xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 78beef62)
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 #define CC(consistency_check)						\
23 ({									\
24 	bool failed = (consistency_check);				\
25 	if (failed)							\
26 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
27 	failed;								\
28 })
29 
30 /*
31  * Hyper-V requires all of these, so mark them as supported even though
32  * they are just treated the same as all-context.
33  */
34 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
35 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
36 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
37 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
38 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
39 
40 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
41 
42 enum {
43 	VMX_VMREAD_BITMAP,
44 	VMX_VMWRITE_BITMAP,
45 	VMX_BITMAP_NR
46 };
47 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
48 
49 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
50 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
51 
52 struct shadow_vmcs_field {
53 	u16	encoding;
54 	u16	offset;
55 };
56 static struct shadow_vmcs_field shadow_read_only_fields[] = {
57 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
58 #include "vmcs_shadow_fields.h"
59 };
60 static int max_shadow_read_only_fields =
61 	ARRAY_SIZE(shadow_read_only_fields);
62 
63 static struct shadow_vmcs_field shadow_read_write_fields[] = {
64 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
65 #include "vmcs_shadow_fields.h"
66 };
67 static int max_shadow_read_write_fields =
68 	ARRAY_SIZE(shadow_read_write_fields);
69 
70 static void init_vmcs_shadow_fields(void)
71 {
72 	int i, j;
73 
74 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
75 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
76 
77 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
78 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
79 		u16 field = entry.encoding;
80 
81 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
82 		    (i + 1 == max_shadow_read_only_fields ||
83 		     shadow_read_only_fields[i + 1].encoding != field + 1))
84 			pr_err("Missing field from shadow_read_only_field %x\n",
85 			       field + 1);
86 
87 		clear_bit(field, vmx_vmread_bitmap);
88 		if (field & 1)
89 #ifdef CONFIG_X86_64
90 			continue;
91 #else
92 			entry.offset += sizeof(u32);
93 #endif
94 		shadow_read_only_fields[j++] = entry;
95 	}
96 	max_shadow_read_only_fields = j;
97 
98 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
99 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
100 		u16 field = entry.encoding;
101 
102 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
103 		    (i + 1 == max_shadow_read_write_fields ||
104 		     shadow_read_write_fields[i + 1].encoding != field + 1))
105 			pr_err("Missing field from shadow_read_write_field %x\n",
106 			       field + 1);
107 
108 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
109 			  field <= GUEST_TR_AR_BYTES,
110 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
111 
112 		/*
113 		 * PML and the preemption timer can be emulated, but the
114 		 * processor cannot vmwrite to fields that don't exist
115 		 * on bare metal.
116 		 */
117 		switch (field) {
118 		case GUEST_PML_INDEX:
119 			if (!cpu_has_vmx_pml())
120 				continue;
121 			break;
122 		case VMX_PREEMPTION_TIMER_VALUE:
123 			if (!cpu_has_vmx_preemption_timer())
124 				continue;
125 			break;
126 		case GUEST_INTR_STATUS:
127 			if (!cpu_has_vmx_apicv())
128 				continue;
129 			break;
130 		default:
131 			break;
132 		}
133 
134 		clear_bit(field, vmx_vmwrite_bitmap);
135 		clear_bit(field, vmx_vmread_bitmap);
136 		if (field & 1)
137 #ifdef CONFIG_X86_64
138 			continue;
139 #else
140 			entry.offset += sizeof(u32);
141 #endif
142 		shadow_read_write_fields[j++] = entry;
143 	}
144 	max_shadow_read_write_fields = j;
145 }
146 
147 /*
148  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
149  * set the success or error code of an emulated VMX instruction (as specified
150  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
151  * instruction.
152  */
153 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
154 {
155 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
156 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
157 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
158 	return kvm_skip_emulated_instruction(vcpu);
159 }
160 
161 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
162 {
163 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
164 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
165 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
166 			| X86_EFLAGS_CF);
167 	return kvm_skip_emulated_instruction(vcpu);
168 }
169 
170 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
171 				u32 vm_instruction_error)
172 {
173 	struct vcpu_vmx *vmx = to_vmx(vcpu);
174 
175 	/*
176 	 * failValid writes the error number to the current VMCS, which
177 	 * can't be done if there isn't a current VMCS.
178 	 */
179 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
180 		return nested_vmx_failInvalid(vcpu);
181 
182 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
183 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
184 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
185 			| X86_EFLAGS_ZF);
186 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
187 	/*
188 	 * We don't need to force a shadow sync because
189 	 * VM_INSTRUCTION_ERROR is not shadowed
190 	 */
191 	return kvm_skip_emulated_instruction(vcpu);
192 }
193 
194 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
195 {
196 	/* TODO: not to reset guest simply here. */
197 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
198 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
199 }
200 
201 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
202 {
203 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
204 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
205 	vmx->nested.need_vmcs12_to_shadow_sync = false;
206 }
207 
208 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
209 {
210 	struct vcpu_vmx *vmx = to_vmx(vcpu);
211 
212 	if (!vmx->nested.hv_evmcs)
213 		return;
214 
215 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
216 	vmx->nested.hv_evmcs_vmptr = -1ull;
217 	vmx->nested.hv_evmcs = NULL;
218 }
219 
220 /*
221  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
222  * just stops using VMX.
223  */
224 static void free_nested(struct kvm_vcpu *vcpu)
225 {
226 	struct vcpu_vmx *vmx = to_vmx(vcpu);
227 
228 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
229 		return;
230 
231 	kvm_clear_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
232 
233 	vmx->nested.vmxon = false;
234 	vmx->nested.smm.vmxon = false;
235 	free_vpid(vmx->nested.vpid02);
236 	vmx->nested.posted_intr_nv = -1;
237 	vmx->nested.current_vmptr = -1ull;
238 	if (enable_shadow_vmcs) {
239 		vmx_disable_shadow_vmcs(vmx);
240 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
241 		free_vmcs(vmx->vmcs01.shadow_vmcs);
242 		vmx->vmcs01.shadow_vmcs = NULL;
243 	}
244 	kfree(vmx->nested.cached_vmcs12);
245 	vmx->nested.cached_vmcs12 = NULL;
246 	kfree(vmx->nested.cached_shadow_vmcs12);
247 	vmx->nested.cached_shadow_vmcs12 = NULL;
248 	/* Unpin physical memory we referred to in the vmcs02 */
249 	if (vmx->nested.apic_access_page) {
250 		kvm_release_page_dirty(vmx->nested.apic_access_page);
251 		vmx->nested.apic_access_page = NULL;
252 	}
253 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
254 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
255 	vmx->nested.pi_desc = NULL;
256 
257 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
258 
259 	nested_release_evmcs(vcpu);
260 
261 	free_loaded_vmcs(&vmx->nested.vmcs02);
262 }
263 
264 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
265 				     struct loaded_vmcs *prev)
266 {
267 	struct vmcs_host_state *dest, *src;
268 
269 	if (unlikely(!vmx->guest_state_loaded))
270 		return;
271 
272 	src = &prev->host_state;
273 	dest = &vmx->loaded_vmcs->host_state;
274 
275 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
276 	dest->ldt_sel = src->ldt_sel;
277 #ifdef CONFIG_X86_64
278 	dest->ds_sel = src->ds_sel;
279 	dest->es_sel = src->es_sel;
280 #endif
281 }
282 
283 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
284 {
285 	struct vcpu_vmx *vmx = to_vmx(vcpu);
286 	struct loaded_vmcs *prev;
287 	int cpu;
288 
289 	if (vmx->loaded_vmcs == vmcs)
290 		return;
291 
292 	cpu = get_cpu();
293 	prev = vmx->loaded_vmcs;
294 	vmx->loaded_vmcs = vmcs;
295 	vmx_vcpu_load_vmcs(vcpu, cpu);
296 	vmx_sync_vmcs_host_state(vmx, prev);
297 	put_cpu();
298 
299 	vmx_segment_cache_clear(vmx);
300 }
301 
302 /*
303  * Ensure that the current vmcs of the logical processor is the
304  * vmcs01 of the vcpu before calling free_nested().
305  */
306 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
307 {
308 	vcpu_load(vcpu);
309 	vmx_leave_nested(vcpu);
310 	vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
311 	free_nested(vcpu);
312 	vcpu_put(vcpu);
313 }
314 
315 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
316 		struct x86_exception *fault)
317 {
318 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
319 	struct vcpu_vmx *vmx = to_vmx(vcpu);
320 	u32 exit_reason;
321 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
322 
323 	if (vmx->nested.pml_full) {
324 		exit_reason = EXIT_REASON_PML_FULL;
325 		vmx->nested.pml_full = false;
326 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
327 	} else if (fault->error_code & PFERR_RSVD_MASK)
328 		exit_reason = EXIT_REASON_EPT_MISCONFIG;
329 	else
330 		exit_reason = EXIT_REASON_EPT_VIOLATION;
331 
332 	nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
333 	vmcs12->guest_physical_address = fault->address;
334 }
335 
336 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
337 {
338 	WARN_ON(mmu_is_nested(vcpu));
339 
340 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
341 	kvm_init_shadow_ept_mmu(vcpu,
342 			to_vmx(vcpu)->nested.msrs.ept_caps &
343 			VMX_EPT_EXECUTE_ONLY_BIT,
344 			nested_ept_ad_enabled(vcpu),
345 			nested_ept_get_cr3(vcpu));
346 	vcpu->arch.mmu->set_cr3           = vmx_set_cr3;
347 	vcpu->arch.mmu->get_cr3           = nested_ept_get_cr3;
348 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
349 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
350 
351 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
352 }
353 
354 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
355 {
356 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
357 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
358 }
359 
360 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
361 					    u16 error_code)
362 {
363 	bool inequality, bit;
364 
365 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
366 	inequality =
367 		(error_code & vmcs12->page_fault_error_code_mask) !=
368 		 vmcs12->page_fault_error_code_match;
369 	return inequality ^ bit;
370 }
371 
372 
373 /*
374  * KVM wants to inject page-faults which it got to the guest. This function
375  * checks whether in a nested guest, we need to inject them to L1 or L2.
376  */
377 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
378 {
379 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
380 	unsigned int nr = vcpu->arch.exception.nr;
381 	bool has_payload = vcpu->arch.exception.has_payload;
382 	unsigned long payload = vcpu->arch.exception.payload;
383 
384 	if (nr == PF_VECTOR) {
385 		if (vcpu->arch.exception.nested_apf) {
386 			*exit_qual = vcpu->arch.apf.nested_apf_token;
387 			return 1;
388 		}
389 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
390 						    vcpu->arch.exception.error_code)) {
391 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
392 			return 1;
393 		}
394 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
395 		if (nr == DB_VECTOR) {
396 			if (!has_payload) {
397 				payload = vcpu->arch.dr6;
398 				payload &= ~(DR6_FIXED_1 | DR6_BT);
399 				payload ^= DR6_RTM;
400 			}
401 			*exit_qual = payload;
402 		} else
403 			*exit_qual = 0;
404 		return 1;
405 	}
406 
407 	return 0;
408 }
409 
410 
411 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
412 		struct x86_exception *fault)
413 {
414 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
415 
416 	WARN_ON(!is_guest_mode(vcpu));
417 
418 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
419 		!to_vmx(vcpu)->nested.nested_run_pending) {
420 		vmcs12->vm_exit_intr_error_code = fault->error_code;
421 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
422 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
423 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
424 				  fault->address);
425 	} else {
426 		kvm_inject_page_fault(vcpu, fault);
427 	}
428 }
429 
430 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
431 {
432 	return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
433 }
434 
435 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
436 					       struct vmcs12 *vmcs12)
437 {
438 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
439 		return 0;
440 
441 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
442 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
443 		return -EINVAL;
444 
445 	return 0;
446 }
447 
448 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
449 						struct vmcs12 *vmcs12)
450 {
451 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
452 		return 0;
453 
454 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
455 		return -EINVAL;
456 
457 	return 0;
458 }
459 
460 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
461 						struct vmcs12 *vmcs12)
462 {
463 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
464 		return 0;
465 
466 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
467 		return -EINVAL;
468 
469 	return 0;
470 }
471 
472 /*
473  * Check if MSR is intercepted for L01 MSR bitmap.
474  */
475 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
476 {
477 	unsigned long *msr_bitmap;
478 	int f = sizeof(unsigned long);
479 
480 	if (!cpu_has_vmx_msr_bitmap())
481 		return true;
482 
483 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
484 
485 	if (msr <= 0x1fff) {
486 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
487 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
488 		msr &= 0x1fff;
489 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
490 	}
491 
492 	return true;
493 }
494 
495 /*
496  * If a msr is allowed by L0, we should check whether it is allowed by L1.
497  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
498  */
499 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
500 					       unsigned long *msr_bitmap_nested,
501 					       u32 msr, int type)
502 {
503 	int f = sizeof(unsigned long);
504 
505 	/*
506 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
507 	 * have the write-low and read-high bitmap offsets the wrong way round.
508 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
509 	 */
510 	if (msr <= 0x1fff) {
511 		if (type & MSR_TYPE_R &&
512 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
513 			/* read-low */
514 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
515 
516 		if (type & MSR_TYPE_W &&
517 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
518 			/* write-low */
519 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
520 
521 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
522 		msr &= 0x1fff;
523 		if (type & MSR_TYPE_R &&
524 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
525 			/* read-high */
526 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
527 
528 		if (type & MSR_TYPE_W &&
529 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
530 			/* write-high */
531 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
532 
533 	}
534 }
535 
536 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap) {
537 	int msr;
538 
539 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
540 		unsigned word = msr / BITS_PER_LONG;
541 
542 		msr_bitmap[word] = ~0;
543 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
544 	}
545 }
546 
547 /*
548  * Merge L0's and L1's MSR bitmap, return false to indicate that
549  * we do not use the hardware.
550  */
551 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
552 						 struct vmcs12 *vmcs12)
553 {
554 	int msr;
555 	unsigned long *msr_bitmap_l1;
556 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
557 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
558 
559 	/* Nothing to do if the MSR bitmap is not in use.  */
560 	if (!cpu_has_vmx_msr_bitmap() ||
561 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
562 		return false;
563 
564 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
565 		return false;
566 
567 	msr_bitmap_l1 = (unsigned long *)map->hva;
568 
569 	/*
570 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
571 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
572 	 * the x2APIC MSR range and selectively disable them below.
573 	 */
574 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
575 
576 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
577 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
578 			/*
579 			 * L0 need not intercept reads for MSRs between 0x800
580 			 * and 0x8ff, it just lets the processor take the value
581 			 * from the virtual-APIC page; take those 256 bits
582 			 * directly from the L1 bitmap.
583 			 */
584 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
585 				unsigned word = msr / BITS_PER_LONG;
586 
587 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
588 			}
589 		}
590 
591 		nested_vmx_disable_intercept_for_msr(
592 			msr_bitmap_l1, msr_bitmap_l0,
593 			X2APIC_MSR(APIC_TASKPRI),
594 			MSR_TYPE_R | MSR_TYPE_W);
595 
596 		if (nested_cpu_has_vid(vmcs12)) {
597 			nested_vmx_disable_intercept_for_msr(
598 				msr_bitmap_l1, msr_bitmap_l0,
599 				X2APIC_MSR(APIC_EOI),
600 				MSR_TYPE_W);
601 			nested_vmx_disable_intercept_for_msr(
602 				msr_bitmap_l1, msr_bitmap_l0,
603 				X2APIC_MSR(APIC_SELF_IPI),
604 				MSR_TYPE_W);
605 		}
606 	}
607 
608 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
609 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
610 					     MSR_FS_BASE, MSR_TYPE_RW);
611 
612 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
613 					     MSR_GS_BASE, MSR_TYPE_RW);
614 
615 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
616 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
617 
618 	/*
619 	 * Checking the L0->L1 bitmap is trying to verify two things:
620 	 *
621 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
622 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
623 	 *    from the L12 MSR bitmap that is too permissive.
624 	 * 2. That L1 or L2s have actually used the MSR. This avoids
625 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
626 	 *    works properly because we only update the L01 MSR bitmap lazily.
627 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
628 	 *    updated to reflect this when L1 (or its L2s) actually write to
629 	 *    the MSR.
