xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision b830f94f)
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 "trace.h"
14 #include "x86.h"
15 
16 static bool __read_mostly enable_shadow_vmcs = 1;
17 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
18 
19 static bool __read_mostly nested_early_check = 0;
20 module_param(nested_early_check, bool, S_IRUGO);
21 
22 /*
23  * Hyper-V requires all of these, so mark them as supported even though
24  * they are just treated the same as all-context.
25  */
26 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
27 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
28 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
29 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
30 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
31 
32 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
33 
34 enum {
35 	VMX_VMREAD_BITMAP,
36 	VMX_VMWRITE_BITMAP,
37 	VMX_BITMAP_NR
38 };
39 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
40 
41 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
42 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
43 
44 struct shadow_vmcs_field {
45 	u16	encoding;
46 	u16	offset;
47 };
48 static struct shadow_vmcs_field shadow_read_only_fields[] = {
49 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
50 #include "vmcs_shadow_fields.h"
51 };
52 static int max_shadow_read_only_fields =
53 	ARRAY_SIZE(shadow_read_only_fields);
54 
55 static struct shadow_vmcs_field shadow_read_write_fields[] = {
56 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
57 #include "vmcs_shadow_fields.h"
58 };
59 static int max_shadow_read_write_fields =
60 	ARRAY_SIZE(shadow_read_write_fields);
61 
62 static void init_vmcs_shadow_fields(void)
63 {
64 	int i, j;
65 
66 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
67 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
68 
69 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
70 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
71 		u16 field = entry.encoding;
72 
73 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
74 		    (i + 1 == max_shadow_read_only_fields ||
75 		     shadow_read_only_fields[i + 1].encoding != field + 1))
76 			pr_err("Missing field from shadow_read_only_field %x\n",
77 			       field + 1);
78 
79 		clear_bit(field, vmx_vmread_bitmap);
80 		if (field & 1)
81 #ifdef CONFIG_X86_64
82 			continue;
83 #else
84 			entry.offset += sizeof(u32);
85 #endif
86 		shadow_read_only_fields[j++] = entry;
87 	}
88 	max_shadow_read_only_fields = j;
89 
90 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
91 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
92 		u16 field = entry.encoding;
93 
94 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
95 		    (i + 1 == max_shadow_read_write_fields ||
96 		     shadow_read_write_fields[i + 1].encoding != field + 1))
97 			pr_err("Missing field from shadow_read_write_field %x\n",
98 			       field + 1);
99 
100 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
101 			  field <= GUEST_TR_AR_BYTES,
102 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
103 
104 		/*
105 		 * PML and the preemption timer can be emulated, but the
106 		 * processor cannot vmwrite to fields that don't exist
107 		 * on bare metal.
108 		 */
109 		switch (field) {
110 		case GUEST_PML_INDEX:
111 			if (!cpu_has_vmx_pml())
112 				continue;
113 			break;
114 		case VMX_PREEMPTION_TIMER_VALUE:
115 			if (!cpu_has_vmx_preemption_timer())
116 				continue;
117 			break;
118 		case GUEST_INTR_STATUS:
119 			if (!cpu_has_vmx_apicv())
120 				continue;
121 			break;
122 		default:
123 			break;
124 		}
125 
126 		clear_bit(field, vmx_vmwrite_bitmap);
127 		clear_bit(field, vmx_vmread_bitmap);
128 		if (field & 1)
129 #ifdef CONFIG_X86_64
130 			continue;
131 #else
132 			entry.offset += sizeof(u32);
133 #endif
134 		shadow_read_write_fields[j++] = entry;
135 	}
136 	max_shadow_read_write_fields = j;
137 }
138 
139 /*
140  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
141  * set the success or error code of an emulated VMX instruction (as specified
142  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
143  * instruction.
144  */
145 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
146 {
147 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
148 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
149 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
150 	return kvm_skip_emulated_instruction(vcpu);
151 }
152 
153 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
154 {
155 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
156 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
157 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
158 			| X86_EFLAGS_CF);
159 	return kvm_skip_emulated_instruction(vcpu);
160 }
161 
162 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
163 				u32 vm_instruction_error)
164 {
165 	struct vcpu_vmx *vmx = to_vmx(vcpu);
166 
167 	/*
168 	 * failValid writes the error number to the current VMCS, which
169 	 * can't be done if there isn't a current VMCS.
170 	 */
171 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
172 		return nested_vmx_failInvalid(vcpu);
173 
174 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
175 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
176 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
177 			| X86_EFLAGS_ZF);
178 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
179 	/*
180 	 * We don't need to force a shadow sync because
181 	 * VM_INSTRUCTION_ERROR is not shadowed
182 	 */
183 	return kvm_skip_emulated_instruction(vcpu);
184 }
185 
186 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
187 {
188 	/* TODO: not to reset guest simply here. */
189 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
190 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
191 }
192 
193 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
194 {
195 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
196 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
197 	vmx->nested.need_vmcs12_to_shadow_sync = false;
198 }
199 
200 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
201 {
202 	struct vcpu_vmx *vmx = to_vmx(vcpu);
203 
204 	if (!vmx->nested.hv_evmcs)
205 		return;
206 
207 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
208 	vmx->nested.hv_evmcs_vmptr = -1ull;
209 	vmx->nested.hv_evmcs = NULL;
210 }
211 
212 /*
213  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
214  * just stops using VMX.
215  */
216 static void free_nested(struct kvm_vcpu *vcpu)
217 {
218 	struct vcpu_vmx *vmx = to_vmx(vcpu);
219 
220 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
221 		return;
222 
223 	vmx->nested.vmxon = false;
224 	vmx->nested.smm.vmxon = false;
225 	free_vpid(vmx->nested.vpid02);
226 	vmx->nested.posted_intr_nv = -1;
227 	vmx->nested.current_vmptr = -1ull;
228 	if (enable_shadow_vmcs) {
229 		vmx_disable_shadow_vmcs(vmx);
230 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
231 		free_vmcs(vmx->vmcs01.shadow_vmcs);
232 		vmx->vmcs01.shadow_vmcs = NULL;
233 	}
234 	kfree(vmx->nested.cached_vmcs12);
235 	kfree(vmx->nested.cached_shadow_vmcs12);
236 	/* Unpin physical memory we referred to in the vmcs02 */
237 	if (vmx->nested.apic_access_page) {
238 		kvm_release_page_dirty(vmx->nested.apic_access_page);
239 		vmx->nested.apic_access_page = NULL;
240 	}
241 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
242 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
243 	vmx->nested.pi_desc = NULL;
244 
245 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
246 
247 	nested_release_evmcs(vcpu);
248 
249 	free_loaded_vmcs(&vmx->nested.vmcs02);
250 }
251 
252 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
253 				     struct loaded_vmcs *prev)
254 {
255 	struct vmcs_host_state *dest, *src;
256 
257 	if (unlikely(!vmx->guest_state_loaded))
258 		return;
259 
260 	src = &prev->host_state;
261 	dest = &vmx->loaded_vmcs->host_state;
262 
263 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
264 	dest->ldt_sel = src->ldt_sel;
265 #ifdef CONFIG_X86_64
266 	dest->ds_sel = src->ds_sel;
267 	dest->es_sel = src->es_sel;
268 #endif
269 }
270 
271 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
272 {
273 	struct vcpu_vmx *vmx = to_vmx(vcpu);
274 	struct loaded_vmcs *prev;
275 	int cpu;
276 
277 	if (vmx->loaded_vmcs == vmcs)
278 		return;
279 
280 	cpu = get_cpu();
281 	prev = vmx->loaded_vmcs;
282 	vmx->loaded_vmcs = vmcs;
283 	vmx_vcpu_load_vmcs(vcpu, cpu);
284 	vmx_sync_vmcs_host_state(vmx, prev);
285 	put_cpu();
286 
287 	vmx_segment_cache_clear(vmx);
288 }
289 
290 /*
291  * Ensure that the current vmcs of the logical processor is the
292  * vmcs01 of the vcpu before calling free_nested().
293  */
294 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
295 {
296 	vcpu_load(vcpu);
297 	vmx_leave_nested(vcpu);
298 	vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
299 	free_nested(vcpu);
300 	vcpu_put(vcpu);
301 }
302 
303 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
304 		struct x86_exception *fault)
305 {
306 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
307 	struct vcpu_vmx *vmx = to_vmx(vcpu);
308 	u32 exit_reason;
309 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
310 
311 	if (vmx->nested.pml_full) {
312 		exit_reason = EXIT_REASON_PML_FULL;
313 		vmx->nested.pml_full = false;
314 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
315 	} else if (fault->error_code & PFERR_RSVD_MASK)
316 		exit_reason = EXIT_REASON_EPT_MISCONFIG;
317 	else
318 		exit_reason = EXIT_REASON_EPT_VIOLATION;
319 
320 	nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
321 	vmcs12->guest_physical_address = fault->address;
322 }
323 
324 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
325 {
326 	WARN_ON(mmu_is_nested(vcpu));
327 
328 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
329 	kvm_init_shadow_ept_mmu(vcpu,
330 			to_vmx(vcpu)->nested.msrs.ept_caps &
331 			VMX_EPT_EXECUTE_ONLY_BIT,
332 			nested_ept_ad_enabled(vcpu),
333 			nested_ept_get_cr3(vcpu));
334 	vcpu->arch.mmu->set_cr3           = vmx_set_cr3;
335 	vcpu->arch.mmu->get_cr3           = nested_ept_get_cr3;
336 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
337 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
338 
339 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
340 }
341 
342 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
343 {
344 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
345 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
346 }
347 
348 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
349 					    u16 error_code)
350 {
351 	bool inequality, bit;
352 
353 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
354 	inequality =
355 		(error_code & vmcs12->page_fault_error_code_mask) !=
356 		 vmcs12->page_fault_error_code_match;
357 	return inequality ^ bit;
358 }
359 
360 
361 /*
362  * KVM wants to inject page-faults which it got to the guest. This function
363  * checks whether in a nested guest, we need to inject them to L1 or L2.
364  */
365 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
366 {
367 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
368 	unsigned int nr = vcpu->arch.exception.nr;
369 	bool has_payload = vcpu->arch.exception.has_payload;
370 	unsigned long payload = vcpu->arch.exception.payload;
371 
372 	if (nr == PF_VECTOR) {
373 		if (vcpu->arch.exception.nested_apf) {
374 			*exit_qual = vcpu->arch.apf.nested_apf_token;
375 			return 1;
376 		}
377 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
378 						    vcpu->arch.exception.error_code)) {
379 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
380 			return 1;
381 		}
382 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
383 		if (nr == DB_VECTOR) {
384 			if (!has_payload) {
385 				payload = vcpu->arch.dr6;
386 				payload &= ~(DR6_FIXED_1 | DR6_BT);
387 				payload ^= DR6_RTM;
388 			}
389 			*exit_qual = payload;
390 		} else
391 			*exit_qual = 0;
392 		return 1;
393 	}
394 
395 	return 0;
396 }
397 
398 
399 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
400 		struct x86_exception *fault)
401 {
402 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
403 
404 	WARN_ON(!is_guest_mode(vcpu));
405 
406 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
407 		!to_vmx(vcpu)->nested.nested_run_pending) {
408 		vmcs12->vm_exit_intr_error_code = fault->error_code;
409 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
410 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
411 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
412 				  fault->address);
413 	} else {
414 		kvm_inject_page_fault(vcpu, fault);
415 	}
416 }
417 
418 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
419 {
420 	return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
421 }
422 
423 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
424 					       struct vmcs12 *vmcs12)
425 {
426 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
427 		return 0;
428 
429 	if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
430 	    !page_address_valid(vcpu, vmcs12->io_bitmap_b))
431 		return -EINVAL;
432 
433 	return 0;
434 }
435 
436 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
437 						struct vmcs12 *vmcs12)
438 {
439 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
440 		return 0;
441 
442 	if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
443 		return -EINVAL;
444 
445 	return 0;
446 }
447 
448 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
449 						struct vmcs12 *vmcs12)
450 {
451 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
452 		return 0;
453 
454 	if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
455 		return -EINVAL;
456 
457 	return 0;
458 }
459 
460 /*
461  * Check if MSR is intercepted for L01 MSR bitmap.
462  */
463 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
464 {
465 	unsigned long *msr_bitmap;
466 	int f = sizeof(unsigned long);
467 
468 	if (!cpu_has_vmx_msr_bitmap())
469 		return true;
470 
471 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
472 
473 	if (msr <= 0x1fff) {
474 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
475 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
476 		msr &= 0x1fff;
477 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
478 	}
479 
480 	return true;
481 }
482 
483 /*
484  * If a msr is allowed by L0, we should check whether it is allowed by L1.
485  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
486  */
487 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
488 					       unsigned long *msr_bitmap_nested,
489 					       u32 msr, int type)
490 {
491 	int f = sizeof(unsigned long);
492 
493 	/*
494 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
495 	 * have the write-low and read-high bitmap offsets the wrong way round.
496 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
497 	 */
498 	if (msr <= 0x1fff) {
499 		if (type & MSR_TYPE_R &&
500 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
501 			/* read-low */
502 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
503 
504 		if (type & MSR_TYPE_W &&
505 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
506 			/* write-low */
507 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
508 
509 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
510 		msr &= 0x1fff;
511 		if (type & MSR_TYPE_R &&
512 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
513 			/* read-high */
514 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
515 
516 		if (type & MSR_TYPE_W &&
517 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
518 			/* write-high */
519 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
520 
521 	}
522 }
523 
524 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap) {
525 	int msr;
526 
527 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
528 		unsigned word = msr / BITS_PER_LONG;
529 
530 		msr_bitmap[word] = ~0;
531 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
532 	}
533 }
534 
535 /*
536  * Merge L0's and L1's MSR bitmap, return false to indicate that
537  * we do not use the hardware.
538  */
539 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
540 						 struct vmcs12 *vmcs12)
541 {
542 	int msr;
543 	unsigned long *msr_bitmap_l1;
544 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
545 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
546 
547 	/* Nothing to do if the MSR bitmap is not in use.  */
548 	if (!cpu_has_vmx_msr_bitmap() ||
549 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
550 		return false;
551 
552 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
553 		return false;
554 
555 	msr_bitmap_l1 = (unsigned long *)map->hva;
556 
557 	/*
558 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
559 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
560 	 * the x2APIC MSR range and selectively disable them below.
561 	 */
562 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
563 
564 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
565 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
566 			/*
567 			 * L0 need not intercept reads for MSRs between 0x800
568 			 * and 0x8ff, it just lets the processor take the value
569 			 * from the virtual-APIC page; take those 256 bits
570 			 * directly from the L1 bitmap.
571 			 */
572 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
573 				unsigned word = msr / BITS_PER_LONG;
574 
575 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
576 			}
577 		}
578 
579 		nested_vmx_disable_intercept_for_msr(
580 			msr_bitmap_l1, msr_bitmap_l0,
581 			X2APIC_MSR(APIC_TASKPRI),
582 			MSR_TYPE_R | MSR_TYPE_W);
583 
584 		if (nested_cpu_has_vid(vmcs12)) {
585 			nested_vmx_disable_intercept_for_msr(
586 				msr_bitmap_l1, msr_bitmap_l0,
587 				X2APIC_MSR(APIC_EOI),
588 				MSR_TYPE_W);
589 			nested_vmx_disable_intercept_for_msr(
590 				msr_bitmap_l1, msr_bitmap_l0,
591 				X2APIC_MSR(APIC_SELF_IPI),
592 				MSR_TYPE_W);
593 		}
594 	}
595 
596 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
597 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
598 					     MSR_FS_BASE, MSR_TYPE_RW);
599 
600 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
601 					     MSR_GS_BASE, MSR_TYPE_RW);
602 
603 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
604 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
605 
606 	/*
607 	 * Checking the L0->L1 bitmap is trying to verify two things:
608 	 *
609 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
610 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
611 	 *    from the L12 MSR bitmap that is too permissive.
612 	 * 2. That L1 or L2s have actually used the MSR. This avoids
613 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
614 	 *    works properly because we only update the L01 MSR bitmap lazily.
615 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
616 	 *    updated to reflect this when L1 (or its L2s) actually write to
617 	 *    the MSR.
618 	 */
619 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
620 		nested_vmx_disable_intercept_for_msr(
621 					msr_bitmap_l1, msr_bitmap_l0,
622 					MSR_IA32_SPEC_CTRL,
623 					MSR_TYPE_R | MSR_TYPE_W);
624 
625 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
626 		nested_vmx_disable_intercept_for_msr(
627 					msr_bitmap_l1, msr_bitmap_l0,
628 					MSR_IA32_PRED_CMD,
629 					MSR_TYPE_W);
630 
631 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
632 
633 	return true;
634 }
635 
636 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
637 				       struct vmcs12 *vmcs12)
638 {
639 	struct kvm_host_map map;
640 	struct vmcs12 *shadow;
641 
642 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
643 	    vmcs12->vmcs_link_pointer == -1ull)
644 		return;
645 
646 	shadow = get_shadow_vmcs12(vcpu);
647 
648 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
649 		return;
650 
651 	memcpy(shadow, map.hva, VMCS12_SIZE);
652 	kvm_vcpu_unmap(vcpu, &map, false);
653 }
654 
655 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
656 					      struct vmcs12 *vmcs12)
657 {
658 	struct vcpu_vmx *vmx = to_vmx(vcpu);
659 
660 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
661 	    vmcs12->vmcs_link_pointer == -1ull)
662 		return;
663 
664 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
665 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
666 }
667 
668 /*
669  * In nested virtualization, check if L1 has set
670  * VM_EXIT_ACK_INTR_ON_EXIT
671  */
672 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
673 {
674 	return get_vmcs12(vcpu)->vm_exit_controls &
675 		VM_EXIT_ACK_INTR_ON_EXIT;
676 }
677 
678 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
679 {
680 	return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
681 }
682 
683 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
684 					  struct vmcs12 *vmcs12)
685 {
686 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
687 	    !page_address_valid(vcpu, vmcs12->apic_access_addr))
688 		return -EINVAL;
689 	else
690 		return 0;
691 }
692 
693 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
694 					   struct vmcs12 *vmcs12)
695 {
696 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
697 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
698 	    !nested_cpu_has_vid(vmcs12) &&
699 	    !nested_cpu_has_posted_intr(vmcs12))
700 		return 0;
701 
702 	/*
703 	 * If virtualize x2apic mode is enabled,
704 	 * virtualize apic access must be disabled.
705 	 */
706 	if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
707 	    nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
708 		return -EINVAL;
709 
710 	/*
711 	 * If virtual interrupt delivery is enabled,
712 	 * we must exit on external interrupts.
713 	 */
714 	if (nested_cpu_has_vid(vmcs12) &&
715 	   !nested_exit_on_intr(vcpu))
716 		return -EINVAL;
717 
718 	/*
719 	 * bits 15:8 should be zero in posted_intr_nv,
720 	 * the descriptor address has been already checked
721 	 * in nested_get_vmcs12_pages.
722 	 *
723 	 * bits 5:0 of posted_intr_desc_addr should be zero.
724 	 */
725 	if (nested_cpu_has_posted_intr(vmcs12) &&
726 	   (!nested_cpu_has_vid(vmcs12) ||
727 	    !nested_exit_intr_ack_set(vcpu) ||
728 	    (vmcs12->posted_intr_nv & 0xff00) ||
729 	    (vmcs12->posted_intr_desc_addr & 0x3f) ||
730 	    (vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu))))
731 		return -EINVAL;
732 
733 	/* tpr shadow is needed by all apicv features. */
734 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
735 		return -EINVAL;
736 
737 	return 0;
738 }
739 
740 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
741 				       u32 count, u64 addr)
742 {
743 	int maxphyaddr;
744 
745 	if (count == 0)
746 		return 0;
747 	maxphyaddr = cpuid_maxphyaddr(vcpu);
748 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
749 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
750 		return -EINVAL;
751 
752 	return 0;
753 }
754 
755 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
756 						     struct vmcs12 *vmcs12)
757 {
758 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_load_count,
759 					vmcs12->vm_exit_msr_load_addr) ||
760 	    nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_store_count,
761 					vmcs12->vm_exit_msr_store_addr))
762 		return -EINVAL;
763 
764 	return 0;
765 }
766 
767 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
768                                                       struct vmcs12 *vmcs12)
769 {
770 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_entry_msr_load_count,
771                                         vmcs12->vm_entry_msr_load_addr))
772                 return -EINVAL;
773 
774 	return 0;
775 }
776 
777 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
778 					 struct vmcs12 *vmcs12)
779 {
780 	if (!nested_cpu_has_pml(vmcs12))
781 		return 0;
782 
783 	if (!nested_cpu_has_ept(vmcs12) ||
784 	    !page_address_valid(vcpu, vmcs12->pml_address))
785 		return -EINVAL;
786 
787 	return 0;
788 }
789 
790 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
791 							struct vmcs12 *vmcs12)
792 {
793 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
794 	    !nested_cpu_has_ept(vmcs12))
795 		return -EINVAL;
796 	return 0;
797 }
798 
799 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
800 							 struct vmcs12 *vmcs12)
801 {
802 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
803 	    !nested_cpu_has_ept(vmcs12))
804 		return -EINVAL;
805 	return 0;
806 }
807 
808 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
809 						 struct vmcs12 *vmcs12)
810 {
811 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
812 		return 0;
813 
814 	if (!page_address_valid(vcpu, vmcs12->vmread_bitmap) ||
815 	    !page_address_valid(vcpu, vmcs12->vmwrite_bitmap))
816 		return -EINVAL;
817 
818 	return 0;
819 }
820 
821 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
822 				       struct vmx_msr_entry *e)
823 {
824 	/* x2APIC MSR accesses are not allowed */
825 	if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
826 		return -EINVAL;
827 	if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
828 	    e->index == MSR_IA32_UCODE_REV)
829 		return -EINVAL;
830 	if (e->reserved != 0)
831 		return -EINVAL;
832 	return 0;
833 }
834 
835 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
836 				     struct vmx_msr_entry *e)
837 {
838 	if (e->index == MSR_FS_BASE ||
839 	    e->index == MSR_GS_BASE ||
840 	    e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
841 	    nested_vmx_msr_check_common(vcpu, e))
842 		return -EINVAL;
843 	return 0;
844 }
845 
846 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
847 				      struct vmx_msr_entry *e)
848 {
849 	if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
850 	    nested_vmx_msr_check_common(vcpu, e))
851 		return -EINVAL;
852 	return 0;
853 }
854 
855 /*
856  * Load guest's/host's msr at nested entry/exit.
