556 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3557 			vcpu->arch.cr4_guest_owned_bits));
3558 }
3559 
3560 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3561 				      struct vmcs12 *vmcs12)
3562 {
3563 	u32 idt_vectoring;
3564 	unsigned int nr;
3565 
3566 	if (vcpu->arch.exception.injected) {
3567 		nr = vcpu->arch.exception.nr;
3568 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3569 
3570 		if (kvm_exception_is_soft(nr)) {
3571 			vmcs12->vm_exit_instruction_len =
3572 				vcpu->arch.event_exit_inst_len;
3573 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3574 		} else
3575 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3576 
3577 		if (vcpu->arch.exception.has_error_code) {
3578 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3579 			vmcs12->idt_vectoring_error_code =
3580 				vcpu->arch.exception.error_code;
3581 		}
3582 
3583 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3584 	} else if (vcpu->arch.nmi_injected) {
3585 		vmcs12->idt_vectoring_info_field =
3586 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3587 	} else if (vcpu->arch.interrupt.injected) {
3588 		nr = vcpu->arch.interrupt.nr;
3589 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3590 
3591 		if (vcpu->arch.interrupt.soft) {
3592 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3593 			vmcs12->vm_entry_instruction_len =
3594 				vcpu->arch.event_exit_inst_len;
3595 		} else
3596 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3597 
3598 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3599 	}
3600 }
3601 
3602 
3603 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3604 {
3605 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3606 	gfn_t gfn;
3607 
3608 	/*
3609 	 * Don't need to mark the APIC access page dirty; it is never
3610 	 * written to by the CPU during APIC virtualization.
3611 	 */
3612 
3613 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3614 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3615 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3616 	}
3617 
3618 	if (nested_cpu_has_posted_intr(vmcs12)) {
3619 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3620 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3621 	}
3622 }
3623 
3624 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3625 {
3626 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3627 	int max_irr;
3628 	void *vapic_page;
3629 	u16 status;
3630 
3631 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3632 		return;
3633 
3634 	vmx->nested.pi_pending = false;
3635 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3636 		return;
3637 
3638 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3639 	if (max_irr != 256) {
3640 		vapic_page = vmx->nested.virtual_apic_map.hva;
3641 		if (!vapic_page)
3642 			return;
3643 
3644 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3645 			vapic_page, &max_irr);
3646 		status = vmcs_read16(GUEST_INTR_STATUS);
3647 		if ((u8)max_irr > ((u8)status & 0xff)) {
3648 			status &= ~0xff;
3649 			status |= (u8)max_irr;
3650 			vmcs_write16(GUEST_INTR_STATUS, status);
3651 		}
3652 	}
3653 
3654 	nested_mark_vmcs12_pages_dirty(vcpu);
3655 }
3656 
3657 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3658 					       unsigned long exit_qual)
3659 {
3660 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3661 	unsigned int nr = vcpu->arch.exception.nr;
3662 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3663 
3664 	if (vcpu->arch.exception.has_error_code) {
3665 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3666 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3667 	}
3668 
3669 	if (kvm_exception_is_soft(nr))
3670 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3671 	else
3672 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3673 
3674 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3675 	    vmx_get_nmi_mask(vcpu))
3676 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3677 
3678 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3679 }
3680 
3681 /*
3682  * Returns true if a debug trap is pending delivery.
3683  *
3684  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3685  * exception may be inferred from the presence of an exception payload.
3686  */
3687 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3688 {
3689 	return vcpu->arch.exception.pending &&
3690 			vcpu->arch.exception.nr == DB_VECTOR &&
3691 			vcpu->arch.exception.payload;
3692 }
3693 
3694 /*
3695  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3696  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3697  * represents these debug traps with a payload that is said to be compatible
3698  * with the 'pending debug exceptions' field, write the payload to the VMCS
3699  * field if a VM-exit is delivered before the debug trap.
3700  */
3701 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3702 {
3703 	if (vmx_pending_dbg_trap(vcpu))
3704 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3705 			    vcpu->arch.exception.payload);
3706 }
3707 
3708 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3709 {
3710 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3711 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3712 }
3713 
3714 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3715 {
3716 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3717 	unsigned long exit_qual;
3718 	bool block_nested_events =
3719 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3720 	bool mtf_pending = vmx->nested.mtf_pending;
3721 	struct kvm_lapic *apic = vcpu->arch.apic;
3722 
3723 	/*
3724 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3725 	 * this state is discarded.
3726 	 */
3727 	if (!block_nested_events)
3728 		vmx->nested.mtf_pending = false;
3729 
3730 	if (lapic_in_kernel(vcpu) &&
3731 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3732 		if (block_nested_events)
3733 			return -EBUSY;
3734 		nested_vmx_update_pending_dbg(vcpu);
3735 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3736 		nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3737 		return 0;
3738 	}
3739 
3740 	/*
3741 	 * Process any exceptions that are not debug traps before MTF.
3742 	 */
3743 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3744 		if (block_nested_events)
3745 			return -EBUSY;
3746 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3747 			goto no_vmexit;
3748 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3749 		return 0;
3750 	}
3751 
3752 	if (mtf_pending) {
3753 		if (block_nested_events)
3754 			return -EBUSY;
3755 		nested_vmx_update_pending_dbg(vcpu);
3756 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3757 		return 0;
3758 	}
3759 
3760 	if (vcpu->arch.exception.pending) {
3761 		if (block_nested_events)
3762 			return -EBUSY;
3763 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3764 			goto no_vmexit;
3765 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3766 		return 0;
3767 	}
3768 
3769 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3770 		if (block_nested_events)
3771 			return -EBUSY;
3772 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3773 		return 0;
3774 	}
3775 
3776 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3777 		if (block_nested_events)
3778 			return -EBUSY;
3779 		goto no_vmexit;
3780 	}
3781 
3782 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3783 		if (block_nested_events)
3784 			return -EBUSY;
3785 		if (!nested_exit_on_nmi(vcpu))
3786 			goto no_vmexit;
3787 
3788 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3789 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3790 				  INTR_INFO_VALID_MASK, 0);
3791 		/*
3792 		 * The NMI-triggered VM exit counts as injection:
3793 		 * clear this one and block further NMIs.
3794 		 */
3795 		vcpu->arch.nmi_pending = 0;
3796 		vmx_set_nmi_mask(vcpu, true);
3797 		return 0;
3798 	}
3799 
3800 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3801 		if (block_nested_events)
3802 			return -EBUSY;
3803 		if (!nested_exit_on_intr(vcpu))
3804 			goto no_vmexit;
3805 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3806 		return 0;
3807 	}
3808 
3809 no_vmexit:
3810 	vmx_complete_nested_posted_interrupt(vcpu);
3811 	return 0;
3812 }
3813 
3814 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3815 {
3816 	ktime_t remaining =
3817 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3818 	u64 value;
3819 
3820 	if (ktime_to_ns(remaining) <= 0)
3821 		return 0;
3822 
3823 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3824 	do_div(value, 1000000);
3825 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3826 }
3827 
3828 static bool is_vmcs12_ext_field(unsigned long field)
3829 {
3830 	switch (field) {
3831 	case GUEST_ES_SELECTOR:
3832 	case GUEST_CS_SELECTOR:
3833 	case GUEST_SS_SELECTOR:
3834 	case GUEST_DS_SELECTOR:
3835 	case GUEST_FS_SELECTOR:
3836 	case GUEST_GS_SELECTOR:
3837 	case GUEST_LDTR_SELECTOR:
3838 	case GUEST_TR_SELECTOR:
3839 	case GUEST_ES_LIMIT:
3840 	case GUEST_CS_LIMIT:
3841 	case GUEST_SS_LIMIT:
3842 	case GUEST_DS_LIMIT:
3843 	case GUEST_FS_LIMIT:
3844 	case GUEST_GS_LIMIT:
3845 	case GUEST_LDTR_LIMIT:
3846 	case GUEST_TR_LIMIT:
3847 	case GUEST_GDTR_LIMIT:
3848 	case GUEST_IDTR_LIMIT:
3849 	case GUEST_ES_AR_BYTES:
3850 	case GUEST_DS_AR_BYTES:
3851 	case GUEST_FS_AR_BYTES:
3852 	case GUEST_GS_AR_BYTES:
3853 	case GUEST_LDTR_AR_BYTES:
3854 	case GUEST_TR_AR_BYTES:
3855 	case GUEST_ES_BASE:
3856 	case GUEST_CS_BASE:
3857 	case GUEST_SS_BASE:
3858 	case GUEST_DS_BASE:
3859 	case GUEST_FS_BASE:
3860 	case GUEST_GS_BASE:
3861 	case GUEST_LDTR_BASE:
3862 	case GUEST_TR_BASE:
3863 	case GUEST_GDTR_BASE:
3864 	case GUEST_IDTR_BASE:
3865 	case GUEST_PENDING_DBG_EXCEPTIONS:
3866 	case GUEST_BNDCFGS:
3867 		return true;
3868 	default:
3869 		break;
3870 	}
3871 
3872 	return false;
3873 }
3874 
3875 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3876 				       struct vmcs12 *vmcs12)
3877 {
3878 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3879 
3880 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3881 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3882 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3883 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3884 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3885 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3886 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3887 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3888 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3889 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3890 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3891 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3892 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3893 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3894 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3895 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3896 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3897 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3898 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3899 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3900 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3901 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3902 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3903 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3904 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3905 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3906 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3907 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3908 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3909 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3910 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3911 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3912 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3913 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3914 	vmcs12->guest_pending_dbg_exceptions =
3915 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3916 	if (kvm_mpx_supported())
3917 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3918 
3919 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
3920 }
3921 
3922 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3923 				       struct vmcs12 *vmcs12)
3924 {
3925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3926 	int cpu;
3927 
3928 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
3929 		return;
3930 
3931 
3932 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
3933 
3934 	cpu = get_cpu();
3935 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
3936 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
3937 
3938 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3939 
3940 	vmx->loaded_vmcs = &vmx->vmcs01;
3941 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
3942 	put_cpu();
3943 }
3944 
3945 /*
3946  * Update the guest state fields of vmcs12 to reflect changes that
3947  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3948  * VM-entry controls is also updated, since this is really a guest
3949  * state bit.)
3950  */
3951 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3952 {
3953 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3954 
3955 	if (vmx->nested.hv_evmcs)
3956 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3957 
3958 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = !vmx->nested.hv_evmcs;
3959 
3960 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3961 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3962 
3963 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
3964 	vmcs12->guest_rip = kvm_rip_read(vcpu);
3965 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3966 
3967 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3968 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3969 
3970 	vmcs12->guest_interruptibility_info =
3971 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3972 
3973 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3974 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3975 	else
3976 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3977 
3978 	if (nested_cpu_has_preemption_timer(vmcs12) &&
3979 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
3980 	    !vmx->nested.nested_run_pending)
3981 		vmcs12->vmx_preemption_timer_value =
3982 			vmx_get_preemption_timer_value(vcpu);
3983 
3984 	/*
3985 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3986 	 * own CR3 without exiting. If it has changed it, we must keep it.