630 	 */
631 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
632 		nested_vmx_disable_intercept_for_msr(
633 					msr_bitmap_l1, msr_bitmap_l0,
634 					MSR_IA32_SPEC_CTRL,
635 					MSR_TYPE_R | MSR_TYPE_W);
636 
637 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
638 		nested_vmx_disable_intercept_for_msr(
639 					msr_bitmap_l1, msr_bitmap_l0,
640 					MSR_IA32_PRED_CMD,
641 					MSR_TYPE_W);
642 
643 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
644 
645 	return true;
646 }
647 
648 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
649 				       struct vmcs12 *vmcs12)
650 {
651 	struct kvm_host_map map;
652 	struct vmcs12 *shadow;
653 
654 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
655 	    vmcs12->vmcs_link_pointer == -1ull)
656 		return;
657 
658 	shadow = get_shadow_vmcs12(vcpu);
659 
660 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
661 		return;
662 
663 	memcpy(shadow, map.hva, VMCS12_SIZE);
664 	kvm_vcpu_unmap(vcpu, &map, false);
665 }
666 
667 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
668 					      struct vmcs12 *vmcs12)
669 {
670 	struct vcpu_vmx *vmx = to_vmx(vcpu);
671 
672 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
673 	    vmcs12->vmcs_link_pointer == -1ull)
674 		return;
675 
676 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
677 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
678 }
679 
680 /*
681  * In nested virtualization, check if L1 has set
682  * VM_EXIT_ACK_INTR_ON_EXIT
683  */
684 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
685 {
686 	return get_vmcs12(vcpu)->vm_exit_controls &
687 		VM_EXIT_ACK_INTR_ON_EXIT;
688 }
689 
690 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
691 {
692 	return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
693 }
694 
695 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
696 					  struct vmcs12 *vmcs12)
697 {
698 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
699 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
700 		return -EINVAL;
701 	else
702 		return 0;
703 }
704 
705 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
706 					   struct vmcs12 *vmcs12)
707 {
708 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
709 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
710 	    !nested_cpu_has_vid(vmcs12) &&
711 	    !nested_cpu_has_posted_intr(vmcs12))
712 		return 0;
713 
714 	/*
715 	 * If virtualize x2apic mode is enabled,
716 	 * virtualize apic access must be disabled.
717 	 */
718 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
719 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
720 		return -EINVAL;
721 
722 	/*
723 	 * If virtual interrupt delivery is enabled,
724 	 * we must exit on external interrupts.
725 	 */
726 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
727 		return -EINVAL;
728 
729 	/*
730 	 * bits 15:8 should be zero in posted_intr_nv,
731 	 * the descriptor address has been already checked
732 	 * in nested_get_vmcs12_pages.
733 	 *
734 	 * bits 5:0 of posted_intr_desc_addr should be zero.
735 	 */
736 	if (nested_cpu_has_posted_intr(vmcs12) &&
737 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
738 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
739 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
740 	    CC((vmcs12->posted_intr_desc_addr & 0x3f)) ||
741 	    CC((vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu)))))
742 		return -EINVAL;
743 
744 	/* tpr shadow is needed by all apicv features. */
745 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
746 		return -EINVAL;
747 
748 	return 0;
749 }
750 
751 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
752 				       u32 count, u64 addr)
753 {
754 	int maxphyaddr;
755 
756 	if (count == 0)
757 		return 0;
758 	maxphyaddr = cpuid_maxphyaddr(vcpu);
759 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
760 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
761 		return -EINVAL;
762 
763 	return 0;
764 }
765 
766 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
767 						     struct vmcs12 *vmcs12)
768 {
769 	if (CC(nested_vmx_check_msr_switch(vcpu,
770 					   vmcs12->vm_exit_msr_load_count,
771 					   vmcs12->vm_exit_msr_load_addr)) ||
772 	    CC(nested_vmx_check_msr_switch(vcpu,
773 					   vmcs12->vm_exit_msr_store_count,
774 					   vmcs12->vm_exit_msr_store_addr)))
775 		return -EINVAL;
776 
777 	return 0;
778 }
779 
780 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
781                                                       struct vmcs12 *vmcs12)
782 {
783 	if (CC(nested_vmx_check_msr_switch(vcpu,
784 					   vmcs12->vm_entry_msr_load_count,
785 					   vmcs12->vm_entry_msr_load_addr)))
786                 return -EINVAL;
787 
788 	return 0;
789 }
790 
791 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
792 					 struct vmcs12 *vmcs12)
793 {
794 	if (!nested_cpu_has_pml(vmcs12))
795 		return 0;
796 
797 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
798 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
799 		return -EINVAL;
800 
801 	return 0;
802 }
803 
804 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
805 							struct vmcs12 *vmcs12)
806 {
807 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
808 	       !nested_cpu_has_ept(vmcs12)))
809 		return -EINVAL;
810 	return 0;
811 }
812 
813 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
814 							 struct vmcs12 *vmcs12)
815 {
816 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
817 	       !nested_cpu_has_ept(vmcs12)))
818 		return -EINVAL;
819 	return 0;
820 }
821 
822 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
823 						 struct vmcs12 *vmcs12)
824 {
825 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
826 		return 0;
827 
828 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
829 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
830 		return -EINVAL;
831 
832 	return 0;
833 }
834 
835 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
836 				       struct vmx_msr_entry *e)
837 {
838 	/* x2APIC MSR accesses are not allowed */
839 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
840 		return -EINVAL;
841 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
842 	    CC(e->index == MSR_IA32_UCODE_REV))
843 		return -EINVAL;
844 	if (CC(e->reserved != 0))
845 		return -EINVAL;
846 	return 0;
847 }
848 
849 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
850 				     struct vmx_msr_entry *e)
851 {
852 	if (CC(e->index == MSR_FS_BASE) ||
853 	    CC(e->index == MSR_GS_BASE) ||
854 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
855 	    nested_vmx_msr_check_common(vcpu, e))
856 		return -EINVAL;
857 	return 0;
858 }
859 
860 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
861 				      struct vmx_msr_entry *e)
862 {
863 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
864 	    nested_vmx_msr_check_common(vcpu, e))
865 		return -EINVAL;
866 	return 0;
867 }
868 
869 /*
870  * Load guest's/host's msr at nested entry/exit.
871  * return 0 for success, entry index for failure.
872  */
873 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
874 {
875 	u32 i;
876 	struct vmx_msr_entry e;
877 
878 	for (i = 0; i < count; i++) {
879 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
880 					&e, sizeof(e))) {
881 			pr_debug_ratelimited(
882 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
883 				__func__, i, gpa + i * sizeof(e));
884 			goto fail;
885 		}
886 		if (nested_vmx_load_msr_check(vcpu, &e)) {
887 			pr_debug_ratelimited(
888 				"%s check failed (%u, 0x%x, 0x%x)\n",
889 				__func__, i, e.index, e.reserved);
890 			goto fail;
891 		}
892 		if (kvm_set_msr(vcpu, e.index, e.value)) {
893 			pr_debug_ratelimited(
894 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
895 				__func__, i, e.index, e.value);
896 			goto fail;
897 		}
898 	}
899 	return 0;
900 fail:
901 	return i + 1;
902 }
903 
904 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
905 {
906 	u64 data;
907 	u32 i;
908 	struct vmx_msr_entry e;
909 
910 	for (i = 0; i < count; i++) {
911 		if (kvm_vcpu_read_guest(vcpu,
912 					gpa + i * sizeof(e),
913 					&e, 2 * sizeof(u32))) {
914 			pr_debug_ratelimited(
915 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
916 				__func__, i, gpa + i * sizeof(e));
917 			return -EINVAL;
918 		}
919 		if (nested_vmx_store_msr_check(vcpu, &e)) {
920 			pr_debug_ratelimited(
921 				"%s check failed (%u, 0x%x, 0x%x)\n",
922 				__func__, i, e.index, e.reserved);
923 			return -EINVAL;
924 		}
925 		if (kvm_get_msr(vcpu, e.index, &data)) {
926 			pr_debug_ratelimited(
927 				"%s cannot read MSR (%u, 0x%x)\n",
928 				__func__, i, e.index);
929 			return -EINVAL;
930 		}
931 		if (kvm_vcpu_write_guest(vcpu,
932 					 gpa + i * sizeof(e) +
933 					     offsetof(struct vmx_msr_entry, value),
934 					 &data, sizeof(data))) {
935 			pr_debug_ratelimited(
936 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
937 				__func__, i, e.index, data);
938 			return -EINVAL;
939 		}
940 	}
941 	return 0;
942 }
943 
944 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
945 {
946 	unsigned long invalid_mask;
947 
948 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
949 	return (val & invalid_mask) == 0;
950 }
951 
952 /*
953  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
954  * emulating VM entry into a guest with EPT enabled.
955  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
956  * is assigned to entry_failure_code on failure.
957  */
958 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
959 			       u32 *entry_failure_code)
960 {
961 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
962 		if (CC(!nested_cr3_valid(vcpu, cr3))) {
963 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
964 			return -EINVAL;
965 		}
966 
967 		/*
968 		 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
969 		 * must not be dereferenced.
970 		 */
971 		if (is_pae_paging(vcpu) && !nested_ept) {
972 			if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
973 				*entry_failure_code = ENTRY_FAIL_PDPTE;
974 				return -EINVAL;
975 			}
976 		}
977 	}
978 
979 	if (!nested_ept)
980 		kvm_mmu_new_cr3(vcpu, cr3, false);
981 
982 	vcpu->arch.cr3 = cr3;
983 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
984 
985 	kvm_init_mmu(vcpu, false);
986 
987 	return 0;
988 }
989 
990 /*
991  * Returns if KVM is able to config CPU to tag TLB entries
992  * populated by L2 differently than TLB entries populated
993  * by L1.
994  *
995  * If L1 uses EPT, then TLB entries are tagged with different EPTP.
996  *
997  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
998  * with different VPID (L1 entries are tagged with vmx->vpid
999  * while L2 entries are tagged with vmx->nested.vpid02).
1000  */
1001 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1002 {
1003 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1004 
1005 	return nested_cpu_has_ept(vmcs12) ||
1006 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1007 }
1008 
1009 static u16 nested_get_vpid02(struct kvm_vcpu *vcpu)
1010 {
1011 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1012 
1013 	return vmx->nested.vpid02 ? vmx->nested.vpid02 : vmx->vpid;
1014 }
1015 
1016 
1017 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1018 {
1019 	return fixed_bits_valid(control, low, high);
1020 }
1021 
1022 static inline u64 vmx_control_msr(u32 low, u32 high)
1023 {
1024 	return low | ((u64)high << 32);
1025 }
1026 
1027 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1028 {
1029 	superset &= mask;
1030 	subset &= mask;
1031 
1032 	return (superset | subset) == superset;
1033 }
1034 
1035 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1036 {
1037 	const u64 feature_and_reserved =
1038 		/* feature (except bit 48; see below) */
1039 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1040 		/* reserved */
1041 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1042 	u64 vmx_basic = vmx->nested.msrs.basic;
1043 
1044 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1045 		return -EINVAL;
1046 
1047 	/*
1048 	 * KVM does not emulate a version of VMX that constrains physical
1049 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1050 	 */
1051 	if (data & BIT_ULL(48))
1052 		return -EINVAL;
1053 
1054 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1055 	    vmx_basic_vmcs_revision_id(data))
1056 		return -EINVAL;
1057 
1058 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1059 		return -EINVAL;
1060 
1061 	vmx->nested.msrs.basic = data;
1062 	return 0;
1063 }
1064 
1065 static int
1066 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1067 {
1068 	u64 supported;
1069 	u32 *lowp, *highp;
1070 
1071 	switch (msr_index) {
1072 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1073 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1074 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1075 		break;
1076 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1077 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1078 		highp = &vmx->nested.msrs.procbased_ctls_high;
1079 		break;
1080 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1081 		lowp = &vmx->nested.msrs.exit_ctls_low;
1082 		highp = &vmx->nested.msrs.exit_ctls_high;
1083 		break;
1084 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1085 		lowp = &vmx->nested.msrs.entry_ctls_low;
1086 		highp = &vmx->nested.msrs.entry_ctls_high;
1087 		break;
1088 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1089 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1090 		highp = &vmx->nested.msrs.secondary_ctls_high;
1091 		break;
1092 	default:
1093 		BUG();
1094 	}
1095 
1096 	supported = vmx_control_msr(*lowp, *highp);
1097 
1098 	/* Check must-be-1 bits are still 1. */
1099 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1100 		return -EINVAL;
1101 
1102 	/* Check must-be-0 bits are still 0. */
1103 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1104 		return -EINVAL;
1105 
1106 	*lowp = data;
1107 	*highp = data >> 32;
1108 	return 0;
1109 }
1110 
1111 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1112 {
1113 	const u64 feature_and_reserved_bits =
1114 		/* feature */
1115 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1116 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1117 		/* reserved */
1118 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1119 	u64 vmx_misc;
1120 
1121 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1122 				   vmx->nested.msrs.misc_high);
1123 
1124 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1125 		return -EINVAL;
1126 
1127 	if ((vmx->nested.msrs.pinbased_ctls_high &
1128 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1129 	    vmx_misc_preemption_timer_rate(data) !=
1130 	    vmx_misc_preemption_timer_rate(vmx_misc))
1131 		return -EINVAL;
1132 
1133 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1134 		return -EINVAL;
1135 
1136 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1137 		return -EINVAL;
1138 
1139 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1140 		return -EINVAL;
1141 
1142 	vmx->nested.msrs.misc_low = data;
1143 	vmx->nested.msrs.misc_high = data >> 32;
1144 
1145 	return 0;
1146 }
1147 
1148 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1149 {
1150 	u64 vmx_ept_vpid_cap;
1151 
1152 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1153 					   vmx->nested.msrs.vpid_caps);
1154 
1155 	/* Every bit is either reserved or a feature bit. */
1156 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1157 		return -EINVAL;
1158 
1159 	vmx->nested.msrs.ept_caps = data;
1160 	vmx->nested.msrs.vpid_caps = data >> 32;
1161 	return 0;
1162 }
1163 
1164 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1165 {
1166 	u64 *msr;
1167 
1168 	switch (msr_index) {
1169 	case MSR_IA32_VMX_CR0_FIXED0:
1170 		msr = &vmx->nested.msrs.cr0_fixed0;
1171 		break;
1172 	case MSR_IA32_VMX_CR4_FIXED0:
1173 		msr = &vmx->nested.msrs.cr4_fixed0;
1174 		break;
1175 	default:
1176 		BUG();
1177 	}
1178 
1179 	/*
1180 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1181 	 * must be 1 in the restored value.
1182 	 */
1183 	if (!is_bitwise_subset(data, *msr, -1ULL))
1184 		return -EINVAL;
1185 
1186 	*msr = data;
1187 	return 0;
1188 }
1189 
1190 /*
1191  * Called when userspace is restoring VMX MSRs.
1192  *
1193  * Returns 0 on success, non-0 otherwise.
1194  */
1195 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1196 {
1197 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1198 
1199 	/*
1200 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1201 	 * is in VMX operation.
1202 	 */
1203 	if (vmx->nested.vmxon)
1204 		return -EBUSY;
1205 
1206 	switch (msr_index) {
1207 	case MSR_IA32_VMX_BASIC:
1208 		return vmx_restore_vmx_basic(vmx, data);
1209 	case MSR_IA32_VMX_PINBASED_CTLS:
1210 	case MSR_IA32_VMX_PROCBASED_CTLS:
1211 	case MSR_IA32_VMX_EXIT_CTLS:
1212 	case MSR_IA32_VMX_ENTRY_CTLS:
1213 		/*
1214 		 * The "non-true" VMX capability MSRs are generated from the
1215 		 * "true" MSRs, so we do not support restoring them directly.
1216 		 *
1217 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1218 		 * should restore the "true" MSRs with the must-be-1 bits
1219 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1220 		 * DEFAULT SETTINGS".
1221 		 */
1222 		return -EINVAL;
1223 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1224 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1225 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1226 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1227 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1228 		return vmx_restore_control_msr(vmx, msr_index, data);
1229 	case MSR_IA32_VMX_MISC:
1230 		return vmx_restore_vmx_misc(vmx, data);
1231 	case MSR_IA32_VMX_CR0_FIXED0:
1232 	case MSR_IA32_VMX_CR4_FIXED0:
1233 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1234 	case MSR_IA32_VMX_CR0_FIXED1:
1235 	case MSR_IA32_VMX_CR4_FIXED1:
1236 		/*
1237 		 * These MSRs are generated based on the vCPU's CPUID, so we
1238 		 * do not support restoring them directly.
1239 		 */
1240 		return -EINVAL;
1241 	case MSR_IA32_VMX_EPT_VPID_CAP:
1242 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1243 	case MSR_IA32_VMX_VMCS_ENUM:
1244 		vmx->nested.msrs.vmcs_enum = data;
1245 		return 0;
1246 	case MSR_IA32_VMX_VMFUNC:
1247 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1248 			return -EINVAL;
1249 		vmx->nested.msrs.vmfunc_controls = data;
1250 		return 0;
1251 	default:
1252 		/*
1253 		 * The rest of the VMX capability MSRs do not support restore.