857  * return 0 for success, entry index for failure.
858  */
859 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
860 {
861 	u32 i;
862 	struct vmx_msr_entry e;
863 	struct msr_data msr;
864 
865 	msr.host_initiated = false;
866 	for (i = 0; i < count; i++) {
867 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
868 					&e, sizeof(e))) {
869 			pr_debug_ratelimited(
870 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
871 				__func__, i, gpa + i * sizeof(e));
872 			goto fail;
873 		}
874 		if (nested_vmx_load_msr_check(vcpu, &e)) {
875 			pr_debug_ratelimited(
876 				"%s check failed (%u, 0x%x, 0x%x)\n",
877 				__func__, i, e.index, e.reserved);
878 			goto fail;
879 		}
880 		msr.index = e.index;
881 		msr.data = e.value;
882 		if (kvm_set_msr(vcpu, &msr)) {
883 			pr_debug_ratelimited(
884 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
885 				__func__, i, e.index, e.value);
886 			goto fail;
887 		}
888 	}
889 	return 0;
890 fail:
891 	return i + 1;
892 }
893 
894 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
895 {
896 	u32 i;
897 	struct vmx_msr_entry e;
898 
899 	for (i = 0; i < count; i++) {
900 		struct msr_data msr_info;
901 		if (kvm_vcpu_read_guest(vcpu,
902 					gpa + i * sizeof(e),
903 					&e, 2 * sizeof(u32))) {
904 			pr_debug_ratelimited(
905 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
906 				__func__, i, gpa + i * sizeof(e));
907 			return -EINVAL;
908 		}
909 		if (nested_vmx_store_msr_check(vcpu, &e)) {
910 			pr_debug_ratelimited(
911 				"%s check failed (%u, 0x%x, 0x%x)\n",
912 				__func__, i, e.index, e.reserved);
913 			return -EINVAL;
914 		}
915 		msr_info.host_initiated = false;
916 		msr_info.index = e.index;
917 		if (kvm_get_msr(vcpu, &msr_info)) {
918 			pr_debug_ratelimited(
919 				"%s cannot read MSR (%u, 0x%x)\n",
920 				__func__, i, e.index);
921 			return -EINVAL;
922 		}
923 		if (kvm_vcpu_write_guest(vcpu,
924 					 gpa + i * sizeof(e) +
925 					     offsetof(struct vmx_msr_entry, value),
926 					 &msr_info.data, sizeof(msr_info.data))) {
927 			pr_debug_ratelimited(
928 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
929 				__func__, i, e.index, msr_info.data);
930 			return -EINVAL;
931 		}
932 	}
933 	return 0;
934 }
935 
936 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
937 {
938 	unsigned long invalid_mask;
939 
940 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
941 	return (val & invalid_mask) == 0;
942 }
943 
944 /*
945  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
946  * emulating VM entry into a guest with EPT enabled.
947  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
948  * is assigned to entry_failure_code on failure.
949  */
950 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
951 			       u32 *entry_failure_code)
952 {
953 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
954 		if (!nested_cr3_valid(vcpu, cr3)) {
955 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
956 			return -EINVAL;
957 		}
958 
959 		/*
960 		 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
961 		 * must not be dereferenced.
962 		 */
963 		if (is_pae_paging(vcpu) && !nested_ept) {
964 			if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
965 				*entry_failure_code = ENTRY_FAIL_PDPTE;
966 				return -EINVAL;
967 			}
968 		}
969 	}
970 
971 	if (!nested_ept)
972 		kvm_mmu_new_cr3(vcpu, cr3, false);
973 
974 	vcpu->arch.cr3 = cr3;
975 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
976 
977 	kvm_init_mmu(vcpu, false);
978 
979 	return 0;
980 }
981 
982 /*
983  * Returns if KVM is able to config CPU to tag TLB entries
984  * populated by L2 differently than TLB entries populated
985  * by L1.
986  *
987  * If L1 uses EPT, then TLB entries are tagged with different EPTP.
988  *
989  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
990  * with different VPID (L1 entries are tagged with vmx->vpid
991  * while L2 entries are tagged with vmx->nested.vpid02).
992  */
993 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
994 {
995 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
996 
997 	return nested_cpu_has_ept(vmcs12) ||
998 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
999 }
1000 
1001 static u16 nested_get_vpid02(struct kvm_vcpu *vcpu)
1002 {
1003 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1004 
1005 	return vmx->nested.vpid02 ? vmx->nested.vpid02 : vmx->vpid;
1006 }
1007 
1008 
1009 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1010 {
1011 	return fixed_bits_valid(control, low, high);
1012 }
1013 
1014 static inline u64 vmx_control_msr(u32 low, u32 high)
1015 {
1016 	return low | ((u64)high << 32);
1017 }
1018 
1019 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1020 {
1021 	superset &= mask;
1022 	subset &= mask;
1023 
1024 	return (superset | subset) == superset;
1025 }
1026 
1027 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1028 {
1029 	const u64 feature_and_reserved =
1030 		/* feature (except bit 48; see below) */
1031 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1032 		/* reserved */
1033 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1034 	u64 vmx_basic = vmx->nested.msrs.basic;
1035 
1036 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1037 		return -EINVAL;
1038 
1039 	/*
1040 	 * KVM does not emulate a version of VMX that constrains physical
1041 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1042 	 */
1043 	if (data & BIT_ULL(48))
1044 		return -EINVAL;
1045 
1046 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1047 	    vmx_basic_vmcs_revision_id(data))
1048 		return -EINVAL;
1049 
1050 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1051 		return -EINVAL;
1052 
1053 	vmx->nested.msrs.basic = data;
1054 	return 0;
1055 }
1056 
1057 static int
1058 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1059 {
1060 	u64 supported;
1061 	u32 *lowp, *highp;
1062 
1063 	switch (msr_index) {
1064 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1065 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1066 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1067 		break;
1068 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1069 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1070 		highp = &vmx->nested.msrs.procbased_ctls_high;
1071 		break;
1072 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1073 		lowp = &vmx->nested.msrs.exit_ctls_low;
1074 		highp = &vmx->nested.msrs.exit_ctls_high;
1075 		break;
1076 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1077 		lowp = &vmx->nested.msrs.entry_ctls_low;
1078 		highp = &vmx->nested.msrs.entry_ctls_high;
1079 		break;
1080 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1081 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1082 		highp = &vmx->nested.msrs.secondary_ctls_high;
1083 		break;
1084 	default:
1085 		BUG();
1086 	}
1087 
1088 	supported = vmx_control_msr(*lowp, *highp);
1089 
1090 	/* Check must-be-1 bits are still 1. */
1091 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1092 		return -EINVAL;
1093 
1094 	/* Check must-be-0 bits are still 0. */
1095 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1096 		return -EINVAL;
1097 
1098 	*lowp = data;
1099 	*highp = data >> 32;
1100 	return 0;
1101 }
1102 
1103 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1104 {
1105 	const u64 feature_and_reserved_bits =
1106 		/* feature */
1107 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1108 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1109 		/* reserved */
1110 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1111 	u64 vmx_misc;
1112 
1113 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1114 				   vmx->nested.msrs.misc_high);
1115 
1116 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1117 		return -EINVAL;
1118 
1119 	if ((vmx->nested.msrs.pinbased_ctls_high &
1120 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1121 	    vmx_misc_preemption_timer_rate(data) !=
1122 	    vmx_misc_preemption_timer_rate(vmx_misc))
1123 		return -EINVAL;
1124 
1125 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1126 		return -EINVAL;
1127 
1128 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1129 		return -EINVAL;
1130 
1131 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1132 		return -EINVAL;
1133 
1134 	vmx->nested.msrs.misc_low = data;
1135 	vmx->nested.msrs.misc_high = data >> 32;
1136 
1137 	return 0;
1138 }
1139 
1140 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1141 {
1142 	u64 vmx_ept_vpid_cap;
1143 
1144 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1145 					   vmx->nested.msrs.vpid_caps);
1146 
1147 	/* Every bit is either reserved or a feature bit. */
1148 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1149 		return -EINVAL;
1150 
1151 	vmx->nested.msrs.ept_caps = data;
1152 	vmx->nested.msrs.vpid_caps = data >> 32;
1153 	return 0;
1154 }
1155 
1156 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1157 {
1158 	u64 *msr;
1159 
1160 	switch (msr_index) {
1161 	case MSR_IA32_VMX_CR0_FIXED0:
1162 		msr = &vmx->nested.msrs.cr0_fixed0;
1163 		break;
1164 	case MSR_IA32_VMX_CR4_FIXED0:
1165 		msr = &vmx->nested.msrs.cr4_fixed0;
1166 		break;
1167 	default:
1168 		BUG();
1169 	}
1170 
1171 	/*
1172 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1173 	 * must be 1 in the restored value.
1174 	 */
1175 	if (!is_bitwise_subset(data, *msr, -1ULL))
1176 		return -EINVAL;
1177 
1178 	*msr = data;
1179 	return 0;
1180 }
1181 
1182 /*
1183  * Called when userspace is restoring VMX MSRs.
1184  *
1185  * Returns 0 on success, non-0 otherwise.
1186  */
1187 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1188 {
1189 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1190 
1191 	/*
1192 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1193 	 * is in VMX operation.
1194 	 */
1195 	if (vmx->nested.vmxon)
1196 		return -EBUSY;
1197 
1198 	switch (msr_index) {
1199 	case MSR_IA32_VMX_BASIC:
1200 		return vmx_restore_vmx_basic(vmx, data);
1201 	case MSR_IA32_VMX_PINBASED_CTLS:
1202 	case MSR_IA32_VMX_PROCBASED_CTLS:
1203 	case MSR_IA32_VMX_EXIT_CTLS:
1204 	case MSR_IA32_VMX_ENTRY_CTLS:
1205 		/*
1206 		 * The "non-true" VMX capability MSRs are generated from the
1207 		 * "true" MSRs, so we do not support restoring them directly.
1208 		 *
1209 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1210 		 * should restore the "true" MSRs with the must-be-1 bits
1211 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1212 		 * DEFAULT SETTINGS".
1213 		 */
1214 		return -EINVAL;
1215 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1216 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1217 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1218 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1219 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1220 		return vmx_restore_control_msr(vmx, msr_index, data);
1221 	case MSR_IA32_VMX_MISC:
1222 		return vmx_restore_vmx_misc(vmx, data);
1223 	case MSR_IA32_VMX_CR0_FIXED0:
1224 	case MSR_IA32_VMX_CR4_FIXED0:
1225 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1226 	case MSR_IA32_VMX_CR0_FIXED1:
1227 	case MSR_IA32_VMX_CR4_FIXED1:
1228 		/*
1229 		 * These MSRs are generated based on the vCPU's CPUID, so we
1230 		 * do not support restoring them directly.
1231 		 */
1232 		return -EINVAL;
1233 	case MSR_IA32_VMX_EPT_VPID_CAP:
1234 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1235 	case MSR_IA32_VMX_VMCS_ENUM:
1236 		vmx->nested.msrs.vmcs_enum = data;
1237 		return 0;
1238 	case MSR_IA32_VMX_VMFUNC:
1239 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1240 			return -EINVAL;
1241 		vmx->nested.msrs.vmfunc_controls = data;
1242 		return 0;
1243 	default:
1244 		/*
1245 		 * The rest of the VMX capability MSRs do not support restore.
1246 		 */
1247 		return -EINVAL;
1248 	}
1249 }
1250 
1251 /* Returns 0 on success, non-0 otherwise. */
1252 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1253 {
1254 	switch (msr_index) {
1255 	case MSR_IA32_VMX_BASIC:
1256 		*pdata = msrs->basic;
1257 		break;
1258 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1259 	case MSR_IA32_VMX_PINBASED_CTLS:
1260 		*pdata = vmx_control_msr(
1261 			msrs->pinbased_ctls_low,
1262 			msrs->pinbased_ctls_high);
1263 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1264 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1265 		break;
1266 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1267 	case MSR_IA32_VMX_PROCBASED_CTLS:
1268 		*pdata = vmx_control_msr(
1269 			msrs->procbased_ctls_low,
1270 			msrs->procbased_ctls_high);
1271 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1272 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1273 		break;
1274 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1275 	case MSR_IA32_VMX_EXIT_CTLS:
1276 		*pdata = vmx_control_msr(
1277 			msrs->exit_ctls_low,
1278 			msrs->exit_ctls_high);
1279 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1280 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1281 		break;
1282 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1283 	case MSR_IA32_VMX_ENTRY_CTLS:
1284 		*pdata = vmx_control_msr(
1285 			msrs->entry_ctls_low,
1286 			msrs->entry_ctls_high);
1287 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1288 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1289 		break;
1290 	case MSR_IA32_VMX_MISC:
1291 		*pdata = vmx_control_msr(
1292 			msrs->misc_low,
1293 			msrs->misc_high);
1294 		break;
1295 	case MSR_IA32_VMX_CR0_FIXED0:
1296 		*pdata = msrs->cr0_fixed0;
1297 		break;
1298 	case MSR_IA32_VMX_CR0_FIXED1:
1299 		*pdata = msrs->cr0_fixed1;
1300 		break;
1301 	case MSR_IA32_VMX_CR4_FIXED0:
1302 		*pdata = msrs->cr4_fixed0;
1303 		break;
1304 	case MSR_IA32_VMX_CR4_FIXED1:
1305 		*pdata = msrs->cr4_fixed1;
1306 		break;
1307 	case MSR_IA32_VMX_VMCS_ENUM:
1308 		*pdata = msrs->vmcs_enum;
1309 		break;
1310 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1311 		*pdata = vmx_control_msr(
1312 			msrs->secondary_ctls_low,
1313 			msrs->secondary_ctls_high);
1314 		break;
1315 	case MSR_IA32_VMX_EPT_VPID_CAP:
1316 		*pdata = msrs->ept_caps |
1317 			((u64)msrs->vpid_caps << 32);
1318 		break;
1319 	case MSR_IA32_VMX_VMFUNC:
1320 		*pdata = msrs->vmfunc_controls;
1321 		break;
1322 	default:
1323 		return 1;
1324 	}
1325 
1326 	return 0;
1327 }
1328 
1329 /*
1330  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1331  * been modified by the L1 guest.  Note, "writable" in this context means
1332  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1333  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1334  * VM-exit information fields (which are actually writable if the vCPU is
1335  * configured to support "VMWRITE to any supported field in the VMCS").
1336  */
1337 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1338 {
1339 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1340 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1341 	struct shadow_vmcs_field field;
1342 	unsigned long val;
1343 	int i;
1344 
1345 	if (WARN_ON(!shadow_vmcs))
1346 		return;
1347 
1348 	preempt_disable();
1349 
1350 	vmcs_load(shadow_vmcs);
1351 
1352 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1353 		field = shadow_read_write_fields[i];
1354 		val = __vmcs_readl(field.encoding);
1355 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1356 	}
1357 
1358 	vmcs_clear(shadow_vmcs);
1359 	vmcs_load(vmx->loaded_vmcs->vmcs);
1360 
1361 	preempt_enable();
1362 }
1363 
1364 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1365 {
1366 	const struct shadow_vmcs_field *fields[] = {
1367 		shadow_read_write_fields,
1368 		shadow_read_only_fields
1369 	};
1370 	const int max_fields[] = {
1371 		max_shadow_read_write_fields,
1372 		max_shadow_read_only_fields
1373 	};
1374 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1375 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1376 	struct shadow_vmcs_field field;
1377 	unsigned long val;
1378 	int i, q;
1379 
1380 	if (WARN_ON(!shadow_vmcs))
1381 		return;
1382 
1383 	vmcs_load(shadow_vmcs);
1384 
1385 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1386 		for (i = 0; i < max_fields[q]; i++) {
1387 			field = fields[q][i];
1388 			val = vmcs12_read_any(vmcs12, field.encoding,
1389 					      field.offset);
1390 			__vmcs_writel(field.encoding, val);
1391 		}
1392 	}
1393 
1394 	vmcs_clear(shadow_vmcs);
1395 	vmcs_load(vmx->loaded_vmcs->vmcs);
1396 }
1397 
1398 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1399 {
1400 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1401 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1402 
1403 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1404 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1405 	vmcs12->guest_rip = evmcs->guest_rip;
1406 
1407 	if (unlikely(!(evmcs->hv_clean_fields &
1408 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1409 		vmcs12->guest_rsp = evmcs->guest_rsp;
1410 		vmcs12->guest_rflags = evmcs->guest_rflags;
1411 		vmcs12->guest_interruptibility_info =
1412 			evmcs->guest_interruptibility_info;
1413 	}
1414 
1415 	if (unlikely(!(evmcs->hv_clean_fields &
1416 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1417 		vmcs12->cpu_based_vm_exec_control =
1418 			evmcs->cpu_based_vm_exec_control;
1419 	}
1420 
1421 	if (unlikely(!(evmcs->hv_clean_fields &
1422 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1423 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1424 	}
1425 
1426 	if (unlikely(!(evmcs->hv_clean_fields &
1427 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1428 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1429 	}
1430 
1431 	if (unlikely(!(evmcs->hv_clean_fields &
1432 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1433 		vmcs12->vm_entry_intr_info_field =
1434 			evmcs->vm_entry_intr_info_field;
1435 		vmcs12->vm_entry_exception_error_code =
1436 			evmcs->vm_entry_exception_error_code;
1437 		vmcs12->vm_entry_instruction_len =
1438 			evmcs->vm_entry_instruction_len;
1439 	}
1440 
1441 	if (unlikely(!(evmcs->hv_clean_fields &
1442 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1443 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1444 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1445 		vmcs12->host_cr0 = evmcs->host_cr0;
1446 		vmcs12->host_cr3 = evmcs->host_cr3;
1447 		vmcs12->host_cr4 = evmcs->host_cr4;
1448 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1449 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1450 		vmcs12->host_rip = evmcs->host_rip;
1451 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1452 		vmcs12->host_es_selector = evmcs->host_es_selector;
1453 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1454 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1455 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1456 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1457 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1458 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1459 	}
1460 
1461 	if (unlikely(!(evmcs->hv_clean_fields &
1462 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1463 		vmcs12->pin_based_vm_exec_control =
1464 			evmcs->pin_based_vm_exec_control;
1465 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1466 		vmcs12->secondary_vm_exec_control =
1467 			evmcs->secondary_vm_exec_control;
1468 	}
1469 
1470 	if (unlikely(!(evmcs->hv_clean_fields &
1471 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1472 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1473 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1474 	}
1475 
1476 	if (unlikely(!(evmcs->hv_clean_fields &
1477 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1478 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1479 	}
1480 
1481 	if (unlikely(!(evmcs->hv_clean_fields &
1482 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1483 		vmcs12->guest_es_base = evmcs->guest_es_base;
1484 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1485 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1486 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1487 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1488 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1489 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1490 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1491 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1492 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1493 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1494 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1495 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1496 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1497 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1498 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1499 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1500 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1501 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1502 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1503 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1504 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1505 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1506 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1507 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1508 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1509 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1510 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1511 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1512 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1513 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1514 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1515 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1516 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1517 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1518 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1519 	}
1520 
1521 	if (unlikely(!(evmcs->hv_clean_fields &
1522 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1523 		vmcs12->tsc_offset = evmcs->tsc_offset;
1524 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1525 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1526 	}
1527 
1528 	if (unlikely(!(evmcs->hv_clean_fields &
1529 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1530 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1531 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1532 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1533 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1534 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1535 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1536 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1537 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1538 	}
1539 
1540 	if (unlikely(!(evmcs->hv_clean_fields &
1541 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1542 		vmcs12->host_fs_base = evmcs->host_fs_base;
1543 		vmcs12->host_gs_base = evmcs->host_gs_base;
1544 		vmcs12->host_tr_base = evmcs->host_tr_base;
1545 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1546 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1547 		vmcs12->host_rsp = evmcs->host_rsp;
1548 	}
1549 
1550 	if (unlikely(!(evmcs->hv_clean_fields &
1551 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1552 		vmcs12->ept_pointer = evmcs->ept_pointer;
1553 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1554 	}
1555 
1556 	if (unlikely(!(evmcs->hv_clean_fields &
1557 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1558 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1559 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1560 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1561 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1562 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1563 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1564 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1565 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1566 		vmcs12->guest_pending_dbg_exceptions =
1567 			evmcs->guest_pending_dbg_exceptions;
1568 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1569 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1570 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1571 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1572 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1573 	}
1574 
1575 	/*
1576 	 * Not used?