3987 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3988 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3989 	 *
3990 	 * Additionally, restore L2's PDPTR to vmcs12.
3991 	 */
3992 	if (enable_ept) {
3993 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3994 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3995 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3996 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3997 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3998 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3999 		}
4000 	}
4001 
4002 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4003 
4004 	if (nested_cpu_has_vid(vmcs12))
4005 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4006 
4007 	vmcs12->vm_entry_controls =
4008 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4009 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4010 
4011 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4012 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4013 
4014 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4015 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4016 }
4017 
4018 /*
4019  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4020  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4021  * and this function updates it to reflect the changes to the guest state while
4022  * L2 was running (and perhaps made some exits which were handled directly by L0
4023  * without going back to L1), and to reflect the exit reason.
4024  * Note that we do not have to copy here all VMCS fields, just those that
4025  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4026  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4027  * which already writes to vmcs12 directly.
4028  */
4029 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4030 			   u32 vm_exit_reason, u32 exit_intr_info,
4031 			   unsigned long exit_qualification)
4032 {
4033 	/* update exit information fields: */
4034 	vmcs12->vm_exit_reason = vm_exit_reason;
4035 	vmcs12->exit_qualification = exit_qualification;
4036 	vmcs12->vm_exit_intr_info = exit_intr_info;
4037 
4038 	vmcs12->idt_vectoring_info_field = 0;
4039 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4040 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4041 
4042 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4043 		vmcs12->launch_state = 1;
4044 
4045 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4046 		 * instead of reading the real value. */
4047 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4048 
4049 		/*
4050 		 * Transfer the event that L0 or L1 may wanted to inject into
4051 		 * L2 to IDT_VECTORING_INFO_FIELD.
4052 		 */
4053 		vmcs12_save_pending_event(vcpu, vmcs12);
4054 
4055 		/*
4056 		 * According to spec, there's no need to store the guest's
4057 		 * MSRs if the exit is due to a VM-entry failure that occurs
4058 		 * during or after loading the guest state. Since this exit
4059 		 * does not fall in that category, we need to save the MSRs.
4060 		 */
4061 		if (nested_vmx_store_msr(vcpu,
4062 					 vmcs12->vm_exit_msr_store_addr,
4063 					 vmcs12->vm_exit_msr_store_count))
4064 			nested_vmx_abort(vcpu,
4065 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4066 	}
4067 
4068 	/*
4069 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4070 	 * preserved above and would only end up incorrectly in L1.
4071 	 */
4072 	vcpu->arch.nmi_injected = false;
4073 	kvm_clear_exception_queue(vcpu);
4074 	kvm_clear_interrupt_queue(vcpu);
4075 }
4076 
4077 /*
4078  * A part of what we need to when the nested L2 guest exits and we want to
4079  * run its L1 parent, is to reset L1's guest state to the host state specified
4080  * in vmcs12.
4081  * This function is to be called not only on normal nested exit, but also on
4082  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4083  * Failures During or After Loading Guest State").
4084  * This function should be called when the active VMCS is L1's (vmcs01).
4085  */
4086 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4087 				   struct vmcs12 *vmcs12)
4088 {
4089 	enum vm_entry_failure_code ignored;
4090 	struct kvm_segment seg;
4091 
4092 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4093 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4094 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4095 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4096 	else
4097 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4098 	vmx_set_efer(vcpu, vcpu->arch.efer);
4099 
4100 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4101 	kvm_rip_write(vcpu, vmcs12->host_rip);
4102 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4103 	vmx_set_interrupt_shadow(vcpu, 0);
4104 
4105 	/*
4106 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4107 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4108 	 *
4109 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4110 	 * (KVM doesn't change it);
4111 	 */
4112 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
4113 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4114 
4115 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4116 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4117 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4118 
4119 	nested_ept_uninit_mmu_context(vcpu);
4120 
4121 	/*
4122 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4123 	 * couldn't have changed.
4124 	 */
4125 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &ignored))
4126 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4127 
4128 	if (!enable_ept)
4129 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
4130 
4131 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4132 
4133 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4134 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4135 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4136 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4137 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4138 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4139 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4140 
4141 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4142 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4143 		vmcs_write64(GUEST_BNDCFGS, 0);
4144 
4145 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4146 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4147 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4148 	}
4149 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4150 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4151 					 vmcs12->host_ia32_perf_global_ctrl));
4152 
4153 	/* Set L1 segment info according to Intel SDM
4154 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4155 	seg = (struct kvm_segment) {
4156 		.base = 0,
4157 		.limit = 0xFFFFFFFF,
4158 		.selector = vmcs12->host_cs_selector,
4159 		.type = 11,
4160 		.present = 1,
4161 		.s = 1,
4162 		.g = 1
4163 	};
4164 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4165 		seg.l = 1;
4166 	else
4167 		seg.db = 1;
4168 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4169 	seg = (struct kvm_segment) {
4170 		.base = 0,
4171 		.limit = 0xFFFFFFFF,
4172 		.type = 3,
4173 		.present = 1,
4174 		.s = 1,
4175 		.db = 1,
4176 		.g = 1
4177 	};
4178 	seg.selector = vmcs12->host_ds_selector;
4179 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4180 	seg.selector = vmcs12->host_es_selector;
4181 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4182 	seg.selector = vmcs12->host_ss_selector;
4183 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4184 	seg.selector = vmcs12->host_fs_selector;
4185 	seg.base = vmcs12->host_fs_base;
4186 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4187 	seg.selector = vmcs12->host_gs_selector;
4188 	seg.base = vmcs12->host_gs_base;
4189 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4190 	seg = (struct kvm_segment) {
4191 		.base = vmcs12->host_tr_base,
4192 		.limit = 0x67,
4193 		.selector = vmcs12->host_tr_selector,
4194 		.type = 11,
4195 		.present = 1
4196 	};
4197 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4198 
4199 	kvm_set_dr(vcpu, 7, 0x400);
4200 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4201 
4202 	if (cpu_has_vmx_msr_bitmap())
4203 		vmx_update_msr_bitmap(vcpu);
4204 
4205 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4206 				vmcs12->vm_exit_msr_load_count))
4207 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4208 }
4209 
4210 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4211 {
4212 	struct shared_msr_entry *efer_msr;
4213 	unsigned int i;
4214 
4215 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4216 		return vmcs_read64(GUEST_IA32_EFER);
4217 
4218 	if (cpu_has_load_ia32_efer())
4219 		return host_efer;
4220 
4221 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4222 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4223 			return vmx->msr_autoload.guest.val[i].value;
4224 	}
4225 
4226 	efer_msr = find_msr_entry(vmx, MSR_EFER);
4227 	if (efer_msr)
4228 		return efer_msr->data;
4229 
4230 	return host_efer;
4231 }
4232 
4233 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4234 {
4235 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4236 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4237 	struct vmx_msr_entry g, h;
4238 	gpa_t gpa;
4239 	u32 i, j;
4240 
4241 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4242 
4243 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4244 		/*
4245 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4246 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4247 		 * and vcpu->arch.dr7 is not squirreled away before the
4248 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4249 		 */
4250 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4251 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4252 		else
4253 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4254 	}
4255 
4256 	/*
4257 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4258 	 * handle a variety of side effects to KVM's software model.
4259 	 */
4260 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4261 
4262 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
4263 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4264 
4265 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4266 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4267 
4268 	nested_ept_uninit_mmu_context(vcpu);
4269 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4270 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4271 
4272 	/*
4273 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4274 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4275 	 * VMFail, like everything else we just need to ensure our
4276 	 * software model is up-to-date.
4277 	 */
4278 	if (enable_ept && is_pae_paging(vcpu))
4279 		ept_save_pdptrs(vcpu);
4280 
4281 	kvm_mmu_reset_context(vcpu);
4282 
4283 	if (cpu_has_vmx_msr_bitmap())
4284 		vmx_update_msr_bitmap(vcpu);
4285 
4286 	/*
4287 	 * This nasty bit of open coding is a compromise between blindly
4288 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4289 	 * of VMFail), leaving the nested VM's MSRs in the software model
4290 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4291 	 * expensive since the lists are unbound by hardware).  For each
4292 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4293 	 * list, reload it from the exit load list if it exists and differs
4294 	 * from the guest value.  The intent is to stuff host state as
4295 	 * silently as possible, not to fully process the exit load list.
4296 	 */
4297 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4298 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4299 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4300 			pr_debug_ratelimited(
4301 				"%s read MSR index failed (%u, 0x%08llx)\n",
4302 				__func__, i, gpa);
4303 			goto vmabort;
4304 		}
4305 
4306 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4307 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4308 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4309 				pr_debug_ratelimited(
4310 					"%s read MSR failed (%u, 0x%08llx)\n",
4311 					__func__, j, gpa);
4312 				goto vmabort;
4313 			}
4314 			if (h.index != g.index)
4315 				continue;
4316 			if (h.value == g.value)
4317 				break;
4318 
4319 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4320 				pr_debug_ratelimited(
4321 					"%s check failed (%u, 0x%x, 0x%x)\n",
4322 					__func__, j, h.index, h.reserved);
4323 				goto vmabort;
4324 			}
4325 
4326 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4327 				pr_debug_ratelimited(
4328 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4329 					__func__, j, h.index, h.value);
4330 				goto vmabort;
4331 			}
4332 		}
4333 	}
4334 
4335 	return;
4336 
4337 vmabort:
4338 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4339 }
4340 
4341 /*
4342  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4343  * and modify vmcs12 to make it see what it would expect to see there if
4344  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4345  */
4346 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4347 		       u32 exit_intr_info, unsigned long exit_qualification)
4348 {
4349 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4350 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4351 
4352 	/* trying to cancel vmlaunch/vmresume is a bug */
4353 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4354 
4355 	/* Service the TLB flush request for L2 before switching to L1. */
4356 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
4357 		kvm_vcpu_flush_tlb_current(vcpu);
4358 
4359 	leave_guest_mode(vcpu);
4360 
4361 	if (nested_cpu_has_preemption_timer(vmcs12))
4362 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4363 
4364 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
4365 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
4366 
4367 	if (likely(!vmx->fail)) {
4368 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4369 
4370 		if (vm_exit_reason != -1)
4371 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4372 				       exit_intr_info, exit_qualification);
4373 
4374 		/*
4375 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4376 		 * also be used to capture vmcs12 cache as part of
4377 		 * capturing nVMX state for snapshot (migration).