1254 		 */
1255 		return -EINVAL;
1256 	}
1257 }
1258 
1259 /* Returns 0 on success, non-0 otherwise. */
1260 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1261 {
1262 	switch (msr_index) {
1263 	case MSR_IA32_VMX_BASIC:
1264 		*pdata = msrs->basic;
1265 		break;
1266 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1267 	case MSR_IA32_VMX_PINBASED_CTLS:
1268 		*pdata = vmx_control_msr(
1269 			msrs->pinbased_ctls_low,
1270 			msrs->pinbased_ctls_high);
1271 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1272 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1273 		break;
1274 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1275 	case MSR_IA32_VMX_PROCBASED_CTLS:
1276 		*pdata = vmx_control_msr(
1277 			msrs->procbased_ctls_low,
1278 			msrs->procbased_ctls_high);
1279 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1280 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1281 		break;
1282 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1283 	case MSR_IA32_VMX_EXIT_CTLS:
1284 		*pdata = vmx_control_msr(
1285 			msrs->exit_ctls_low,
1286 			msrs->exit_ctls_high);
1287 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1288 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1289 		break;
1290 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1291 	case MSR_IA32_VMX_ENTRY_CTLS:
1292 		*pdata = vmx_control_msr(
1293 			msrs->entry_ctls_low,
1294 			msrs->entry_ctls_high);
1295 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1296 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1297 		break;
1298 	case MSR_IA32_VMX_MISC:
1299 		*pdata = vmx_control_msr(
1300 			msrs->misc_low,
1301 			msrs->misc_high);
1302 		break;
1303 	case MSR_IA32_VMX_CR0_FIXED0:
1304 		*pdata = msrs->cr0_fixed0;
1305 		break;
1306 	case MSR_IA32_VMX_CR0_FIXED1:
1307 		*pdata = msrs->cr0_fixed1;
1308 		break;
1309 	case MSR_IA32_VMX_CR4_FIXED0:
1310 		*pdata = msrs->cr4_fixed0;
1311 		break;
1312 	case MSR_IA32_VMX_CR4_FIXED1:
1313 		*pdata = msrs->cr4_fixed1;
1314 		break;
1315 	case MSR_IA32_VMX_VMCS_ENUM:
1316 		*pdata = msrs->vmcs_enum;
1317 		break;
1318 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1319 		*pdata = vmx_control_msr(
1320 			msrs->secondary_ctls_low,
1321 			msrs->secondary_ctls_high);
1322 		break;
1323 	case MSR_IA32_VMX_EPT_VPID_CAP:
1324 		*pdata = msrs->ept_caps |
1325 			((u64)msrs->vpid_caps << 32);
1326 		break;
1327 	case MSR_IA32_VMX_VMFUNC:
1328 		*pdata = msrs->vmfunc_controls;
1329 		break;
1330 	default:
1331 		return 1;
1332 	}
1333 
1334 	return 0;
1335 }
1336 
1337 /*
1338  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1339  * been modified by the L1 guest.  Note, "writable" in this context means
1340  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1341  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1342  * VM-exit information fields (which are actually writable if the vCPU is
1343  * configured to support "VMWRITE to any supported field in the VMCS").
1344  */
1345 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1346 {
1347 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1348 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1349 	struct shadow_vmcs_field field;
1350 	unsigned long val;
1351 	int i;
1352 
1353 	if (WARN_ON(!shadow_vmcs))
1354 		return;
1355 
1356 	preempt_disable();
1357 
1358 	vmcs_load(shadow_vmcs);
1359 
1360 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1361 		field = shadow_read_write_fields[i];
1362 		val = __vmcs_readl(field.encoding);
1363 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1364 	}
1365 
1366 	vmcs_clear(shadow_vmcs);
1367 	vmcs_load(vmx->loaded_vmcs->vmcs);
1368 
1369 	preempt_enable();
1370 }
1371 
1372 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1373 {
1374 	const struct shadow_vmcs_field *fields[] = {
1375 		shadow_read_write_fields,
1376 		shadow_read_only_fields
1377 	};
1378 	const int max_fields[] = {
1379 		max_shadow_read_write_fields,
1380 		max_shadow_read_only_fields
1381 	};
1382 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1383 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1384 	struct shadow_vmcs_field field;
1385 	unsigned long val;
1386 	int i, q;
1387 
1388 	if (WARN_ON(!shadow_vmcs))
1389 		return;
1390 
1391 	vmcs_load(shadow_vmcs);
1392 
1393 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1394 		for (i = 0; i < max_fields[q]; i++) {
1395 			field = fields[q][i];
1396 			val = vmcs12_read_any(vmcs12, field.encoding,
1397 					      field.offset);
1398 			__vmcs_writel(field.encoding, val);
1399 		}
1400 	}
1401 
1402 	vmcs_clear(shadow_vmcs);
1403 	vmcs_load(vmx->loaded_vmcs->vmcs);
1404 }
1405 
1406 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1407 {
1408 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1409 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1410 
1411 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1412 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1413 	vmcs12->guest_rip = evmcs->guest_rip;
1414 
1415 	if (unlikely(!(evmcs->hv_clean_fields &
1416 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1417 		vmcs12->guest_rsp = evmcs->guest_rsp;
1418 		vmcs12->guest_rflags = evmcs->guest_rflags;
1419 		vmcs12->guest_interruptibility_info =
1420 			evmcs->guest_interruptibility_info;
1421 	}
1422 
1423 	if (unlikely(!(evmcs->hv_clean_fields &
1424 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1425 		vmcs12->cpu_based_vm_exec_control =
1426 			evmcs->cpu_based_vm_exec_control;
1427 	}
1428 
1429 	if (unlikely(!(evmcs->hv_clean_fields &
1430 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1431 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1432 	}
1433 
1434 	if (unlikely(!(evmcs->hv_clean_fields &
1435 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1436 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1437 	}
1438 
1439 	if (unlikely(!(evmcs->hv_clean_fields &
1440 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1441 		vmcs12->vm_entry_intr_info_field =
1442 			evmcs->vm_entry_intr_info_field;
1443 		vmcs12->vm_entry_exception_error_code =
1444 			evmcs->vm_entry_exception_error_code;
1445 		vmcs12->vm_entry_instruction_len =
1446 			evmcs->vm_entry_instruction_len;
1447 	}
1448 
1449 	if (unlikely(!(evmcs->hv_clean_fields &
1450 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1451 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1452 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1453 		vmcs12->host_cr0 = evmcs->host_cr0;
1454 		vmcs12->host_cr3 = evmcs->host_cr3;
1455 		vmcs12->host_cr4 = evmcs->host_cr4;
1456 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1457 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1458 		vmcs12->host_rip = evmcs->host_rip;
1459 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1460 		vmcs12->host_es_selector = evmcs->host_es_selector;
1461 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1462 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1463 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1464 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1465 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1466 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1467 	}
1468 
1469 	if (unlikely(!(evmcs->hv_clean_fields &
1470 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1471 		vmcs12->pin_based_vm_exec_control =
1472 			evmcs->pin_based_vm_exec_control;
1473 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1474 		vmcs12->secondary_vm_exec_control =
1475 			evmcs->secondary_vm_exec_control;
1476 	}
1477 
1478 	if (unlikely(!(evmcs->hv_clean_fields &
1479 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1480 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1481 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1482 	}
1483 
1484 	if (unlikely(!(evmcs->hv_clean_fields &
1485 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1486 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1487 	}
1488 
1489 	if (unlikely(!(evmcs->hv_clean_fields &
1490 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1491 		vmcs12->guest_es_base = evmcs->guest_es_base;
1492 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1493 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1494 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1495 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1496 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1497 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1498 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1499 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1500 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1501 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1502 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1503 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1504 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1505 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1506 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1507 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1508 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1509 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1510 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1511 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1512 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1513 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1514 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1515 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1516 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1517 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1518 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1519 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1520 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1521 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1522 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1523 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1524 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1525 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1526 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1527 	}
1528 
1529 	if (unlikely(!(evmcs->hv_clean_fields &
1530 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1531 		vmcs12->tsc_offset = evmcs->tsc_offset;
1532 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1533 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1534 	}
1535 
1536 	if (unlikely(!(evmcs->hv_clean_fields &
1537 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1538 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1539 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1540 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1541 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1542 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1543 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1544 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1545 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1546 	}
1547 
1548 	if (unlikely(!(evmcs->hv_clean_fields &
1549 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1550 		vmcs12->host_fs_base = evmcs->host_fs_base;
1551 		vmcs12->host_gs_base = evmcs->host_gs_base;
1552 		vmcs12->host_tr_base = evmcs->host_tr_base;
1553 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1554 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1555 		vmcs12->host_rsp = evmcs->host_rsp;
1556 	}
1557 
1558 	if (unlikely(!(evmcs->hv_clean_fields &
1559 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1560 		vmcs12->ept_pointer = evmcs->ept_pointer;
1561 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1562 	}
1563 
1564 	if (unlikely(!(evmcs->hv_clean_fields &
1565 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1566 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1567 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1568 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1569 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1570 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1571 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1572 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1573 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1574 		vmcs12->guest_pending_dbg_exceptions =
1575 			evmcs->guest_pending_dbg_exceptions;
1576 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1577 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1578 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1579 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1580 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1581 	}
1582 
1583 	/*
1584 	 * Not used?
1585 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1586 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1587 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1588 	 * vmcs12->cr3_target_value0 = evmcs->cr3_target_value0;
1589 	 * vmcs12->cr3_target_value1 = evmcs->cr3_target_value1;
1590 	 * vmcs12->cr3_target_value2 = evmcs->cr3_target_value2;
1591 	 * vmcs12->cr3_target_value3 = evmcs->cr3_target_value3;
1592 	 * vmcs12->page_fault_error_code_mask =
1593 	 *		evmcs->page_fault_error_code_mask;
1594 	 * vmcs12->page_fault_error_code_match =
1595 	 *		evmcs->page_fault_error_code_match;
1596 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1597 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1598 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1599 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1600 	 */
1601 
1602 	/*
1603 	 * Read only fields:
1604 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1605 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1606 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1607 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1608 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1609 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1610 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1611 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1612 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1613 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1614 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1615 	 *
1616 	 * Not present in struct vmcs12:
1617 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1618 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1619 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1620 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1621 	 */
1622 
1623 	return 0;
1624 }
1625 
1626 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1627 {
1628 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1629 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1630 
1631 	/*
1632 	 * Should not be changed by KVM:
1633 	 *
1634 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1635 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1636 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1637 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1638 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1639 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1640 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1641 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1642 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1643 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1644 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1645 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1646 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1647 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1648 	 * evmcs->host_rip = vmcs12->host_rip;
1649 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1650 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1651 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1652 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1653 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1654 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1655 	 * evmcs->host_rsp = vmcs12->host_rsp;
1656 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1657 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1658 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1659 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1660 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1661 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1662 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1663 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1664 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1665 	 * evmcs->cr3_target_value0 = vmcs12->cr3_target_value0;
1666 	 * evmcs->cr3_target_value1 = vmcs12->cr3_target_value1;
1667 	 * evmcs->cr3_target_value2 = vmcs12->cr3_target_value2;
1668 	 * evmcs->cr3_target_value3 = vmcs12->cr3_target_value3;
1669 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1670 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1671 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1672 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1673 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1674 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1675 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1676 	 * evmcs->page_fault_error_code_mask =
1677 	 *		vmcs12->page_fault_error_code_mask;
1678 	 * evmcs->page_fault_error_code_match =
1679 	 *		vmcs12->page_fault_error_code_match;
1680 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1681 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1682 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1683 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1684 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1685 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1686 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1687 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1688 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1689 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1690 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1691 	 *
1692 	 * Not present in struct vmcs12:
1693 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1694 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1695 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1696 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1697 	 */
1698 
1699 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1700 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1701 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1702 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1703 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1704 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1705 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1706 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1707 
1708 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1709 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1710 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1711 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1712 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1713 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1714 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1715 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1716 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1717 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1718 
1719 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1720 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1721 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1722 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1723 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1724 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1725 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1726 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1727 
1728 	evmcs->guest_es_base = vmcs12->guest_es_base;
1729 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1730 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1731 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1732 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1733 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1734 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1735 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1736 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1737 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1738 
1739 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1740 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1741 
1742 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1743 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1744 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1745 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1746 
1747 	evmcs->guest_pending_dbg_exceptions =
1748 		vmcs12->guest_pending_dbg_exceptions;
1749 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1750 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1751 
1752 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1753 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1754 
1755 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1756 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1757 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1758 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1759 
1760 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1761 
1762 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1763 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1764 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1765 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1766 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1767 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1768 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1769 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1770 
1771 	evmcs->exit_qualification = vmcs12->exit_qualification;
1772 
1773 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1774 	evmcs->guest_rsp = vmcs12->guest_rsp;
1775 	evmcs->guest_rflags = vmcs12->guest_rflags;
1776 
1777 	evmcs->guest_interruptibility_info =
1778 		vmcs12->guest_interruptibility_info;
1779 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1780 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1781 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1782 	evmcs->vm_entry_exception_error_code =
1783 		vmcs12->vm_entry_exception_error_code;
1784 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1785 
1786 	evmcs->guest_rip = vmcs12->guest_rip;
1787 
1788 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1789 
1790 	return 0;
1791 }
1792 
1793 /*
1794  * This is an equivalent of the nested hypervisor executing the vmptrld
1795  * instruction.
1796  */
1797 static int nested_vmx_handle_enlightened_vmptrld(struct kvm_vcpu *vcpu,
1798 						 bool from_launch)
1799 {
1800 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1801 	bool evmcs_gpa_changed = false;
1802 	u64 evmcs_gpa;
1803 
1804 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1805 		return 1;
1806 
1807 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa))
1808 		return 1;
1809 
1810 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1811 		if (!vmx->nested.hv_evmcs)
1812 			vmx->nested.current_vmptr = -1ull;
1813 
1814 		nested_release_evmcs(vcpu);
1815 
1816 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1817 				 &vmx->nested.hv_evmcs_map))
1818 			return 0;
1819 
1820 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1821 
1822 		/*
1823 		 * Currently, KVM only supports eVMCS version 1
1824 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1825 		 * value to first u32 field of eVMCS which should specify eVMCS
1826 		 * VersionNumber.
1827 		 *
1828 		 * Guest should be aware of supported eVMCS versions by host by
1829 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1830 		 * expected to set this CPUID leaf according to the value
1831 		 * returned in vmcs_version from nested_enable_evmcs().
1832 		 *
1833 		 * However, it turns out that Microsoft Hyper-V fails to comply
1834 		 * to their own invented interface: When Hyper-V use eVMCS, it
1835 		 * just sets first u32 field of eVMCS to revision_id specified
1836 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1837 		 * which is one of the supported versions specified in
1838 		 * CPUID.0x4000000A.EAX[0:15].
1839 		 *
1840 		 * To overcome Hyper-V bug, we accept here either a supported
1841 		 * eVMCS version or VMCS12 revision_id as valid values for first
1842 		 * u32 field of eVMCS.
1843 		 */
1844 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1845 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1846 			nested_release_evmcs(vcpu);
1847 			return 0;
1848 		}
1849 
1850 		vmx->nested.dirty_vmcs12 = true;
1851 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
1852 
1853 		evmcs_gpa_changed = true;
1854 		/*
1855 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
1856 		 * reloaded from guest's memory (read only fields, fields not
1857 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
1858 		 * are no leftovers.
1859 		 */
1860 		if (from_launch) {
1861 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1862 			memset(vmcs12, 0, sizeof(*vmcs12));
1863 			vmcs12->hdr.revision_id = VMCS12_REVISION;
1864 		}
1865 
1866 	}
1867 
1868 	/*
1869 	 * Clean fields data can't de used on VMLAUNCH and when we switch
1870 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
1871 	 */
1872 	if (from_launch || evmcs_gpa_changed)
1873 		vmx->nested.hv_evmcs->hv_clean_fields &=
1874 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1875 
1876 	return 1;
1877 }
1878 
1879 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
1880 {
1881 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1882 
1883 	/*
1884 	 * hv_evmcs may end up being not mapped after migration (when
1885 	 * L2 was running), map it here to make sure vmcs12 changes are
1886 	 * properly reflected.