1577 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1578 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1579 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1580 	 * vmcs12->cr3_target_value0 = evmcs->cr3_target_value0;
1581 	 * vmcs12->cr3_target_value1 = evmcs->cr3_target_value1;
1582 	 * vmcs12->cr3_target_value2 = evmcs->cr3_target_value2;
1583 	 * vmcs12->cr3_target_value3 = evmcs->cr3_target_value3;
1584 	 * vmcs12->page_fault_error_code_mask =
1585 	 *		evmcs->page_fault_error_code_mask;
1586 	 * vmcs12->page_fault_error_code_match =
1587 	 *		evmcs->page_fault_error_code_match;
1588 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1589 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1590 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1591 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1592 	 */
1593 
1594 	/*
1595 	 * Read only fields:
1596 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1597 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1598 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1599 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1600 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1601 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1602 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1603 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1604 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1605 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1606 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1607 	 *
1608 	 * Not present in struct vmcs12:
1609 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1610 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1611 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1612 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1613 	 */
1614 
1615 	return 0;
1616 }
1617 
1618 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1619 {
1620 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1621 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1622 
1623 	/*
1624 	 * Should not be changed by KVM:
1625 	 *
1626 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1627 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1628 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1629 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1630 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1631 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1632 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1633 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1634 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1635 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1636 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1637 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1638 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1639 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1640 	 * evmcs->host_rip = vmcs12->host_rip;
1641 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1642 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1643 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1644 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1645 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1646 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1647 	 * evmcs->host_rsp = vmcs12->host_rsp;
1648 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1649 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1650 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1651 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1652 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1653 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1654 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1655 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1656 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1657 	 * evmcs->cr3_target_value0 = vmcs12->cr3_target_value0;
1658 	 * evmcs->cr3_target_value1 = vmcs12->cr3_target_value1;
1659 	 * evmcs->cr3_target_value2 = vmcs12->cr3_target_value2;
1660 	 * evmcs->cr3_target_value3 = vmcs12->cr3_target_value3;
1661 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1662 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1663 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1664 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1665 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1666 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1667 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1668 	 * evmcs->page_fault_error_code_mask =
1669 	 *		vmcs12->page_fault_error_code_mask;
1670 	 * evmcs->page_fault_error_code_match =
1671 	 *		vmcs12->page_fault_error_code_match;
1672 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1673 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1674 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1675 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1676 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1677 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1678 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1679 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1680 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1681 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1682 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1683 	 *
1684 	 * Not present in struct vmcs12:
1685 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1686 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1687 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1688 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1689 	 */
1690 
1691 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1692 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1693 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1694 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1695 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1696 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1697 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1698 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1699 
1700 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1701 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1702 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1703 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1704 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1705 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1706 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1707 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1708 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1709 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1710 
1711 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1712 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1713 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1714 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1715 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1716 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1717 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1718 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1719 
1720 	evmcs->guest_es_base = vmcs12->guest_es_base;
1721 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1722 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1723 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1724 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1725 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1726 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1727 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1728 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1729 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1730 
1731 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1732 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1733 
1734 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1735 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1736 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1737 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1738 
1739 	evmcs->guest_pending_dbg_exceptions =
1740 		vmcs12->guest_pending_dbg_exceptions;
1741 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1742 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1743 
1744 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1745 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1746 
1747 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1748 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1749 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1750 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1751 
1752 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1753 
1754 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1755 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1756 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1757 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1758 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1759 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1760 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1761 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1762 
1763 	evmcs->exit_qualification = vmcs12->exit_qualification;
1764 
1765 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1766 	evmcs->guest_rsp = vmcs12->guest_rsp;
1767 	evmcs->guest_rflags = vmcs12->guest_rflags;
1768 
1769 	evmcs->guest_interruptibility_info =
1770 		vmcs12->guest_interruptibility_info;
1771 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1772 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1773 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1774 	evmcs->vm_entry_exception_error_code =
1775 		vmcs12->vm_entry_exception_error_code;
1776 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1777 
1778 	evmcs->guest_rip = vmcs12->guest_rip;
1779 
1780 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1781 
1782 	return 0;
1783 }
1784 
1785 /*
1786  * This is an equivalent of the nested hypervisor executing the vmptrld
1787  * instruction.
1788  */
1789 static int nested_vmx_handle_enlightened_vmptrld(struct kvm_vcpu *vcpu,
1790 						 bool from_launch)
1791 {
1792 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1793 	bool evmcs_gpa_changed = false;
1794 	u64 evmcs_gpa;
1795 
1796 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1797 		return 1;
1798 
1799 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa))
1800 		return 1;
1801 
1802 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1803 		if (!vmx->nested.hv_evmcs)
1804 			vmx->nested.current_vmptr = -1ull;
1805 
1806 		nested_release_evmcs(vcpu);
1807 
1808 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1809 				 &vmx->nested.hv_evmcs_map))
1810 			return 0;
1811 
1812 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1813 
1814 		/*
1815 		 * Currently, KVM only supports eVMCS version 1
1816 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1817 		 * value to first u32 field of eVMCS which should specify eVMCS
1818 		 * VersionNumber.
1819 		 *
1820 		 * Guest should be aware of supported eVMCS versions by host by
1821 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1822 		 * expected to set this CPUID leaf according to the value
1823 		 * returned in vmcs_version from nested_enable_evmcs().
1824 		 *
1825 		 * However, it turns out that Microsoft Hyper-V fails to comply
1826 		 * to their own invented interface: When Hyper-V use eVMCS, it
1827 		 * just sets first u32 field of eVMCS to revision_id specified
1828 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1829 		 * which is one of the supported versions specified in
1830 		 * CPUID.0x4000000A.EAX[0:15].
1831 		 *
1832 		 * To overcome Hyper-V bug, we accept here either a supported
1833 		 * eVMCS version or VMCS12 revision_id as valid values for first
1834 		 * u32 field of eVMCS.
1835 		 */
1836 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1837 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1838 			nested_release_evmcs(vcpu);
1839 			return 0;
1840 		}
1841 
1842 		vmx->nested.dirty_vmcs12 = true;
1843 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
1844 
1845 		evmcs_gpa_changed = true;
1846 		/*
1847 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
1848 		 * reloaded from guest's memory (read only fields, fields not
1849 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
1850 		 * are no leftovers.
1851 		 */
1852 		if (from_launch) {
1853 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1854 			memset(vmcs12, 0, sizeof(*vmcs12));
1855 			vmcs12->hdr.revision_id = VMCS12_REVISION;
1856 		}
1857 
1858 	}
1859 
1860 	/*
1861 	 * Clean fields data can't de used on VMLAUNCH and when we switch
1862 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
1863 	 */
1864 	if (from_launch || evmcs_gpa_changed)
1865 		vmx->nested.hv_evmcs->hv_clean_fields &=
1866 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1867 
1868 	return 1;
1869 }
1870 
1871 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
1872 {
1873 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1874 
1875 	/*
1876 	 * hv_evmcs may end up being not mapped after migration (when
1877 	 * L2 was running), map it here to make sure vmcs12 changes are
1878 	 * properly reflected.
1879 	 */
1880 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs)
1881 		nested_vmx_handle_enlightened_vmptrld(vcpu, false);
1882 
1883 	if (vmx->nested.hv_evmcs) {
1884 		copy_vmcs12_to_enlightened(vmx);
1885 		/* All fields are clean */
1886 		vmx->nested.hv_evmcs->hv_clean_fields |=
1887 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1888 	} else {
1889 		copy_vmcs12_to_shadow(vmx);
1890 	}
1891 
1892 	vmx->nested.need_vmcs12_to_shadow_sync = false;
1893 }
1894 
1895 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
1896 {
1897 	struct vcpu_vmx *vmx =
1898 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
1899 
1900 	vmx->nested.preemption_timer_expired = true;
1901 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
1902 	kvm_vcpu_kick(&vmx->vcpu);
1903 
1904 	return HRTIMER_NORESTART;
1905 }
1906 
1907 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
1908 {
1909 	u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
1910 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1911 
1912 	/*
1913 	 * A timer value of zero is architecturally guaranteed to cause
1914 	 * a VMExit prior to executing any instructions in the guest.
1915 	 */
1916 	if (preemption_timeout == 0) {
1917 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
1918 		return;
1919 	}
1920 
1921 	if (vcpu->arch.virtual_tsc_khz == 0)
1922 		return;
1923 
1924 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
1925 	preemption_timeout *= 1000000;
1926 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
1927 	hrtimer_start(&vmx->nested.preemption_timer,
1928 		      ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
1929 }
1930 
1931 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1932 {
1933 	if (vmx->nested.nested_run_pending &&
1934 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
1935 		return vmcs12->guest_ia32_efer;
1936 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
1937 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
1938 	else
1939 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
1940 }
1941 
1942 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
1943 {
1944 	/*
1945 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
1946 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
1947 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
1948 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
1949 	 */
1950 	if (vmx->nested.vmcs02_initialized)
1951 		return;
1952 	vmx->nested.vmcs02_initialized = true;
1953 
1954 	/*
1955 	 * We don't care what the EPTP value is we just need to guarantee
1956 	 * it's valid so we don't get a false positive when doing early
1957 	 * consistency checks.
1958 	 */
1959 	if (enable_ept && nested_early_check)
1960 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
1961 
1962 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
1963 	if (cpu_has_vmx_vmfunc())
1964 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
1965 
1966 	if (cpu_has_vmx_posted_intr())
1967 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
1968 
1969 	if (cpu_has_vmx_msr_bitmap())
1970 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
1971 
1972 	/*
1973 	 * The PML address never changes, so it is constant in vmcs02.
1974 	 * Conceptually we want to copy the PML index from vmcs01 here,
1975 	 * and then back to vmcs01 on nested vmexit.  But since we flush
1976 	 * the log and reset GUEST_PML_INDEX on each vmexit, the PML
1977 	 * index is also effectively constant in vmcs02.
1978 	 */
1979 	if (enable_pml) {
1980 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
1981 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
1982 	}
1983 
1984 	if (cpu_has_vmx_encls_vmexit())
1985 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
1986 
1987 	/*
1988 	 * Set the MSR load/store lists to match L0's settings.  Only the
1989 	 * addresses are constant (for vmcs02), the counts can change based
1990 	 * on L2's behavior, e.g. switching to/from long mode.
1991 	 */
1992 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1993 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
1994 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
1995 
1996 	vmx_set_constant_host_state(vmx);
1997 }
1998 
1999 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2000 				      struct vmcs12 *vmcs12)
2001 {
2002 	prepare_vmcs02_constant_state(vmx);
2003 
2004 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2005 
2006 	if (enable_vpid) {
2007 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2008 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2009 		else
2010 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2011 	}
2012 }
2013 
2014 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2015 {
2016 	u32 exec_control, vmcs12_exec_ctrl;
2017 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2018 
2019 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
2020 		prepare_vmcs02_early_rare(vmx, vmcs12);
2021 
2022 	/*
2023 	 * PIN CONTROLS
2024 	 */
2025 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2026 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2027 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2028 
2029 	/* Posted interrupts setting is only taken from vmcs12.  */
2030 	if (nested_cpu_has_posted_intr(vmcs12)) {
2031 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2032 		vmx->nested.pi_pending = false;
2033 	} else {
2034 		exec_control &= ~PIN_BASED_POSTED_INTR;
2035 	}
2036 	pin_controls_set(vmx, exec_control);
2037 
2038 	/*
2039 	 * EXEC CONTROLS
2040 	 */
2041 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2042 	exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2043 	exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2044 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2045 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2046 
2047 	if (exec_control & CPU_BASED_TPR_SHADOW)
2048 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2049 #ifdef CONFIG_X86_64
2050 	else
2051 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2052 				CPU_BASED_CR8_STORE_EXITING;
2053 #endif
2054 
2055 	/*
2056 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2057 	 * for I/O port accesses.
2058 	 */
2059 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2060 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2061 
2062 	/*
2063 	 * This bit will be computed in nested_get_vmcs12_pages, because
2064 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2065 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2066 	 * only set/clear this bit.
2067 	 */
2068 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2069 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2070 
2071 	exec_controls_set(vmx, exec_control);
2072 
2073 	/*
2074 	 * SECONDARY EXEC CONTROLS
2075 	 */
2076 	if (cpu_has_secondary_exec_ctrls()) {
2077 		exec_control = vmx->secondary_exec_control;
2078 
2079 		/* Take the following fields only from vmcs12 */
2080 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2081 				  SECONDARY_EXEC_ENABLE_INVPCID |
2082 				  SECONDARY_EXEC_RDTSCP |
2083 				  SECONDARY_EXEC_XSAVES |
2084 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2085 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2086 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2087 		if (nested_cpu_has(vmcs12,
2088 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2089 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2090 				~SECONDARY_EXEC_ENABLE_PML;
2091 			exec_control |= vmcs12_exec_ctrl;
2092 		}
2093 
2094 		/* VMCS shadowing for L2 is emulated for now */
2095 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2096 
2097 		/*
2098 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2099 		 * will not have to rewrite the controls just for this bit.
2100 		 */
2101 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2102 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2103 			exec_control |= SECONDARY_EXEC_DESC;
2104 
2105 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2106 			vmcs_write16(GUEST_INTR_STATUS,
2107 				vmcs12->guest_intr_status);
2108 
2109 		secondary_exec_controls_set(vmx, exec_control);
2110 	}
2111 
2112 	/*
2113 	 * ENTRY CONTROLS
2114 	 *
2115 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2116 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2117 	 * on the related bits (if supported by the CPU) in the hope that
2118 	 * we can avoid VMWrites during vmx_set_efer().
2119 	 */
2120 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2121 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2122 	if (cpu_has_load_ia32_efer()) {
2123 		if (guest_efer & EFER_LMA)
2124 			exec_control |= VM_ENTRY_IA32E_MODE;
2125 		if (guest_efer != host_efer)
2126 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2127 	}
2128 	vm_entry_controls_set(vmx, exec_control);
2129 
2130 	/*
2131 	 * EXIT CONTROLS
2132 	 *
2133 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2134 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2135 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2136 	 */
2137 	exec_control = vmx_vmexit_ctrl();
2138 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2139 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2140 	vm_exit_controls_set(vmx, exec_control);
2141 
2142 	/*
2143 	 * Interrupt/Exception Fields
2144 	 */
2145 	if (vmx->nested.nested_run_pending) {
2146 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2147 			     vmcs12->vm_entry_intr_info_field);
2148 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2149 			     vmcs12->vm_entry_exception_error_code);
2150 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2151 			     vmcs12->vm_entry_instruction_len);
2152 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2153 			     vmcs12->guest_interruptibility_info);
2154 		vmx->loaded_vmcs->nmi_known_unmasked =
2155 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2156 	} else {
2157 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2158 	}
2159 }
2160 
2161 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2162 {
2163 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2164 
2165 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2166 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2167 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2168 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2169 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2170 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2171 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2172 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2173 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2174 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2175 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2176 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2177 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2178 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2179 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2180 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2181 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2182 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2183 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2184 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2185 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2186 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2187 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2188 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2189 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2190 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2191 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2192 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2193 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2194 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2195 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2196 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2197 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2198 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2199 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2200 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2201 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2202 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2203 	}
2204 
2205 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2206 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2207 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2208 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2209 			    vmcs12->guest_pending_dbg_exceptions);
2210 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2211 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2212 
2213 		/*
2214 		 * L1 may access the L2's PDPTR, so save them to construct
2215 		 * vmcs12
2216 		 */
2217 		if (enable_ept) {
2218 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2219 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2220 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2221 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2222 		}
2223 
2224 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2225 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2226 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2227 	}
2228 
2229 	if (nested_cpu_has_xsaves(vmcs12))
2230 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2231 
2232 	/*
2233 	 * Whether page-faults are trapped is determined by a combination of
2234 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2235 	 * If enable_ept, L0 doesn't care about page faults and we should
2236 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2237 	 * care about (at least some) page faults, and because it is not easy
2238 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2239 	 * to exit on each and every L2 page fault. This is done by setting
2240 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2241 	 * Note that below we don't need special code to set EB.PF beyond the
2242 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2243 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2244 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2245 	 */
2246 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2247 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2248 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2249 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2250 
2251 	if (cpu_has_vmx_apicv()) {
2252 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2253 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2254 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2255 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2256 	}
2257 
2258 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2259 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2260 
2261 	set_cr4_guest_host_mask(vmx);
2262 }
2263 
2264 /*
2265  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2266  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2267  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2268  * guest in a way that will both be appropriate to L1's requests, and our
2269  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2270  * function also has additional necessary side-effects, like setting various
2271  * vcpu->arch fields.
2272  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2273  * is assigned to entry_failure_code on failure.
2274  */
2275 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2276 			  u32 *entry_failure_code)
2277 {
2278 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2279 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2280 	bool load_guest_pdptrs_vmcs12 = false;
2281 
2282 	if (vmx->nested.dirty_vmcs12 || hv_evmcs) {
2283 		prepare_vmcs02_rare(vmx, vmcs12);
2284 		vmx->nested.dirty_vmcs12 = false;
2285 
2286 		load_guest_pdptrs_vmcs12 = !hv_evmcs ||
2287 			!(hv_evmcs->hv_clean_fields &
2288 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2289 	}
2290 
2291 	if (vmx->nested.nested_run_pending &&
2292 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2293 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2294 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2295 	} else {
2296 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2297 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2298 	}
2299 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2300 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2301 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2302 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2303 
2304 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2305 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2306 	 * trap. Note that CR0.TS also needs updating - we do this later.
2307 	 */
2308 	update_exception_bitmap(vcpu);
2309 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2310 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2311 
2312 	if (vmx->nested.nested_run_pending &&
2313 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2314 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2315 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2316 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2317 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2318 	}
2319 
2320 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2321 
2322 	if (kvm_has_tsc_control)
2323 		decache_tsc_multiplier(vmx);
2324 
2325 	if (enable_vpid) {
2326 		/*
2327 		 * There is no direct mapping between vpid02 and vpid12, the
2328 		 * vpid02 is per-vCPU for L0 and reused while the value of
2329 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2330 		 * The vpid12 is allocated by L1 for L2, so it will not
2331 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2332 		 * even if spawn a lot of nested vCPUs.
2333 		 */
2334 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2335 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2336 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2337 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2338 			}
2339 		} else {
2340 			/*
2341 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2342 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2343 			 * flush until vmcs02->eptp is fully updated by
2344 			 * KVM_REQ_LOAD_CR3. Note that this assumes
2345 			 * KVM_REQ_TLB_FLUSH is evaluated after
2346 			 * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
2347 			 */
2348 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2349 		}
2350 	}
2351 
2352 	if (nested_cpu_has_ept(vmcs12))
2353 		nested_ept_init_mmu_context(vcpu);
2354 	else if (nested_cpu_has2(vmcs12,
2355 				 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2356 		vmx_flush_tlb(vcpu, true);
2357 
2358 	/*
2359 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2360 	 * bits which we consider mandatory enabled.
2361 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2362 	 * the specifications by L1; It's not enough to take
2363 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2364 	 * have more bits than L1 expected.
2365 	 */
2366 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2367 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2368 
2369 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2370 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2371 
2372 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2373 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2374 	vmx_set_efer(vcpu, vcpu->arch.efer);
2375 
2376 	/*
2377 	 * Guest state is invalid and unrestricted guest is disabled,
2378 	 * which means L1 attempted VMEntry to L2 with invalid state.
2379 	 * Fail the VMEntry.