4378 		 *
4379 		 * Otherwise, this flush will dirty guest memory at a
4380 		 * point it is already assumed by user-space to be
4381 		 * immutable.
4382 		 */
4383 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4384 	} else {
4385 		/*
4386 		 * The only expected VM-instruction error is "VM entry with
4387 		 * invalid control field(s)." Anything else indicates a
4388 		 * problem with L0.  And we should never get here with a
4389 		 * VMFail of any type if early consistency checks are enabled.
4390 		 */
4391 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4392 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4393 		WARN_ON_ONCE(nested_early_check);
4394 	}
4395 
4396 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4397 
4398 	/* Update any VMCS fields that might have changed while L2 ran */
4399 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4400 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4401 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4402 	if (vmx->nested.l1_tpr_threshold != -1)
4403 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4404 
4405 	if (kvm_has_tsc_control)
4406 		decache_tsc_multiplier(vmx);
4407 
4408 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4409 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4410 		vmx_set_virtual_apic_mode(vcpu);
4411 	}
4412 
4413 	/* Unpin physical memory we referred to in vmcs02 */
4414 	if (vmx->nested.apic_access_page) {
4415 		kvm_release_page_clean(vmx->nested.apic_access_page);
4416 		vmx->nested.apic_access_page = NULL;
4417 	}
4418 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4419 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4420 	vmx->nested.pi_desc = NULL;
4421 
4422 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4423 		vmx->nested.reload_vmcs01_apic_access_page = false;
4424 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4425 	}
4426 
4427 	if ((vm_exit_reason != -1) &&
4428 	    (enable_shadow_vmcs || vmx->nested.hv_evmcs))
4429 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4430 
4431 	/* in case we halted in L2 */
4432 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4433 
4434 	if (likely(!vmx->fail)) {
4435 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4436 		    nested_exit_intr_ack_set(vcpu)) {
4437 			int irq = kvm_cpu_get_interrupt(vcpu);
4438 			WARN_ON(irq < 0);
4439 			vmcs12->vm_exit_intr_info = irq |
4440 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4441 		}
4442 
4443 		if (vm_exit_reason != -1)
4444 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4445 						       vmcs12->exit_qualification,
4446 						       vmcs12->idt_vectoring_info_field,
4447 						       vmcs12->vm_exit_intr_info,
4448 						       vmcs12->vm_exit_intr_error_code,
4449 						       KVM_ISA_VMX);
4450 
4451 		load_vmcs12_host_state(vcpu, vmcs12);
4452 
4453 		return;
4454 	}
4455 
4456 	/*
4457 	 * After an early L2 VM-entry failure, we're now back
4458 	 * in L1 which thinks it just finished a VMLAUNCH or
4459 	 * VMRESUME instruction, so we need to set the failure
4460 	 * flag and the VM-instruction error field of the VMCS
4461 	 * accordingly, and skip the emulated instruction.
4462 	 */
4463 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4464 
4465 	/*
4466 	 * Restore L1's host state to KVM's software model.  We're here
4467 	 * because a consistency check was caught by hardware, which
4468 	 * means some amount of guest state has been propagated to KVM's
4469 	 * model and needs to be unwound to the host's state.
4470 	 */
4471 	nested_vmx_restore_host_state(vcpu);
4472 
4473 	vmx->fail = 0;
4474 }
4475 
4476 /*
4477  * Decode the memory-address operand of a vmx instruction, as recorded on an
4478  * exit caused by such an instruction (run by a guest hypervisor).
4479  * On success, returns 0. When the operand is invalid, returns 1 and throws
4480  * #UD, #GP, or #SS.
4481  */
4482 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4483 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4484 {
4485 	gva_t off;
4486 	bool exn;
4487 	struct kvm_segment s;
4488 
4489 	/*
4490 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4491 	 * Execution", on an exit, vmx_instruction_info holds most of the
4492 	 * addressing components of the operand. Only the displacement part
4493 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4494 	 * For how an actual address is calculated from all these components,
4495 	 * refer to Vol. 1, "Operand Addressing".
4496 	 */
4497 	int  scaling = vmx_instruction_info & 3;
4498 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4499 	bool is_reg = vmx_instruction_info & (1u << 10);
4500 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4501 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4502 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4503 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4504 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4505 
4506 	if (is_reg) {
4507 		kvm_queue_exception(vcpu, UD_VECTOR);
4508 		return 1;
4509 	}
4510 
4511 	/* Addr = segment_base + offset */
4512 	/* offset = base + [index * scale] + displacement */
4513 	off = exit_qualification; /* holds the displacement */
4514 	if (addr_size == 1)
4515 		off = (gva_t)sign_extend64(off, 31);
4516 	else if (addr_size == 0)
4517 		off = (gva_t)sign_extend64(off, 15);
4518 	if (base_is_valid)
4519 		off += kvm_register_read(vcpu, base_reg);
4520 	if (index_is_valid)
4521 		off += kvm_register_read(vcpu, index_reg) << scaling;
4522 	vmx_get_segment(vcpu, &s, seg_reg);
4523 
4524 	/*
4525 	 * The effective address, i.e. @off, of a memory operand is truncated
4526 	 * based on the address size of the instruction.  Note that this is
4527 	 * the *effective address*, i.e. the address prior to accounting for
4528 	 * the segment's base.
4529 	 */
4530 	if (addr_size == 1) /* 32 bit */
4531 		off &= 0xffffffff;
4532 	else if (addr_size == 0) /* 16 bit */
4533 		off &= 0xffff;
4534 
4535 	/* Checks for #GP/#SS exceptions. */
4536 	exn = false;
4537 	if (is_long_mode(vcpu)) {
4538 		/*
4539 		 * The virtual/linear address is never truncated in 64-bit
4540 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4541 		 * address when using FS/GS with a non-zero base.
4542 		 */
4543 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4544 			*ret = s.base + off;
4545 		else
4546 			*ret = off;
4547 
4548 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4549 		 * non-canonical form. This is the only check on the memory
4550 		 * destination for long mode!
4551 		 */
4552 		exn = is_noncanonical_address(*ret, vcpu);
4553 	} else {
4554 		/*
4555 		 * When not in long mode, the virtual/linear address is
4556 		 * unconditionally truncated to 32 bits regardless of the
4557 		 * address size.
4558 		 */
4559 		*ret = (s.base + off) & 0xffffffff;
4560 
4561 		/* Protected mode: apply checks for segment validity in the
4562 		 * following order:
4563 		 * - segment type check (#GP(0) may be thrown)
4564 		 * - usability check (#GP(0)/#SS(0))
4565 		 * - limit check (#GP(0)/#SS(0))
4566 		 */
4567 		if (wr)
4568 			/* #GP(0) if the destination operand is located in a
4569 			 * read-only data segment or any code segment.
4570 			 */
4571 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4572 		else
4573 			/* #GP(0) if the source operand is located in an
4574 			 * execute-only code segment
4575 			 */
4576 			exn = ((s.type & 0xa) == 8);
4577 		if (exn) {
4578 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4579 			return 1;
4580 		}
4581 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4582 		 */
4583 		exn = (s.unusable != 0);
4584 
4585 		/*
4586 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4587 		 * outside the segment limit.  All CPUs that support VMX ignore
4588 		 * limit checks for flat segments, i.e. segments with base==0,
4589 		 * limit==0xffffffff and of type expand-up data or code.
4590 		 */
4591 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4592 		     ((s.type & 8) || !(s.type & 4))))
4593 			exn = exn || ((u64)off + len - 1 > s.limit);
4594 	}
4595 	if (exn) {
4596 		kvm_queue_exception_e(vcpu,
4597 				      seg_reg == VCPU_SREG_SS ?
4598 						SS_VECTOR : GP_VECTOR,
4599 				      0);
4600 		return 1;
4601 	}
4602 
4603 	return 0;
4604 }
4605 
4606 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4607 {
4608 	struct vcpu_vmx *vmx;
4609 
4610 	if (!nested_vmx_allowed(vcpu))
4611 		return;
4612 
4613 	vmx = to_vmx(vcpu);
4614 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4615 		vmx->nested.msrs.entry_ctls_high |=
4616 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4617 		vmx->nested.msrs.exit_ctls_high |=
4618 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4619 	} else {
4620 		vmx->nested.msrs.entry_ctls_high &=
4621 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4622 		vmx->nested.msrs.exit_ctls_high &=
4623 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4624 	}
4625 }
4626 
4627 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4628 				int *ret)
4629 {
4630 	gva_t gva;
4631 	struct x86_exception e;
4632 	int r;
4633 
4634 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4635 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4636 				sizeof(*vmpointer), &gva)) {
4637 		*ret = 1;
4638 		return -EINVAL;
4639 	}
4640 
4641 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4642 	if (r != X86EMUL_CONTINUE) {
4643 		*ret = vmx_handle_memory_failure(vcpu, r, &e);
4644 		return -EINVAL;
4645 	}
4646 
4647 	return 0;
4648 }
4649 
4650 /*
4651  * Allocate a shadow VMCS and associate it with the currently loaded
4652  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4653  * VMCS is also VMCLEARed, so that it is ready for use.
4654  */
4655 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4656 {
4657 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4658 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4659 
4660 	/*
4661 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4662 	 * executes VMXON and free it when L1 executes VMXOFF.
4663 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4664 	 * here when vmcs01 already have an allocated shadow vmcs.
4665 	 */
4666 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4667 
4668 	if (!loaded_vmcs->shadow_vmcs) {
4669 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4670 		if (loaded_vmcs->shadow_vmcs)
4671 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4672 	}
4673 	return loaded_vmcs->shadow_vmcs;
4674 }
4675 
4676 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4677 {
4678 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4679 	int r;
4680 
4681 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4682 	if (r < 0)
4683 		goto out_vmcs02;
4684 
4685 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4686 	if (!vmx->nested.cached_vmcs12)
4687 		goto out_cached_vmcs12;
4688 
4689 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4690 	if (!vmx->nested.cached_shadow_vmcs12)
4691 		goto out_cached_shadow_vmcs12;
4692 
4693 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4694 		goto out_shadow_vmcs;
4695 
4696 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4697 		     HRTIMER_MODE_ABS_PINNED);
4698 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4699 
4700 	vmx->nested.vpid02 = allocate_vpid();
4701 
4702 	vmx->nested.vmcs02_initialized = false;
4703 	vmx->nested.vmxon = true;
4704 
4705 	if (vmx_pt_mode_is_host_guest()) {
4706 		vmx->pt_desc.guest.ctl = 0;
4707 		pt_update_intercept_for_msr(vmx);
4708 	}
4709 
4710 	return 0;
4711 
4712 out_shadow_vmcs:
4713 	kfree(vmx->nested.cached_shadow_vmcs12);
4714 
4715 out_cached_shadow_vmcs12:
4716 	kfree(vmx->nested.cached_vmcs12);
4717 
4718 out_cached_vmcs12:
4719 	free_loaded_vmcs(&vmx->nested.vmcs02);
4720 
4721 out_vmcs02:
4722 	return -ENOMEM;
4723 }
4724 
4725 /*
4726  * Emulate the VMXON instruction.