1887 	 */
1888 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs)
1889 		nested_vmx_handle_enlightened_vmptrld(vcpu, false);
1890 
1891 	if (vmx->nested.hv_evmcs) {
1892 		copy_vmcs12_to_enlightened(vmx);
1893 		/* All fields are clean */
1894 		vmx->nested.hv_evmcs->hv_clean_fields |=
1895 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1896 	} else {
1897 		copy_vmcs12_to_shadow(vmx);
1898 	}
1899 
1900 	vmx->nested.need_vmcs12_to_shadow_sync = false;
1901 }
1902 
1903 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
1904 {
1905 	struct vcpu_vmx *vmx =
1906 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
1907 
1908 	vmx->nested.preemption_timer_expired = true;
1909 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
1910 	kvm_vcpu_kick(&vmx->vcpu);
1911 
1912 	return HRTIMER_NORESTART;
1913 }
1914 
1915 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
1916 {
1917 	u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
1918 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1919 
1920 	/*
1921 	 * A timer value of zero is architecturally guaranteed to cause
1922 	 * a VMExit prior to executing any instructions in the guest.
1923 	 */
1924 	if (preemption_timeout == 0) {
1925 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
1926 		return;
1927 	}
1928 
1929 	if (vcpu->arch.virtual_tsc_khz == 0)
1930 		return;
1931 
1932 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
1933 	preemption_timeout *= 1000000;
1934 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
1935 	hrtimer_start(&vmx->nested.preemption_timer,
1936 		      ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
1937 }
1938 
1939 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1940 {
1941 	if (vmx->nested.nested_run_pending &&
1942 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
1943 		return vmcs12->guest_ia32_efer;
1944 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
1945 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
1946 	else
1947 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
1948 }
1949 
1950 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
1951 {
1952 	/*
1953 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
1954 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
1955 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
1956 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
1957 	 */
1958 	if (vmx->nested.vmcs02_initialized)
1959 		return;
1960 	vmx->nested.vmcs02_initialized = true;
1961 
1962 	/*
1963 	 * We don't care what the EPTP value is we just need to guarantee
1964 	 * it's valid so we don't get a false positive when doing early
1965 	 * consistency checks.
1966 	 */
1967 	if (enable_ept && nested_early_check)
1968 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
1969 
1970 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
1971 	if (cpu_has_vmx_vmfunc())
1972 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
1973 
1974 	if (cpu_has_vmx_posted_intr())
1975 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
1976 
1977 	if (cpu_has_vmx_msr_bitmap())
1978 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
1979 
1980 	/*
1981 	 * The PML address never changes, so it is constant in vmcs02.
1982 	 * Conceptually we want to copy the PML index from vmcs01 here,
1983 	 * and then back to vmcs01 on nested vmexit.  But since we flush
1984 	 * the log and reset GUEST_PML_INDEX on each vmexit, the PML
1985 	 * index is also effectively constant in vmcs02.
1986 	 */
1987 	if (enable_pml) {
1988 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
1989 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
1990 	}
1991 
1992 	if (cpu_has_vmx_encls_vmexit())
1993 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
1994 
1995 	/*
1996 	 * Set the MSR load/store lists to match L0's settings.  Only the
1997 	 * addresses are constant (for vmcs02), the counts can change based
1998 	 * on L2's behavior, e.g. switching to/from long mode.
1999 	 */
2000 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2001 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2002 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2003 
2004 	vmx_set_constant_host_state(vmx);
2005 }
2006 
2007 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2008 				      struct vmcs12 *vmcs12)
2009 {
2010 	prepare_vmcs02_constant_state(vmx);
2011 
2012 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2013 
2014 	if (enable_vpid) {
2015 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2016 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2017 		else
2018 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2019 	}
2020 }
2021 
2022 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2023 {
2024 	u32 exec_control, vmcs12_exec_ctrl;
2025 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2026 
2027 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
2028 		prepare_vmcs02_early_rare(vmx, vmcs12);
2029 
2030 	/*
2031 	 * PIN CONTROLS
2032 	 */
2033 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2034 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2035 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2036 
2037 	/* Posted interrupts setting is only taken from vmcs12.  */
2038 	if (nested_cpu_has_posted_intr(vmcs12)) {
2039 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2040 		vmx->nested.pi_pending = false;
2041 	} else {
2042 		exec_control &= ~PIN_BASED_POSTED_INTR;
2043 	}
2044 	pin_controls_set(vmx, exec_control);
2045 
2046 	/*
2047 	 * EXEC CONTROLS
2048 	 */
2049 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2050 	exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2051 	exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2052 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2053 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2054 
2055 	if (exec_control & CPU_BASED_TPR_SHADOW)
2056 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2057 #ifdef CONFIG_X86_64
2058 	else
2059 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2060 				CPU_BASED_CR8_STORE_EXITING;
2061 #endif
2062 
2063 	/*
2064 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2065 	 * for I/O port accesses.
2066 	 */
2067 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2068 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2069 
2070 	/*
2071 	 * This bit will be computed in nested_get_vmcs12_pages, because
2072 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2073 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2074 	 * only set/clear this bit.
2075 	 */
2076 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2077 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2078 
2079 	exec_controls_set(vmx, exec_control);
2080 
2081 	/*
2082 	 * SECONDARY EXEC CONTROLS
2083 	 */
2084 	if (cpu_has_secondary_exec_ctrls()) {
2085 		exec_control = vmx->secondary_exec_control;
2086 
2087 		/* Take the following fields only from vmcs12 */
2088 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2089 				  SECONDARY_EXEC_ENABLE_INVPCID |
2090 				  SECONDARY_EXEC_RDTSCP |
2091 				  SECONDARY_EXEC_XSAVES |
2092 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2093 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2094 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2095 		if (nested_cpu_has(vmcs12,
2096 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2097 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2098 				~SECONDARY_EXEC_ENABLE_PML;
2099 			exec_control |= vmcs12_exec_ctrl;
2100 		}
2101 
2102 		/* VMCS shadowing for L2 is emulated for now */
2103 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2104 
2105 		/*
2106 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2107 		 * will not have to rewrite the controls just for this bit.
2108 		 */
2109 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2110 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2111 			exec_control |= SECONDARY_EXEC_DESC;
2112 
2113 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2114 			vmcs_write16(GUEST_INTR_STATUS,
2115 				vmcs12->guest_intr_status);
2116 
2117 		secondary_exec_controls_set(vmx, exec_control);
2118 	}
2119 
2120 	/*
2121 	 * ENTRY CONTROLS
2122 	 *
2123 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2124 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2125 	 * on the related bits (if supported by the CPU) in the hope that
2126 	 * we can avoid VMWrites during vmx_set_efer().
2127 	 */
2128 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2129 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2130 	if (cpu_has_load_ia32_efer()) {
2131 		if (guest_efer & EFER_LMA)
2132 			exec_control |= VM_ENTRY_IA32E_MODE;
2133 		if (guest_efer != host_efer)
2134 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2135 	}
2136 	vm_entry_controls_set(vmx, exec_control);
2137 
2138 	/*
2139 	 * EXIT CONTROLS
2140 	 *
2141 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2142 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2143 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2144 	 */
2145 	exec_control = vmx_vmexit_ctrl();
2146 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2147 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2148 	vm_exit_controls_set(vmx, exec_control);
2149 
2150 	/*
2151 	 * Interrupt/Exception Fields
2152 	 */
2153 	if (vmx->nested.nested_run_pending) {
2154 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2155 			     vmcs12->vm_entry_intr_info_field);
2156 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2157 			     vmcs12->vm_entry_exception_error_code);
2158 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2159 			     vmcs12->vm_entry_instruction_len);
2160 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2161 			     vmcs12->guest_interruptibility_info);
2162 		vmx->loaded_vmcs->nmi_known_unmasked =
2163 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2164 	} else {
2165 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2166 	}
2167 }
2168 
2169 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2170 {
2171 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2172 
2173 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2174 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2175 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2176 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2177 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2178 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2179 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2180 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2181 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2182 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2183 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2184 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2185 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2186 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2187 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2188 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2189 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2190 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2191 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2192 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2193 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2194 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2195 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2196 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2197 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2198 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2199 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2200 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2201 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2202 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2203 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2204 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2205 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2206 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2207 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2208 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2209 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2210 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2211 	}
2212 
2213 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2214 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2215 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2216 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2217 			    vmcs12->guest_pending_dbg_exceptions);
2218 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2219 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2220 
2221 		/*
2222 		 * L1 may access the L2's PDPTR, so save them to construct
2223 		 * vmcs12
2224 		 */
2225 		if (enable_ept) {
2226 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2227 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2228 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2229 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2230 		}
2231 
2232 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2233 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2234 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2235 	}
2236 
2237 	if (nested_cpu_has_xsaves(vmcs12))
2238 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2239 
2240 	/*
2241 	 * Whether page-faults are trapped is determined by a combination of
2242 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2243 	 * If enable_ept, L0 doesn't care about page faults and we should
2244 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2245 	 * care about (at least some) page faults, and because it is not easy
2246 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2247 	 * to exit on each and every L2 page fault. This is done by setting
2248 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2249 	 * Note that below we don't need special code to set EB.PF beyond the
2250 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2251 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2252 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2253 	 */
2254 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2255 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2256 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2257 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2258 
2259 	if (cpu_has_vmx_apicv()) {
2260 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2261 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2262 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2263 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2264 	}
2265 
2266 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2267 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2268 
2269 	set_cr4_guest_host_mask(vmx);
2270 }
2271 
2272 /*
2273  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2274  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2275  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2276  * guest in a way that will both be appropriate to L1's requests, and our
2277  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2278  * function also has additional necessary side-effects, like setting various
2279  * vcpu->arch fields.
2280  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2281  * is assigned to entry_failure_code on failure.
2282  */
2283 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2284 			  u32 *entry_failure_code)
2285 {
2286 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2287 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2288 	bool load_guest_pdptrs_vmcs12 = false;
2289 
2290 	if (vmx->nested.dirty_vmcs12 || hv_evmcs) {
2291 		prepare_vmcs02_rare(vmx, vmcs12);
2292 		vmx->nested.dirty_vmcs12 = false;
2293 
2294 		load_guest_pdptrs_vmcs12 = !hv_evmcs ||
2295 			!(hv_evmcs->hv_clean_fields &
2296 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2297 	}
2298 
2299 	if (vmx->nested.nested_run_pending &&
2300 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2301 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2302 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2303 	} else {
2304 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2305 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2306 	}
2307 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2308 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2309 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2310 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2311 
2312 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2313 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2314 	 * trap. Note that CR0.TS also needs updating - we do this later.
2315 	 */
2316 	update_exception_bitmap(vcpu);
2317 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2318 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2319 
2320 	if (vmx->nested.nested_run_pending &&
2321 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2322 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2323 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2324 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2325 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2326 	}
2327 
2328 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2329 
2330 	if (kvm_has_tsc_control)
2331 		decache_tsc_multiplier(vmx);
2332 
2333 	if (enable_vpid) {
2334 		/*
2335 		 * There is no direct mapping between vpid02 and vpid12, the
2336 		 * vpid02 is per-vCPU for L0 and reused while the value of
2337 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2338 		 * The vpid12 is allocated by L1 for L2, so it will not
2339 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2340 		 * even if spawn a lot of nested vCPUs.
2341 		 */
2342 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2343 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2344 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2345 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2346 			}
2347 		} else {
2348 			/*
2349 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2350 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2351 			 * flush until vmcs02->eptp is fully updated by
2352 			 * KVM_REQ_LOAD_CR3. Note that this assumes
2353 			 * KVM_REQ_TLB_FLUSH is evaluated after
2354 			 * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
2355 			 */
2356 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2357 		}
2358 	}
2359 
2360 	if (nested_cpu_has_ept(vmcs12))
2361 		nested_ept_init_mmu_context(vcpu);
2362 	else if (nested_cpu_has2(vmcs12,
2363 				 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2364 		vmx_flush_tlb(vcpu, true);
2365 
2366 	/*
2367 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2368 	 * bits which we consider mandatory enabled.
2369 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2370 	 * the specifications by L1; It's not enough to take
2371 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2372 	 * have more bits than L1 expected.
2373 	 */
2374 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2375 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2376 
2377 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2378 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2379 
2380 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2381 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2382 	vmx_set_efer(vcpu, vcpu->arch.efer);
2383 
2384 	/*
2385 	 * Guest state is invalid and unrestricted guest is disabled,
2386 	 * which means L1 attempted VMEntry to L2 with invalid state.
2387 	 * Fail the VMEntry.
2388 	 */
2389 	if (vmx->emulation_required) {
2390 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2391 		return -EINVAL;
2392 	}
2393 
2394 	/* Shadow page tables on either EPT or shadow page tables. */
2395 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2396 				entry_failure_code))
2397 		return -EINVAL;
2398 
2399 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2400 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2401 	    is_pae_paging(vcpu)) {
2402 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2403 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2404 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2405 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2406 	}
2407 
2408 	if (!enable_ept)
2409 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2410 
2411 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2412 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2413 	return 0;
2414 }
2415 
2416 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2417 {
2418 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2419 	       nested_cpu_has_virtual_nmis(vmcs12)))
2420 		return -EINVAL;
2421 
2422 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2423 	       nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING)))
2424 		return -EINVAL;
2425 
2426 	return 0;
2427 }
2428 
2429 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
2430 {
2431 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2432 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2433 
2434 	/* Check for memory type validity */
2435 	switch (address & VMX_EPTP_MT_MASK) {
2436 	case VMX_EPTP_MT_UC:
2437 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2438 			return false;
2439 		break;
2440 	case VMX_EPTP_MT_WB:
2441 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2442 			return false;
2443 		break;
2444 	default:
2445 		return false;
2446 	}
2447 
2448 	/* only 4 levels page-walk length are valid */
2449 	if (CC((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4))
2450 		return false;
2451 
2452 	/* Reserved bits should not be set */
2453 	if (CC(address >> maxphyaddr || ((address >> 7) & 0x1f)))
2454 		return false;
2455 
2456 	/* AD, if set, should be supported */
2457 	if (address & VMX_EPTP_AD_ENABLE_BIT) {
2458 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2459 			return false;
2460 	}
2461 
2462 	return true;
2463 }
2464 
2465 /*
2466  * Checks related to VM-Execution Control Fields
2467  */
2468 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2469                                               struct vmcs12 *vmcs12)
2470 {
2471 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2472 
2473 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2474 				   vmx->nested.msrs.pinbased_ctls_low,
2475 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2476 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2477 				   vmx->nested.msrs.procbased_ctls_low,
2478 				   vmx->nested.msrs.procbased_ctls_high)))
2479 		return -EINVAL;
2480 
2481 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2482 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2483 				   vmx->nested.msrs.secondary_ctls_low,
2484 				   vmx->nested.msrs.secondary_ctls_high)))
2485 		return -EINVAL;
2486 
2487 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2488 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2489 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2490 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2491 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2492 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2493 	    nested_vmx_check_nmi_controls(vmcs12) ||
2494 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2495 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2496 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2497 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2498 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2499 		return -EINVAL;
2500 
2501 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2502 	    nested_cpu_has_save_preemption_timer(vmcs12))
2503 		return -EINVAL;
2504 
2505 	if (nested_cpu_has_ept(vmcs12) &&
2506 	    CC(!valid_ept_address(vcpu, vmcs12->ept_pointer)))
2507 		return -EINVAL;
2508 
2509 	if (nested_cpu_has_vmfunc(vmcs12)) {
2510 		if (CC(vmcs12->vm_function_control &
2511 		       ~vmx->nested.msrs.vmfunc_controls))
2512 			return -EINVAL;
2513 
2514 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2515 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2516 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2517 				return -EINVAL;
2518 		}
2519 	}
2520 
2521 	return 0;
2522 }
2523 
2524 /*
2525  * Checks related to VM-Exit Control Fields
2526  */
2527 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2528                                          struct vmcs12 *vmcs12)
2529 {
2530 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2531 
2532 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2533 				    vmx->nested.msrs.exit_ctls_low,
2534 				    vmx->nested.msrs.exit_ctls_high)) ||
2535 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2536 		return -EINVAL;
2537 
2538 	return 0;
2539 }
2540 
2541 /*
2542  * Checks related to VM-Entry Control Fields
2543  */
2544 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2545 					  struct vmcs12 *vmcs12)
2546 {
2547 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2548 
2549 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2550 				    vmx->nested.msrs.entry_ctls_low,
2551 				    vmx->nested.msrs.entry_ctls_high)))
2552 		return -EINVAL;
2553 
2554 	/*
2555 	 * From the Intel SDM, volume 3:
2556 	 * Fields relevant to VM-entry event injection must be set properly.
2557 	 * These fields are the VM-entry interruption-information field, the
2558 	 * VM-entry exception error code, and the VM-entry instruction length.