2380 	 */
2381 	if (vmx->emulation_required) {
2382 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2383 		return -EINVAL;
2384 	}
2385 
2386 	/* Shadow page tables on either EPT or shadow page tables. */
2387 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2388 				entry_failure_code))
2389 		return -EINVAL;
2390 
2391 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2392 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2393 	    is_pae_paging(vcpu)) {
2394 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2395 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2396 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2397 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2398 	}
2399 
2400 	if (!enable_ept)
2401 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2402 
2403 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2404 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2405 	return 0;
2406 }
2407 
2408 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2409 {
2410 	if (!nested_cpu_has_nmi_exiting(vmcs12) &&
2411 	    nested_cpu_has_virtual_nmis(vmcs12))
2412 		return -EINVAL;
2413 
2414 	if (!nested_cpu_has_virtual_nmis(vmcs12) &&
2415 	    nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING))
2416 		return -EINVAL;
2417 
2418 	return 0;
2419 }
2420 
2421 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
2422 {
2423 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2424 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2425 
2426 	/* Check for memory type validity */
2427 	switch (address & VMX_EPTP_MT_MASK) {
2428 	case VMX_EPTP_MT_UC:
2429 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT))
2430 			return false;
2431 		break;
2432 	case VMX_EPTP_MT_WB:
2433 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT))
2434 			return false;
2435 		break;
2436 	default:
2437 		return false;
2438 	}
2439 
2440 	/* only 4 levels page-walk length are valid */
2441 	if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
2442 		return false;
2443 
2444 	/* Reserved bits should not be set */
2445 	if (address >> maxphyaddr || ((address >> 7) & 0x1f))
2446 		return false;
2447 
2448 	/* AD, if set, should be supported */
2449 	if (address & VMX_EPTP_AD_ENABLE_BIT) {
2450 		if (!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT))
2451 			return false;
2452 	}
2453 
2454 	return true;
2455 }
2456 
2457 /*
2458  * Checks related to VM-Execution Control Fields
2459  */
2460 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2461                                               struct vmcs12 *vmcs12)
2462 {
2463 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2464 
2465 	if (!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2466 				vmx->nested.msrs.pinbased_ctls_low,
2467 				vmx->nested.msrs.pinbased_ctls_high) ||
2468 	    !vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2469 				vmx->nested.msrs.procbased_ctls_low,
2470 				vmx->nested.msrs.procbased_ctls_high))
2471 		return -EINVAL;
2472 
2473 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2474 	    !vmx_control_verify(vmcs12->secondary_vm_exec_control,
2475 				 vmx->nested.msrs.secondary_ctls_low,
2476 				 vmx->nested.msrs.secondary_ctls_high))
2477 		return -EINVAL;
2478 
2479 	if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu) ||
2480 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2481 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2482 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2483 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2484 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2485 	    nested_vmx_check_nmi_controls(vmcs12) ||
2486 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2487 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2488 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2489 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2490 	    (nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2491 		return -EINVAL;
2492 
2493 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2494 	    nested_cpu_has_save_preemption_timer(vmcs12))
2495 		return -EINVAL;
2496 
2497 	if (nested_cpu_has_ept(vmcs12) &&
2498 	    !valid_ept_address(vcpu, vmcs12->ept_pointer))
2499 		return -EINVAL;
2500 
2501 	if (nested_cpu_has_vmfunc(vmcs12)) {
2502 		if (vmcs12->vm_function_control &
2503 		    ~vmx->nested.msrs.vmfunc_controls)
2504 			return -EINVAL;
2505 
2506 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2507 			if (!nested_cpu_has_ept(vmcs12) ||
2508 			    !page_address_valid(vcpu, vmcs12->eptp_list_address))
2509 				return -EINVAL;
2510 		}
2511 	}
2512 
2513 	return 0;
2514 }
2515 
2516 /*
2517  * Checks related to VM-Exit Control Fields
2518  */
2519 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2520                                          struct vmcs12 *vmcs12)
2521 {
2522 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2523 
2524 	if (!vmx_control_verify(vmcs12->vm_exit_controls,
2525 				vmx->nested.msrs.exit_ctls_low,
2526 				vmx->nested.msrs.exit_ctls_high) ||
2527 	    nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12))
2528 		return -EINVAL;
2529 
2530 	return 0;
2531 }
2532 
2533 /*
2534  * Checks related to VM-Entry Control Fields
2535  */
2536 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2537 					  struct vmcs12 *vmcs12)
2538 {
2539 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2540 
2541 	if (!vmx_control_verify(vmcs12->vm_entry_controls,
2542 				vmx->nested.msrs.entry_ctls_low,
2543 				vmx->nested.msrs.entry_ctls_high))
2544 		return -EINVAL;
2545 
2546 	/*
2547 	 * From the Intel SDM, volume 3:
2548 	 * Fields relevant to VM-entry event injection must be set properly.
2549 	 * These fields are the VM-entry interruption-information field, the
2550 	 * VM-entry exception error code, and the VM-entry instruction length.
2551 	 */
2552 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2553 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2554 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2555 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2556 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2557 		bool should_have_error_code;
2558 		bool urg = nested_cpu_has2(vmcs12,
2559 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2560 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2561 
2562 		/* VM-entry interruption-info field: interruption type */
2563 		if (intr_type == INTR_TYPE_RESERVED ||
2564 		    (intr_type == INTR_TYPE_OTHER_EVENT &&
2565 		     !nested_cpu_supports_monitor_trap_flag(vcpu)))
2566 			return -EINVAL;
2567 
2568 		/* VM-entry interruption-info field: vector */
2569 		if ((intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2570 		    (intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2571 		    (intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2572 			return -EINVAL;
2573 
2574 		/* VM-entry interruption-info field: deliver error code */
2575 		should_have_error_code =
2576 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2577 			x86_exception_has_error_code(vector);
2578 		if (has_error_code != should_have_error_code)
2579 			return -EINVAL;
2580 
2581 		/* VM-entry exception error code */
2582 		if (has_error_code &&
2583 		    vmcs12->vm_entry_exception_error_code & GENMASK(31, 15))
2584 			return -EINVAL;
2585 
2586 		/* VM-entry interruption-info field: reserved bits */
2587 		if (intr_info & INTR_INFO_RESVD_BITS_MASK)
2588 			return -EINVAL;
2589 
2590 		/* VM-entry instruction length */
2591 		switch (intr_type) {
2592 		case INTR_TYPE_SOFT_EXCEPTION:
2593 		case INTR_TYPE_SOFT_INTR:
2594 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2595 			if ((vmcs12->vm_entry_instruction_len > 15) ||
2596 			    (vmcs12->vm_entry_instruction_len == 0 &&
2597 			     !nested_cpu_has_zero_length_injection(vcpu)))
2598 				return -EINVAL;
2599 		}
2600 	}
2601 
2602 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2603 		return -EINVAL;
2604 
2605 	return 0;
2606 }
2607 
2608 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2609 				     struct vmcs12 *vmcs12)
2610 {
2611 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2612 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2613 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2614 		return -EINVAL;
2615 
2616 	return 0;
2617 }
2618 
2619 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2620 				       struct vmcs12 *vmcs12)
2621 {
2622 	bool ia32e;
2623 
2624 	if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
2625 	    !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
2626 	    !nested_cr3_valid(vcpu, vmcs12->host_cr3))
2627 		return -EINVAL;
2628 
2629 	if (is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu) ||
2630 	    is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu))
2631 		return -EINVAL;
2632 
2633 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2634 	    !kvm_pat_valid(vmcs12->host_ia32_pat))
2635 		return -EINVAL;
2636 
2637 	ia32e = (vmcs12->vm_exit_controls &
2638 		 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
2639 
2640 	if (vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK) ||
2641 	    vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK) ||
2642 	    vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK) ||
2643 	    vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK) ||
2644 	    vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK) ||
2645 	    vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK) ||
2646 	    vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK) ||
2647 	    vmcs12->host_cs_selector == 0 ||
2648 	    vmcs12->host_tr_selector == 0 ||
2649 	    (vmcs12->host_ss_selector == 0 && !ia32e))
2650 		return -EINVAL;
2651 
2652 #ifdef CONFIG_X86_64
2653 	if (is_noncanonical_address(vmcs12->host_fs_base, vcpu) ||
2654 	    is_noncanonical_address(vmcs12->host_gs_base, vcpu) ||
2655 	    is_noncanonical_address(vmcs12->host_gdtr_base, vcpu) ||
2656 	    is_noncanonical_address(vmcs12->host_idtr_base, vcpu) ||
2657 	    is_noncanonical_address(vmcs12->host_tr_base, vcpu))
2658 		return -EINVAL;
2659 #endif
2660 
2661 	/*
2662 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2663 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2664 	 * the values of the LMA and LME bits in the field must each be that of
2665 	 * the host address-space size VM-exit control.
2666 	 */
2667 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2668 		if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
2669 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
2670 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
2671 			return -EINVAL;
2672 	}
2673 
2674 	return 0;
2675 }
2676 
2677 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2678 					  struct vmcs12 *vmcs12)
2679 {
2680 	int r = 0;
2681 	struct vmcs12 *shadow;
2682 	struct kvm_host_map map;
2683 
2684 	if (vmcs12->vmcs_link_pointer == -1ull)
2685 		return 0;
2686 
2687 	if (!page_address_valid(vcpu, vmcs12->vmcs_link_pointer))
2688 		return -EINVAL;
2689 
2690 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
2691 		return -EINVAL;
2692 
2693 	shadow = map.hva;
2694 
2695 	if (shadow->hdr.revision_id != VMCS12_REVISION ||
2696 	    shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12))
2697 		r = -EINVAL;
2698 
2699 	kvm_vcpu_unmap(vcpu, &map, false);
2700 	return r;
2701 }
2702 
2703 /*
2704  * Checks related to Guest Non-register State
2705  */
2706 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2707 {
2708 	if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2709 	    vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
2710 		return -EINVAL;
2711 
2712 	return 0;
2713 }
2714 
2715 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2716 					struct vmcs12 *vmcs12,
2717 					u32 *exit_qual)
2718 {
2719 	bool ia32e;
2720 
2721 	*exit_qual = ENTRY_FAIL_DEFAULT;
2722 
2723 	if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
2724 	    !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
2725 		return -EINVAL;
2726 
2727 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2728 	    !kvm_pat_valid(vmcs12->guest_ia32_pat))
2729 		return -EINVAL;
2730 
2731 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2732 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2733 		return -EINVAL;
2734 	}
2735 
2736 	/*
2737 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2738 	 * are performed on the field for the IA32_EFER MSR:
2739 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2740 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2741 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2742 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2743 	 *   CR0.PG) is 1.
2744 	 */
2745 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2746 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2747 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2748 		if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
2749 		    ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
2750 		    ((vmcs12->guest_cr0 & X86_CR0_PG) &&
2751 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
2752 			return -EINVAL;
2753 	}
2754 
2755 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2756 	    (is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
2757 	     (vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
2758 		return -EINVAL;
2759 
2760 	if (nested_check_guest_non_reg_state(vmcs12))
2761 		return -EINVAL;
2762 
2763 	return 0;
2764 }
2765 
2766 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2767 {
2768 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2769 	unsigned long cr3, cr4;
2770 	bool vm_fail;
2771 
2772 	if (!nested_early_check)
2773 		return 0;
2774 
2775 	if (vmx->msr_autoload.host.nr)
2776 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2777 	if (vmx->msr_autoload.guest.nr)
2778 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2779 
2780 	preempt_disable();
2781 
2782 	vmx_prepare_switch_to_guest(vcpu);
2783 
2784 	/*
2785 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2786 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2787 	 * be written (by preparve_vmcs02()) before the "real" VMEnter, i.e.
2788 	 * there is no need to preserve other bits or save/restore the field.
2789 	 */
2790 	vmcs_writel(GUEST_RFLAGS, 0);
2791 
2792 	cr3 = __get_current_cr3_fast();
2793 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2794 		vmcs_writel(HOST_CR3, cr3);
2795 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2796 	}
2797 
2798 	cr4 = cr4_read_shadow();
2799 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2800 		vmcs_writel(HOST_CR4, cr4);
2801 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2802 	}
2803 
2804 	asm(
2805 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2806 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2807 		"je 1f \n\t"
2808 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
2809 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2810 		"1: \n\t"
2811 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2812 
2813 		/* Check if vmlaunch or vmresume is needed */
2814 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
2815 
2816 		/*
2817 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
2818 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
2819 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
2820 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
2821 		 */
2822 		"call vmx_vmenter\n\t"
2823 
2824 		CC_SET(be)
2825 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
2826 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
2827 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
2828 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
2829 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
2830 		[wordsize]"i"(sizeof(ulong))
2831 	      : "memory"
2832 	);
2833 
2834 	if (vmx->msr_autoload.host.nr)
2835 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2836 	if (vmx->msr_autoload.guest.nr)
2837 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2838 
2839 	if (vm_fail) {
2840 		preempt_enable();
2841 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
2842 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
2843 		return 1;
2844 	}
2845 
2846 	/*
2847 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
2848 	 */
2849 	local_irq_enable();
2850 	if (hw_breakpoint_active())
2851 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
2852 	preempt_enable();
2853 
2854 	/*
2855 	 * A non-failing VMEntry means we somehow entered guest mode with
2856 	 * an illegal RIP, and that's just the tip of the iceberg.  There
2857 	 * is no telling what memory has been modified or what state has
2858 	 * been exposed to unknown code.  Hitting this all but guarantees
2859 	 * a (very critical) hardware issue.
2860 	 */
2861 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
2862 		VMX_EXIT_REASONS_FAILED_VMENTRY));
2863 
2864 	return 0;
2865 }
2866 
2867 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
2868 						 struct vmcs12 *vmcs12);
2869 
2870 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
2871 {
2872 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2873 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2874 	struct kvm_host_map *map;
2875 	struct page *page;
2876 	u64 hpa;
2877 
2878 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2879 		/*
2880 		 * Translate L1 physical address to host physical
2881 		 * address for vmcs02. Keep the page pinned, so this
2882 		 * physical address remains valid. We keep a reference
2883 		 * to it so we can release it later.
2884 		 */
2885 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
2886 			kvm_release_page_dirty(vmx->nested.apic_access_page);
2887 			vmx->nested.apic_access_page = NULL;
2888 		}
2889 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
2890 		/*
2891 		 * If translation failed, no matter: This feature asks
2892 		 * to exit when accessing the given address, and if it
2893 		 * can never be accessed, this feature won't do
2894 		 * anything anyway.
2895 		 */
2896 		if (!is_error_page(page)) {
2897 			vmx->nested.apic_access_page = page;
2898 			hpa = page_to_phys(vmx->nested.apic_access_page);
2899 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
2900 		} else {
2901 			secondary_exec_controls_clearbit(vmx,
2902 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
2903 		}
2904 	}
2905 
2906 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
2907 		map = &vmx->nested.virtual_apic_map;
2908 
2909 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
2910 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
2911 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
2912 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
2913 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2914 			/*
2915 			 * The processor will never use the TPR shadow, simply
2916 			 * clear the bit from the execution control.  Such a
2917 			 * configuration is useless, but it happens in tests.
2918 			 * For any other configuration, failing the vm entry is
2919 			 * _not_ what the processor does but it's basically the
2920 			 * only possibility we have.
2921 			 */
2922 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
2923 		} else {
2924 			/*
2925 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
2926 			 * force VM-Entry to fail.
2927 			 */
2928 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
2929 		}
2930 	}
2931 
2932 	if (nested_cpu_has_posted_intr(vmcs12)) {
2933 		map = &vmx->nested.pi_desc_map;
2934 
2935 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
2936 			vmx->nested.pi_desc =
2937 				(struct pi_desc *)(((void *)map->hva) +
2938 				offset_in_page(vmcs12->posted_intr_desc_addr));
2939 			vmcs_write64(POSTED_INTR_DESC_ADDR,
2940 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
2941 		}
2942 	}
2943 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
2944 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
2945 	else
2946 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
2947 }
2948 
2949 /*
2950  * Intel's VMX Instruction Reference specifies a common set of prerequisites
2951  * for running VMX instructions (except VMXON, whose prerequisites are
2952  * slightly different). It also specifies what exception to inject otherwise.
2953  * Note that many of these exceptions have priority over VM exits, so they
2954  * don't have to be checked again here.
2955  */
2956 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
2957 {
2958 	if (!to_vmx(vcpu)->nested.vmxon) {
2959 		kvm_queue_exception(vcpu, UD_VECTOR);
2960 		return 0;
2961 	}
2962 
2963 	if (vmx_get_cpl(vcpu)) {
2964 		kvm_inject_gp(vcpu, 0);
2965 		return 0;
2966 	}
2967 
2968 	return 1;
2969 }
2970 
2971 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
2972 {
2973 	u8 rvi = vmx_get_rvi();
2974 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
2975 
2976 	return ((rvi & 0xf0) > (vppr & 0xf0));
2977 }
2978 
2979 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
2980 				   struct vmcs12 *vmcs12);
2981 
2982 /*
2983  * If from_vmentry is false, this is being called from state restore (either RSM
2984  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
2985 + *
2986 + * Returns:
2987 + *   0 - success, i.e. proceed with actual VMEnter
2988 + *   1 - consistency check VMExit
2989 + *  -1 - consistency check VMFail
2990  */
2991 int nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
2992 {
2993 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2994 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2995 	bool evaluate_pending_interrupts;
2996 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
2997 	u32 exit_qual;
2998 
2999 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3000 		(CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_VIRTUAL_NMI_PENDING);
3001 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3002 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3003 
3004 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3005 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3006 	if (kvm_mpx_supported() &&
3007 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3008 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3009 
3010 	/*
3011 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3012 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3013 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3014 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3015 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3016 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3017 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3018 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3019 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3020 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3021 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3022 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3023 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3024 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3025 	 */
3026 	if (!enable_ept && !nested_early_check)
3027 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3028 
3029 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3030 
3031 	prepare_vmcs02_early(vmx, vmcs12);
3032 
3033 	if (from_vmentry) {
3034 		nested_get_vmcs12_pages(vcpu);
3035 
3036 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3037 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3038 			return -1;
3039 		}
3040 
3041 		if (nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
3042 			goto vmentry_fail_vmexit;
3043 	}
3044 
3045 	enter_guest_mode(vcpu);
3046 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3047 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3048 
3049 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
3050 		goto vmentry_fail_vmexit_guest_mode;
3051 
3052 	if (from_vmentry) {
3053 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3054 		exit_qual = nested_vmx_load_msr(vcpu,
3055 						vmcs12->vm_entry_msr_load_addr,
3056 						vmcs12->vm_entry_msr_load_count);
3057 		if (exit_qual)
3058 			goto vmentry_fail_vmexit_guest_mode;
3059 	} else {
3060 		/*
3061 		 * The MMU is not initialized to point at the right entities yet and
3062 		 * "get pages" would need to read data from the guest (i.e. we will
3063 		 * need to perform gpa to hpa translation). Request a call
3064 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3065 		 * have already been set at vmentry time and should not be reset.
3066 		 */
3067 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3068 	}
3069 
3070 	/*
3071 	 * If L1 had a pending IRQ/NMI until it executed
3072 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3073 	 * disallowed (e.g. interrupts disabled), L0 needs to
3074 	 * evaluate if this pending event should cause an exit from L2
3075 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3076 	 * intercept EXTERNAL_INTERRUPT).
3077 	 *
3078 	 * Usually this would be handled by the processor noticing an
3079 	 * IRQ/NMI window request, or checking RVI during evaluation of
3080 	 * pending virtual interrupts.  However, this setting was done
3081 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3082 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3083 	 */
3084 	if (unlikely(evaluate_pending_interrupts))
3085 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3086 
3087 	/*
3088 	 * Do not start the preemption timer hrtimer until after we know
3089 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3090 	 * the timer.
3091 	 */
3092 	vmx->nested.preemption_timer_expired = false;
3093 	if (nested_cpu_has_preemption_timer(vmcs12))
3094 		vmx_start_preemption_timer(vcpu);
3095 
3096 	/*
3097 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3098 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3099 	 * returned as far as L1 is concerned. It will only return (and set
3100 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3101 	 */
3102 	return 0;
3103 
3104 	/*
3105 	 * A failed consistency check that leads to a VMExit during L1's
3106 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3107 	 * 26.7 "VM-entry failures during or after loading guest state".
3108 	 */
3109 vmentry_fail_vmexit_guest_mode:
3110 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3111 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3112 	leave_guest_mode(vcpu);
3113 
3114 vmentry_fail_vmexit:
3115 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3116 
3117 	if (!from_vmentry)
3118 		return 1;
3119 
3120 	load_vmcs12_host_state(vcpu, vmcs12);
3121 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3122 	vmcs12->exit_qualification = exit_qual;
3123 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3124 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3125 	return 1;
3126 }
3127 
3128 /*
3129  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3130  * for running an L2 nested guest.
3131  */
3132 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3133 {
3134 	struct vmcs12 *vmcs12;
3135 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3136 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3137 	int ret;
3138 
3139 	if (!nested_vmx_check_permission(vcpu))
3140 		return 1;
3141 
3142 	if (!nested_vmx_handle_enlightened_vmptrld(vcpu, launch))
3143 		return 1;
3144 
3145 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3146 		return nested_vmx_failInvalid(vcpu);
3147 
3148 	vmcs12 = get_vmcs12(vcpu);
3149 
3150 	/*
3151 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3152 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3153 	 * rather than RFLAGS.ZF, and no error number is stored to the
3154 	 * VM-instruction error field.