4727  * Currently, we just remember that VMX is active, and do not save or even
4728  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4729  * do not currently need to store anything in that guest-allocated memory
4730  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4731  * argument is different from the VMXON pointer (which the spec says they do).
4732  */
4733 static int handle_vmon(struct kvm_vcpu *vcpu)
4734 {
4735 	int ret;
4736 	gpa_t vmptr;
4737 	uint32_t revision;
4738 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4739 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4740 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4741 
4742 	/*
4743 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4744 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4745 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4746 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4747 	 * have already been checked by hardware, prior to the VM-exit for
4748 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4749 	 * that bit set to 1 in non-root mode.
4750 	 */
4751 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4752 		kvm_queue_exception(vcpu, UD_VECTOR);
4753 		return 1;
4754 	}
4755 
4756 	/* CPL=0 must be checked manually. */
4757 	if (vmx_get_cpl(vcpu)) {
4758 		kvm_inject_gp(vcpu, 0);
4759 		return 1;
4760 	}
4761 
4762 	if (vmx->nested.vmxon)
4763 		return nested_vmx_failValid(vcpu,
4764 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4765 
4766 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4767 			!= VMXON_NEEDED_FEATURES) {
4768 		kvm_inject_gp(vcpu, 0);
4769 		return 1;
4770 	}
4771 
4772 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4773 		return ret;
4774 
4775 	/*
4776 	 * SDM 3: 24.11.5
4777 	 * The first 4 bytes of VMXON region contain the supported
4778 	 * VMCS revision identifier
4779 	 *
4780 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4781 	 * which replaces physical address width with 32
4782 	 */
4783 	if (!page_address_valid(vcpu, vmptr))
4784 		return nested_vmx_failInvalid(vcpu);
4785 
4786 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4787 	    revision != VMCS12_REVISION)
4788 		return nested_vmx_failInvalid(vcpu);
4789 
4790 	vmx->nested.vmxon_ptr = vmptr;
4791 	ret = enter_vmx_operation(vcpu);
4792 	if (ret)
4793 		return ret;
4794 
4795 	return nested_vmx_succeed(vcpu);
4796 }
4797 
4798 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4799 {
4800 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4801 
4802 	if (vmx->nested.current_vmptr == -1ull)
4803 		return;
4804 
4805 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4806 
4807 	if (enable_shadow_vmcs) {
4808 		/* copy to memory all shadowed fields in case
4809 		   they were modified */
4810 		copy_shadow_to_vmcs12(vmx);
4811 		vmx_disable_shadow_vmcs(vmx);
4812 	}
4813 	vmx->nested.posted_intr_nv = -1;
4814 
4815 	/* Flush VMCS12 to guest memory */
4816 	kvm_vcpu_write_guest_page(vcpu,
4817 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4818 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4819 
4820 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4821 
4822 	vmx->nested.current_vmptr = -1ull;
4823 }
4824 
4825 /* Emulate the VMXOFF instruction */
4826 static int handle_vmoff(struct kvm_vcpu *vcpu)
4827 {
4828 	if (!nested_vmx_check_permission(vcpu))
4829 		return 1;
4830 
4831 	free_nested(vcpu);
4832 
4833 	/* Process a latched INIT during time CPU was in VMX operation */
4834 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4835 
4836 	return nested_vmx_succeed(vcpu);
4837 }
4838 
4839 /* Emulate the VMCLEAR instruction */
4840 static int handle_vmclear(struct kvm_vcpu *vcpu)
4841 {
4842 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4843 	u32 zero = 0;
4844 	gpa_t vmptr;
4845 	u64 evmcs_gpa;
4846 	int r;
4847 
4848 	if (!nested_vmx_check_permission(vcpu))
4849 		return 1;
4850 
4851 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
4852 		return r;
4853 
4854 	if (!page_address_valid(vcpu, vmptr))
4855 		return nested_vmx_failValid(vcpu,
4856 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4857 
4858 	if (vmptr == vmx->nested.vmxon_ptr)
4859 		return nested_vmx_failValid(vcpu,
4860 			VMXERR_VMCLEAR_VMXON_POINTER);
4861 
4862 	/*
4863 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4864 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4865 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4866 	 * writing to the non-existent 'launch_state' field. The area doesn't
4867 	 * have to be the currently active EVMCS on the calling CPU and there's
4868 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
4869 	 * state. It is possible that the area will stay mapped as
4870 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
4871 	 */
4872 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
4873 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
4874 		if (vmptr == vmx->nested.current_vmptr)
4875 			nested_release_vmcs12(vcpu);
4876 
4877 		kvm_vcpu_write_guest(vcpu,
4878 				     vmptr + offsetof(struct vmcs12,
4879 						      launch_state),
4880 				     &zero, sizeof(zero));
4881 	}
4882 
4883 	return nested_vmx_succeed(vcpu);
4884 }
4885 
4886 /* Emulate the VMLAUNCH instruction */
4887 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4888 {
4889 	return nested_vmx_run(vcpu, true);
4890 }
4891 
4892 /* Emulate the VMRESUME instruction */
4893 static int handle_vmresume(struct kvm_vcpu *vcpu)
4894 {
4895 
4896 	return nested_vmx_run(vcpu, false);
4897 }
4898 
4899 static int handle_vmread(struct kvm_vcpu *vcpu)
4900 {
4901 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4902 						    : get_vmcs12(vcpu);
4903 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
4904 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4905 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4906 	struct x86_exception e;
4907 	unsigned long field;
4908 	u64 value;
4909 	gva_t gva = 0;
4910 	short offset;
4911 	int len, r;
4912 
4913 	if (!nested_vmx_check_permission(vcpu))
4914 		return 1;
4915 
4916 	/*
4917 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
4918 	 * any VMREAD sets the ALU flags for VMfailInvalid.
4919 	 */
4920 	if (vmx->nested.current_vmptr == -1ull ||
4921 	    (is_guest_mode(vcpu) &&
4922 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
4923 		return nested_vmx_failInvalid(vcpu);
4924 
4925 	/* Decode instruction info and find the field to read */
4926 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
4927 
4928 	offset = vmcs_field_to_offset(field);
4929 	if (offset < 0)
4930 		return nested_vmx_failValid(vcpu,
4931 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4932 
4933 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
4934 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4935 
4936 	/* Read the field, zero-extended to a u64 value */
4937 	value = vmcs12_read_any(vmcs12, field, offset);
4938 
4939 	/*
4940 	 * Now copy part of this value to register or memory, as requested.
4941 	 * Note that the number of bits actually copied is 32 or 64 depending
4942 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4943 	 */
4944 	if (instr_info & BIT(10)) {
4945 		kvm_register_writel(vcpu, (((instr_info) >> 3) & 0xf), value);
4946 	} else {
4947 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4948 		if (get_vmx_mem_address(vcpu, exit_qualification,
4949 					instr_info, true, len, &gva))
4950 			return 1;
4951 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4952 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
4953 		if (r != X86EMUL_CONTINUE)
4954 			return vmx_handle_memory_failure(vcpu, r, &e);
4955 	}
4956 
4957 	return nested_vmx_succeed(vcpu);
4958 }
4959 
4960 static bool is_shadow_field_rw(unsigned long field)
4961 {
4962 	switch (field) {
4963 #define SHADOW_FIELD_RW(x, y) case x:
4964 #include "vmcs_shadow_fields.h"
4965 		return true;
4966 	default:
4967 		break;
4968 	}
4969 	return false;
4970 }
4971 
4972 static bool is_shadow_field_ro(unsigned long field)
4973 {
4974 	switch (field) {
4975 #define SHADOW_FIELD_RO(x, y) case x:
4976 #include "vmcs_shadow_fields.h"
4977 		return true;
4978 	default:
4979 		break;
4980 	}
4981 	return false;
4982 }
4983 
4984 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4985 {
4986 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4987 						    : get_vmcs12(vcpu);
4988 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
4989 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4990 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4991 	struct x86_exception e;
4992 	unsigned long field;
4993 	short offset;
4994 	gva_t gva;
4995 	int len, r;
4996 
4997 	/*
4998 	 * The value to write might be 32 or 64 bits, depending on L1's long
4999 	 * mode, and eventually we need to write that into a field of several
5000 	 * possible lengths. The code below first zero-extends the value to 64
5001 	 * bit (value), and then copies only the appropriate number of
5002 	 * bits into the vmcs12 field.
5003 	 */
5004 	u64 value = 0;
5005 
5006 	if (!nested_vmx_check_permission(vcpu))
5007 		return 1;
5008 
5009 	/*
5010 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
5011 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5012 	 */
5013 	if (vmx->nested.current_vmptr == -1ull ||
5014 	    (is_guest_mode(vcpu) &&
5015 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
5016 		return nested_vmx_failInvalid(vcpu);
5017 
5018 	if (instr_info & BIT(10))
5019 		value = kvm_register_readl(vcpu, (((instr_info) >> 3) & 0xf));
5020 	else {
5021 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5022 		if (get_vmx_mem_address(vcpu, exit_qualification,
5023 					instr_info, false, len, &gva))
5024 			return 1;
5025 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5026 		if (r != X86EMUL_CONTINUE)
5027 			return vmx_handle_memory_failure(vcpu, r, &e);
5028 	}
5029 
5030 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
5031 
5032 	offset = vmcs_field_to_offset(field);
5033 	if (offset < 0)
5034 		return nested_vmx_failValid(vcpu,
5035 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5036 
5037 	/*
5038 	 * If the vCPU supports "VMWRITE to any supported field in the
5039 	 * VMCS," then the "read-only" fields are actually read/write.
5040 	 */
5041 	if (vmcs_field_readonly(field) &&
5042 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5043 		return nested_vmx_failValid(vcpu,
5044 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5045 
5046 	/*
5047 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5048 	 * vmcs12, else we may crush a field or consume a stale value.
5049 	 */
5050 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5051 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5052 
5053 	/*
5054 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5055 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5056 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5057 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5058 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5059 	 * the stripped down value, L2 sees the full value as stored by KVM).
5060 	 */
5061 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5062 		value &= 0x1f0ff;
5063 
5064 	vmcs12_write_any(vmcs12, field, offset, value);
5065 
5066 	/*
5067 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5068 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5069 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5070 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5071 	 */
5072 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5073 		/*
5074 		 * L1 can read these fields without exiting, ensure the
5075 		 * shadow VMCS is up-to-date.