2559 	 */
2560 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2561 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2562 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2563 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2564 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2565 		bool should_have_error_code;
2566 		bool urg = nested_cpu_has2(vmcs12,
2567 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2568 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2569 
2570 		/* VM-entry interruption-info field: interruption type */
2571 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2572 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2573 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2574 			return -EINVAL;
2575 
2576 		/* VM-entry interruption-info field: vector */
2577 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2578 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2579 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2580 			return -EINVAL;
2581 
2582 		/* VM-entry interruption-info field: deliver error code */
2583 		should_have_error_code =
2584 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2585 			x86_exception_has_error_code(vector);
2586 		if (CC(has_error_code != should_have_error_code))
2587 			return -EINVAL;
2588 
2589 		/* VM-entry exception error code */
2590 		if (CC(has_error_code &&
2591 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 15)))
2592 			return -EINVAL;
2593 
2594 		/* VM-entry interruption-info field: reserved bits */
2595 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2596 			return -EINVAL;
2597 
2598 		/* VM-entry instruction length */
2599 		switch (intr_type) {
2600 		case INTR_TYPE_SOFT_EXCEPTION:
2601 		case INTR_TYPE_SOFT_INTR:
2602 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2603 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2604 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2605 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2606 				return -EINVAL;
2607 		}
2608 	}
2609 
2610 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2611 		return -EINVAL;
2612 
2613 	return 0;
2614 }
2615 
2616 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2617 				     struct vmcs12 *vmcs12)
2618 {
2619 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2620 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2621 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2622 		return -EINVAL;
2623 
2624 	return 0;
2625 }
2626 
2627 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2628 				       struct vmcs12 *vmcs12)
2629 {
2630 	bool ia32e;
2631 
2632 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2633 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2634 	    CC(!nested_cr3_valid(vcpu, vmcs12->host_cr3)))
2635 		return -EINVAL;
2636 
2637 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2638 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2639 		return -EINVAL;
2640 
2641 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2642 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2643 		return -EINVAL;
2644 
2645 	ia32e = (vmcs12->vm_exit_controls &
2646 		 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
2647 
2648 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2649 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2650 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2651 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2652 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2653 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2654 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2655 	    CC(vmcs12->host_cs_selector == 0) ||
2656 	    CC(vmcs12->host_tr_selector == 0) ||
2657 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2658 		return -EINVAL;
2659 
2660 #ifdef CONFIG_X86_64
2661 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2662 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2663 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2664 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2665 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)))
2666 		return -EINVAL;
2667 #endif
2668 
2669 	/*
2670 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2671 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2672 	 * the values of the LMA and LME bits in the field must each be that of
2673 	 * the host address-space size VM-exit control.
2674 	 */
2675 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2676 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2677 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2678 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2679 			return -EINVAL;
2680 	}
2681 
2682 	return 0;
2683 }
2684 
2685 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2686 					  struct vmcs12 *vmcs12)
2687 {
2688 	int r = 0;
2689 	struct vmcs12 *shadow;
2690 	struct kvm_host_map map;
2691 
2692 	if (vmcs12->vmcs_link_pointer == -1ull)
2693 		return 0;
2694 
2695 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2696 		return -EINVAL;
2697 
2698 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2699 		return -EINVAL;
2700 
2701 	shadow = map.hva;
2702 
2703 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2704 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2705 		r = -EINVAL;
2706 
2707 	kvm_vcpu_unmap(vcpu, &map, false);
2708 	return r;
2709 }
2710 
2711 /*
2712  * Checks related to Guest Non-register State
2713  */
2714 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2715 {
2716 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2717 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT))
2718 		return -EINVAL;
2719 
2720 	return 0;
2721 }
2722 
2723 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2724 					struct vmcs12 *vmcs12,
2725 					u32 *exit_qual)
2726 {
2727 	bool ia32e;
2728 
2729 	*exit_qual = ENTRY_FAIL_DEFAULT;
2730 
2731 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2732 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2733 		return -EINVAL;
2734 
2735 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2736 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2737 		return -EINVAL;
2738 
2739 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2740 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2741 		return -EINVAL;
2742 	}
2743 
2744 	/*
2745 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2746 	 * are performed on the field for the IA32_EFER MSR:
2747 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2748 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2749 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2750 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2751 	 *   CR0.PG) is 1.
2752 	 */
2753 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2754 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2755 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2756 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
2757 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
2758 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
2759 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
2760 			return -EINVAL;
2761 	}
2762 
2763 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2764 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
2765 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
2766 		return -EINVAL;
2767 
2768 	if (nested_check_guest_non_reg_state(vmcs12))
2769 		return -EINVAL;
2770 
2771 	return 0;
2772 }
2773 
2774 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2775 {
2776 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2777 	unsigned long cr3, cr4;
2778 	bool vm_fail;
2779 
2780 	if (!nested_early_check)
2781 		return 0;
2782 
2783 	if (vmx->msr_autoload.host.nr)
2784 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2785 	if (vmx->msr_autoload.guest.nr)
2786 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2787 
2788 	preempt_disable();
2789 
2790 	vmx_prepare_switch_to_guest(vcpu);
2791 
2792 	/*
2793 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2794 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2795 	 * be written (by preparve_vmcs02()) before the "real" VMEnter, i.e.
2796 	 * there is no need to preserve other bits or save/restore the field.
2797 	 */
2798 	vmcs_writel(GUEST_RFLAGS, 0);
2799 
2800 	cr3 = __get_current_cr3_fast();
2801 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2802 		vmcs_writel(HOST_CR3, cr3);
2803 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2804 	}
2805 
2806 	cr4 = cr4_read_shadow();
2807 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2808 		vmcs_writel(HOST_CR4, cr4);
2809 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2810 	}
2811 
2812 	asm(
2813 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2814 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2815 		"je 1f \n\t"
2816 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
2817 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2818 		"1: \n\t"
2819 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2820 
2821 		/* Check if vmlaunch or vmresume is needed */
2822 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
2823 
2824 		/*
2825 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
2826 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
2827 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
2828 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
2829 		 */
2830 		"call vmx_vmenter\n\t"
2831 
2832 		CC_SET(be)
2833 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
2834 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
2835 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
2836 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
2837 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
2838 		[wordsize]"i"(sizeof(ulong))
2839 	      : "memory"
2840 	);
2841 
2842 	if (vmx->msr_autoload.host.nr)
2843 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2844 	if (vmx->msr_autoload.guest.nr)
2845 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2846 
2847 	if (vm_fail) {
2848 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
2849 
2850 		preempt_enable();
2851 
2852 		trace_kvm_nested_vmenter_failed(
2853 			"early hardware check VM-instruction error: ", error);
2854 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
2855 		return 1;
2856 	}
2857 
2858 	/*
2859 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
2860 	 */
2861 	local_irq_enable();
2862 	if (hw_breakpoint_active())
2863 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
2864 	preempt_enable();
2865 
2866 	/*
2867 	 * A non-failing VMEntry means we somehow entered guest mode with
2868 	 * an illegal RIP, and that's just the tip of the iceberg.  There
2869 	 * is no telling what memory has been modified or what state has
2870 	 * been exposed to unknown code.  Hitting this all but guarantees
2871 	 * a (very critical) hardware issue.
2872 	 */
2873 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
2874 		VMX_EXIT_REASONS_FAILED_VMENTRY));
2875 
2876 	return 0;
2877 }
2878 
2879 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
2880 						 struct vmcs12 *vmcs12);
2881 
2882 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
2883 {
2884 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2885 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2886 	struct kvm_host_map *map;
2887 	struct page *page;
2888 	u64 hpa;
2889 
2890 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2891 		/*
2892 		 * Translate L1 physical address to host physical
2893 		 * address for vmcs02. Keep the page pinned, so this
2894 		 * physical address remains valid. We keep a reference
2895 		 * to it so we can release it later.
2896 		 */
2897 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
2898 			kvm_release_page_dirty(vmx->nested.apic_access_page);
2899 			vmx->nested.apic_access_page = NULL;
2900 		}
2901 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
2902 		/*
2903 		 * If translation failed, no matter: This feature asks
2904 		 * to exit when accessing the given address, and if it
2905 		 * can never be accessed, this feature won't do
2906 		 * anything anyway.
2907 		 */
2908 		if (!is_error_page(page)) {
2909 			vmx->nested.apic_access_page = page;
2910 			hpa = page_to_phys(vmx->nested.apic_access_page);
2911 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
2912 		} else {
2913 			secondary_exec_controls_clearbit(vmx,
2914 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
2915 		}
2916 	}
2917 
2918 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
2919 		map = &vmx->nested.virtual_apic_map;
2920 
2921 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
2922 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
2923 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
2924 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
2925 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2926 			/*
2927 			 * The processor will never use the TPR shadow, simply
2928 			 * clear the bit from the execution control.  Such a
2929 			 * configuration is useless, but it happens in tests.
2930 			 * For any other configuration, failing the vm entry is
2931 			 * _not_ what the processor does but it's basically the
2932 			 * only possibility we have.
2933 			 */
2934 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
2935 		} else {
2936 			/*
2937 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
2938 			 * force VM-Entry to fail.
2939 			 */
2940 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
2941 		}
2942 	}
2943 
2944 	if (nested_cpu_has_posted_intr(vmcs12)) {
2945 		map = &vmx->nested.pi_desc_map;
2946 
2947 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
2948 			vmx->nested.pi_desc =
2949 				(struct pi_desc *)(((void *)map->hva) +
2950 				offset_in_page(vmcs12->posted_intr_desc_addr));
2951 			vmcs_write64(POSTED_INTR_DESC_ADDR,
2952 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
2953 		}
2954 	}
2955 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
2956 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
2957 	else
2958 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
2959 }
2960 
2961 /*
2962  * Intel's VMX Instruction Reference specifies a common set of prerequisites
2963  * for running VMX instructions (except VMXON, whose prerequisites are
2964  * slightly different). It also specifies what exception to inject otherwise.
2965  * Note that many of these exceptions have priority over VM exits, so they
2966  * don't have to be checked again here.
2967  */
2968 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
2969 {
2970 	if (!to_vmx(vcpu)->nested.vmxon) {
2971 		kvm_queue_exception(vcpu, UD_VECTOR);
2972 		return 0;
2973 	}
2974 
2975 	if (vmx_get_cpl(vcpu)) {
2976 		kvm_inject_gp(vcpu, 0);
2977 		return 0;
2978 	}
2979 
2980 	return 1;
2981 }
2982 
2983 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
2984 {
2985 	u8 rvi = vmx_get_rvi();
2986 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
2987 
2988 	return ((rvi & 0xf0) > (vppr & 0xf0));
2989 }
2990 
2991 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
2992 				   struct vmcs12 *vmcs12);
2993 
2994 /*
2995  * If from_vmentry is false, this is being called from state restore (either RSM
2996  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
2997 + *
2998 + * Returns:
2999 + *   0 - success, i.e. proceed with actual VMEnter
3000 + *   1 - consistency check VMExit
3001 + *  -1 - consistency check VMFail
3002  */
3003 int nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
3004 {
3005 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3006 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3007 	bool evaluate_pending_interrupts;
3008 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
3009 	u32 exit_qual;
3010 
3011 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3012 		(CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_VIRTUAL_NMI_PENDING);
3013 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3014 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3015 
3016 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3017 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3018 	if (kvm_mpx_supported() &&
3019 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3020 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3021 
3022 	/*
3023 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3024 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3025 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3026 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3027 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3028 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3029 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3030 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3031 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3032 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3033 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3034 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3035 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3036 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3037 	 */
3038 	if (!enable_ept && !nested_early_check)
3039 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3040 
3041 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3042 
3043 	prepare_vmcs02_early(vmx, vmcs12);
3044 
3045 	if (from_vmentry) {
3046 		nested_get_vmcs12_pages(vcpu);
3047 
3048 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3049 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3050 			return -1;
3051 		}
3052 
3053 		if (nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
3054 			goto vmentry_fail_vmexit;
3055 	}
3056 
3057 	enter_guest_mode(vcpu);
3058 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3059 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3060 
3061 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
3062 		goto vmentry_fail_vmexit_guest_mode;
3063 
3064 	if (from_vmentry) {
3065 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3066 		exit_qual = nested_vmx_load_msr(vcpu,
3067 						vmcs12->vm_entry_msr_load_addr,
3068 						vmcs12->vm_entry_msr_load_count);
3069 		if (exit_qual)
3070 			goto vmentry_fail_vmexit_guest_mode;
3071 	} else {
3072 		/*
3073 		 * The MMU is not initialized to point at the right entities yet and
3074 		 * "get pages" would need to read data from the guest (i.e. we will
3075 		 * need to perform gpa to hpa translation). Request a call
3076 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3077 		 * have already been set at vmentry time and should not be reset.
3078 		 */
3079 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3080 	}
3081 
3082 	/*
3083 	 * If L1 had a pending IRQ/NMI until it executed
3084 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3085 	 * disallowed (e.g. interrupts disabled), L0 needs to
3086 	 * evaluate if this pending event should cause an exit from L2
3087 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3088 	 * intercept EXTERNAL_INTERRUPT).
3089 	 *
3090 	 * Usually this would be handled by the processor noticing an
3091 	 * IRQ/NMI window request, or checking RVI during evaluation of
3092 	 * pending virtual interrupts.  However, this setting was done
3093 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3094 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3095 	 */
3096 	if (unlikely(evaluate_pending_interrupts))
3097 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3098 
3099 	/*
3100 	 * Do not start the preemption timer hrtimer until after we know
3101 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3102 	 * the timer.
3103 	 */
3104 	vmx->nested.preemption_timer_expired = false;
3105 	if (nested_cpu_has_preemption_timer(vmcs12))
3106 		vmx_start_preemption_timer(vcpu);
3107 
3108 	/*
3109 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3110 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3111 	 * returned as far as L1 is concerned. It will only return (and set
3112 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3113 	 */
3114 	return 0;
3115 
3116 	/*
3117 	 * A failed consistency check that leads to a VMExit during L1's
3118 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3119 	 * 26.7 "VM-entry failures during or after loading guest state".
3120 	 */
3121 vmentry_fail_vmexit_guest_mode:
3122 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3123 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3124 	leave_guest_mode(vcpu);
3125 
3126 vmentry_fail_vmexit:
3127 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3128 
3129 	if (!from_vmentry)
3130 		return 1;
3131 
3132 	load_vmcs12_host_state(vcpu, vmcs12);
3133 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3134 	vmcs12->exit_qualification = exit_qual;
3135 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3136 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3137 	return 1;
3138 }
3139 
3140 /*
3141  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3142  * for running an L2 nested guest.
3143  */
3144 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3145 {
3146 	struct vmcs12 *vmcs12;
3147 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3148 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3149 	int ret;
3150 
3151 	if (!nested_vmx_check_permission(vcpu))
3152 		return 1;
3153 
3154 	if (!nested_vmx_handle_enlightened_vmptrld(vcpu, launch))
3155 		return 1;
3156 
3157 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3158 		return nested_vmx_failInvalid(vcpu);
3159 
3160 	vmcs12 = get_vmcs12(vcpu);
3161 
3162 	/*
3163 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3164 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3165 	 * rather than RFLAGS.ZF, and no error number is stored to the
3166 	 * VM-instruction error field.
3167 	 */
3168 	if (vmcs12->hdr.shadow_vmcs)
3169 		return nested_vmx_failInvalid(vcpu);
3170 
3171 	if (vmx->nested.hv_evmcs) {
3172 		copy_enlightened_to_vmcs12(vmx);
3173 		/* Enlightened VMCS doesn't have launch state */
3174 		vmcs12->launch_state = !launch;
3175 	} else if (enable_shadow_vmcs) {
3176 		copy_shadow_to_vmcs12(vmx);
3177 	}
3178 
3179 	/*
3180 	 * The nested entry process starts with enforcing various prerequisites
3181 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3182 	 * they fail: As the SDM explains, some conditions should cause the
3183 	 * instruction to fail, while others will cause the instruction to seem
3184 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3185 	 * To speed up the normal (success) code path, we should avoid checking
3186 	 * for misconfigurations which will anyway be caught by the processor
3187 	 * when using the merged vmcs02.
3188 	 */
3189 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3190 		return nested_vmx_failValid(vcpu,
3191 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3192 
3193 	if (vmcs12->launch_state == launch)
3194 		return nested_vmx_failValid(vcpu,
3195 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3196 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3197 
3198 	if (nested_vmx_check_controls(vcpu, vmcs12))
3199 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3200 
3201 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3202 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3203 
3204 	/*
3205 	 * We're finally done with prerequisite checking, and can start with
3206 	 * the nested entry.