3155 	 */
3156 	if (vmcs12->hdr.shadow_vmcs)
3157 		return nested_vmx_failInvalid(vcpu);
3158 
3159 	if (vmx->nested.hv_evmcs) {
3160 		copy_enlightened_to_vmcs12(vmx);
3161 		/* Enlightened VMCS doesn't have launch state */
3162 		vmcs12->launch_state = !launch;
3163 	} else if (enable_shadow_vmcs) {
3164 		copy_shadow_to_vmcs12(vmx);
3165 	}
3166 
3167 	/*
3168 	 * The nested entry process starts with enforcing various prerequisites
3169 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3170 	 * they fail: As the SDM explains, some conditions should cause the
3171 	 * instruction to fail, while others will cause the instruction to seem
3172 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3173 	 * To speed up the normal (success) code path, we should avoid checking
3174 	 * for misconfigurations which will anyway be caught by the processor
3175 	 * when using the merged vmcs02.
3176 	 */
3177 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3178 		return nested_vmx_failValid(vcpu,
3179 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3180 
3181 	if (vmcs12->launch_state == launch)
3182 		return nested_vmx_failValid(vcpu,
3183 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3184 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3185 
3186 	if (nested_vmx_check_controls(vcpu, vmcs12))
3187 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3188 
3189 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3190 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3191 
3192 	/*
3193 	 * We're finally done with prerequisite checking, and can start with
3194 	 * the nested entry.
3195 	 */
3196 	vmx->nested.nested_run_pending = 1;
3197 	ret = nested_vmx_enter_non_root_mode(vcpu, true);
3198 	vmx->nested.nested_run_pending = !ret;
3199 	if (ret > 0)
3200 		return 1;
3201 	else if (ret)
3202 		return nested_vmx_failValid(vcpu,
3203 			VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3204 
3205 	/* Hide L1D cache contents from the nested guest.  */
3206 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3207 
3208 	/*
3209 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3210 	 * also be used as part of restoring nVMX state for
3211 	 * snapshot restore (migration).
3212 	 *
3213 	 * In this flow, it is assumed that vmcs12 cache was
3214 	 * trasferred as part of captured nVMX state and should
3215 	 * therefore not be read from guest memory (which may not
3216 	 * exist on destination host yet).
3217 	 */
3218 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3219 
3220 	/*
3221 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3222 	 * awakened by event injection or by an NMI-window VM-exit or
3223 	 * by an interrupt-window VM-exit, halt the vcpu.
3224 	 */
3225 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3226 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3227 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_NMI_PENDING) &&
3228 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_INTR_PENDING) &&
3229 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3230 		vmx->nested.nested_run_pending = 0;
3231 		return kvm_vcpu_halt(vcpu);
3232 	}
3233 	return 1;
3234 }
3235 
3236 /*
3237  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3238  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
3239  * This function returns the new value we should put in vmcs12.guest_cr0.
3240  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3241  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3242  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3243  *     didn't trap the bit, because if L1 did, so would L0).
3244  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3245  *     been modified by L2, and L1 knows it. So just leave the old value of
3246  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3247  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3248  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3249  *     changed these bits, and therefore they need to be updated, but L0
3250  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3251  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3252  */
3253 static inline unsigned long
3254 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3255 {
3256 	return
3257 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3258 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3259 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3260 			vcpu->arch.cr0_guest_owned_bits));
3261 }
3262 
3263 static inline unsigned long
3264 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3265 {
3266 	return
3267 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3268 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3269 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3270 			vcpu->arch.cr4_guest_owned_bits));
3271 }
3272 
3273 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3274 				      struct vmcs12 *vmcs12)
3275 {
3276 	u32 idt_vectoring;
3277 	unsigned int nr;
3278 
3279 	if (vcpu->arch.exception.injected) {
3280 		nr = vcpu->arch.exception.nr;
3281 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3282 
3283 		if (kvm_exception_is_soft(nr)) {
3284 			vmcs12->vm_exit_instruction_len =
3285 				vcpu->arch.event_exit_inst_len;
3286 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3287 		} else
3288 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3289 
3290 		if (vcpu->arch.exception.has_error_code) {
3291 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3292 			vmcs12->idt_vectoring_error_code =
3293 				vcpu->arch.exception.error_code;
3294 		}
3295 
3296 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3297 	} else if (vcpu->arch.nmi_injected) {
3298 		vmcs12->idt_vectoring_info_field =
3299 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3300 	} else if (vcpu->arch.interrupt.injected) {
3301 		nr = vcpu->arch.interrupt.nr;
3302 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3303 
3304 		if (vcpu->arch.interrupt.soft) {
3305 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3306 			vmcs12->vm_entry_instruction_len =
3307 				vcpu->arch.event_exit_inst_len;
3308 		} else
3309 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3310 
3311 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3312 	}
3313 }
3314 
3315 
3316 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3317 {
3318 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3319 	gfn_t gfn;
3320 
3321 	/*
3322 	 * Don't need to mark the APIC access page dirty; it is never
3323 	 * written to by the CPU during APIC virtualization.
3324 	 */
3325 
3326 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3327 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3328 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3329 	}
3330 
3331 	if (nested_cpu_has_posted_intr(vmcs12)) {
3332 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3333 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3334 	}
3335 }
3336 
3337 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3338 {
3339 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3340 	int max_irr;
3341 	void *vapic_page;
3342 	u16 status;
3343 
3344 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3345 		return;
3346 
3347 	vmx->nested.pi_pending = false;
3348 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3349 		return;
3350 
3351 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3352 	if (max_irr != 256) {
3353 		vapic_page = vmx->nested.virtual_apic_map.hva;
3354 		if (!vapic_page)
3355 			return;
3356 
3357 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3358 			vapic_page, &max_irr);
3359 		status = vmcs_read16(GUEST_INTR_STATUS);
3360 		if ((u8)max_irr > ((u8)status & 0xff)) {
3361 			status &= ~0xff;
3362 			status |= (u8)max_irr;
3363 			vmcs_write16(GUEST_INTR_STATUS, status);
3364 		}
3365 	}
3366 
3367 	nested_mark_vmcs12_pages_dirty(vcpu);
3368 }
3369 
3370 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3371 					       unsigned long exit_qual)
3372 {
3373 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3374 	unsigned int nr = vcpu->arch.exception.nr;
3375 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3376 
3377 	if (vcpu->arch.exception.has_error_code) {
3378 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3379 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3380 	}
3381 
3382 	if (kvm_exception_is_soft(nr))
3383 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3384 	else
3385 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3386 
3387 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3388 	    vmx_get_nmi_mask(vcpu))
3389 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3390 
3391 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3392 }
3393 
3394 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
3395 {
3396 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3397 	unsigned long exit_qual;
3398 	bool block_nested_events =
3399 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3400 
3401 	if (vcpu->arch.exception.pending &&
3402 		nested_vmx_check_exception(vcpu, &exit_qual)) {
3403 		if (block_nested_events)
3404 			return -EBUSY;
3405 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3406 		return 0;
3407 	}
3408 
3409 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3410 	    vmx->nested.preemption_timer_expired) {
3411 		if (block_nested_events)
3412 			return -EBUSY;
3413 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3414 		return 0;
3415 	}
3416 
3417 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3418 		if (block_nested_events)
3419 			return -EBUSY;
3420 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3421 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3422 				  INTR_INFO_VALID_MASK, 0);
3423 		/*
3424 		 * The NMI-triggered VM exit counts as injection:
3425 		 * clear this one and block further NMIs.
3426 		 */
3427 		vcpu->arch.nmi_pending = 0;
3428 		vmx_set_nmi_mask(vcpu, true);
3429 		return 0;
3430 	}
3431 
3432 	if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
3433 	    nested_exit_on_intr(vcpu)) {
3434 		if (block_nested_events)
3435 			return -EBUSY;
3436 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3437 		return 0;
3438 	}
3439 
3440 	vmx_complete_nested_posted_interrupt(vcpu);
3441 	return 0;
3442 }
3443 
3444 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3445 {
3446 	ktime_t remaining =
3447 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3448 	u64 value;
3449 
3450 	if (ktime_to_ns(remaining) <= 0)
3451 		return 0;
3452 
3453 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3454 	do_div(value, 1000000);
3455 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3456 }
3457 
3458 static bool is_vmcs12_ext_field(unsigned long field)
3459 {
3460 	switch (field) {
3461 	case GUEST_ES_SELECTOR:
3462 	case GUEST_CS_SELECTOR:
3463 	case GUEST_SS_SELECTOR:
3464 	case GUEST_DS_SELECTOR:
3465 	case GUEST_FS_SELECTOR:
3466 	case GUEST_GS_SELECTOR:
3467 	case GUEST_LDTR_SELECTOR:
3468 	case GUEST_TR_SELECTOR:
3469 	case GUEST_ES_LIMIT:
3470 	case GUEST_CS_LIMIT:
3471 	case GUEST_SS_LIMIT:
3472 	case GUEST_DS_LIMIT:
3473 	case GUEST_FS_LIMIT:
3474 	case GUEST_GS_LIMIT:
3475 	case GUEST_LDTR_LIMIT:
3476 	case GUEST_TR_LIMIT:
3477 	case GUEST_GDTR_LIMIT:
3478 	case GUEST_IDTR_LIMIT:
3479 	case GUEST_ES_AR_BYTES:
3480 	case GUEST_DS_AR_BYTES:
3481 	case GUEST_FS_AR_BYTES:
3482 	case GUEST_GS_AR_BYTES:
3483 	case GUEST_LDTR_AR_BYTES:
3484 	case GUEST_TR_AR_BYTES:
3485 	case GUEST_ES_BASE:
3486 	case GUEST_CS_BASE:
3487 	case GUEST_SS_BASE:
3488 	case GUEST_DS_BASE:
3489 	case GUEST_FS_BASE:
3490 	case GUEST_GS_BASE:
3491 	case GUEST_LDTR_BASE:
3492 	case GUEST_TR_BASE:
3493 	case GUEST_GDTR_BASE:
3494 	case GUEST_IDTR_BASE:
3495 	case GUEST_PENDING_DBG_EXCEPTIONS:
3496 	case GUEST_BNDCFGS:
3497 		return true;
3498 	default:
3499 		break;
3500 	}
3501 
3502 	return false;
3503 }
3504 
3505 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3506 				       struct vmcs12 *vmcs12)
3507 {
3508 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3509 
3510 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3511 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3512 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3513 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3514 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3515 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3516 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3517 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3518 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3519 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3520 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3521 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3522 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3523 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3524 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3525 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3526 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3527 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3528 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3529 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3530 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3531 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3532 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3533 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3534 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3535 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3536 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3537 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3538 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3539 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3540 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3541 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3542 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3543 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3544 	vmcs12->guest_pending_dbg_exceptions =
3545 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3546 	if (kvm_mpx_supported())
3547 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3548 
3549 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
3550 }
3551 
3552 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3553 				       struct vmcs12 *vmcs12)
3554 {
3555 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3556 	int cpu;
3557 
3558 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
3559 		return;
3560 
3561 
3562 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
3563 
3564 	cpu = get_cpu();
3565 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
3566 	vmx_vcpu_load(&vmx->vcpu, cpu);
3567 
3568 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3569 
3570 	vmx->loaded_vmcs = &vmx->vmcs01;
3571 	vmx_vcpu_load(&vmx->vcpu, cpu);
3572 	put_cpu();
3573 }
3574 
3575 /*
3576  * Update the guest state fields of vmcs12 to reflect changes that
3577  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3578  * VM-entry controls is also updated, since this is really a guest
3579  * state bit.)
3580  */
3581 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3582 {
3583 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3584 
3585 	if (vmx->nested.hv_evmcs)
3586 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3587 
3588 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = !vmx->nested.hv_evmcs;
3589 
3590 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3591 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3592 
3593 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
3594 	vmcs12->guest_rip = kvm_rip_read(vcpu);
3595 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3596 
3597 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3598 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3599 
3600 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3601 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3602 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3603 
3604 	vmcs12->guest_interruptibility_info =
3605 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3606 
3607 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3608 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3609 	else
3610 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3611 
3612 	if (nested_cpu_has_preemption_timer(vmcs12) &&
3613 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3614 			vmcs12->vmx_preemption_timer_value =
3615 				vmx_get_preemption_timer_value(vcpu);
3616 
3617 	/*
3618 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3619 	 * own CR3 without exiting. If it has changed it, we must keep it.
3620 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3621 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3622 	 *
3623 	 * Additionally, restore L2's PDPTR to vmcs12.
3624 	 */
3625 	if (enable_ept) {
3626 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3627 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3628 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3629 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3630 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3631 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3632 		}
3633 	}
3634 
3635 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3636 
3637 	if (nested_cpu_has_vid(vmcs12))
3638 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3639 
3640 	vmcs12->vm_entry_controls =
3641 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3642 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3643 
3644 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
3645 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3646 
3647 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3648 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3649 }
3650 
3651 /*
3652  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3653  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3654  * and this function updates it to reflect the changes to the guest state while
3655  * L2 was running (and perhaps made some exits which were handled directly by L0
3656  * without going back to L1), and to reflect the exit reason.
3657  * Note that we do not have to copy here all VMCS fields, just those that
3658  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3659  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3660  * which already writes to vmcs12 directly.
3661  */
3662 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3663 			   u32 exit_reason, u32 exit_intr_info,
3664 			   unsigned long exit_qualification)
3665 {
3666 	/* update exit information fields: */
3667 	vmcs12->vm_exit_reason = exit_reason;
3668 	vmcs12->exit_qualification = exit_qualification;
3669 	vmcs12->vm_exit_intr_info = exit_intr_info;
3670 
3671 	vmcs12->idt_vectoring_info_field = 0;
3672 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3673 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3674 
3675 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3676 		vmcs12->launch_state = 1;
3677 
3678 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3679 		 * instead of reading the real value. */
3680 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3681 
3682 		/*
3683 		 * Transfer the event that L0 or L1 may wanted to inject into
3684 		 * L2 to IDT_VECTORING_INFO_FIELD.
3685 		 */
3686 		vmcs12_save_pending_event(vcpu, vmcs12);
3687 
3688 		/*
3689 		 * According to spec, there's no need to store the guest's
3690 		 * MSRs if the exit is due to a VM-entry failure that occurs
3691 		 * during or after loading the guest state. Since this exit
3692 		 * does not fall in that category, we need to save the MSRs.
3693 		 */
3694 		if (nested_vmx_store_msr(vcpu,
3695 					 vmcs12->vm_exit_msr_store_addr,
3696 					 vmcs12->vm_exit_msr_store_count))
3697 			nested_vmx_abort(vcpu,
3698 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3699 	}
3700 
3701 	/*
3702 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3703 	 * preserved above and would only end up incorrectly in L1.
3704 	 */
3705 	vcpu->arch.nmi_injected = false;
3706 	kvm_clear_exception_queue(vcpu);
3707 	kvm_clear_interrupt_queue(vcpu);
3708 }
3709 
3710 /*
3711  * A part of what we need to when the nested L2 guest exits and we want to
3712  * run its L1 parent, is to reset L1's guest state to the host state specified
3713  * in vmcs12.
3714  * This function is to be called not only on normal nested exit, but also on
3715  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3716  * Failures During or After Loading Guest State").
3717  * This function should be called when the active VMCS is L1's (vmcs01).
3718  */
3719 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3720 				   struct vmcs12 *vmcs12)
3721 {
3722 	struct kvm_segment seg;
3723 	u32 entry_failure_code;
3724 
3725 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
3726 		vcpu->arch.efer = vmcs12->host_ia32_efer;
3727 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3728 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
3729 	else
3730 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
3731 	vmx_set_efer(vcpu, vcpu->arch.efer);
3732 
3733 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
3734 	kvm_rip_write(vcpu, vmcs12->host_rip);
3735 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
3736 	vmx_set_interrupt_shadow(vcpu, 0);
3737 
3738 	/*
3739 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
3740 	 * actually changed, because vmx_set_cr0 refers to efer set above.
3741 	 *
3742 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
3743 	 * (KVM doesn't change it);
3744 	 */
3745 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3746 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
3747 
3748 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
3749 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3750 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
3751 
3752 	nested_ept_uninit_mmu_context(vcpu);
3753 
3754 	/*
3755 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
3756 	 * couldn't have changed.
3757 	 */
3758 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
3759 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
3760 
3761 	if (!enable_ept)
3762 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
3763 
3764 	/*
3765 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
3766 	 * VMEntry/VMExit. Thus, no need to flush TLB.
3767 	 *
3768 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
3769 	 * flushed on every VMEntry/VMExit.
3770 	 *
3771 	 * Otherwise, we can preserve TLB entries as long as we are
3772 	 * able to tag L1 TLB entries differently than L2 TLB entries.
3773 	 *
3774 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
3775 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
3776 	 * has been set by KVM_REQ_LOAD_CR3.
3777 	 */
3778 	if (enable_vpid &&
3779 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
3780 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3781 	}
3782 
3783 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
3784 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
3785 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
3786 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
3787 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
3788 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
3789 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
3790 
3791 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
3792 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
3793 		vmcs_write64(GUEST_BNDCFGS, 0);
3794 
3795 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
3796 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
3797 		vcpu->arch.pat = vmcs12->host_ia32_pat;
3798 	}
3799 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
3800 		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
3801 			vmcs12->host_ia32_perf_global_ctrl);
3802 
3803 	/* Set L1 segment info according to Intel SDM
3804 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
3805 	seg = (struct kvm_segment) {
3806 		.base = 0,
3807 		.limit = 0xFFFFFFFF,
3808 		.selector = vmcs12->host_cs_selector,
3809 		.type = 11,
3810 		.present = 1,
3811 		.s = 1,
3812 		.g = 1
3813 	};
3814 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3815 		seg.l = 1;
3816 	else
3817 		seg.db = 1;
3818 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
3819 	seg = (struct kvm_segment) {
3820 		.base = 0,
3821 		.limit = 0xFFFFFFFF,
3822 		.type = 3,
3823 		.present = 1,
3824 		.s = 1,
3825 		.db = 1,
3826 		.g = 1
3827 	};
3828 	seg.selector = vmcs12->host_ds_selector;
3829 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
3830 	seg.selector = vmcs12->host_es_selector;
3831 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
3832 	seg.selector = vmcs12->host_ss_selector;
3833 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
3834 	seg.selector = vmcs12->host_fs_selector;
3835 	seg.base = vmcs12->host_fs_base;
3836 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
3837 	seg.selector = vmcs12->host_gs_selector;
3838 	seg.base = vmcs12->host_gs_base;
3839 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
3840 	seg = (struct kvm_segment) {
3841 		.base = vmcs12->host_tr_base,
3842 		.limit = 0x67,
3843 		.selector = vmcs12->host_tr_selector,
3844 		.type = 11,
3845 		.present = 1
3846 	};
3847 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
3848 
3849 	kvm_set_dr(vcpu, 7, 0x400);
3850 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3851 
3852 	if (cpu_has_vmx_msr_bitmap())
3853 		vmx_update_msr_bitmap(vcpu);
3854 
3855 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
3856 				vmcs12->vm_exit_msr_load_count))
3857 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3858 }
3859 
3860 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
3861 {
3862 	struct shared_msr_entry *efer_msr;
3863 	unsigned int i;
3864 
3865 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
3866 		return vmcs_read64(GUEST_IA32_EFER);
3867 
3868 	if (cpu_has_load_ia32_efer())
3869 		return host_efer;
3870 
3871 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
3872 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
3873 			return vmx->msr_autoload.guest.val[i].value;
3874 	}
3875 
3876 	efer_msr = find_msr_entry(vmx, MSR_EFER);
3877 	if (efer_msr)
3878 		return efer_msr->data;
3879 
3880 	return host_efer;
3881 }
3882 
3883 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
3884 {
3885 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3886 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3887 	struct vmx_msr_entry g, h;
3888 	struct msr_data msr;
3889 	gpa_t gpa;
3890 	u32 i, j;
3891 
3892 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
3893 
3894 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
3895 		/*
3896 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
3897 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
3898 		 * and vcpu->arch.dr7 is not squirreled away before the
3899 		 * nested VMENTER (not worth adding a variable in nested_vmx).
3900 		 */
3901 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
3902 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
3903 		else
3904 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
3905 	}
3906 
3907 	/*
3908 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
3909 	 * handle a variety of side effects to KVM's software model.
3910 	 */
3911 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
3912 
3913 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3914 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
3915 
3916 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3917 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
3918 
3919 	nested_ept_uninit_mmu_context(vcpu);
3920 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3921 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3922 
3923 	/*
3924 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
3925 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
3926 	 * VMFail, like everything else we just need to ensure our
3927 	 * software model is up-to-date.
3928 	 */
3929 	if (enable_ept)
3930 		ept_save_pdptrs(vcpu);
3931 
3932 	kvm_mmu_reset_context(vcpu);
3933 
3934 	if (cpu_has_vmx_msr_bitmap())
3935 		vmx_update_msr_bitmap(vcpu);
3936 
3937 	/*
3938 	 * This nasty bit of open coding is a compromise between blindly
3939 	 * loading L1's MSRs using the exit load lists (incorrect emulation
3940 	 * of VMFail), leaving the nested VM's MSRs in the software model
3941 	 * (incorrect behavior) and snapshotting the modified MSRs (too
3942 	 * expensive since the lists are unbound by hardware).  For each
3943 	 * MSR that was (prematurely) loaded from the nested VMEntry load
3944 	 * list, reload it from the exit load list if it exists and differs
3945 	 * from the guest value.  The intent is to stuff host state as
3946 	 * silently as possible, not to fully process the exit load list.