5076 		 */
5077 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5078 			preempt_disable();
5079 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5080 
5081 			__vmcs_writel(field, value);
5082 
5083 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5084 			vmcs_load(vmx->loaded_vmcs->vmcs);
5085 			preempt_enable();
5086 		}
5087 		vmx->nested.dirty_vmcs12 = true;
5088 	}
5089 
5090 	return nested_vmx_succeed(vcpu);
5091 }
5092 
5093 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5094 {
5095 	vmx->nested.current_vmptr = vmptr;
5096 	if (enable_shadow_vmcs) {
5097 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5098 		vmcs_write64(VMCS_LINK_POINTER,
5099 			     __pa(vmx->vmcs01.shadow_vmcs));
5100 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5101 	}
5102 	vmx->nested.dirty_vmcs12 = true;
5103 }
5104 
5105 /* Emulate the VMPTRLD instruction */
5106 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5107 {
5108 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5109 	gpa_t vmptr;
5110 	int r;
5111 
5112 	if (!nested_vmx_check_permission(vcpu))
5113 		return 1;
5114 
5115 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5116 		return r;
5117 
5118 	if (!page_address_valid(vcpu, vmptr))
5119 		return nested_vmx_failValid(vcpu,
5120 			VMXERR_VMPTRLD_INVALID_ADDRESS);
5121 
5122 	if (vmptr == vmx->nested.vmxon_ptr)
5123 		return nested_vmx_failValid(vcpu,
5124 			VMXERR_VMPTRLD_VMXON_POINTER);
5125 
5126 	/* Forbid normal VMPTRLD if Enlightened version was used */
5127 	if (vmx->nested.hv_evmcs)
5128 		return 1;
5129 
5130 	if (vmx->nested.current_vmptr != vmptr) {
5131 		struct kvm_host_map map;
5132 		struct vmcs12 *new_vmcs12;
5133 
5134 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
5135 			/*
5136 			 * Reads from an unbacked page return all 1s,
5137 			 * which means that the 32 bits located at the
5138 			 * given physical address won't match the required
5139 			 * VMCS12_REVISION identifier.
5140 			 */
5141 			return nested_vmx_failValid(vcpu,
5142 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5143 		}
5144 
5145 		new_vmcs12 = map.hva;
5146 
5147 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5148 		    (new_vmcs12->hdr.shadow_vmcs &&
5149 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5150 			kvm_vcpu_unmap(vcpu, &map, false);
5151 			return nested_vmx_failValid(vcpu,
5152 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5153 		}
5154 
5155 		nested_release_vmcs12(vcpu);
5156 
5157 		/*
5158 		 * Load VMCS12 from guest memory since it is not already
5159 		 * cached.
5160 		 */
5161 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
5162 		kvm_vcpu_unmap(vcpu, &map, false);
5163 
5164 		set_current_vmptr(vmx, vmptr);
5165 	}
5166 
5167 	return nested_vmx_succeed(vcpu);
5168 }
5169 
5170 /* Emulate the VMPTRST instruction */
5171 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5172 {
5173 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5174 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5175 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5176 	struct x86_exception e;
5177 	gva_t gva;
5178 	int r;
5179 
5180 	if (!nested_vmx_check_permission(vcpu))
5181 		return 1;
5182 
5183 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
5184 		return 1;
5185 
5186 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5187 				true, sizeof(gpa_t), &gva))
5188 		return 1;
5189 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5190 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5191 					sizeof(gpa_t), &e);
5192 	if (r != X86EMUL_CONTINUE)
5193 		return vmx_handle_memory_failure(vcpu, r, &e);
5194 
5195 	return nested_vmx_succeed(vcpu);
5196 }
5197 
5198 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
5199 
5200 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
5201 {
5202 	return VALID_PAGE(root_hpa) &&
5203 		((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
5204 }
5205 
5206 /* Emulate the INVEPT instruction */
5207 static int handle_invept(struct kvm_vcpu *vcpu)
5208 {
5209 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5210 	u32 vmx_instruction_info, types;
5211 	unsigned long type, roots_to_free;
5212 	struct kvm_mmu *mmu;
5213 	gva_t gva;
5214 	struct x86_exception e;
5215 	struct {
5216 		u64 eptp, gpa;
5217 	} operand;
5218 	int i, r;
5219 
5220 	if (!(vmx->nested.msrs.secondary_ctls_high &
5221 	      SECONDARY_EXEC_ENABLE_EPT) ||
5222 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5223 		kvm_queue_exception(vcpu, UD_VECTOR);
5224 		return 1;
5225 	}
5226 
5227 	if (!nested_vmx_check_permission(vcpu))
5228 		return 1;
5229 
5230 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5231 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5232 
5233 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5234 
5235 	if (type >= 32 || !(types & (1 << type)))
5236 		return nested_vmx_failValid(vcpu,
5237 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5238 
5239 	/* According to the Intel VMX instruction reference, the memory
5240 	 * operand is read even if it isn't needed (e.g., for type==global)
5241 	 */
5242 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5243 			vmx_instruction_info, false, sizeof(operand), &gva))
5244 		return 1;
5245 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5246 	if (r != X86EMUL_CONTINUE)
5247 		return vmx_handle_memory_failure(vcpu, r, &e);
5248 
5249 	/*
5250 	 * Nested EPT roots are always held through guest_mmu,
5251 	 * not root_mmu.
5252 	 */
5253 	mmu = &vcpu->arch.guest_mmu;
5254 
5255 	switch (type) {
5256 	case VMX_EPT_EXTENT_CONTEXT:
5257 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5258 			return nested_vmx_failValid(vcpu,
5259 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5260 
5261 		roots_to_free = 0;
5262 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
5263 					    operand.eptp))
5264 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5265 
5266 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5267 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5268 						    mmu->prev_roots[i].pgd,
5269 						    operand.eptp))
5270 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5271 		}
5272 		break;
5273 	case VMX_EPT_EXTENT_GLOBAL:
5274 		roots_to_free = KVM_MMU_ROOTS_ALL;
5275 		break;
5276 	default:
5277 		BUG();
5278 		break;
5279 	}
5280 
5281 	if (roots_to_free)
5282 		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
5283 
5284 	return nested_vmx_succeed(vcpu);
5285 }
5286 
5287 static int handle_invvpid(struct kvm_vcpu *vcpu)
5288 {
5289 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5290 	u32 vmx_instruction_info;
5291 	unsigned long type, types;
5292 	gva_t gva;
5293 	struct x86_exception e;
5294 	struct {
5295 		u64 vpid;
5296 		u64 gla;
5297 	} operand;
5298 	u16 vpid02;
5299 	int r;
5300 
5301 	if (!(vmx->nested.msrs.secondary_ctls_high &
5302 	      SECONDARY_EXEC_ENABLE_VPID) ||
5303 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5304 		kvm_queue_exception(vcpu, UD_VECTOR);
5305 		return 1;
5306 	}
5307 
5308 	if (!nested_vmx_check_permission(vcpu))
5309 		return 1;
5310 
5311 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5312 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5313 
5314 	types = (vmx->nested.msrs.vpid_caps &
5315 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5316 
5317 	if (type >= 32 || !(types & (1 << type)))
5318 		return nested_vmx_failValid(vcpu,
5319 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5320 
5321 	/* according to the intel vmx instruction reference, the memory
5322 	 * operand is read even if it isn't needed (e.g., for type==global)
5323 	 */
5324 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5325 			vmx_instruction_info, false, sizeof(operand), &gva))
5326 		return 1;
5327 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5328 	if (r != X86EMUL_CONTINUE)
5329 		return vmx_handle_memory_failure(vcpu, r, &e);
5330 
5331 	if (operand.vpid >> 16)
5332 		return nested_vmx_failValid(vcpu,
5333 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5334 
5335 	vpid02 = nested_get_vpid02(vcpu);
5336 	switch (type) {
5337 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5338 		if (!operand.vpid ||
5339 		    is_noncanonical_address(operand.gla, vcpu))
5340 			return nested_vmx_failValid(vcpu,
5341 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5342 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5343 		break;
5344 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5345 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5346 		if (!operand.vpid)
5347 			return nested_vmx_failValid(vcpu,
5348 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5349 		vpid_sync_context(vpid02);
5350 		break;
5351 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5352 		vpid_sync_context(vpid02);
5353 		break;
5354 	default:
5355 		WARN_ON_ONCE(1);
5356 		return kvm_skip_emulated_instruction(vcpu);
5357 	}
5358 
5359 	/*
5360 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5361 	 * linear mappings for L2 (tagged with L2's VPID).  Free all roots as
5362 	 * VPIDs are not tracked in the MMU role.
5363 	 *
5364 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5365 	 * an MMU when EPT is disabled.
5366 	 *
5367 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5368 	 */
5369 	if (!enable_ept)
5370 		kvm_mmu_free_roots(vcpu, &vcpu->arch.root_mmu,
5371 				   KVM_MMU_ROOTS_ALL);
5372 
5373 	return nested_vmx_succeed(vcpu);
5374 }
5375 
5376 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5377 				     struct vmcs12 *vmcs12)
5378 {
5379 	u32 index = kvm_rcx_read(vcpu);
5380 	u64 new_eptp;
5381 	bool accessed_dirty;
5382 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
5383 
5384 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
5385 	    !nested_cpu_has_ept(vmcs12))
5386 		return 1;
5387 
5388 	if (index >= VMFUNC_EPTP_ENTRIES)
5389 		return 1;
5390 
5391 
5392 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5393 				     &new_eptp, index * 8, 8))
5394 		return 1;
5395 
5396 	accessed_dirty = !!(new_eptp & VMX_EPTP_AD_ENABLE_BIT);
5397 
5398 	/*
5399 	 * If the (L2) guest does a vmfunc to the currently
5400 	 * active ept pointer, we don't have to do anything else
5401 	 */
5402 	if (vmcs12->ept_pointer != new_eptp) {
5403 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5404 			return 1;
5405 
5406 		kvm_mmu_unload(vcpu);
5407 		mmu->ept_ad = accessed_dirty;
5408 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
5409 		vmcs12->ept_pointer = new_eptp;
5410 		/*
5411 		 * TODO: Check what's the correct approach in case
5412 		 * mmu reload fails. Currently, we just let the next
5413 		 * reload potentially fail
5414 		 */
5415 		kvm_mmu_reload(vcpu);
5416 	}
5417 
5418 	return 0;
5419 }
5420 
5421 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5422 {
5423 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5424 	struct vmcs12 *vmcs12;
5425 	u32 function = kvm_rax_read(vcpu);
5426 
5427 	/*
5428 	 * VMFUNC is only supported for nested guests, but we always enable the
5429 	 * secondary control for simplicity; for non-nested mode, fake that we
5430 	 * didn't by injecting #UD.