3207 	 */
3208 	vmx->nested.nested_run_pending = 1;
3209 	ret = nested_vmx_enter_non_root_mode(vcpu, true);
3210 	vmx->nested.nested_run_pending = !ret;
3211 	if (ret > 0)
3212 		return 1;
3213 	else if (ret)
3214 		return nested_vmx_failValid(vcpu,
3215 			VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3216 
3217 	/* Hide L1D cache contents from the nested guest.  */
3218 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3219 
3220 	/*
3221 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3222 	 * also be used as part of restoring nVMX state for
3223 	 * snapshot restore (migration).
3224 	 *
3225 	 * In this flow, it is assumed that vmcs12 cache was
3226 	 * trasferred as part of captured nVMX state and should
3227 	 * therefore not be read from guest memory (which may not
3228 	 * exist on destination host yet).
3229 	 */
3230 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3231 
3232 	/*
3233 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3234 	 * awakened by event injection or by an NMI-window VM-exit or
3235 	 * by an interrupt-window VM-exit, halt the vcpu.
3236 	 */
3237 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3238 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3239 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_NMI_PENDING) &&
3240 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_INTR_PENDING) &&
3241 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3242 		vmx->nested.nested_run_pending = 0;
3243 		return kvm_vcpu_halt(vcpu);
3244 	}
3245 	return 1;
3246 }
3247 
3248 /*
3249  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3250  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
3251  * This function returns the new value we should put in vmcs12.guest_cr0.
3252  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3253  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3254  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3255  *     didn't trap the bit, because if L1 did, so would L0).
3256  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3257  *     been modified by L2, and L1 knows it. So just leave the old value of
3258  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3259  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3260  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3261  *     changed these bits, and therefore they need to be updated, but L0
3262  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3263  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3264  */
3265 static inline unsigned long
3266 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3267 {
3268 	return
3269 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3270 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3271 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3272 			vcpu->arch.cr0_guest_owned_bits));
3273 }
3274 
3275 static inline unsigned long
3276 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3277 {
3278 	return
3279 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3280 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3281 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3282 			vcpu->arch.cr4_guest_owned_bits));
3283 }
3284 
3285 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3286 				      struct vmcs12 *vmcs12)
3287 {
3288 	u32 idt_vectoring;
3289 	unsigned int nr;
3290 
3291 	if (vcpu->arch.exception.injected) {
3292 		nr = vcpu->arch.exception.nr;
3293 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3294 
3295 		if (kvm_exception_is_soft(nr)) {
3296 			vmcs12->vm_exit_instruction_len =
3297 				vcpu->arch.event_exit_inst_len;
3298 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3299 		} else
3300 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3301 
3302 		if (vcpu->arch.exception.has_error_code) {
3303 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3304 			vmcs12->idt_vectoring_error_code =
3305 				vcpu->arch.exception.error_code;
3306 		}
3307 
3308 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3309 	} else if (vcpu->arch.nmi_injected) {
3310 		vmcs12->idt_vectoring_info_field =
3311 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3312 	} else if (vcpu->arch.interrupt.injected) {
3313 		nr = vcpu->arch.interrupt.nr;
3314 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3315 
3316 		if (vcpu->arch.interrupt.soft) {
3317 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3318 			vmcs12->vm_entry_instruction_len =
3319 				vcpu->arch.event_exit_inst_len;
3320 		} else
3321 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3322 
3323 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3324 	}
3325 }
3326 
3327 
3328 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3329 {
3330 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3331 	gfn_t gfn;
3332 
3333 	/*
3334 	 * Don't need to mark the APIC access page dirty; it is never
3335 	 * written to by the CPU during APIC virtualization.
3336 	 */
3337 
3338 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3339 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3340 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3341 	}
3342 
3343 	if (nested_cpu_has_posted_intr(vmcs12)) {
3344 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3345 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3346 	}
3347 }
3348 
3349 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3350 {
3351 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3352 	int max_irr;
3353 	void *vapic_page;
3354 	u16 status;
3355 
3356 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3357 		return;
3358 
3359 	vmx->nested.pi_pending = false;
3360 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3361 		return;
3362 
3363 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3364 	if (max_irr != 256) {
3365 		vapic_page = vmx->nested.virtual_apic_map.hva;
3366 		if (!vapic_page)
3367 			return;
3368 
3369 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3370 			vapic_page, &max_irr);
3371 		status = vmcs_read16(GUEST_INTR_STATUS);
3372 		if ((u8)max_irr > ((u8)status & 0xff)) {
3373 			status &= ~0xff;
3374 			status |= (u8)max_irr;
3375 			vmcs_write16(GUEST_INTR_STATUS, status);
3376 		}
3377 	}
3378 
3379 	nested_mark_vmcs12_pages_dirty(vcpu);
3380 }
3381 
3382 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3383 					       unsigned long exit_qual)
3384 {
3385 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3386 	unsigned int nr = vcpu->arch.exception.nr;
3387 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3388 
3389 	if (vcpu->arch.exception.has_error_code) {
3390 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3391 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3392 	}
3393 
3394 	if (kvm_exception_is_soft(nr))
3395 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3396 	else
3397 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3398 
3399 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3400 	    vmx_get_nmi_mask(vcpu))
3401 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3402 
3403 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3404 }
3405 
3406 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
3407 {
3408 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3409 	unsigned long exit_qual;
3410 	bool block_nested_events =
3411 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3412 	struct kvm_lapic *apic = vcpu->arch.apic;
3413 
3414 	if (lapic_in_kernel(vcpu) &&
3415 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3416 		if (block_nested_events)
3417 			return -EBUSY;
3418 		nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3419 		return 0;
3420 	}
3421 
3422 	if (vcpu->arch.exception.pending &&
3423 		nested_vmx_check_exception(vcpu, &exit_qual)) {
3424 		if (block_nested_events)
3425 			return -EBUSY;
3426 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3427 		return 0;
3428 	}
3429 
3430 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3431 	    vmx->nested.preemption_timer_expired) {
3432 		if (block_nested_events)
3433 			return -EBUSY;
3434 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3435 		return 0;
3436 	}
3437 
3438 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3439 		if (block_nested_events)
3440 			return -EBUSY;
3441 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3442 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3443 				  INTR_INFO_VALID_MASK, 0);
3444 		/*
3445 		 * The NMI-triggered VM exit counts as injection:
3446 		 * clear this one and block further NMIs.
3447 		 */
3448 		vcpu->arch.nmi_pending = 0;
3449 		vmx_set_nmi_mask(vcpu, true);
3450 		return 0;
3451 	}
3452 
3453 	if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
3454 	    nested_exit_on_intr(vcpu)) {
3455 		if (block_nested_events)
3456 			return -EBUSY;
3457 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3458 		return 0;
3459 	}
3460 
3461 	vmx_complete_nested_posted_interrupt(vcpu);
3462 	return 0;
3463 }
3464 
3465 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3466 {
3467 	ktime_t remaining =
3468 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3469 	u64 value;
3470 
3471 	if (ktime_to_ns(remaining) <= 0)
3472 		return 0;
3473 
3474 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3475 	do_div(value, 1000000);
3476 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3477 }
3478 
3479 static bool is_vmcs12_ext_field(unsigned long field)
3480 {
3481 	switch (field) {
3482 	case GUEST_ES_SELECTOR:
3483 	case GUEST_CS_SELECTOR:
3484 	case GUEST_SS_SELECTOR:
3485 	case GUEST_DS_SELECTOR:
3486 	case GUEST_FS_SELECTOR:
3487 	case GUEST_GS_SELECTOR:
3488 	case GUEST_LDTR_SELECTOR:
3489 	case GUEST_TR_SELECTOR:
3490 	case GUEST_ES_LIMIT:
3491 	case GUEST_CS_LIMIT:
3492 	case GUEST_SS_LIMIT:
3493 	case GUEST_DS_LIMIT:
3494 	case GUEST_FS_LIMIT:
3495 	case GUEST_GS_LIMIT:
3496 	case GUEST_LDTR_LIMIT:
3497 	case GUEST_TR_LIMIT:
3498 	case GUEST_GDTR_LIMIT:
3499 	case GUEST_IDTR_LIMIT:
3500 	case GUEST_ES_AR_BYTES:
3501 	case GUEST_DS_AR_BYTES:
3502 	case GUEST_FS_AR_BYTES:
3503 	case GUEST_GS_AR_BYTES:
3504 	case GUEST_LDTR_AR_BYTES:
3505 	case GUEST_TR_AR_BYTES:
3506 	case GUEST_ES_BASE:
3507 	case GUEST_CS_BASE:
3508 	case GUEST_SS_BASE:
3509 	case GUEST_DS_BASE:
3510 	case GUEST_FS_BASE:
3511 	case GUEST_GS_BASE:
3512 	case GUEST_LDTR_BASE:
3513 	case GUEST_TR_BASE:
3514 	case GUEST_GDTR_BASE:
3515 	case GUEST_IDTR_BASE:
3516 	case GUEST_PENDING_DBG_EXCEPTIONS:
3517 	case GUEST_BNDCFGS:
3518 		return true;
3519 	default:
3520 		break;
3521 	}
3522 
3523 	return false;
3524 }
3525 
3526 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3527 				       struct vmcs12 *vmcs12)
3528 {
3529 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3530 
3531 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3532 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3533 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3534 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3535 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3536 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3537 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3538 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3539 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3540 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3541 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3542 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3543 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3544 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3545 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3546 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3547 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3548 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3549 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3550 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3551 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3552 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3553 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3554 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3555 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3556 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3557 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3558 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3559 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3560 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3561 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3562 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3563 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3564 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3565 	vmcs12->guest_pending_dbg_exceptions =
3566 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3567 	if (kvm_mpx_supported())
3568 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3569 
3570 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
3571 }
3572 
3573 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3574 				       struct vmcs12 *vmcs12)
3575 {
3576 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3577 	int cpu;
3578 
3579 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
3580 		return;
3581 
3582 
3583 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
3584 
3585 	cpu = get_cpu();
3586 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
3587 	vmx_vcpu_load(&vmx->vcpu, cpu);
3588 
3589 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3590 
3591 	vmx->loaded_vmcs = &vmx->vmcs01;
3592 	vmx_vcpu_load(&vmx->vcpu, cpu);
3593 	put_cpu();
3594 }
3595 
3596 /*
3597  * Update the guest state fields of vmcs12 to reflect changes that
3598  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3599  * VM-entry controls is also updated, since this is really a guest
3600  * state bit.)
3601  */
3602 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3603 {
3604 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3605 
3606 	if (vmx->nested.hv_evmcs)
3607 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3608 
3609 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = !vmx->nested.hv_evmcs;
3610 
3611 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3612 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3613 
3614 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
3615 	vmcs12->guest_rip = kvm_rip_read(vcpu);
3616 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3617 
3618 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3619 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3620 
3621 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3622 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3623 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3624 
3625 	vmcs12->guest_interruptibility_info =
3626 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3627 
3628 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3629 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3630 	else
3631 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3632 
3633 	if (nested_cpu_has_preemption_timer(vmcs12) &&
3634 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3635 			vmcs12->vmx_preemption_timer_value =
3636 				vmx_get_preemption_timer_value(vcpu);
3637 
3638 	/*
3639 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3640 	 * own CR3 without exiting. If it has changed it, we must keep it.
3641 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3642 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3643 	 *
3644 	 * Additionally, restore L2's PDPTR to vmcs12.
3645 	 */
3646 	if (enable_ept) {
3647 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3648 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3649 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3650 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3651 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3652 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3653 		}
3654 	}
3655 
3656 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3657 
3658 	if (nested_cpu_has_vid(vmcs12))
3659 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3660 
3661 	vmcs12->vm_entry_controls =
3662 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3663 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3664 
3665 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
3666 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3667 
3668 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3669 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3670 }
3671 
3672 /*
3673  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3674  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3675  * and this function updates it to reflect the changes to the guest state while
3676  * L2 was running (and perhaps made some exits which were handled directly by L0
3677  * without going back to L1), and to reflect the exit reason.
3678  * Note that we do not have to copy here all VMCS fields, just those that
3679  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3680  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3681  * which already writes to vmcs12 directly.
3682  */
3683 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3684 			   u32 exit_reason, u32 exit_intr_info,
3685 			   unsigned long exit_qualification)
3686 {
3687 	/* update exit information fields: */
3688 	vmcs12->vm_exit_reason = exit_reason;
3689 	vmcs12->exit_qualification = exit_qualification;
3690 	vmcs12->vm_exit_intr_info = exit_intr_info;
3691 
3692 	vmcs12->idt_vectoring_info_field = 0;
3693 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3694 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3695 
3696 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3697 		vmcs12->launch_state = 1;
3698 
3699 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3700 		 * instead of reading the real value. */
3701 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3702 
3703 		/*
3704 		 * Transfer the event that L0 or L1 may wanted to inject into
3705 		 * L2 to IDT_VECTORING_INFO_FIELD.
3706 		 */
3707 		vmcs12_save_pending_event(vcpu, vmcs12);
3708 
3709 		/*
3710 		 * According to spec, there's no need to store the guest's
3711 		 * MSRs if the exit is due to a VM-entry failure that occurs
3712 		 * during or after loading the guest state. Since this exit
3713 		 * does not fall in that category, we need to save the MSRs.
3714 		 */
3715 		if (nested_vmx_store_msr(vcpu,
3716 					 vmcs12->vm_exit_msr_store_addr,
3717 					 vmcs12->vm_exit_msr_store_count))
3718 			nested_vmx_abort(vcpu,
3719 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3720 	}
3721 
3722 	/*
3723 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3724 	 * preserved above and would only end up incorrectly in L1.
3725 	 */
3726 	vcpu->arch.nmi_injected = false;
3727 	kvm_clear_exception_queue(vcpu);
3728 	kvm_clear_interrupt_queue(vcpu);
3729 }
3730 
3731 /*
3732  * A part of what we need to when the nested L2 guest exits and we want to
3733  * run its L1 parent, is to reset L1's guest state to the host state specified
3734  * in vmcs12.
3735  * This function is to be called not only on normal nested exit, but also on
3736  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3737  * Failures During or After Loading Guest State").
3738  * This function should be called when the active VMCS is L1's (vmcs01).
3739  */
3740 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3741 				   struct vmcs12 *vmcs12)
3742 {
3743 	struct kvm_segment seg;
3744 	u32 entry_failure_code;
3745 
3746 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
3747 		vcpu->arch.efer = vmcs12->host_ia32_efer;
3748 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3749 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
3750 	else
3751 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
3752 	vmx_set_efer(vcpu, vcpu->arch.efer);
3753 
3754 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
3755 	kvm_rip_write(vcpu, vmcs12->host_rip);
3756 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
3757 	vmx_set_interrupt_shadow(vcpu, 0);
3758 
3759 	/*
3760 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
3761 	 * actually changed, because vmx_set_cr0 refers to efer set above.
3762 	 *
3763 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
3764 	 * (KVM doesn't change it);
3765 	 */
3766 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3767 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
3768 
3769 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
3770 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3771 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
3772 
3773 	nested_ept_uninit_mmu_context(vcpu);
3774 
3775 	/*
3776 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
3777 	 * couldn't have changed.
3778 	 */
3779 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
3780 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
3781 
3782 	if (!enable_ept)
3783 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
3784 
3785 	/*
3786 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
3787 	 * VMEntry/VMExit. Thus, no need to flush TLB.
3788 	 *
3789 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
3790 	 * flushed on every VMEntry/VMExit.
3791 	 *
3792 	 * Otherwise, we can preserve TLB entries as long as we are
3793 	 * able to tag L1 TLB entries differently than L2 TLB entries.
3794 	 *
3795 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
3796 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
3797 	 * has been set by KVM_REQ_LOAD_CR3.