3947 	 */
3948 	msr.host_initiated = false;
3949 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
3950 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
3951 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
3952 			pr_debug_ratelimited(
3953 				"%s read MSR index failed (%u, 0x%08llx)\n",
3954 				__func__, i, gpa);
3955 			goto vmabort;
3956 		}
3957 
3958 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
3959 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
3960 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
3961 				pr_debug_ratelimited(
3962 					"%s read MSR failed (%u, 0x%08llx)\n",
3963 					__func__, j, gpa);
3964 				goto vmabort;
3965 			}
3966 			if (h.index != g.index)
3967 				continue;
3968 			if (h.value == g.value)
3969 				break;
3970 
3971 			if (nested_vmx_load_msr_check(vcpu, &h)) {
3972 				pr_debug_ratelimited(
3973 					"%s check failed (%u, 0x%x, 0x%x)\n",
3974 					__func__, j, h.index, h.reserved);
3975 				goto vmabort;
3976 			}
3977 
3978 			msr.index = h.index;
3979 			msr.data = h.value;
3980 			if (kvm_set_msr(vcpu, &msr)) {
3981 				pr_debug_ratelimited(
3982 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
3983 					__func__, j, h.index, h.value);
3984 				goto vmabort;
3985 			}
3986 		}
3987 	}
3988 
3989 	return;
3990 
3991 vmabort:
3992 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3993 }
3994 
3995 /*
3996  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
3997  * and modify vmcs12 to make it see what it would expect to see there if
3998  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
3999  */
4000 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
4001 		       u32 exit_intr_info, unsigned long exit_qualification)
4002 {
4003 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4004 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4005 
4006 	/* trying to cancel vmlaunch/vmresume is a bug */
4007 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4008 
4009 	leave_guest_mode(vcpu);
4010 
4011 	if (nested_cpu_has_preemption_timer(vmcs12))
4012 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4013 
4014 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
4015 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
4016 
4017 	if (likely(!vmx->fail)) {
4018 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4019 
4020 		if (exit_reason != -1)
4021 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
4022 				       exit_qualification);
4023 
4024 		/*
4025 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4026 		 * also be used to capture vmcs12 cache as part of
4027 		 * capturing nVMX state for snapshot (migration).
4028 		 *
4029 		 * Otherwise, this flush will dirty guest memory at a
4030 		 * point it is already assumed by user-space to be
4031 		 * immutable.
4032 		 */
4033 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4034 	} else {
4035 		/*
4036 		 * The only expected VM-instruction error is "VM entry with
4037 		 * invalid control field(s)." Anything else indicates a
4038 		 * problem with L0.  And we should never get here with a
4039 		 * VMFail of any type if early consistency checks are enabled.
4040 		 */
4041 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4042 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4043 		WARN_ON_ONCE(nested_early_check);
4044 	}
4045 
4046 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4047 
4048 	/* Update any VMCS fields that might have changed while L2 ran */
4049 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4050 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4051 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4052 
4053 	if (kvm_has_tsc_control)
4054 		decache_tsc_multiplier(vmx);
4055 
4056 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4057 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4058 		vmx_set_virtual_apic_mode(vcpu);
4059 	} else if (!nested_cpu_has_ept(vmcs12) &&
4060 		   nested_cpu_has2(vmcs12,
4061 				   SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
4062 		vmx_flush_tlb(vcpu, true);
4063 	}
4064 
4065 	/* Unpin physical memory we referred to in vmcs02 */
4066 	if (vmx->nested.apic_access_page) {
4067 		kvm_release_page_dirty(vmx->nested.apic_access_page);
4068 		vmx->nested.apic_access_page = NULL;
4069 	}
4070 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4071 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4072 	vmx->nested.pi_desc = NULL;
4073 
4074 	/*
4075 	 * We are now running in L2, mmu_notifier will force to reload the
4076 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
4077 	 */
4078 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4079 
4080 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
4081 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4082 
4083 	/* in case we halted in L2 */
4084 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4085 
4086 	if (likely(!vmx->fail)) {
4087 		/*
4088 		 * TODO: SDM says that with acknowledge interrupt on
4089 		 * exit, bit 31 of the VM-exit interrupt information
4090 		 * (valid interrupt) is always set to 1 on
4091 		 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
4092 		 * need kvm_cpu_has_interrupt().  See the commit
4093 		 * message for details.
4094 		 */
4095 		if (nested_exit_intr_ack_set(vcpu) &&
4096 		    exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4097 		    kvm_cpu_has_interrupt(vcpu)) {
4098 			int irq = kvm_cpu_get_interrupt(vcpu);
4099 			WARN_ON(irq < 0);
4100 			vmcs12->vm_exit_intr_info = irq |
4101 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4102 		}
4103 
4104 		if (exit_reason != -1)
4105 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4106 						       vmcs12->exit_qualification,
4107 						       vmcs12->idt_vectoring_info_field,
4108 						       vmcs12->vm_exit_intr_info,
4109 						       vmcs12->vm_exit_intr_error_code,
4110 						       KVM_ISA_VMX);
4111 
4112 		load_vmcs12_host_state(vcpu, vmcs12);
4113 
4114 		return;
4115 	}
4116 
4117 	/*
4118 	 * After an early L2 VM-entry failure, we're now back
4119 	 * in L1 which thinks it just finished a VMLAUNCH or
4120 	 * VMRESUME instruction, so we need to set the failure
4121 	 * flag and the VM-instruction error field of the VMCS
4122 	 * accordingly, and skip the emulated instruction.
4123 	 */
4124 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4125 
4126 	/*
4127 	 * Restore L1's host state to KVM's software model.  We're here
4128 	 * because a consistency check was caught by hardware, which
4129 	 * means some amount of guest state has been propagated to KVM's
4130 	 * model and needs to be unwound to the host's state.
4131 	 */
4132 	nested_vmx_restore_host_state(vcpu);
4133 
4134 	vmx->fail = 0;
4135 }
4136 
4137 /*
4138  * Decode the memory-address operand of a vmx instruction, as recorded on an
4139  * exit caused by such an instruction (run by a guest hypervisor).
4140  * On success, returns 0. When the operand is invalid, returns 1 and throws
4141  * #UD or #GP.
4142  */
4143 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4144 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4145 {
4146 	gva_t off;
4147 	bool exn;
4148 	struct kvm_segment s;
4149 
4150 	/*
4151 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4152 	 * Execution", on an exit, vmx_instruction_info holds most of the
4153 	 * addressing components of the operand. Only the displacement part
4154 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4155 	 * For how an actual address is calculated from all these components,
4156 	 * refer to Vol. 1, "Operand Addressing".
4157 	 */
4158 	int  scaling = vmx_instruction_info & 3;
4159 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4160 	bool is_reg = vmx_instruction_info & (1u << 10);
4161 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4162 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4163 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4164 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4165 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4166 
4167 	if (is_reg) {
4168 		kvm_queue_exception(vcpu, UD_VECTOR);
4169 		return 1;
4170 	}
4171 
4172 	/* Addr = segment_base + offset */
4173 	/* offset = base + [index * scale] + displacement */
4174 	off = exit_qualification; /* holds the displacement */
4175 	if (addr_size == 1)
4176 		off = (gva_t)sign_extend64(off, 31);
4177 	else if (addr_size == 0)
4178 		off = (gva_t)sign_extend64(off, 15);
4179 	if (base_is_valid)
4180 		off += kvm_register_read(vcpu, base_reg);
4181 	if (index_is_valid)
4182 		off += kvm_register_read(vcpu, index_reg)<<scaling;
4183 	vmx_get_segment(vcpu, &s, seg_reg);
4184 
4185 	/*
4186 	 * The effective address, i.e. @off, of a memory operand is truncated
4187 	 * based on the address size of the instruction.  Note that this is
4188 	 * the *effective address*, i.e. the address prior to accounting for
4189 	 * the segment's base.
4190 	 */
4191 	if (addr_size == 1) /* 32 bit */
4192 		off &= 0xffffffff;
4193 	else if (addr_size == 0) /* 16 bit */
4194 		off &= 0xffff;
4195 
4196 	/* Checks for #GP/#SS exceptions. */
4197 	exn = false;
4198 	if (is_long_mode(vcpu)) {
4199 		/*
4200 		 * The virtual/linear address is never truncated in 64-bit
4201 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4202 		 * address when using FS/GS with a non-zero base.
4203 		 */
4204 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4205 			*ret = s.base + off;
4206 		else
4207 			*ret = off;
4208 
4209 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4210 		 * non-canonical form. This is the only check on the memory
4211 		 * destination for long mode!
4212 		 */
4213 		exn = is_noncanonical_address(*ret, vcpu);
4214 	} else {
4215 		/*
4216 		 * When not in long mode, the virtual/linear address is
4217 		 * unconditionally truncated to 32 bits regardless of the
4218 		 * address size.
4219 		 */
4220 		*ret = (s.base + off) & 0xffffffff;
4221 
4222 		/* Protected mode: apply checks for segment validity in the
4223 		 * following order:
4224 		 * - segment type check (#GP(0) may be thrown)
4225 		 * - usability check (#GP(0)/#SS(0))
4226 		 * - limit check (#GP(0)/#SS(0))
4227 		 */
4228 		if (wr)
4229 			/* #GP(0) if the destination operand is located in a
4230 			 * read-only data segment or any code segment.
4231 			 */
4232 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4233 		else
4234 			/* #GP(0) if the source operand is located in an
4235 			 * execute-only code segment
4236 			 */
4237 			exn = ((s.type & 0xa) == 8);
4238 		if (exn) {
4239 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4240 			return 1;
4241 		}
4242 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4243 		 */
4244 		exn = (s.unusable != 0);
4245 
4246 		/*
4247 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4248 		 * outside the segment limit.  All CPUs that support VMX ignore
4249 		 * limit checks for flat segments, i.e. segments with base==0,
4250 		 * limit==0xffffffff and of type expand-up data or code.
4251 		 */
4252 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4253 		     ((s.type & 8) || !(s.type & 4))))
4254 			exn = exn || ((u64)off + len - 1 > s.limit);
4255 	}
4256 	if (exn) {
4257 		kvm_queue_exception_e(vcpu,
4258 				      seg_reg == VCPU_SREG_SS ?
4259 						SS_VECTOR : GP_VECTOR,
4260 				      0);
4261 		return 1;
4262 	}
4263 
4264 	return 0;
4265 }
4266 
4267 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4268 {
4269 	gva_t gva;
4270 	struct x86_exception e;
4271 
4272 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4273 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4274 				sizeof(*vmpointer), &gva))
4275 		return 1;
4276 
4277 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4278 		kvm_inject_page_fault(vcpu, &e);
4279 		return 1;
4280 	}
4281 
4282 	return 0;
4283 }
4284 
4285 /*
4286  * Allocate a shadow VMCS and associate it with the currently loaded
4287  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4288  * VMCS is also VMCLEARed, so that it is ready for use.
4289  */
4290 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4291 {
4292 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4293 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4294 
4295 	/*
4296 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4297 	 * executes VMXON and free it when L1 executes VMXOFF.
4298 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4299 	 * here when vmcs01 already have an allocated shadow vmcs.
4300 	 */
4301 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4302 
4303 	if (!loaded_vmcs->shadow_vmcs) {
4304 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4305 		if (loaded_vmcs->shadow_vmcs)
4306 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4307 	}
4308 	return loaded_vmcs->shadow_vmcs;
4309 }
4310 
4311 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4312 {
4313 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4314 	int r;
4315 
4316 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4317 	if (r < 0)
4318 		goto out_vmcs02;
4319 
4320 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4321 	if (!vmx->nested.cached_vmcs12)
4322 		goto out_cached_vmcs12;
4323 
4324 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4325 	if (!vmx->nested.cached_shadow_vmcs12)
4326 		goto out_cached_shadow_vmcs12;
4327 
4328 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4329 		goto out_shadow_vmcs;
4330 
4331 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4332 		     HRTIMER_MODE_REL_PINNED);
4333 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4334 
4335 	vmx->nested.vpid02 = allocate_vpid();
4336 
4337 	vmx->nested.vmcs02_initialized = false;
4338 	vmx->nested.vmxon = true;
4339 
4340 	if (pt_mode == PT_MODE_HOST_GUEST) {
4341 		vmx->pt_desc.guest.ctl = 0;
4342 		pt_update_intercept_for_msr(vmx);
4343 	}
4344 
4345 	return 0;
4346 
4347 out_shadow_vmcs:
4348 	kfree(vmx->nested.cached_shadow_vmcs12);
4349 
4350 out_cached_shadow_vmcs12:
4351 	kfree(vmx->nested.cached_vmcs12);
4352 
4353 out_cached_vmcs12:
4354 	free_loaded_vmcs(&vmx->nested.vmcs02);
4355 
4356 out_vmcs02:
4357 	return -ENOMEM;
4358 }
4359 
4360 /*
4361  * Emulate the VMXON instruction.
4362  * Currently, we just remember that VMX is active, and do not save or even
4363  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4364  * do not currently need to store anything in that guest-allocated memory
4365  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4366  * argument is different from the VMXON pointer (which the spec says they do).
4367  */
4368 static int handle_vmon(struct kvm_vcpu *vcpu)
4369 {
4370 	int ret;
4371 	gpa_t vmptr;
4372 	uint32_t revision;
4373 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4374 	const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
4375 		| FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
4376 
4377 	/*
4378 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4379 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4380 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4381 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4382 	 * have already been checked by hardware, prior to the VM-exit for
4383 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4384 	 * that bit set to 1 in non-root mode.
4385 	 */
4386 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4387 		kvm_queue_exception(vcpu, UD_VECTOR);
4388 		return 1;
4389 	}
4390 
4391 	/* CPL=0 must be checked manually. */
4392 	if (vmx_get_cpl(vcpu)) {
4393 		kvm_inject_gp(vcpu, 0);
4394 		return 1;
4395 	}
4396 
4397 	if (vmx->nested.vmxon)
4398 		return nested_vmx_failValid(vcpu,
4399 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4400 
4401 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4402 			!= VMXON_NEEDED_FEATURES) {
4403 		kvm_inject_gp(vcpu, 0);
4404 		return 1;
4405 	}
4406 
4407 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4408 		return 1;
4409 
4410 	/*
4411 	 * SDM 3: 24.11.5
4412 	 * The first 4 bytes of VMXON region contain the supported
4413 	 * VMCS revision identifier
4414 	 *
4415 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4416 	 * which replaces physical address width with 32
4417 	 */
4418 	if (!page_address_valid(vcpu, vmptr))
4419 		return nested_vmx_failInvalid(vcpu);
4420 
4421 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4422 	    revision != VMCS12_REVISION)
4423 		return nested_vmx_failInvalid(vcpu);
4424 
4425 	vmx->nested.vmxon_ptr = vmptr;
4426 	ret = enter_vmx_operation(vcpu);
4427 	if (ret)
4428 		return ret;
4429 
4430 	return nested_vmx_succeed(vcpu);
4431 }
4432 
4433 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4434 {
4435 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4436 
4437 	if (vmx->nested.current_vmptr == -1ull)
4438 		return;
4439 
4440 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4441 
4442 	if (enable_shadow_vmcs) {
4443 		/* copy to memory all shadowed fields in case
4444 		   they were modified */
4445 		copy_shadow_to_vmcs12(vmx);
4446 		vmx_disable_shadow_vmcs(vmx);
4447 	}
4448 	vmx->nested.posted_intr_nv = -1;
4449 
4450 	/* Flush VMCS12 to guest memory */
4451 	kvm_vcpu_write_guest_page(vcpu,
4452 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4453 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4454 
4455 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4456 
4457 	vmx->nested.current_vmptr = -1ull;
4458 }
4459 
4460 /* Emulate the VMXOFF instruction */
4461 static int handle_vmoff(struct kvm_vcpu *vcpu)
4462 {
4463 	if (!nested_vmx_check_permission(vcpu))
4464 		return 1;
4465 	free_nested(vcpu);
4466 	return nested_vmx_succeed(vcpu);
4467 }
4468 
4469 /* Emulate the VMCLEAR instruction */
4470 static int handle_vmclear(struct kvm_vcpu *vcpu)
4471 {
4472 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4473 	u32 zero = 0;
4474 	gpa_t vmptr;
4475 	u64 evmcs_gpa;
4476 
4477 	if (!nested_vmx_check_permission(vcpu))
4478 		return 1;
4479 
4480 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4481 		return 1;
4482 
4483 	if (!page_address_valid(vcpu, vmptr))
4484 		return nested_vmx_failValid(vcpu,
4485 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4486 
4487 	if (vmptr == vmx->nested.vmxon_ptr)
4488 		return nested_vmx_failValid(vcpu,
4489 			VMXERR_VMCLEAR_VMXON_POINTER);
4490 
4491 	/*
4492 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4493 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4494 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4495 	 * writing to the non-existent 'launch_state' field. The area doesn't
4496 	 * have to be the currently active EVMCS on the calling CPU and there's
4497 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
4498 	 * state. It is possible that the area will stay mapped as
4499 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
4500 	 */
4501 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
4502 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
4503 		if (vmptr == vmx->nested.current_vmptr)
4504 			nested_release_vmcs12(vcpu);
4505 
4506 		kvm_vcpu_write_guest(vcpu,
4507 				     vmptr + offsetof(struct vmcs12,
4508 						      launch_state),
4509 				     &zero, sizeof(zero));
4510 	}
4511 
4512 	return nested_vmx_succeed(vcpu);
4513 }
4514 
4515 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4516 
4517 /* Emulate the VMLAUNCH instruction */
4518 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4519 {
4520 	return nested_vmx_run(vcpu, true);
4521 }
4522 
4523 /* Emulate the VMRESUME instruction */
4524 static int handle_vmresume(struct kvm_vcpu *vcpu)
4525 {
4526 
4527 	return nested_vmx_run(vcpu, false);
4528 }
4529 
4530 static int handle_vmread(struct kvm_vcpu *vcpu)
4531 {
4532 	unsigned long field;
4533 	u64 field_value;
4534 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4535 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4536 	int len;
4537 	gva_t gva = 0;
4538 	struct vmcs12 *vmcs12;
4539 	short offset;
4540 
4541 	if (!nested_vmx_check_permission(vcpu))
4542 		return 1;
4543 
4544 	if (to_vmx(vcpu)->nested.current_vmptr == -1ull)
4545 		return nested_vmx_failInvalid(vcpu);
4546 
4547 	if (!is_guest_mode(vcpu))
4548 		vmcs12 = get_vmcs12(vcpu);
4549 	else {
4550 		/*
4551 		 * When vmcs->vmcs_link_pointer is -1ull, any VMREAD
4552 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4553 		 */
4554 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4555 			return nested_vmx_failInvalid(vcpu);
4556 		vmcs12 = get_shadow_vmcs12(vcpu);
4557 	}
4558 
4559 	/* Decode instruction info and find the field to read */
4560 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4561 
4562 	offset = vmcs_field_to_offset(field);
4563 	if (offset < 0)
4564 		return nested_vmx_failValid(vcpu,
4565 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4566 
4567 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
4568 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4569 
4570 	/* Read the field, zero-extended to a u64 field_value */
4571 	field_value = vmcs12_read_any(vmcs12, field, offset);
4572 
4573 	/*
4574 	 * Now copy part of this value to register or memory, as requested.
4575 	 * Note that the number of bits actually copied is 32 or 64 depending
4576 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4577 	 */
4578 	if (vmx_instruction_info & (1u << 10)) {
4579 		kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
4580 			field_value);
4581 	} else {
4582 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4583 		if (get_vmx_mem_address(vcpu, exit_qualification,
4584 				vmx_instruction_info, true, len, &gva))
4585 			return 1;
4586 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4587 		kvm_write_guest_virt_system(vcpu, gva, &field_value, len, NULL);
4588 	}
4589 
4590 	return nested_vmx_succeed(vcpu);
4591 }
4592 
4593 static bool is_shadow_field_rw(unsigned long field)
4594 {
4595 	switch (field) {
4596 #define SHADOW_FIELD_RW(x, y) case x:
4597 #include "vmcs_shadow_fields.h"
4598 		return true;
4599 	default:
4600 		break;
4601 	}
4602 	return false;
4603 }
4604 
4605 static bool is_shadow_field_ro(unsigned long field)
4606 {
4607 	switch (field) {
4608 #define SHADOW_FIELD_RO(x, y) case x:
4609 #include "vmcs_shadow_fields.h"
4610 		return true;
4611 	default:
4612 		break;
4613 	}
4614 	return false;
4615 }
4616 
4617 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4618 {
4619 	unsigned long field;
4620 	int len;
4621 	gva_t gva;
4622 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4623 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4624 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4625 
4626 	/* The value to write might be 32 or 64 bits, depending on L1's long
4627 	 * mode, and eventually we need to write that into a field of several
4628 	 * possible lengths. The code below first zero-extends the value to 64
4629 	 * bit (field_value), and then copies only the appropriate number of
4630 	 * bits into the vmcs12 field.