5431 	 */
5432 	if (!is_guest_mode(vcpu)) {
5433 		kvm_queue_exception(vcpu, UD_VECTOR);
5434 		return 1;
5435 	}
5436 
5437 	vmcs12 = get_vmcs12(vcpu);
5438 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
5439 		goto fail;
5440 
5441 	switch (function) {
5442 	case 0:
5443 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5444 			goto fail;
5445 		break;
5446 	default:
5447 		goto fail;
5448 	}
5449 	return kvm_skip_emulated_instruction(vcpu);
5450 
5451 fail:
5452 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
5453 			  vmx_get_intr_info(vcpu),
5454 			  vmx_get_exit_qual(vcpu));
5455 	return 1;
5456 }
5457 
5458 /*
5459  * Return true if an IO instruction with the specified port and size should cause
5460  * a VM-exit into L1.
5461  */
5462 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5463 				 int size)
5464 {
5465 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5466 	gpa_t bitmap, last_bitmap;
5467 	u8 b;
5468 
5469 	last_bitmap = (gpa_t)-1;
5470 	b = -1;
5471 
5472 	while (size > 0) {
5473 		if (port < 0x8000)
5474 			bitmap = vmcs12->io_bitmap_a;
5475 		else if (port < 0x10000)
5476 			bitmap = vmcs12->io_bitmap_b;
5477 		else
5478 			return true;
5479 		bitmap += (port & 0x7fff) / 8;
5480 
5481 		if (last_bitmap != bitmap)
5482 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5483 				return true;
5484 		if (b & (1 << (port & 7)))
5485 			return true;
5486 
5487 		port++;
5488 		size--;
5489 		last_bitmap = bitmap;
5490 	}
5491 
5492 	return false;
5493 }
5494 
5495 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5496 				       struct vmcs12 *vmcs12)
5497 {
5498 	unsigned long exit_qualification;
5499 	unsigned short port;
5500 	int size;
5501 
5502 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5503 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5504 
5505 	exit_qualification = vmx_get_exit_qual(vcpu);
5506 
5507 	port = exit_qualification >> 16;
5508 	size = (exit_qualification & 7) + 1;
5509 
5510 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5511 }
5512 
5513 /*
5514  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5515  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5516  * disinterest in the current event (read or write a specific MSR) by using an
5517  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5518  */
5519 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5520 	struct vmcs12 *vmcs12, u32 exit_reason)
5521 {
5522 	u32 msr_index = kvm_rcx_read(vcpu);
5523 	gpa_t bitmap;
5524 
5525 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5526 		return true;
5527 
5528 	/*
5529 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5530 	 * for the four combinations of read/write and low/high MSR numbers.
5531 	 * First we need to figure out which of the four to use:
5532 	 */
5533 	bitmap = vmcs12->msr_bitmap;
5534 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5535 		bitmap += 2048;
5536 	if (msr_index >= 0xc0000000) {
5537 		msr_index -= 0xc0000000;
5538 		bitmap += 1024;
5539 	}
5540 
5541 	/* Then read the msr_index'th bit from this bitmap: */
5542 	if (msr_index < 1024*8) {
5543 		unsigned char b;
5544 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5545 			return true;
5546 		return 1 & (b >> (msr_index & 7));
5547 	} else
5548 		return true; /* let L1 handle the wrong parameter */
5549 }
5550 
5551 /*
5552  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5553  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5554  * intercept (via guest_host_mask etc.) the current event.
5555  */
5556 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5557 	struct vmcs12 *vmcs12)
5558 {
5559 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5560 	int cr = exit_qualification & 15;
5561 	int reg;
5562 	unsigned long val;
5563 
5564 	switch ((exit_qualification >> 4) & 3) {
5565 	case 0: /* mov to cr */
5566 		reg = (exit_qualification >> 8) & 15;
5567 		val = kvm_register_readl(vcpu, reg);
5568 		switch (cr) {
5569 		case 0:
5570 			if (vmcs12->cr0_guest_host_mask &
5571 			    (val ^ vmcs12->cr0_read_shadow))
5572 				return true;
5573 			break;
5574 		case 3:
5575 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5576 				return true;
5577 			break;
5578 		case 4:
5579 			if (vmcs12->cr4_guest_host_mask &
5580 			    (vmcs12->cr4_read_shadow ^ val))
5581 				return true;
5582 			break;
5583 		case 8:
5584 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5585 				return true;
5586 			break;
5587 		}
5588 		break;
5589 	case 2: /* clts */
5590 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5591 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5592 			return true;
5593 		break;
5594 	case 1: /* mov from cr */
5595 		switch (cr) {
5596 		case 3:
5597 			if (vmcs12->cpu_based_vm_exec_control &
5598 			    CPU_BASED_CR3_STORE_EXITING)
5599 				return true;
5600 			break;
5601 		case 8:
5602 			if (vmcs12->cpu_based_vm_exec_control &
5603 			    CPU_BASED_CR8_STORE_EXITING)
5604 				return true;
5605 			break;
5606 		}
5607 		break;
5608 	case 3: /* lmsw */
5609 		/*
5610 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5611 		 * cr0. Other attempted changes are ignored, with no exit.
5612 		 */
5613 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5614 		if (vmcs12->cr0_guest_host_mask & 0xe &
5615 		    (val ^ vmcs12->cr0_read_shadow))
5616 			return true;
5617 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5618 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5619 		    (val & 0x1))
5620 			return true;
5621 		break;
5622 	}
5623 	return false;
5624 }
5625 
5626 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5627 	struct vmcs12 *vmcs12, gpa_t bitmap)
5628 {
5629 	u32 vmx_instruction_info;
5630 	unsigned long field;
5631 	u8 b;
5632 
5633 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5634 		return true;
5635 
5636 	/* Decode instruction info and find the field to access */
5637 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5638 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5639 
5640 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5641 	if (field >> 15)
5642 		return true;
5643 
5644 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5645 		return true;
5646 
5647 	return 1 & (b >> (field & 7));
5648 }
5649 
5650 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5651 {
5652 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5653 
5654 	if (nested_cpu_has_mtf(vmcs12))
5655 		return true;
5656 
5657 	/*
5658 	 * An MTF VM-exit may be injected into the guest by setting the
5659 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5660 	 * is the case regardless of the 'monitor trap flag' VM-execution
5661 	 * control.
5662 	 */
5663 	return entry_intr_info == (INTR_INFO_VALID_MASK
5664 				   | INTR_TYPE_OTHER_EVENT);
5665 }
5666 
5667 /*
5668  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5669  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5670  */
5671 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu, u32 exit_reason)
5672 {
5673 	u32 intr_info;
5674 
5675 	switch ((u16)exit_reason) {
5676 	case EXIT_REASON_EXCEPTION_NMI:
5677 		intr_info = vmx_get_intr_info(vcpu);
5678 		if (is_nmi(intr_info))
5679 			return true;
5680 		else if (is_page_fault(intr_info))
5681 			return vcpu->arch.apf.host_apf_flags || !enable_ept;
5682 		else if (is_debug(intr_info) &&
5683 			 vcpu->guest_debug &
5684 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5685 			return true;
5686 		else if (is_breakpoint(intr_info) &&
5687 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5688 			return true;
5689 		return false;
5690 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5691 		return true;
5692 	case EXIT_REASON_MCE_DURING_VMENTRY:
5693 		return true;
5694 	case EXIT_REASON_EPT_VIOLATION:
5695 		/*
5696 		 * L0 always deals with the EPT violation. If nested EPT is
5697 		 * used, and the nested mmu code discovers that the address is
5698 		 * missing in the guest EPT table (EPT12), the EPT violation
5699 		 * will be injected with nested_ept_inject_page_fault()
5700 		 */
5701 		return true;
5702 	case EXIT_REASON_EPT_MISCONFIG:
5703 		/*
5704 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5705 		 * table (shadow on EPT) or a merged EPT table that L0 built
5706 		 * (EPT on EPT). So any problems with the structure of the
5707 		 * table is L0's fault.
5708 		 */
5709 		return true;
5710 	case EXIT_REASON_PREEMPTION_TIMER:
5711 		return true;
5712 	case EXIT_REASON_PML_FULL:
5713 		/* We emulate PML support to L1. */
5714 		return true;
5715 	case EXIT_REASON_VMFUNC:
5716 		/* VM functions are emulated through L2->L0 vmexits. */
5717 		return true;
5718 	case EXIT_REASON_ENCLS:
5719 		/* SGX is never exposed to L1 */
5720 		return true;
5721 	default:
5722 		break;
5723 	}
5724 	return false;
5725 }
5726 
5727 /*
5728  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5729  * is_guest_mode (L2).
5730  */
5731 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu, u32 exit_reason)
5732 {
5733 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5734 	u32 intr_info;
5735 
5736 	switch ((u16)exit_reason) {
5737 	case EXIT_REASON_EXCEPTION_NMI:
5738 		intr_info = vmx_get_intr_info(vcpu);
5739 		if (is_nmi(intr_info))
5740 			return true;
5741 		else if (is_page_fault(intr_info))
5742 			return true;
5743 		return vmcs12->exception_bitmap &
5744 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5745 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5746 		return nested_exit_on_intr(vcpu);
5747 	case EXIT_REASON_TRIPLE_FAULT:
5748 		return true;
5749 	case EXIT_REASON_INTERRUPT_WINDOW:
5750 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5751 	case EXIT_REASON_NMI_WINDOW:
5752 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5753 	case EXIT_REASON_TASK_SWITCH:
5754 		return true;
5755 	case EXIT_REASON_CPUID:
5756 		return true;
5757 	case EXIT_REASON_HLT:
5758 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5759 	case EXIT_REASON_INVD:
5760 		return true;
5761 	case EXIT_REASON_INVLPG:
5762 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5763 	case EXIT_REASON_RDPMC:
5764 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5765 	case EXIT_REASON_RDRAND:
5766 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5767 	case EXIT_REASON_RDSEED:
5768 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5769 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5770 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5771 	case EXIT_REASON_VMREAD:
5772 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5773 			vmcs12->vmread_bitmap);
5774 	case EXIT_REASON_VMWRITE:
5775 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5776 			vmcs12->vmwrite_bitmap);
5777 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5778 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5779 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5780 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5781 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5782 		/*
5783 		 * VMX instructions trap unconditionally. This allows L1 to
5784 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5785 		 */
5786 		return true;
5787 	case EXIT_REASON_CR_ACCESS:
5788 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5789 	case EXIT_REASON_DR_ACCESS:
5790 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5791 	case EXIT_REASON_IO_INSTRUCTION:
5792 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5793 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5794 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5795 	case EXIT_REASON_MSR_READ:
5796 	case EXIT_REASON_MSR_WRITE:
5797 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5798 	case EXIT_REASON_INVALID_STATE:
5799 		return true;
5800 	case EXIT_REASON_MWAIT_INSTRUCTION:
5801 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5802 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5803 		return nested_vmx_exit_handled_mtf(vmcs12);
5804 	case EXIT_REASON_MONITOR_INSTRUCTION:
5805 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5806 	case EXIT_REASON_PAUSE_INSTRUCTION:
5807 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5808 			nested_cpu_has2(vmcs12,
5809 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5810 	case EXIT_REASON_MCE_DURING_VMENTRY:
5811 		return true;
5812 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5813 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5814 	case EXIT_REASON_APIC_ACCESS:
5815 	case EXIT_REASON_APIC_WRITE:
5816 	case EXIT_REASON_EOI_INDUCED:
5817 		/*
5818 		 * The controls for "virtualize APIC accesses," "APIC-
5819 		 * register virtualization," and "virtual-interrupt
5820 		 * delivery" only come from vmcs12.