3798 	 */
3799 	if (enable_vpid &&
3800 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
3801 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3802 	}
3803 
3804 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
3805 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
3806 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
3807 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
3808 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
3809 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
3810 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
3811 
3812 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
3813 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
3814 		vmcs_write64(GUEST_BNDCFGS, 0);
3815 
3816 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
3817 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
3818 		vcpu->arch.pat = vmcs12->host_ia32_pat;
3819 	}
3820 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
3821 		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
3822 			vmcs12->host_ia32_perf_global_ctrl);
3823 
3824 	/* Set L1 segment info according to Intel SDM
3825 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
3826 	seg = (struct kvm_segment) {
3827 		.base = 0,
3828 		.limit = 0xFFFFFFFF,
3829 		.selector = vmcs12->host_cs_selector,
3830 		.type = 11,
3831 		.present = 1,
3832 		.s = 1,
3833 		.g = 1
3834 	};
3835 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3836 		seg.l = 1;
3837 	else
3838 		seg.db = 1;
3839 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
3840 	seg = (struct kvm_segment) {
3841 		.base = 0,
3842 		.limit = 0xFFFFFFFF,
3843 		.type = 3,
3844 		.present = 1,
3845 		.s = 1,
3846 		.db = 1,
3847 		.g = 1
3848 	};
3849 	seg.selector = vmcs12->host_ds_selector;
3850 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
3851 	seg.selector = vmcs12->host_es_selector;
3852 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
3853 	seg.selector = vmcs12->host_ss_selector;
3854 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
3855 	seg.selector = vmcs12->host_fs_selector;
3856 	seg.base = vmcs12->host_fs_base;
3857 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
3858 	seg.selector = vmcs12->host_gs_selector;
3859 	seg.base = vmcs12->host_gs_base;
3860 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
3861 	seg = (struct kvm_segment) {
3862 		.base = vmcs12->host_tr_base,
3863 		.limit = 0x67,
3864 		.selector = vmcs12->host_tr_selector,
3865 		.type = 11,
3866 		.present = 1
3867 	};
3868 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
3869 
3870 	kvm_set_dr(vcpu, 7, 0x400);
3871 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3872 
3873 	if (cpu_has_vmx_msr_bitmap())
3874 		vmx_update_msr_bitmap(vcpu);
3875 
3876 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
3877 				vmcs12->vm_exit_msr_load_count))
3878 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3879 }
3880 
3881 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
3882 {
3883 	struct shared_msr_entry *efer_msr;
3884 	unsigned int i;
3885 
3886 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
3887 		return vmcs_read64(GUEST_IA32_EFER);
3888 
3889 	if (cpu_has_load_ia32_efer())
3890 		return host_efer;
3891 
3892 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
3893 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
3894 			return vmx->msr_autoload.guest.val[i].value;
3895 	}
3896 
3897 	efer_msr = find_msr_entry(vmx, MSR_EFER);
3898 	if (efer_msr)
3899 		return efer_msr->data;
3900 
3901 	return host_efer;
3902 }
3903 
3904 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
3905 {
3906 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3907 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3908 	struct vmx_msr_entry g, h;
3909 	gpa_t gpa;
3910 	u32 i, j;
3911 
3912 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
3913 
3914 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
3915 		/*
3916 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
3917 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
3918 		 * and vcpu->arch.dr7 is not squirreled away before the
3919 		 * nested VMENTER (not worth adding a variable in nested_vmx).
3920 		 */
3921 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
3922 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
3923 		else
3924 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
3925 	}
3926 
3927 	/*
3928 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
3929 	 * handle a variety of side effects to KVM's software model.
3930 	 */
3931 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
3932 
3933 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3934 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
3935 
3936 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3937 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
3938 
3939 	nested_ept_uninit_mmu_context(vcpu);
3940 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3941 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3942 
3943 	/*
3944 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
3945 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
3946 	 * VMFail, like everything else we just need to ensure our
3947 	 * software model is up-to-date.
3948 	 */
3949 	if (enable_ept)
3950 		ept_save_pdptrs(vcpu);
3951 
3952 	kvm_mmu_reset_context(vcpu);
3953 
3954 	if (cpu_has_vmx_msr_bitmap())
3955 		vmx_update_msr_bitmap(vcpu);
3956 
3957 	/*
3958 	 * This nasty bit of open coding is a compromise between blindly
3959 	 * loading L1's MSRs using the exit load lists (incorrect emulation
3960 	 * of VMFail), leaving the nested VM's MSRs in the software model
3961 	 * (incorrect behavior) and snapshotting the modified MSRs (too
3962 	 * expensive since the lists are unbound by hardware).  For each
3963 	 * MSR that was (prematurely) loaded from the nested VMEntry load
3964 	 * list, reload it from the exit load list if it exists and differs
3965 	 * from the guest value.  The intent is to stuff host state as
3966 	 * silently as possible, not to fully process the exit load list.
3967 	 */
3968 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
3969 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
3970 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
3971 			pr_debug_ratelimited(
3972 				"%s read MSR index failed (%u, 0x%08llx)\n",
3973 				__func__, i, gpa);
3974 			goto vmabort;
3975 		}
3976 
3977 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
3978 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
3979 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
3980 				pr_debug_ratelimited(
3981 					"%s read MSR failed (%u, 0x%08llx)\n",
3982 					__func__, j, gpa);
3983 				goto vmabort;
3984 			}
3985 			if (h.index != g.index)
3986 				continue;
3987 			if (h.value == g.value)
3988 				break;
3989 
3990 			if (nested_vmx_load_msr_check(vcpu, &h)) {
3991 				pr_debug_ratelimited(
3992 					"%s check failed (%u, 0x%x, 0x%x)\n",
3993 					__func__, j, h.index, h.reserved);
3994 				goto vmabort;
3995 			}
3996 
3997 			if (kvm_set_msr(vcpu, h.index, h.value)) {
3998 				pr_debug_ratelimited(
3999 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4000 					__func__, j, h.index, h.value);
4001 				goto vmabort;
4002 			}
4003 		}
4004 	}
4005 
4006 	return;
4007 
4008 vmabort:
4009 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4010 }
4011 
4012 /*
4013  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4014  * and modify vmcs12 to make it see what it would expect to see there if
4015  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4016  */
4017 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
4018 		       u32 exit_intr_info, unsigned long exit_qualification)
4019 {
4020 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4021 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4022 
4023 	/* trying to cancel vmlaunch/vmresume is a bug */
4024 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4025 
4026 	leave_guest_mode(vcpu);
4027 
4028 	if (nested_cpu_has_preemption_timer(vmcs12))
4029 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4030 
4031 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
4032 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
4033 
4034 	if (likely(!vmx->fail)) {
4035 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4036 
4037 		if (exit_reason != -1)
4038 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
4039 				       exit_qualification);
4040 
4041 		/*
4042 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4043 		 * also be used to capture vmcs12 cache as part of
4044 		 * capturing nVMX state for snapshot (migration).
4045 		 *
4046 		 * Otherwise, this flush will dirty guest memory at a
4047 		 * point it is already assumed by user-space to be
4048 		 * immutable.
4049 		 */
4050 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4051 	} else {
4052 		/*
4053 		 * The only expected VM-instruction error is "VM entry with
4054 		 * invalid control field(s)." Anything else indicates a
4055 		 * problem with L0.  And we should never get here with a
4056 		 * VMFail of any type if early consistency checks are enabled.
4057 		 */
4058 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4059 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4060 		WARN_ON_ONCE(nested_early_check);
4061 	}
4062 
4063 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4064 
4065 	/* Update any VMCS fields that might have changed while L2 ran */
4066 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4067 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4068 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4069 
4070 	if (kvm_has_tsc_control)
4071 		decache_tsc_multiplier(vmx);
4072 
4073 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4074 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4075 		vmx_set_virtual_apic_mode(vcpu);
4076 	} else if (!nested_cpu_has_ept(vmcs12) &&
4077 		   nested_cpu_has2(vmcs12,
4078 				   SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
4079 		vmx_flush_tlb(vcpu, true);
4080 	}
4081 
4082 	/* Unpin physical memory we referred to in vmcs02 */
4083 	if (vmx->nested.apic_access_page) {
4084 		kvm_release_page_dirty(vmx->nested.apic_access_page);
4085 		vmx->nested.apic_access_page = NULL;
4086 	}
4087 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4088 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4089 	vmx->nested.pi_desc = NULL;
4090 
4091 	/*
4092 	 * We are now running in L2, mmu_notifier will force to reload the
4093 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
4094 	 */
4095 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4096 
4097 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
4098 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4099 
4100 	/* in case we halted in L2 */
4101 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4102 
4103 	if (likely(!vmx->fail)) {
4104 		/*
4105 		 * TODO: SDM says that with acknowledge interrupt on
4106 		 * exit, bit 31 of the VM-exit interrupt information
4107 		 * (valid interrupt) is always set to 1 on
4108 		 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
4109 		 * need kvm_cpu_has_interrupt().  See the commit
4110 		 * message for details.
4111 		 */
4112 		if (nested_exit_intr_ack_set(vcpu) &&
4113 		    exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4114 		    kvm_cpu_has_interrupt(vcpu)) {
4115 			int irq = kvm_cpu_get_interrupt(vcpu);
4116 			WARN_ON(irq < 0);
4117 			vmcs12->vm_exit_intr_info = irq |
4118 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4119 		}
4120 
4121 		if (exit_reason != -1)
4122 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4123 						       vmcs12->exit_qualification,
4124 						       vmcs12->idt_vectoring_info_field,
4125 						       vmcs12->vm_exit_intr_info,
4126 						       vmcs12->vm_exit_intr_error_code,
4127 						       KVM_ISA_VMX);
4128 
4129 		load_vmcs12_host_state(vcpu, vmcs12);
4130 
4131 		return;
4132 	}
4133 
4134 	/*
4135 	 * After an early L2 VM-entry failure, we're now back
4136 	 * in L1 which thinks it just finished a VMLAUNCH or
4137 	 * VMRESUME instruction, so we need to set the failure
4138 	 * flag and the VM-instruction error field of the VMCS
4139 	 * accordingly, and skip the emulated instruction.
4140 	 */
4141 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4142 
4143 	/*
4144 	 * Restore L1's host state to KVM's software model.  We're here
4145 	 * because a consistency check was caught by hardware, which
4146 	 * means some amount of guest state has been propagated to KVM's
4147 	 * model and needs to be unwound to the host's state.
4148 	 */
4149 	nested_vmx_restore_host_state(vcpu);
4150 
4151 	vmx->fail = 0;
4152 }
4153 
4154 /*
4155  * Decode the memory-address operand of a vmx instruction, as recorded on an
4156  * exit caused by such an instruction (run by a guest hypervisor).
4157  * On success, returns 0. When the operand is invalid, returns 1 and throws
4158  * #UD or #GP.
4159  */
4160 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4161 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4162 {
4163 	gva_t off;
4164 	bool exn;
4165 	struct kvm_segment s;
4166 
4167 	/*
4168 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4169 	 * Execution", on an exit, vmx_instruction_info holds most of the
4170 	 * addressing components of the operand. Only the displacement part
4171 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4172 	 * For how an actual address is calculated from all these components,
4173 	 * refer to Vol. 1, "Operand Addressing".
4174 	 */
4175 	int  scaling = vmx_instruction_info & 3;
4176 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4177 	bool is_reg = vmx_instruction_info & (1u << 10);
4178 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4179 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4180 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4181 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4182 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4183 
4184 	if (is_reg) {
4185 		kvm_queue_exception(vcpu, UD_VECTOR);
4186 		return 1;
4187 	}
4188 
4189 	/* Addr = segment_base + offset */
4190 	/* offset = base + [index * scale] + displacement */
4191 	off = exit_qualification; /* holds the displacement */
4192 	if (addr_size == 1)
4193 		off = (gva_t)sign_extend64(off, 31);
4194 	else if (addr_size == 0)
4195 		off = (gva_t)sign_extend64(off, 15);
4196 	if (base_is_valid)
4197 		off += kvm_register_read(vcpu, base_reg);
4198 	if (index_is_valid)
4199 		off += kvm_register_read(vcpu, index_reg)<<scaling;
4200 	vmx_get_segment(vcpu, &s, seg_reg);
4201 
4202 	/*
4203 	 * The effective address, i.e. @off, of a memory operand is truncated
4204 	 * based on the address size of the instruction.  Note that this is
4205 	 * the *effective address*, i.e. the address prior to accounting for
4206 	 * the segment's base.
4207 	 */
4208 	if (addr_size == 1) /* 32 bit */
4209 		off &= 0xffffffff;
4210 	else if (addr_size == 0) /* 16 bit */
4211 		off &= 0xffff;
4212 
4213 	/* Checks for #GP/#SS exceptions. */
4214 	exn = false;
4215 	if (is_long_mode(vcpu)) {
4216 		/*
4217 		 * The virtual/linear address is never truncated in 64-bit
4218 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4219 		 * address when using FS/GS with a non-zero base.
4220 		 */
4221 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4222 			*ret = s.base + off;
4223 		else
4224 			*ret = off;
4225 
4226 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4227 		 * non-canonical form. This is the only check on the memory
4228 		 * destination for long mode!
4229 		 */
4230 		exn = is_noncanonical_address(*ret, vcpu);
4231 	} else {
4232 		/*
4233 		 * When not in long mode, the virtual/linear address is
4234 		 * unconditionally truncated to 32 bits regardless of the
4235 		 * address size.
4236 		 */
4237 		*ret = (s.base + off) & 0xffffffff;
4238 
4239 		/* Protected mode: apply checks for segment validity in the
4240 		 * following order:
4241 		 * - segment type check (#GP(0) may be thrown)
4242 		 * - usability check (#GP(0)/#SS(0))
4243 		 * - limit check (#GP(0)/#SS(0))
4244 		 */
4245 		if (wr)
4246 			/* #GP(0) if the destination operand is located in a
4247 			 * read-only data segment or any code segment.
4248 			 */
4249 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4250 		else
4251 			/* #GP(0) if the source operand is located in an
4252 			 * execute-only code segment
4253 			 */
4254 			exn = ((s.type & 0xa) == 8);
4255 		if (exn) {
4256 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4257 			return 1;
4258 		}
4259 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4260 		 */
4261 		exn = (s.unusable != 0);
4262 
4263 		/*
4264 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4265 		 * outside the segment limit.  All CPUs that support VMX ignore
4266 		 * limit checks for flat segments, i.e. segments with base==0,
4267 		 * limit==0xffffffff and of type expand-up data or code.
4268 		 */
4269 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4270 		     ((s.type & 8) || !(s.type & 4))))
4271 			exn = exn || ((u64)off + len - 1 > s.limit);
4272 	}
4273 	if (exn) {
4274 		kvm_queue_exception_e(vcpu,
4275 				      seg_reg == VCPU_SREG_SS ?
4276 						SS_VECTOR : GP_VECTOR,
4277 				      0);
4278 		return 1;
4279 	}
4280 
4281 	return 0;
4282 }
4283 
4284 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4285 {
4286 	gva_t gva;
4287 	struct x86_exception e;
4288 
4289 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4290 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4291 				sizeof(*vmpointer), &gva))
4292 		return 1;
4293 
4294 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4295 		kvm_inject_page_fault(vcpu, &e);
4296 		return 1;
4297 	}
4298 
4299 	return 0;
4300 }
4301 
4302 /*
4303  * Allocate a shadow VMCS and associate it with the currently loaded
4304  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4305  * VMCS is also VMCLEARed, so that it is ready for use.
4306  */
4307 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4308 {
4309 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4310 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4311 
4312 	/*
4313 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4314 	 * executes VMXON and free it when L1 executes VMXOFF.
4315 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4316 	 * here when vmcs01 already have an allocated shadow vmcs.
4317 	 */
4318 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4319 
4320 	if (!loaded_vmcs->shadow_vmcs) {
4321 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4322 		if (loaded_vmcs->shadow_vmcs)
4323 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4324 	}
4325 	return loaded_vmcs->shadow_vmcs;
4326 }
4327 
4328 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4329 {
4330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4331 	int r;
4332 
4333 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4334 	if (r < 0)
4335 		goto out_vmcs02;
4336 
4337 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4338 	if (!vmx->nested.cached_vmcs12)
4339 		goto out_cached_vmcs12;
4340 
4341 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4342 	if (!vmx->nested.cached_shadow_vmcs12)
4343 		goto out_cached_shadow_vmcs12;
4344 
4345 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4346 		goto out_shadow_vmcs;
4347 
4348 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4349 		     HRTIMER_MODE_REL_PINNED);
4350 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4351 
4352 	vmx->nested.vpid02 = allocate_vpid();
4353 
4354 	vmx->nested.vmcs02_initialized = false;
4355 	vmx->nested.vmxon = true;
4356 
4357 	if (pt_mode == PT_MODE_HOST_GUEST) {
4358 		vmx->pt_desc.guest.ctl = 0;
4359 		pt_update_intercept_for_msr(vmx);
4360 	}
4361 
4362 	return 0;
4363 
4364 out_shadow_vmcs:
4365 	kfree(vmx->nested.cached_shadow_vmcs12);
4366 
4367 out_cached_shadow_vmcs12:
4368 	kfree(vmx->nested.cached_vmcs12);
4369 
4370 out_cached_vmcs12:
4371 	free_loaded_vmcs(&vmx->nested.vmcs02);
4372 
4373 out_vmcs02:
4374 	return -ENOMEM;
4375 }
4376 
4377 /*
4378  * Emulate the VMXON instruction.