4631 	 */
4632 	u64 field_value = 0;
4633 	struct x86_exception e;
4634 	struct vmcs12 *vmcs12;
4635 	short offset;
4636 
4637 	if (!nested_vmx_check_permission(vcpu))
4638 		return 1;
4639 
4640 	if (vmx->nested.current_vmptr == -1ull)
4641 		return nested_vmx_failInvalid(vcpu);
4642 
4643 	if (vmx_instruction_info & (1u << 10))
4644 		field_value = kvm_register_readl(vcpu,
4645 			(((vmx_instruction_info) >> 3) & 0xf));
4646 	else {
4647 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4648 		if (get_vmx_mem_address(vcpu, exit_qualification,
4649 				vmx_instruction_info, false, len, &gva))
4650 			return 1;
4651 		if (kvm_read_guest_virt(vcpu, gva, &field_value, len, &e)) {
4652 			kvm_inject_page_fault(vcpu, &e);
4653 			return 1;
4654 		}
4655 	}
4656 
4657 
4658 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4659 	/*
4660 	 * If the vCPU supports "VMWRITE to any supported field in the
4661 	 * VMCS," then the "read-only" fields are actually read/write.
4662 	 */
4663 	if (vmcs_field_readonly(field) &&
4664 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4665 		return nested_vmx_failValid(vcpu,
4666 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4667 
4668 	if (!is_guest_mode(vcpu)) {
4669 		vmcs12 = get_vmcs12(vcpu);
4670 
4671 		/*
4672 		 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
4673 		 * vmcs12, else we may crush a field or consume a stale value.
4674 		 */
4675 		if (!is_shadow_field_rw(field))
4676 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4677 	} else {
4678 		/*
4679 		 * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
4680 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4681 		 */
4682 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4683 			return nested_vmx_failInvalid(vcpu);
4684 		vmcs12 = get_shadow_vmcs12(vcpu);
4685 	}
4686 
4687 	offset = vmcs_field_to_offset(field);
4688 	if (offset < 0)
4689 		return nested_vmx_failValid(vcpu,
4690 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4691 
4692 	/*
4693 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
4694 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
4695 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
4696 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
4697 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
4698 	 * the stripped down value, L2 sees the full value as stored by KVM).
4699 	 */
4700 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
4701 		field_value &= 0x1f0ff;
4702 
4703 	vmcs12_write_any(vmcs12, field, offset, field_value);
4704 
4705 	/*
4706 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
4707 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
4708 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
4709 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
4710 	 */
4711 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
4712 		/*
4713 		 * L1 can read these fields without exiting, ensure the
4714 		 * shadow VMCS is up-to-date.
4715 		 */
4716 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
4717 			preempt_disable();
4718 			vmcs_load(vmx->vmcs01.shadow_vmcs);
4719 
4720 			__vmcs_writel(field, field_value);
4721 
4722 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
4723 			vmcs_load(vmx->loaded_vmcs->vmcs);
4724 			preempt_enable();
4725 		}
4726 		vmx->nested.dirty_vmcs12 = true;
4727 	}
4728 
4729 	return nested_vmx_succeed(vcpu);
4730 }
4731 
4732 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
4733 {
4734 	vmx->nested.current_vmptr = vmptr;
4735 	if (enable_shadow_vmcs) {
4736 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
4737 		vmcs_write64(VMCS_LINK_POINTER,
4738 			     __pa(vmx->vmcs01.shadow_vmcs));
4739 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4740 	}
4741 	vmx->nested.dirty_vmcs12 = true;
4742 }
4743 
4744 /* Emulate the VMPTRLD instruction */
4745 static int handle_vmptrld(struct kvm_vcpu *vcpu)
4746 {
4747 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4748 	gpa_t vmptr;
4749 
4750 	if (!nested_vmx_check_permission(vcpu))
4751 		return 1;
4752 
4753 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4754 		return 1;
4755 
4756 	if (!page_address_valid(vcpu, vmptr))
4757 		return nested_vmx_failValid(vcpu,
4758 			VMXERR_VMPTRLD_INVALID_ADDRESS);
4759 
4760 	if (vmptr == vmx->nested.vmxon_ptr)
4761 		return nested_vmx_failValid(vcpu,
4762 			VMXERR_VMPTRLD_VMXON_POINTER);
4763 
4764 	/* Forbid normal VMPTRLD if Enlightened version was used */
4765 	if (vmx->nested.hv_evmcs)
4766 		return 1;
4767 
4768 	if (vmx->nested.current_vmptr != vmptr) {
4769 		struct kvm_host_map map;
4770 		struct vmcs12 *new_vmcs12;
4771 
4772 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
4773 			/*
4774 			 * Reads from an unbacked page return all 1s,
4775 			 * which means that the 32 bits located at the
4776 			 * given physical address won't match the required
4777 			 * VMCS12_REVISION identifier.
4778 			 */
4779 			return nested_vmx_failValid(vcpu,
4780 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4781 		}
4782 
4783 		new_vmcs12 = map.hva;
4784 
4785 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
4786 		    (new_vmcs12->hdr.shadow_vmcs &&
4787 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
4788 			kvm_vcpu_unmap(vcpu, &map, false);
4789 			return nested_vmx_failValid(vcpu,
4790 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4791 		}
4792 
4793 		nested_release_vmcs12(vcpu);
4794 
4795 		/*
4796 		 * Load VMCS12 from guest memory since it is not already
4797 		 * cached.
4798 		 */
4799 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
4800 		kvm_vcpu_unmap(vcpu, &map, false);
4801 
4802 		set_current_vmptr(vmx, vmptr);
4803 	}
4804 
4805 	return nested_vmx_succeed(vcpu);
4806 }
4807 
4808 /* Emulate the VMPTRST instruction */
4809 static int handle_vmptrst(struct kvm_vcpu *vcpu)
4810 {
4811 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
4812 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4813 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
4814 	struct x86_exception e;
4815 	gva_t gva;
4816 
4817 	if (!nested_vmx_check_permission(vcpu))
4818 		return 1;
4819 
4820 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
4821 		return 1;
4822 
4823 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
4824 				true, sizeof(gpa_t), &gva))
4825 		return 1;
4826 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
4827 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
4828 					sizeof(gpa_t), &e)) {
4829 		kvm_inject_page_fault(vcpu, &e);
4830 		return 1;
4831 	}
4832 	return nested_vmx_succeed(vcpu);
4833 }
4834 
4835 /* Emulate the INVEPT instruction */
4836 static int handle_invept(struct kvm_vcpu *vcpu)
4837 {
4838 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4839 	u32 vmx_instruction_info, types;
4840 	unsigned long type;
4841 	gva_t gva;
4842 	struct x86_exception e;
4843 	struct {
4844 		u64 eptp, gpa;
4845 	} operand;
4846 
4847 	if (!(vmx->nested.msrs.secondary_ctls_high &
4848 	      SECONDARY_EXEC_ENABLE_EPT) ||
4849 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
4850 		kvm_queue_exception(vcpu, UD_VECTOR);
4851 		return 1;
4852 	}
4853 
4854 	if (!nested_vmx_check_permission(vcpu))
4855 		return 1;
4856 
4857 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4858 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4859 
4860 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
4861 
4862 	if (type >= 32 || !(types & (1 << type)))
4863 		return nested_vmx_failValid(vcpu,
4864 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4865 
4866 	/* According to the Intel VMX instruction reference, the memory
4867 	 * operand is read even if it isn't needed (e.g., for type==global)
4868 	 */
4869 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4870 			vmx_instruction_info, false, sizeof(operand), &gva))
4871 		return 1;
4872 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4873 		kvm_inject_page_fault(vcpu, &e);
4874 		return 1;
4875 	}
4876 
4877 	switch (type) {
4878 	case VMX_EPT_EXTENT_GLOBAL:
4879 	case VMX_EPT_EXTENT_CONTEXT:
4880 	/*
4881 	 * TODO: Sync the necessary shadow EPT roots here, rather than
4882 	 * at the next emulated VM-entry.
4883 	 */
4884 		break;
4885 	default:
4886 		BUG_ON(1);
4887 		break;
4888 	}
4889 
4890 	return nested_vmx_succeed(vcpu);
4891 }
4892 
4893 static int handle_invvpid(struct kvm_vcpu *vcpu)
4894 {
4895 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4896 	u32 vmx_instruction_info;
4897 	unsigned long type, types;
4898 	gva_t gva;
4899 	struct x86_exception e;
4900 	struct {
4901 		u64 vpid;
4902 		u64 gla;
4903 	} operand;
4904 	u16 vpid02;
4905 
4906 	if (!(vmx->nested.msrs.secondary_ctls_high &
4907 	      SECONDARY_EXEC_ENABLE_VPID) ||
4908 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
4909 		kvm_queue_exception(vcpu, UD_VECTOR);
4910 		return 1;
4911 	}
4912 
4913 	if (!nested_vmx_check_permission(vcpu))
4914 		return 1;
4915 
4916 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4917 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4918 
4919 	types = (vmx->nested.msrs.vpid_caps &
4920 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
4921 
4922 	if (type >= 32 || !(types & (1 << type)))
4923 		return nested_vmx_failValid(vcpu,
4924 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4925 
4926 	/* according to the intel vmx instruction reference, the memory
4927 	 * operand is read even if it isn't needed (e.g., for type==global)
4928 	 */
4929 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4930 			vmx_instruction_info, false, sizeof(operand), &gva))
4931 		return 1;
4932 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4933 		kvm_inject_page_fault(vcpu, &e);
4934 		return 1;
4935 	}
4936 	if (operand.vpid >> 16)
4937 		return nested_vmx_failValid(vcpu,
4938 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4939 
4940 	vpid02 = nested_get_vpid02(vcpu);
4941 	switch (type) {
4942 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
4943 		if (!operand.vpid ||
4944 		    is_noncanonical_address(operand.gla, vcpu))
4945 			return nested_vmx_failValid(vcpu,
4946 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4947 		if (cpu_has_vmx_invvpid_individual_addr()) {
4948 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
4949 				vpid02, operand.gla);
4950 		} else
4951 			__vmx_flush_tlb(vcpu, vpid02, false);
4952 		break;
4953 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
4954 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
4955 		if (!operand.vpid)
4956 			return nested_vmx_failValid(vcpu,
4957 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4958 		__vmx_flush_tlb(vcpu, vpid02, false);
4959 		break;
4960 	case VMX_VPID_EXTENT_ALL_CONTEXT:
4961 		__vmx_flush_tlb(vcpu, vpid02, false);
4962 		break;
4963 	default:
4964 		WARN_ON_ONCE(1);
4965 		return kvm_skip_emulated_instruction(vcpu);
4966 	}
4967 
4968 	return nested_vmx_succeed(vcpu);
4969 }
4970 
4971 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
4972 				     struct vmcs12 *vmcs12)
4973 {
4974 	u32 index = kvm_rcx_read(vcpu);
4975 	u64 address;
4976 	bool accessed_dirty;
4977 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4978 
4979 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
4980 	    !nested_cpu_has_ept(vmcs12))
4981 		return 1;
4982 
4983 	if (index >= VMFUNC_EPTP_ENTRIES)
4984 		return 1;
4985 
4986 
4987 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
4988 				     &address, index * 8, 8))
4989 		return 1;
4990 
4991 	accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
4992 
4993 	/*
4994 	 * If the (L2) guest does a vmfunc to the currently
4995 	 * active ept pointer, we don't have to do anything else
4996 	 */
4997 	if (vmcs12->ept_pointer != address) {
4998 		if (!valid_ept_address(vcpu, address))
4999 			return 1;
5000 
5001 		kvm_mmu_unload(vcpu);
5002 		mmu->ept_ad = accessed_dirty;
5003 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
5004 		vmcs12->ept_pointer = address;
5005 		/*
5006 		 * TODO: Check what's the correct approach in case
5007 		 * mmu reload fails. Currently, we just let the next
5008 		 * reload potentially fail
5009 		 */
5010 		kvm_mmu_reload(vcpu);
5011 	}
5012 
5013 	return 0;
5014 }
5015 
5016 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5017 {
5018 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5019 	struct vmcs12 *vmcs12;
5020 	u32 function = kvm_rax_read(vcpu);
5021 
5022 	/*
5023 	 * VMFUNC is only supported for nested guests, but we always enable the
5024 	 * secondary control for simplicity; for non-nested mode, fake that we
5025 	 * didn't by injecting #UD.
5026 	 */
5027 	if (!is_guest_mode(vcpu)) {
5028 		kvm_queue_exception(vcpu, UD_VECTOR);
5029 		return 1;
5030 	}
5031 
5032 	vmcs12 = get_vmcs12(vcpu);
5033 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
5034 		goto fail;
5035 
5036 	switch (function) {
5037 	case 0:
5038 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5039 			goto fail;
5040 		break;
5041 	default:
5042 		goto fail;
5043 	}
5044 	return kvm_skip_emulated_instruction(vcpu);
5045 
5046 fail:
5047 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
5048 			  vmcs_read32(VM_EXIT_INTR_INFO),
5049 			  vmcs_readl(EXIT_QUALIFICATION));
5050 	return 1;
5051 }
5052 
5053 
5054 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5055 				       struct vmcs12 *vmcs12)
5056 {
5057 	unsigned long exit_qualification;
5058 	gpa_t bitmap, last_bitmap;
5059 	unsigned int port;
5060 	int size;
5061 	u8 b;
5062 
5063 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5064 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5065 
5066 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5067 
5068 	port = exit_qualification >> 16;
5069 	size = (exit_qualification & 7) + 1;
5070 
5071 	last_bitmap = (gpa_t)-1;
5072 	b = -1;
5073 
5074 	while (size > 0) {
5075 		if (port < 0x8000)
5076 			bitmap = vmcs12->io_bitmap_a;
5077 		else if (port < 0x10000)
5078 			bitmap = vmcs12->io_bitmap_b;
5079 		else
5080 			return true;
5081 		bitmap += (port & 0x7fff) / 8;
5082 
5083 		if (last_bitmap != bitmap)
5084 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5085 				return true;
5086 		if (b & (1 << (port & 7)))
5087 			return true;
5088 
5089 		port++;
5090 		size--;
5091 		last_bitmap = bitmap;
5092 	}
5093 
5094 	return false;
5095 }
5096 
5097 /*
5098  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5099  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5100  * disinterest in the current event (read or write a specific MSR) by using an
5101  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5102  */
5103 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5104 	struct vmcs12 *vmcs12, u32 exit_reason)
5105 {
5106 	u32 msr_index = kvm_rcx_read(vcpu);
5107 	gpa_t bitmap;
5108 
5109 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5110 		return true;
5111 
5112 	/*
5113 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5114 	 * for the four combinations of read/write and low/high MSR numbers.
5115 	 * First we need to figure out which of the four to use:
5116 	 */
5117 	bitmap = vmcs12->msr_bitmap;
5118 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5119 		bitmap += 2048;
5120 	if (msr_index >= 0xc0000000) {
5121 		msr_index -= 0xc0000000;
5122 		bitmap += 1024;
5123 	}
5124 
5125 	/* Then read the msr_index'th bit from this bitmap: */
5126 	if (msr_index < 1024*8) {
5127 		unsigned char b;
5128 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5129 			return true;
5130 		return 1 & (b >> (msr_index & 7));
5131 	} else
5132 		return true; /* let L1 handle the wrong parameter */
5133 }
5134 
5135 /*
5136  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5137  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5138  * intercept (via guest_host_mask etc.) the current event.
5139  */
5140 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5141 	struct vmcs12 *vmcs12)
5142 {
5143 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5144 	int cr = exit_qualification & 15;
5145 	int reg;
5146 	unsigned long val;
5147 
5148 	switch ((exit_qualification >> 4) & 3) {
5149 	case 0: /* mov to cr */
5150 		reg = (exit_qualification >> 8) & 15;
5151 		val = kvm_register_readl(vcpu, reg);
5152 		switch (cr) {
5153 		case 0:
5154 			if (vmcs12->cr0_guest_host_mask &
5155 			    (val ^ vmcs12->cr0_read_shadow))
5156 				return true;
5157 			break;
5158 		case 3:
5159 			if ((vmcs12->cr3_target_count >= 1 &&
5160 					vmcs12->cr3_target_value0 == val) ||
5161 				(vmcs12->cr3_target_count >= 2 &&
5162 					vmcs12->cr3_target_value1 == val) ||
5163 				(vmcs12->cr3_target_count >= 3 &&
5164 					vmcs12->cr3_target_value2 == val) ||
5165 				(vmcs12->cr3_target_count >= 4 &&
5166 					vmcs12->cr3_target_value3 == val))
5167 				return false;
5168 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5169 				return true;
5170 			break;
5171 		case 4:
5172 			if (vmcs12->cr4_guest_host_mask &
5173 			    (vmcs12->cr4_read_shadow ^ val))
5174 				return true;
5175 			break;
5176 		case 8:
5177 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5178 				return true;
5179 			break;
5180 		}
5181 		break;
5182 	case 2: /* clts */
5183 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5184 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5185 			return true;
5186 		break;
5187 	case 1: /* mov from cr */
5188 		switch (cr) {
5189 		case 3:
5190 			if (vmcs12->cpu_based_vm_exec_control &
5191 			    CPU_BASED_CR3_STORE_EXITING)
5192 				return true;
5193 			break;
5194 		case 8:
5195 			if (vmcs12->cpu_based_vm_exec_control &
5196 			    CPU_BASED_CR8_STORE_EXITING)
5197 				return true;
5198 			break;
5199 		}
5200 		break;
5201 	case 3: /* lmsw */
5202 		/*
5203 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5204 		 * cr0. Other attempted changes are ignored, with no exit.
5205 		 */
5206 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5207 		if (vmcs12->cr0_guest_host_mask & 0xe &
5208 		    (val ^ vmcs12->cr0_read_shadow))
5209 			return true;
5210 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5211 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5212 		    (val & 0x1))
5213 			return true;
5214 		break;
5215 	}
5216 	return false;
5217 }
5218 
5219 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5220 	struct vmcs12 *vmcs12, gpa_t bitmap)
5221 {
5222 	u32 vmx_instruction_info;
5223 	unsigned long field;
5224 	u8 b;
5225 
5226 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5227 		return true;
5228 
5229 	/* Decode instruction info and find the field to access */
5230 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5231 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5232 
5233 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5234 	if (field >> 15)
5235 		return true;
5236 
5237 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5238 		return true;
5239 
5240 	return 1 & (b >> (field & 7));
5241 }
5242 
5243 /*
5244  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5245  * should handle it ourselves in L0 (and then continue L2). Only call this
5246  * when in is_guest_mode (L2).
5247  */
5248 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5249 {
5250 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5251 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5252 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5253 
5254 	if (vmx->nested.nested_run_pending)
5255 		return false;
5256 
5257 	if (unlikely(vmx->fail)) {
5258 		pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5259 				    vmcs_read32(VM_INSTRUCTION_ERROR));
5260 		return true;
5261 	}
5262 
5263 	/*
5264 	 * The host physical addresses of some pages of guest memory
5265 	 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5266 	 * Page). The CPU may write to these pages via their host
5267 	 * physical address while L2 is running, bypassing any
5268 	 * address-translation-based dirty tracking (e.g. EPT write
5269 	 * protection).
5270 	 *
5271 	 * Mark them dirty on every exit from L2 to prevent them from
5272 	 * getting out of sync with dirty tracking.