5821 		 */
5822 		return true;
5823 	case EXIT_REASON_INVPCID:
5824 		return
5825 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5826 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5827 	case EXIT_REASON_WBINVD:
5828 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5829 	case EXIT_REASON_XSETBV:
5830 		return true;
5831 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5832 		/*
5833 		 * This should never happen, since it is not possible to
5834 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5835 		 * If if it were, XSS would have to be checked against
5836 		 * the XSS exit bitmap in vmcs12.
5837 		 */
5838 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5839 	case EXIT_REASON_UMWAIT:
5840 	case EXIT_REASON_TPAUSE:
5841 		return nested_cpu_has2(vmcs12,
5842 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
5843 	default:
5844 		return true;
5845 	}
5846 }
5847 
5848 /*
5849  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
5850  * reflected into L1.
5851  */
5852 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
5853 {
5854 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5855 	u32 exit_reason = vmx->exit_reason;
5856 	unsigned long exit_qual;
5857 	u32 exit_intr_info;
5858 
5859 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
5860 
5861 	/*
5862 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
5863 	 * has already loaded L2's state.
5864 	 */
5865 	if (unlikely(vmx->fail)) {
5866 		trace_kvm_nested_vmenter_failed(
5867 			"hardware VM-instruction error: ",
5868 			vmcs_read32(VM_INSTRUCTION_ERROR));
5869 		exit_intr_info = 0;
5870 		exit_qual = 0;
5871 		goto reflect_vmexit;
5872 	}
5873 
5874 	exit_intr_info = vmx_get_intr_info(vcpu);
5875 	exit_qual = vmx_get_exit_qual(vcpu);
5876 
5877 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason, exit_qual,
5878 				vmx->idt_vectoring_info, exit_intr_info,
5879 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5880 				KVM_ISA_VMX);
5881 
5882 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
5883 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
5884 		return false;
5885 
5886 	/* If L1 doesn't want the exit, handle it in L0. */
5887 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
5888 		return false;
5889 
5890 	/*
5891 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
5892 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
5893 	 * need to be synthesized by querying the in-kernel LAPIC, but external
5894 	 * interrupts are never reflected to L1 so it's a non-issue.
5895 	 */
5896 	if ((exit_intr_info &
5897 	     (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
5898 	    (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
5899 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5900 
5901 		vmcs12->vm_exit_intr_error_code =
5902 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5903 	}
5904 
5905 reflect_vmexit:
5906 	nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info, exit_qual);
5907 	return true;
5908 }
5909 
5910 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5911 				struct kvm_nested_state __user *user_kvm_nested_state,
5912 				u32 user_data_size)
5913 {
5914 	struct vcpu_vmx *vmx;
5915 	struct vmcs12 *vmcs12;
5916 	struct kvm_nested_state kvm_state = {
5917 		.flags = 0,
5918 		.format = KVM_STATE_NESTED_FORMAT_VMX,
5919 		.size = sizeof(kvm_state),
5920 		.hdr.vmx.flags = 0,
5921 		.hdr.vmx.vmxon_pa = -1ull,
5922 		.hdr.vmx.vmcs12_pa = -1ull,
5923 		.hdr.vmx.preemption_timer_deadline = 0,
5924 	};
5925 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5926 		&user_kvm_nested_state->data.vmx[0];
5927 
5928 	if (!vcpu)
5929 		return kvm_state.size + sizeof(*user_vmx_nested_state);
5930 
5931 	vmx = to_vmx(vcpu);
5932 	vmcs12 = get_vmcs12(vcpu);
5933 
5934 	if (nested_vmx_allowed(vcpu) &&
5935 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5936 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5937 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
5938 
5939 		if (vmx_has_valid_vmcs12(vcpu)) {
5940 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
5941 
5942 			if (vmx->nested.hv_evmcs)
5943 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5944 
5945 			if (is_guest_mode(vcpu) &&
5946 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5947 			    vmcs12->vmcs_link_pointer != -1ull)
5948 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
5949 		}
5950 
5951 		if (vmx->nested.smm.vmxon)
5952 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5953 
5954 		if (vmx->nested.smm.guest_mode)
5955 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5956 
5957 		if (is_guest_mode(vcpu)) {
5958 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5959 
5960 			if (vmx->nested.nested_run_pending)
5961 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5962 
5963 			if (vmx->nested.mtf_pending)
5964 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
5965 
5966 			if (nested_cpu_has_preemption_timer(vmcs12) &&
5967 			    vmx->nested.has_preemption_timer_deadline) {
5968 				kvm_state.hdr.vmx.flags |=
5969 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
5970 				kvm_state.hdr.vmx.preemption_timer_deadline =
5971 					vmx->nested.preemption_timer_deadline;
5972 			}
5973 		}
5974 	}
5975 
5976 	if (user_data_size < kvm_state.size)
5977 		goto out;
5978 
5979 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5980 		return -EFAULT;
5981 
5982 	if (!vmx_has_valid_vmcs12(vcpu))
5983 		goto out;
5984 
5985 	/*
5986 	 * When running L2, the authoritative vmcs12 state is in the
5987 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5988 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5989 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
5990 	 * vmcs12 state is in the vmcs12 already.
5991 	 */
5992 	if (is_guest_mode(vcpu)) {
5993 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
5994 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5995 	} else if (!vmx->nested.need_vmcs12_to_shadow_sync) {
5996 		if (vmx->nested.hv_evmcs)
5997 			copy_enlightened_to_vmcs12(vmx);
5998 		else if (enable_shadow_vmcs)
5999 			copy_shadow_to_vmcs12(vmx);
6000 	}
6001 
6002 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6003 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6004 
6005 	/*
6006 	 * Copy over the full allocated size of vmcs12 rather than just the size
6007 	 * of the struct.
6008 	 */
6009 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6010 		return -EFAULT;
6011 
6012 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6013 	    vmcs12->vmcs_link_pointer != -1ull) {
6014 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6015 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6016 			return -EFAULT;
6017 	}
6018 out:
6019 	return kvm_state.size;
6020 }
6021 
6022 /*
6023  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6024  */
6025 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6026 {
6027 	if (is_guest_mode(vcpu)) {
6028 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6029 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6030 	}
6031 	free_nested(vcpu);
6032 }
6033 
6034 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6035 				struct kvm_nested_state __user *user_kvm_nested_state,
6036 				struct kvm_nested_state *kvm_state)
6037 {
6038 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6039 	struct vmcs12 *vmcs12;
6040 	enum vm_entry_failure_code ignored;
6041 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6042 		&user_kvm_nested_state->data.vmx[0];
6043 	int ret;
6044 
6045 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6046 		return -EINVAL;
6047 
6048 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
6049 		if (kvm_state->hdr.vmx.smm.flags)
6050 			return -EINVAL;
6051 
6052 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
6053 			return -EINVAL;
6054 
6055 		/*
6056 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6057 		 * enable eVMCS capability on vCPU. However, since then
6058 		 * code was changed such that flag signals vmcs12 should
6059 		 * be copied into eVMCS in guest memory.
6060 		 *
6061 		 * To preserve backwards compatability, allow user
6062 		 * to set this flag even when there is no VMXON region.
6063 		 */
6064 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6065 			return -EINVAL;
6066 	} else {
6067 		if (!nested_vmx_allowed(vcpu))
6068 			return -EINVAL;
6069 
6070 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6071 			return -EINVAL;
6072 	}
6073 
6074 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6075 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6076 		return -EINVAL;
6077 
6078 	if (kvm_state->hdr.vmx.smm.flags &
6079 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6080 		return -EINVAL;
6081 
6082 	/*
6083 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6084 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6085 	 * must be zero.
6086 	 */
6087 	if (is_smm(vcpu) ?
6088 		(kvm_state->flags &
6089 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6090 		: kvm_state->hdr.vmx.smm.flags)
6091 		return -EINVAL;
6092 
6093 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6094 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6095 		return -EINVAL;
6096 
6097 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6098 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6099 			return -EINVAL;
6100 
6101 	vmx_leave_nested(vcpu);
6102 
6103 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
6104 		return 0;
6105 
6106 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6107 	ret = enter_vmx_operation(vcpu);
6108 	if (ret)
6109 		return ret;
6110 
6111 	/* Empty 'VMXON' state is permitted */
6112 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12))
6113 		return 0;
6114 
6115 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
6116 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6117 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6118 			return -EINVAL;
6119 
6120 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6121 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6122 		/*
6123 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6124 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6125 		 * restored yet. EVMCS will be mapped from
6126 		 * nested_get_vmcs12_pages().