5273 	 */
5274 	nested_mark_vmcs12_pages_dirty(vcpu);
5275 
5276 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5277 				vmcs_readl(EXIT_QUALIFICATION),
5278 				vmx->idt_vectoring_info,
5279 				intr_info,
5280 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5281 				KVM_ISA_VMX);
5282 
5283 	switch (exit_reason) {
5284 	case EXIT_REASON_EXCEPTION_NMI:
5285 		if (is_nmi(intr_info))
5286 			return false;
5287 		else if (is_page_fault(intr_info))
5288 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5289 		else if (is_debug(intr_info) &&
5290 			 vcpu->guest_debug &
5291 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5292 			return false;
5293 		else if (is_breakpoint(intr_info) &&
5294 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5295 			return false;
5296 		return vmcs12->exception_bitmap &
5297 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5298 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5299 		return false;
5300 	case EXIT_REASON_TRIPLE_FAULT:
5301 		return true;
5302 	case EXIT_REASON_PENDING_INTERRUPT:
5303 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
5304 	case EXIT_REASON_NMI_WINDOW:
5305 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
5306 	case EXIT_REASON_TASK_SWITCH:
5307 		return true;
5308 	case EXIT_REASON_CPUID:
5309 		return true;
5310 	case EXIT_REASON_HLT:
5311 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5312 	case EXIT_REASON_INVD:
5313 		return true;
5314 	case EXIT_REASON_INVLPG:
5315 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5316 	case EXIT_REASON_RDPMC:
5317 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5318 	case EXIT_REASON_RDRAND:
5319 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5320 	case EXIT_REASON_RDSEED:
5321 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5322 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5323 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5324 	case EXIT_REASON_VMREAD:
5325 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5326 			vmcs12->vmread_bitmap);
5327 	case EXIT_REASON_VMWRITE:
5328 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5329 			vmcs12->vmwrite_bitmap);
5330 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5331 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5332 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5333 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5334 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5335 		/*
5336 		 * VMX instructions trap unconditionally. This allows L1 to
5337 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5338 		 */
5339 		return true;
5340 	case EXIT_REASON_CR_ACCESS:
5341 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5342 	case EXIT_REASON_DR_ACCESS:
5343 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5344 	case EXIT_REASON_IO_INSTRUCTION:
5345 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5346 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5347 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5348 	case EXIT_REASON_MSR_READ:
5349 	case EXIT_REASON_MSR_WRITE:
5350 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5351 	case EXIT_REASON_INVALID_STATE:
5352 		return true;
5353 	case EXIT_REASON_MWAIT_INSTRUCTION:
5354 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5355 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5356 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
5357 	case EXIT_REASON_MONITOR_INSTRUCTION:
5358 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5359 	case EXIT_REASON_PAUSE_INSTRUCTION:
5360 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5361 			nested_cpu_has2(vmcs12,
5362 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5363 	case EXIT_REASON_MCE_DURING_VMENTRY:
5364 		return false;
5365 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5366 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5367 	case EXIT_REASON_APIC_ACCESS:
5368 	case EXIT_REASON_APIC_WRITE:
5369 	case EXIT_REASON_EOI_INDUCED:
5370 		/*
5371 		 * The controls for "virtualize APIC accesses," "APIC-
5372 		 * register virtualization," and "virtual-interrupt
5373 		 * delivery" only come from vmcs12.
5374 		 */
5375 		return true;
5376 	case EXIT_REASON_EPT_VIOLATION:
5377 		/*
5378 		 * L0 always deals with the EPT violation. If nested EPT is
5379 		 * used, and the nested mmu code discovers that the address is
5380 		 * missing in the guest EPT table (EPT12), the EPT violation
5381 		 * will be injected with nested_ept_inject_page_fault()
5382 		 */
5383 		return false;
5384 	case EXIT_REASON_EPT_MISCONFIG:
5385 		/*
5386 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5387 		 * table (shadow on EPT) or a merged EPT table that L0 built
5388 		 * (EPT on EPT). So any problems with the structure of the
5389 		 * table is L0's fault.
5390 		 */
5391 		return false;
5392 	case EXIT_REASON_INVPCID:
5393 		return
5394 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5395 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5396 	case EXIT_REASON_WBINVD:
5397 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5398 	case EXIT_REASON_XSETBV:
5399 		return true;
5400 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5401 		/*
5402 		 * This should never happen, since it is not possible to
5403 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5404 		 * If if it were, XSS would have to be checked against
5405 		 * the XSS exit bitmap in vmcs12.
5406 		 */
5407 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5408 	case EXIT_REASON_PREEMPTION_TIMER:
5409 		return false;
5410 	case EXIT_REASON_PML_FULL:
5411 		/* We emulate PML support to L1. */
5412 		return false;
5413 	case EXIT_REASON_VMFUNC:
5414 		/* VM functions are emulated through L2->L0 vmexits. */
5415 		return false;
5416 	case EXIT_REASON_ENCLS:
5417 		/* SGX is never exposed to L1 */
5418 		return false;
5419 	default:
5420 		return true;
5421 	}
5422 }
5423 
5424 
5425 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5426 				struct kvm_nested_state __user *user_kvm_nested_state,
5427 				u32 user_data_size)
5428 {
5429 	struct vcpu_vmx *vmx;
5430 	struct vmcs12 *vmcs12;
5431 	struct kvm_nested_state kvm_state = {
5432 		.flags = 0,
5433 		.format = KVM_STATE_NESTED_FORMAT_VMX,
5434 		.size = sizeof(kvm_state),
5435 		.hdr.vmx.vmxon_pa = -1ull,
5436 		.hdr.vmx.vmcs12_pa = -1ull,
5437 	};
5438 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5439 		&user_kvm_nested_state->data.vmx[0];
5440 
5441 	if (!vcpu)
5442 		return kvm_state.size + sizeof(*user_vmx_nested_state);
5443 
5444 	vmx = to_vmx(vcpu);
5445 	vmcs12 = get_vmcs12(vcpu);
5446 
5447 	if (nested_vmx_allowed(vcpu) &&
5448 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5449 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5450 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
5451 
5452 		if (vmx_has_valid_vmcs12(vcpu)) {
5453 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
5454 
5455 			if (vmx->nested.hv_evmcs)
5456 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5457 
5458 			if (is_guest_mode(vcpu) &&
5459 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5460 			    vmcs12->vmcs_link_pointer != -1ull)
5461 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
5462 		}
5463 
5464 		if (vmx->nested.smm.vmxon)
5465 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5466 
5467 		if (vmx->nested.smm.guest_mode)
5468 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5469 
5470 		if (is_guest_mode(vcpu)) {
5471 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5472 
5473 			if (vmx->nested.nested_run_pending)
5474 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5475 		}
5476 	}
5477 
5478 	if (user_data_size < kvm_state.size)
5479 		goto out;
5480 
5481 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5482 		return -EFAULT;
5483 
5484 	if (!vmx_has_valid_vmcs12(vcpu))
5485 		goto out;
5486 
5487 	/*
5488 	 * When running L2, the authoritative vmcs12 state is in the
5489 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5490 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5491 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
5492 	 * vmcs12 state is in the vmcs12 already.
5493 	 */
5494 	if (is_guest_mode(vcpu)) {
5495 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
5496 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5497 	} else if (!vmx->nested.need_vmcs12_to_shadow_sync) {
5498 		if (vmx->nested.hv_evmcs)
5499 			copy_enlightened_to_vmcs12(vmx);
5500 		else if (enable_shadow_vmcs)
5501 			copy_shadow_to_vmcs12(vmx);
5502 	}
5503 
5504 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
5505 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
5506 
5507 	/*
5508 	 * Copy over the full allocated size of vmcs12 rather than just the size
5509 	 * of the struct.
5510 	 */
5511 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
5512 		return -EFAULT;
5513 
5514 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5515 	    vmcs12->vmcs_link_pointer != -1ull) {
5516 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
5517 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
5518 			return -EFAULT;
5519 	}
5520 
5521 out:
5522 	return kvm_state.size;
5523 }
5524 
5525 /*
5526  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5527  */
5528 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5529 {
5530 	if (is_guest_mode(vcpu)) {
5531 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5532 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5533 	}
5534 	free_nested(vcpu);
5535 }
5536 
5537 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5538 				struct kvm_nested_state __user *user_kvm_nested_state,
5539 				struct kvm_nested_state *kvm_state)
5540 {
5541 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5542 	struct vmcs12 *vmcs12;
5543 	u32 exit_qual;
5544 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5545 		&user_kvm_nested_state->data.vmx[0];
5546 	int ret;
5547 
5548 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
5549 		return -EINVAL;
5550 
5551 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
5552 		if (kvm_state->hdr.vmx.smm.flags)
5553 			return -EINVAL;
5554 
5555 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
5556 			return -EINVAL;
5557 
5558 		/*
5559 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
5560 		 * enable eVMCS capability on vCPU. However, since then
5561 		 * code was changed such that flag signals vmcs12 should
5562 		 * be copied into eVMCS in guest memory.
5563 		 *
5564 		 * To preserve backwards compatability, allow user
5565 		 * to set this flag even when there is no VMXON region.
5566 		 */
5567 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
5568 			return -EINVAL;
5569 	} else {
5570 		if (!nested_vmx_allowed(vcpu))
5571 			return -EINVAL;
5572 
5573 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
5574 			return -EINVAL;
5575 	}
5576 
5577 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5578 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5579 		return -EINVAL;
5580 
5581 	if (kvm_state->hdr.vmx.smm.flags &
5582 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5583 		return -EINVAL;
5584 
5585 	/*
5586 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5587 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5588 	 * must be zero.
5589 	 */
5590 	if (is_smm(vcpu) ?
5591 		(kvm_state->flags &
5592 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
5593 		: kvm_state->hdr.vmx.smm.flags)
5594 		return -EINVAL;
5595 
5596 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5597 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5598 		return -EINVAL;
5599 
5600 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
5601 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
5602 			return -EINVAL;
5603 
5604 	vmx_leave_nested(vcpu);
5605 
5606 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
5607 		return 0;
5608 
5609 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
5610 	ret = enter_vmx_operation(vcpu);
5611 	if (ret)
5612 		return ret;
5613 
5614 	/* Empty 'VMXON' state is permitted */
5615 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12))
5616 		return 0;
5617 
5618 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
5619 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
5620 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
5621 			return -EINVAL;
5622 
5623 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
5624 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5625 		/*
5626 		 * Sync eVMCS upon entry as we may not have
5627 		 * HV_X64_MSR_VP_ASSIST_PAGE set up yet.
5628 		 */
5629 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5630 	} else {
5631 		return -EINVAL;
5632 	}
5633 
5634 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5635 		vmx->nested.smm.vmxon = true;
5636 		vmx->nested.vmxon = false;
5637 
5638 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5639 			vmx->nested.smm.guest_mode = true;
5640 	}
5641 
5642 	vmcs12 = get_vmcs12(vcpu);
5643 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
5644 		return -EFAULT;
5645 
5646 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5647 		return -EINVAL;
5648 
5649 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5650 		return 0;
5651 
5652 	vmx->nested.nested_run_pending =
5653 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5654 
5655 	ret = -EINVAL;
5656 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5657 	    vmcs12->vmcs_link_pointer != -1ull) {
5658 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5659 
5660 		if (kvm_state->size <
5661 		    sizeof(*kvm_state) +
5662 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
5663 			goto error_guest_mode;
5664 
5665 		if (copy_from_user(shadow_vmcs12,
5666 				   user_vmx_nested_state->shadow_vmcs12,
5667 				   sizeof(*shadow_vmcs12))) {
5668 			ret = -EFAULT;
5669 			goto error_guest_mode;
5670 		}
5671 
5672 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5673 		    !shadow_vmcs12->hdr.shadow_vmcs)
5674 			goto error_guest_mode;
5675 	}
5676 
5677 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
5678 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
5679 	    nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
5680 		goto error_guest_mode;
5681 
5682 	vmx->nested.dirty_vmcs12 = true;
5683 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5684 	if (ret)
5685 		goto error_guest_mode;
5686 
5687 	return 0;
5688 
5689 error_guest_mode:
5690 	vmx->nested.nested_run_pending = 0;
5691 	return ret;
5692 }
5693 
5694 void nested_vmx_vcpu_setup(void)
5695 {
5696 	if (enable_shadow_vmcs) {
5697 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5698 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5699 	}
5700 }
5701 
5702 /*
5703  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5704  * returned for the various VMX controls MSRs when nested VMX is enabled.
5705  * The same values should also be used to verify that vmcs12 control fields are
5706  * valid during nested entry from L1 to L2.
5707  * Each of these control msrs has a low and high 32-bit half: A low bit is on
5708  * if the corresponding bit in the (32-bit) control field *must* be on, and a
5709  * bit in the high half is on if the corresponding bit in the control field
5710  * may be on. See also vmx_control_verify().
5711  */
5712 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps,
5713 				bool apicv)
5714 {
5715 	/*
5716 	 * Note that as a general rule, the high half of the MSRs (bits in
5717 	 * the control fields which may be 1) should be initialized by the
5718 	 * intersection of the underlying hardware's MSR (i.e., features which
5719 	 * can be supported) and the list of features we want to expose -
5720 	 * because they are known to be properly supported in our code.
5721 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
5722 	 * be set to 0, meaning that L1 may turn off any of these bits. The
5723 	 * reason is that if one of these bits is necessary, it will appear
5724 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
5725 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
5726 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
5727 	 * These rules have exceptions below.
5728 	 */
5729 
5730 	/* pin-based controls */
5731 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
5732 		msrs->pinbased_ctls_low,
5733 		msrs->pinbased_ctls_high);
5734 	msrs->pinbased_ctls_low |=
5735 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5736 	msrs->pinbased_ctls_high &=
5737 		PIN_BASED_EXT_INTR_MASK |
5738 		PIN_BASED_NMI_EXITING |
5739 		PIN_BASED_VIRTUAL_NMIS |
5740 		(apicv ? PIN_BASED_POSTED_INTR : 0);
5741 	msrs->pinbased_ctls_high |=
5742 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5743 		PIN_BASED_VMX_PREEMPTION_TIMER;
5744 
5745 	/* exit controls */
5746 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
5747 		msrs->exit_ctls_low,
5748 		msrs->exit_ctls_high);
5749 	msrs->exit_ctls_low =
5750 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
5751 
5752 	msrs->exit_ctls_high &=
5753 #ifdef CONFIG_X86_64
5754 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
5755 #endif
5756 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
5757 	msrs->exit_ctls_high |=
5758 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
5759 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
5760 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
5761 
5762 	/* We support free control of debug control saving. */
5763 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
5764 
5765 	/* entry controls */
5766 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
5767 		msrs->entry_ctls_low,
5768 		msrs->entry_ctls_high);
5769 	msrs->entry_ctls_low =
5770 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
5771 	msrs->entry_ctls_high &=
5772 #ifdef CONFIG_X86_64
5773 		VM_ENTRY_IA32E_MODE |
5774 #endif
5775 		VM_ENTRY_LOAD_IA32_PAT;
5776 	msrs->entry_ctls_high |=
5777 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
5778 
5779 	/* We support free control of debug control loading. */
5780 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
5781 
5782 	/* cpu-based controls */
5783 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
5784 		msrs->procbased_ctls_low,
5785 		msrs->procbased_ctls_high);
5786 	msrs->procbased_ctls_low =
5787 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5788 	msrs->procbased_ctls_high &=
5789 		CPU_BASED_VIRTUAL_INTR_PENDING |
5790 		CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
5791 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
5792 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
5793 		CPU_BASED_CR3_STORE_EXITING |
5794 #ifdef CONFIG_X86_64
5795 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
5796 #endif
5797 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
5798 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
5799 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
5800 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
5801 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
5802 	/*
5803 	 * We can allow some features even when not supported by the
5804 	 * hardware. For example, L1 can specify an MSR bitmap - and we
5805 	 * can use it to avoid exits to L1 - even when L0 runs L2
5806 	 * without MSR bitmaps.
5807 	 */
5808 	msrs->procbased_ctls_high |=
5809 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5810 		CPU_BASED_USE_MSR_BITMAPS;
5811 
5812 	/* We support free control of CR3 access interception. */
5813 	msrs->procbased_ctls_low &=
5814 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
5815 
5816 	/*
5817 	 * secondary cpu-based controls.  Do not include those that
5818 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
5819 	 */
5820 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
5821 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
5822 		      msrs->secondary_ctls_low,
5823 		      msrs->secondary_ctls_high);
5824 
5825 	msrs->secondary_ctls_low = 0;
5826 	msrs->secondary_ctls_high &=
5827 		SECONDARY_EXEC_DESC |
5828 		SECONDARY_EXEC_RDTSCP |
5829 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
5830 		SECONDARY_EXEC_WBINVD_EXITING |
5831 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
5832 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
5833 		SECONDARY_EXEC_RDRAND_EXITING |
5834 		SECONDARY_EXEC_ENABLE_INVPCID |
5835 		SECONDARY_EXEC_RDSEED_EXITING |
5836 		SECONDARY_EXEC_XSAVES;
5837 
5838 	/*
5839 	 * We can emulate "VMCS shadowing," even if the hardware
5840 	 * doesn't support it.
5841 	 */
5842 	msrs->secondary_ctls_high |=
5843 		SECONDARY_EXEC_SHADOW_VMCS;
5844 
5845 	if (enable_ept) {
5846 		/* nested EPT: emulate EPT also to L1 */
5847 		msrs->secondary_ctls_high |=
5848 			SECONDARY_EXEC_ENABLE_EPT;
5849 		msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
5850 			 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
5851 		if (cpu_has_vmx_ept_execute_only())
5852 			msrs->ept_caps |=
5853 				VMX_EPT_EXECUTE_ONLY_BIT;
5854 		msrs->ept_caps &= ept_caps;
5855 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
5856 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
5857 			VMX_EPT_1GB_PAGE_BIT;
5858 		if (enable_ept_ad_bits) {
5859 			msrs->secondary_ctls_high |=
5860 				SECONDARY_EXEC_ENABLE_PML;
5861 			msrs->ept_caps |= VMX_EPT_AD_BIT;
5862 		}
5863 	}
5864 
5865 	if (cpu_has_vmx_vmfunc()) {
5866 		msrs->secondary_ctls_high |=
5867 			SECONDARY_EXEC_ENABLE_VMFUNC;
5868 		/*
5869 		 * Advertise EPTP switching unconditionally
5870 		 * since we emulate it
5871 		 */
5872 		if (enable_ept)
5873 			msrs->vmfunc_controls =
5874 				VMX_VMFUNC_EPTP_SWITCHING;
5875 	}
5876 
5877 	/*
5878 	 * Old versions of KVM use the single-context version without
5879 	 * checking for support, so declare that it is supported even
5880 	 * though it is treated as global context.  The alternative is
5881 	 * not failing the single-context invvpid, and it is worse.
5882 	 */
5883 	if (enable_vpid) {
5884 		msrs->secondary_ctls_high |=
5885 			SECONDARY_EXEC_ENABLE_VPID;
5886 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
5887 			VMX_VPID_EXTENT_SUPPORTED_MASK;
5888 	}
5889 
5890 	if (enable_unrestricted_guest)
5891 		msrs->secondary_ctls_high |=
5892 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
5893 
5894 	if (flexpriority_enabled)
5895 		msrs->secondary_ctls_high |=
5896 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5897 
5898 	/* miscellaneous data */
5899 	rdmsr(MSR_IA32_VMX_MISC,
5900 		msrs->misc_low,
5901 		msrs->misc_high);
5902 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
5903 	msrs->misc_low |=
5904 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
5905 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
5906 		VMX_MISC_ACTIVITY_HLT;
5907 	msrs->misc_high = 0;
5908 
5909 	/*
5910 	 * This MSR reports some information about VMX support. We
5911 	 * should return information about the VMX we emulate for the
5912 	 * guest, and the VMCS structure we give it - not about the
5913 	 * VMX support of the underlying hardware.
5914 	 */
5915 	msrs->basic =
5916 		VMCS12_REVISION |
5917 		VMX_BASIC_TRUE_CTLS |
5918 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
5919 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
5920 
5921 	if (cpu_has_vmx_basic_inout())
5922 		msrs->basic |= VMX_BASIC_INOUT;
5923 
5924 	/*
5925 	 * These MSRs specify bits which the guest must keep fixed on
5926 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
5927 	 * We picked the standard core2 setting.
5928 	 */
5929 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
5930 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
5931 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
5932 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
5933 
5934 	/* These MSRs specify bits which the guest must keep fixed off. */
5935 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
5936 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
5937 
5938 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
5939 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
5940 }
5941 
5942 void nested_vmx_hardware_unsetup(void)
5943 {
5944 	int i;
5945 
5946 	if (enable_shadow_vmcs) {
5947 		for (i = 0; i < VMX_BITMAP_NR; i++)
5948 			free_page((unsigned long)vmx_bitmap[i]);
5949 	}
5950 }
5951 
5952 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
5953 {
5954 	int i;
5955 
5956 	if (!cpu_has_vmx_shadow_vmcs())
5957 		enable_shadow_vmcs = 0;
5958 	if (enable_shadow_vmcs) {
5959 		for (i = 0; i < VMX_BITMAP_NR; i++) {
5960 			/*
5961 			 * The vmx_bitmap is not tied to a VM and so should
5962 			 * not be charged to a memcg.
5963 			 */
5964 			vmx_bitmap[i] = (unsigned long *)
5965 				__get_free_page(GFP_KERNEL);
5966 			if (!vmx_bitmap[i]) {
5967 				nested_vmx_hardware_unsetup();
5968 				return -ENOMEM;
5969 			}
5970 		}
5971 
5972 		init_vmcs_shadow_fields();
5973 	}
5974 
5975 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear,
5976 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch,
5977 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld,
5978 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst,
5979 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread,
5980 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume,
5981 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite,
5982 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff,
5983 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon,
5984 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept,
5985 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid,
5986 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc,
5987 
5988 	kvm_x86_ops->check_nested_events = vmx_check_nested_events;
5989 	kvm_x86_ops->get_nested_state = vmx_get_nested_state;
5990 	kvm_x86_ops->set_nested_state = vmx_set_nested_state;
5991 	kvm_x86_ops->get_vmcs12_pages = nested_get_vmcs12_pages,
5992 	kvm_x86_ops->nested_enable_evmcs = nested_enable_evmcs;
5993 	kvm_x86_ops->nested_get_evmcs_version = nested_get_evmcs_version;
5994 
5995 	return 0;
5996 }
5997