6127 		 */
6128 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
6129 	} else {
6130 		return -EINVAL;
6131 	}
6132 
6133 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6134 		vmx->nested.smm.vmxon = true;
6135 		vmx->nested.vmxon = false;
6136 
6137 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6138 			vmx->nested.smm.guest_mode = true;
6139 	}
6140 
6141 	vmcs12 = get_vmcs12(vcpu);
6142 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6143 		return -EFAULT;
6144 
6145 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6146 		return -EINVAL;
6147 
6148 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6149 		return 0;
6150 
6151 	vmx->nested.nested_run_pending =
6152 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6153 
6154 	vmx->nested.mtf_pending =
6155 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6156 
6157 	ret = -EINVAL;
6158 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6159 	    vmcs12->vmcs_link_pointer != -1ull) {
6160 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6161 
6162 		if (kvm_state->size <
6163 		    sizeof(*kvm_state) +
6164 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6165 			goto error_guest_mode;
6166 
6167 		if (copy_from_user(shadow_vmcs12,
6168 				   user_vmx_nested_state->shadow_vmcs12,
6169 				   sizeof(*shadow_vmcs12))) {
6170 			ret = -EFAULT;
6171 			goto error_guest_mode;
6172 		}
6173 
6174 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6175 		    !shadow_vmcs12->hdr.shadow_vmcs)
6176 			goto error_guest_mode;
6177 	}
6178 
6179 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6180 		vmx->nested.has_preemption_timer_deadline = true;
6181 		vmx->nested.preemption_timer_deadline =
6182 			kvm_state->hdr.vmx.preemption_timer_deadline;
6183 	}
6184 
6185 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6186 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6187 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6188 		goto error_guest_mode;
6189 
6190 	vmx->nested.dirty_vmcs12 = true;
6191 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6192 	if (ret)
6193 		goto error_guest_mode;
6194 
6195 	return 0;
6196 
6197 error_guest_mode:
6198 	vmx->nested.nested_run_pending = 0;
6199 	return ret;
6200 }
6201 
6202 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6203 {
6204 	if (enable_shadow_vmcs) {
6205 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6206 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6207 	}
6208 }
6209 
6210 /*
6211  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6212  * returned for the various VMX controls MSRs when nested VMX is enabled.
6213  * The same values should also be used to verify that vmcs12 control fields are
6214  * valid during nested entry from L1 to L2.
6215  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6216  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6217  * bit in the high half is on if the corresponding bit in the control field
6218  * may be on. See also vmx_control_verify().
6219  */
6220 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6221 {
6222 	/*
6223 	 * Note that as a general rule, the high half of the MSRs (bits in
6224 	 * the control fields which may be 1) should be initialized by the
6225 	 * intersection of the underlying hardware's MSR (i.e., features which
6226 	 * can be supported) and the list of features we want to expose -
6227 	 * because they are known to be properly supported in our code.
6228 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6229 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6230 	 * reason is that if one of these bits is necessary, it will appear
6231 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6232 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6233 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6234 	 * These rules have exceptions below.
6235 	 */
6236 
6237 	/* pin-based controls */
6238 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6239 		msrs->pinbased_ctls_low,
6240 		msrs->pinbased_ctls_high);
6241 	msrs->pinbased_ctls_low |=
6242 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6243 	msrs->pinbased_ctls_high &=
6244 		PIN_BASED_EXT_INTR_MASK |
6245 		PIN_BASED_NMI_EXITING |
6246 		PIN_BASED_VIRTUAL_NMIS |
6247 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6248 	msrs->pinbased_ctls_high |=
6249 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6250 		PIN_BASED_VMX_PREEMPTION_TIMER;
6251 
6252 	/* exit controls */
6253 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6254 		msrs->exit_ctls_low,
6255 		msrs->exit_ctls_high);
6256 	msrs->exit_ctls_low =
6257 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6258 
6259 	msrs->exit_ctls_high &=
6260 #ifdef CONFIG_X86_64
6261 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6262 #endif
6263 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
6264 	msrs->exit_ctls_high |=
6265 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6266 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6267 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6268 
6269 	/* We support free control of debug control saving. */
6270 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6271 
6272 	/* entry controls */
6273 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6274 		msrs->entry_ctls_low,
6275 		msrs->entry_ctls_high);
6276 	msrs->entry_ctls_low =
6277 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6278 	msrs->entry_ctls_high &=
6279 #ifdef CONFIG_X86_64
6280 		VM_ENTRY_IA32E_MODE |
6281 #endif
6282 		VM_ENTRY_LOAD_IA32_PAT;
6283 	msrs->entry_ctls_high |=
6284 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6285 
6286 	/* We support free control of debug control loading. */
6287 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6288 
6289 	/* cpu-based controls */
6290 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6291 		msrs->procbased_ctls_low,
6292 		msrs->procbased_ctls_high);
6293 	msrs->procbased_ctls_low =
6294 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6295 	msrs->procbased_ctls_high &=
6296 		CPU_BASED_INTR_WINDOW_EXITING |
6297 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6298 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6299 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6300 		CPU_BASED_CR3_STORE_EXITING |
6301 #ifdef CONFIG_X86_64
6302 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6303 #endif
6304 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6305 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6306 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6307 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6308 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6309 	/*
6310 	 * We can allow some features even when not supported by the
6311 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6312 	 * can use it to avoid exits to L1 - even when L0 runs L2
6313 	 * without MSR bitmaps.
6314 	 */
6315 	msrs->procbased_ctls_high |=
6316 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6317 		CPU_BASED_USE_MSR_BITMAPS;
6318 
6319 	/* We support free control of CR3 access interception. */
6320 	msrs->procbased_ctls_low &=
6321 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6322 
6323 	/*
6324 	 * secondary cpu-based controls.  Do not include those that
6325 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
6326 	 */
6327 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6328 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6329 		      msrs->secondary_ctls_low,
6330 		      msrs->secondary_ctls_high);
6331 
6332 	msrs->secondary_ctls_low = 0;
6333 	msrs->secondary_ctls_high &=
6334 		SECONDARY_EXEC_DESC |
6335 		SECONDARY_EXEC_RDTSCP |
6336 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6337 		SECONDARY_EXEC_WBINVD_EXITING |
6338 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6339 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6340 		SECONDARY_EXEC_RDRAND_EXITING |
6341 		SECONDARY_EXEC_ENABLE_INVPCID |
6342 		SECONDARY_EXEC_RDSEED_EXITING |
6343 		SECONDARY_EXEC_XSAVES;
6344 
6345 	/*
6346 	 * We can emulate "VMCS shadowing," even if the hardware
6347 	 * doesn't support it.
6348 	 */
6349 	msrs->secondary_ctls_high |=
6350 		SECONDARY_EXEC_SHADOW_VMCS;
6351 
6352 	if (enable_ept) {
6353 		/* nested EPT: emulate EPT also to L1 */
6354 		msrs->secondary_ctls_high |=
6355 			SECONDARY_EXEC_ENABLE_EPT;
6356 		msrs->ept_caps =
6357 			VMX_EPT_PAGE_WALK_4_BIT |
6358 			VMX_EPT_PAGE_WALK_5_BIT |
6359 			VMX_EPTP_WB_BIT |
6360 			VMX_EPT_INVEPT_BIT |
6361 			VMX_EPT_EXECUTE_ONLY_BIT;
6362 
6363 		msrs->ept_caps &= ept_caps;
6364 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6365 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6366 			VMX_EPT_1GB_PAGE_BIT;
6367 		if (enable_ept_ad_bits) {
6368 			msrs->secondary_ctls_high |=
6369 				SECONDARY_EXEC_ENABLE_PML;
6370 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6371 		}
6372 	}
6373 
6374 	if (cpu_has_vmx_vmfunc()) {
6375 		msrs->secondary_ctls_high |=
6376 			SECONDARY_EXEC_ENABLE_VMFUNC;
6377 		/*
6378 		 * Advertise EPTP switching unconditionally
6379 		 * since we emulate it
6380 		 */
6381 		if (enable_ept)
6382 			msrs->vmfunc_controls =
6383 				VMX_VMFUNC_EPTP_SWITCHING;
6384 	}
6385 
6386 	/*
6387 	 * Old versions of KVM use the single-context version without
6388 	 * checking for support, so declare that it is supported even
6389 	 * though it is treated as global context.  The alternative is
6390 	 * not failing the single-context invvpid, and it is worse.
6391 	 */
6392 	if (enable_vpid) {
6393 		msrs->secondary_ctls_high |=
6394 			SECONDARY_EXEC_ENABLE_VPID;
6395 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6396 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6397 	}
6398 
6399 	if (enable_unrestricted_guest)
6400 		msrs->secondary_ctls_high |=
6401 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6402 
6403 	if (flexpriority_enabled)
6404 		msrs->secondary_ctls_high |=
6405 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6406 
6407 	/* miscellaneous data */
6408 	rdmsr(MSR_IA32_VMX_MISC,
6409 		msrs->misc_low,
6410 		msrs->misc_high);
6411 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6412 	msrs->misc_low |=
6413 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6414 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6415 		VMX_MISC_ACTIVITY_HLT;
6416 	msrs->misc_high = 0;
6417 
6418 	/*
6419 	 * This MSR reports some information about VMX support. We
6420 	 * should return information about the VMX we emulate for the
6421 	 * guest, and the VMCS structure we give it - not about the
6422 	 * VMX support of the underlying hardware.
6423 	 */
6424 	msrs->basic =
6425 		VMCS12_REVISION |
6426 		VMX_BASIC_TRUE_CTLS |
6427 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6428 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6429 
6430 	if (cpu_has_vmx_basic_inout())
6431 		msrs->basic |= VMX_BASIC_INOUT;
6432 
6433 	/*
6434 	 * These MSRs specify bits which the guest must keep fixed on
6435 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6436 	 * We picked the standard core2 setting.
6437 	 */
6438 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6439 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6440 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6441 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6442 
6443 	/* These MSRs specify bits which the guest must keep fixed off. */
6444 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6445 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6446 
6447 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
6448 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
6449 }
6450 
6451 void nested_vmx_hardware_unsetup(void)
6452 {
6453 	int i;
6454 
6455 	if (enable_shadow_vmcs) {
6456 		for (i = 0; i < VMX_BITMAP_NR; i++)
6457 			free_page((unsigned long)vmx_bitmap[i]);
6458 	}
6459 }
6460 
6461 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6462 {
6463 	int i;
6464 
6465 	if (!cpu_has_vmx_shadow_vmcs())
6466 		enable_shadow_vmcs = 0;
6467 	if (enable_shadow_vmcs) {
6468 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6469 			/*
6470 			 * The vmx_bitmap is not tied to a VM and so should
6471 			 * not be charged to a memcg.
6472 			 */
6473 			vmx_bitmap[i] = (unsigned long *)
6474 				__get_free_page(GFP_KERNEL);
6475 			if (!vmx_bitmap[i]) {
6476 				nested_vmx_hardware_unsetup();
6477 				return -ENOMEM;
6478 			}
6479 		}
6480 
6481 		init_vmcs_shadow_fields();
6482 	}
6483 
6484 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6485 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6486 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6487 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6488 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6489 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6490 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6491 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6492 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6493 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6494 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6495 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6496 
6497 	return 0;
6498 }
6499 
6500 struct kvm_x86_nested_ops vmx_nested_ops = {
6501 	.check_events = vmx_check_nested_events,
6502 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6503 	.get_state = vmx_get_nested_state,
6504 	.set_state = vmx_set_nested_state,
6505 	.get_vmcs12_pages = nested_get_vmcs12_pages,
6506 	.enable_evmcs = nested_enable_evmcs,
6507 	.get_evmcs_version = nested_get_evmcs_version,
6508 };
6509