4379  * Currently, we just remember that VMX is active, and do not save or even
4380  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4381  * do not currently need to store anything in that guest-allocated memory
4382  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4383  * argument is different from the VMXON pointer (which the spec says they do).
4384  */
4385 static int handle_vmon(struct kvm_vcpu *vcpu)
4386 {
4387 	int ret;
4388 	gpa_t vmptr;
4389 	uint32_t revision;
4390 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4391 	const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
4392 		| FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
4393 
4394 	/*
4395 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4396 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4397 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4398 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4399 	 * have already been checked by hardware, prior to the VM-exit for
4400 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4401 	 * that bit set to 1 in non-root mode.
4402 	 */
4403 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4404 		kvm_queue_exception(vcpu, UD_VECTOR);
4405 		return 1;
4406 	}
4407 
4408 	/* CPL=0 must be checked manually. */
4409 	if (vmx_get_cpl(vcpu)) {
4410 		kvm_inject_gp(vcpu, 0);
4411 		return 1;
4412 	}
4413 
4414 	if (vmx->nested.vmxon)
4415 		return nested_vmx_failValid(vcpu,
4416 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4417 
4418 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4419 			!= VMXON_NEEDED_FEATURES) {
4420 		kvm_inject_gp(vcpu, 0);
4421 		return 1;
4422 	}
4423 
4424 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4425 		return 1;
4426 
4427 	/*
4428 	 * SDM 3: 24.11.5
4429 	 * The first 4 bytes of VMXON region contain the supported
4430 	 * VMCS revision identifier
4431 	 *
4432 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4433 	 * which replaces physical address width with 32
4434 	 */
4435 	if (!page_address_valid(vcpu, vmptr))
4436 		return nested_vmx_failInvalid(vcpu);
4437 
4438 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4439 	    revision != VMCS12_REVISION)
4440 		return nested_vmx_failInvalid(vcpu);
4441 
4442 	vmx->nested.vmxon_ptr = vmptr;
4443 	ret = enter_vmx_operation(vcpu);
4444 	if (ret)
4445 		return ret;
4446 
4447 	return nested_vmx_succeed(vcpu);
4448 }
4449 
4450 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4451 {
4452 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4453 
4454 	if (vmx->nested.current_vmptr == -1ull)
4455 		return;
4456 
4457 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4458 
4459 	if (enable_shadow_vmcs) {
4460 		/* copy to memory all shadowed fields in case
4461 		   they were modified */
4462 		copy_shadow_to_vmcs12(vmx);
4463 		vmx_disable_shadow_vmcs(vmx);
4464 	}
4465 	vmx->nested.posted_intr_nv = -1;
4466 
4467 	/* Flush VMCS12 to guest memory */
4468 	kvm_vcpu_write_guest_page(vcpu,
4469 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4470 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4471 
4472 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4473 
4474 	vmx->nested.current_vmptr = -1ull;
4475 }
4476 
4477 /* Emulate the VMXOFF instruction */
4478 static int handle_vmoff(struct kvm_vcpu *vcpu)
4479 {
4480 	if (!nested_vmx_check_permission(vcpu))
4481 		return 1;
4482 
4483 	free_nested(vcpu);
4484 
4485 	/* Process a latched INIT during time CPU was in VMX operation */
4486 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4487 
4488 	return nested_vmx_succeed(vcpu);
4489 }
4490 
4491 /* Emulate the VMCLEAR instruction */
4492 static int handle_vmclear(struct kvm_vcpu *vcpu)
4493 {
4494 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4495 	u32 zero = 0;
4496 	gpa_t vmptr;
4497 	u64 evmcs_gpa;
4498 
4499 	if (!nested_vmx_check_permission(vcpu))
4500 		return 1;
4501 
4502 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4503 		return 1;
4504 
4505 	if (!page_address_valid(vcpu, vmptr))
4506 		return nested_vmx_failValid(vcpu,
4507 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4508 
4509 	if (vmptr == vmx->nested.vmxon_ptr)
4510 		return nested_vmx_failValid(vcpu,
4511 			VMXERR_VMCLEAR_VMXON_POINTER);
4512 
4513 	/*
4514 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4515 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4516 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4517 	 * writing to the non-existent 'launch_state' field. The area doesn't
4518 	 * have to be the currently active EVMCS on the calling CPU and there's
4519 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
4520 	 * state. It is possible that the area will stay mapped as
4521 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
4522 	 */
4523 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
4524 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
4525 		if (vmptr == vmx->nested.current_vmptr)
4526 			nested_release_vmcs12(vcpu);
4527 
4528 		kvm_vcpu_write_guest(vcpu,
4529 				     vmptr + offsetof(struct vmcs12,
4530 						      launch_state),
4531 				     &zero, sizeof(zero));
4532 	}
4533 
4534 	return nested_vmx_succeed(vcpu);
4535 }
4536 
4537 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4538 
4539 /* Emulate the VMLAUNCH instruction */
4540 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4541 {
4542 	return nested_vmx_run(vcpu, true);
4543 }
4544 
4545 /* Emulate the VMRESUME instruction */
4546 static int handle_vmresume(struct kvm_vcpu *vcpu)
4547 {
4548 
4549 	return nested_vmx_run(vcpu, false);
4550 }
4551 
4552 static int handle_vmread(struct kvm_vcpu *vcpu)
4553 {
4554 	unsigned long field;
4555 	u64 field_value;
4556 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4557 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4558 	int len;
4559 	gva_t gva = 0;
4560 	struct vmcs12 *vmcs12;
4561 	struct x86_exception e;
4562 	short offset;
4563 
4564 	if (!nested_vmx_check_permission(vcpu))
4565 		return 1;
4566 
4567 	if (to_vmx(vcpu)->nested.current_vmptr == -1ull)
4568 		return nested_vmx_failInvalid(vcpu);
4569 
4570 	if (!is_guest_mode(vcpu))
4571 		vmcs12 = get_vmcs12(vcpu);
4572 	else {
4573 		/*
4574 		 * When vmcs->vmcs_link_pointer is -1ull, any VMREAD
4575 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4576 		 */
4577 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4578 			return nested_vmx_failInvalid(vcpu);
4579 		vmcs12 = get_shadow_vmcs12(vcpu);
4580 	}
4581 
4582 	/* Decode instruction info and find the field to read */
4583 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4584 
4585 	offset = vmcs_field_to_offset(field);
4586 	if (offset < 0)
4587 		return nested_vmx_failValid(vcpu,
4588 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4589 
4590 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
4591 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4592 
4593 	/* Read the field, zero-extended to a u64 field_value */
4594 	field_value = vmcs12_read_any(vmcs12, field, offset);
4595 
4596 	/*
4597 	 * Now copy part of this value to register or memory, as requested.
4598 	 * Note that the number of bits actually copied is 32 or 64 depending
4599 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4600 	 */
4601 	if (vmx_instruction_info & (1u << 10)) {
4602 		kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
4603 			field_value);
4604 	} else {
4605 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4606 		if (get_vmx_mem_address(vcpu, exit_qualification,
4607 				vmx_instruction_info, true, len, &gva))
4608 			return 1;
4609 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4610 		if (kvm_write_guest_virt_system(vcpu, gva, &field_value, len, &e))
4611 			kvm_inject_page_fault(vcpu, &e);
4612 	}
4613 
4614 	return nested_vmx_succeed(vcpu);
4615 }
4616 
4617 static bool is_shadow_field_rw(unsigned long field)
4618 {
4619 	switch (field) {
4620 #define SHADOW_FIELD_RW(x, y) case x:
4621 #include "vmcs_shadow_fields.h"
4622 		return true;
4623 	default:
4624 		break;
4625 	}
4626 	return false;
4627 }
4628 
4629 static bool is_shadow_field_ro(unsigned long field)
4630 {
4631 	switch (field) {
4632 #define SHADOW_FIELD_RO(x, y) case x:
4633 #include "vmcs_shadow_fields.h"
4634 		return true;
4635 	default:
4636 		break;
4637 	}
4638 	return false;
4639 }
4640 
4641 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4642 {
4643 	unsigned long field;
4644 	int len;
4645 	gva_t gva;
4646 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4647 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4648 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4649 
4650 	/* The value to write might be 32 or 64 bits, depending on L1's long
4651 	 * mode, and eventually we need to write that into a field of several
4652 	 * possible lengths. The code below first zero-extends the value to 64
4653 	 * bit (field_value), and then copies only the appropriate number of
4654 	 * bits into the vmcs12 field.
4655 	 */
4656 	u64 field_value = 0;
4657 	struct x86_exception e;
4658 	struct vmcs12 *vmcs12;
4659 	short offset;
4660 
4661 	if (!nested_vmx_check_permission(vcpu))
4662 		return 1;
4663 
4664 	if (vmx->nested.current_vmptr == -1ull)
4665 		return nested_vmx_failInvalid(vcpu);
4666 
4667 	if (vmx_instruction_info & (1u << 10))
4668 		field_value = kvm_register_readl(vcpu,
4669 			(((vmx_instruction_info) >> 3) & 0xf));
4670 	else {
4671 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4672 		if (get_vmx_mem_address(vcpu, exit_qualification,
4673 				vmx_instruction_info, false, len, &gva))
4674 			return 1;
4675 		if (kvm_read_guest_virt(vcpu, gva, &field_value, len, &e)) {
4676 			kvm_inject_page_fault(vcpu, &e);
4677 			return 1;
4678 		}
4679 	}
4680 
4681 
4682 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4683 	/*
4684 	 * If the vCPU supports "VMWRITE to any supported field in the
4685 	 * VMCS," then the "read-only" fields are actually read/write.
4686 	 */
4687 	if (vmcs_field_readonly(field) &&
4688 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4689 		return nested_vmx_failValid(vcpu,
4690 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4691 
4692 	if (!is_guest_mode(vcpu)) {
4693 		vmcs12 = get_vmcs12(vcpu);
4694 
4695 		/*
4696 		 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
4697 		 * vmcs12, else we may crush a field or consume a stale value.
4698 		 */
4699 		if (!is_shadow_field_rw(field))
4700 			copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4701 	} else {
4702 		/*
4703 		 * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
4704 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4705 		 */
4706 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4707 			return nested_vmx_failInvalid(vcpu);
4708 		vmcs12 = get_shadow_vmcs12(vcpu);
4709 	}
4710 
4711 	offset = vmcs_field_to_offset(field);
4712 	if (offset < 0)
4713 		return nested_vmx_failValid(vcpu,
4714 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4715 
4716 	/*
4717 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
4718 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
4719 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
4720 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
4721 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
4722 	 * the stripped down value, L2 sees the full value as stored by KVM).
4723 	 */
4724 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
4725 		field_value &= 0x1f0ff;
4726 
4727 	vmcs12_write_any(vmcs12, field, offset, field_value);
4728 
4729 	/*
4730 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
4731 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
4732 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
4733 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
4734 	 */
4735 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
4736 		/*
4737 		 * L1 can read these fields without exiting, ensure the
4738 		 * shadow VMCS is up-to-date.
4739 		 */
4740 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
4741 			preempt_disable();
4742 			vmcs_load(vmx->vmcs01.shadow_vmcs);
4743 
4744 			__vmcs_writel(field, field_value);
4745 
4746 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
4747 			vmcs_load(vmx->loaded_vmcs->vmcs);
4748 			preempt_enable();
4749 		}
4750 		vmx->nested.dirty_vmcs12 = true;
4751 	}
4752 
4753 	return nested_vmx_succeed(vcpu);
4754 }
4755 
4756 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
4757 {
4758 	vmx->nested.current_vmptr = vmptr;
4759 	if (enable_shadow_vmcs) {
4760 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
4761 		vmcs_write64(VMCS_LINK_POINTER,
4762 			     __pa(vmx->vmcs01.shadow_vmcs));
4763 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4764 	}
4765 	vmx->nested.dirty_vmcs12 = true;
4766 }
4767 
4768 /* Emulate the VMPTRLD instruction */
4769 static int handle_vmptrld(struct kvm_vcpu *vcpu)
4770 {
4771 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4772 	gpa_t vmptr;
4773 
4774 	if (!nested_vmx_check_permission(vcpu))
4775 		return 1;
4776 
4777 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4778 		return 1;
4779 
4780 	if (!page_address_valid(vcpu, vmptr))
4781 		return nested_vmx_failValid(vcpu,
4782 			VMXERR_VMPTRLD_INVALID_ADDRESS);
4783 
4784 	if (vmptr == vmx->nested.vmxon_ptr)
4785 		return nested_vmx_failValid(vcpu,
4786 			VMXERR_VMPTRLD_VMXON_POINTER);
4787 
4788 	/* Forbid normal VMPTRLD if Enlightened version was used */
4789 	if (vmx->nested.hv_evmcs)
4790 		return 1;
4791 
4792 	if (vmx->nested.current_vmptr != vmptr) {
4793 		struct kvm_host_map map;
4794 		struct vmcs12 *new_vmcs12;
4795 
4796 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
4797 			/*
4798 			 * Reads from an unbacked page return all 1s,
4799 			 * which means that the 32 bits located at the
4800 			 * given physical address won't match the required
4801 			 * VMCS12_REVISION identifier.
4802 			 */
4803 			return nested_vmx_failValid(vcpu,
4804 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4805 		}
4806 
4807 		new_vmcs12 = map.hva;
4808 
4809 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
4810 		    (new_vmcs12->hdr.shadow_vmcs &&
4811 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
4812 			kvm_vcpu_unmap(vcpu, &map, false);
4813 			return nested_vmx_failValid(vcpu,
4814 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4815 		}
4816 
4817 		nested_release_vmcs12(vcpu);
4818 
4819 		/*
4820 		 * Load VMCS12 from guest memory since it is not already
4821 		 * cached.
4822 		 */
4823 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
4824 		kvm_vcpu_unmap(vcpu, &map, false);
4825 
4826 		set_current_vmptr(vmx, vmptr);
4827 	}
4828 
4829 	return nested_vmx_succeed(vcpu);
4830 }
4831 
4832 /* Emulate the VMPTRST instruction */
4833 static int handle_vmptrst(struct kvm_vcpu *vcpu)
4834 {
4835 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
4836 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4837 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
4838 	struct x86_exception e;
4839 	gva_t gva;
4840 
4841 	if (!nested_vmx_check_permission(vcpu))
4842 		return 1;
4843 
4844 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
4845 		return 1;
4846 
4847 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
4848 				true, sizeof(gpa_t), &gva))
4849 		return 1;
4850 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
4851 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
4852 					sizeof(gpa_t), &e)) {
4853 		kvm_inject_page_fault(vcpu, &e);
4854 		return 1;
4855 	}
4856 	return nested_vmx_succeed(vcpu);
4857 }
4858 
4859 /* Emulate the INVEPT instruction */
4860 static int handle_invept(struct kvm_vcpu *vcpu)
4861 {
4862 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4863 	u32 vmx_instruction_info, types;
4864 	unsigned long type;
4865 	gva_t gva;
4866 	struct x86_exception e;
4867 	struct {
4868 		u64 eptp, gpa;
4869 	} operand;
4870 
4871 	if (!(vmx->nested.msrs.secondary_ctls_high &
4872 	      SECONDARY_EXEC_ENABLE_EPT) ||
4873 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
4874 		kvm_queue_exception(vcpu, UD_VECTOR);
4875 		return 1;
4876 	}
4877 
4878 	if (!nested_vmx_check_permission(vcpu))
4879 		return 1;
4880 
4881 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4882 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4883 
4884 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
4885 
4886 	if (type >= 32 || !(types & (1 << type)))
4887 		return nested_vmx_failValid(vcpu,
4888 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4889 
4890 	/* According to the Intel VMX instruction reference, the memory
4891 	 * operand is read even if it isn't needed (e.g., for type==global)
4892 	 */
4893 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4894 			vmx_instruction_info, false, sizeof(operand), &gva))
4895 		return 1;
4896 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4897 		kvm_inject_page_fault(vcpu, &e);
4898 		return 1;
4899 	}
4900 
4901 	switch (type) {
4902 	case VMX_EPT_EXTENT_GLOBAL:
4903 	case VMX_EPT_EXTENT_CONTEXT:
4904 	/*
4905 	 * TODO: Sync the necessary shadow EPT roots here, rather than
4906 	 * at the next emulated VM-entry.
4907 	 */
4908 		break;
4909 	default:
4910 		BUG_ON(1);
4911 		break;
4912 	}
4913 
4914 	return nested_vmx_succeed(vcpu);
4915 }
4916 
4917 static int handle_invvpid(struct kvm_vcpu *vcpu)
4918 {
4919 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4920 	u32 vmx_instruction_info;
4921 	unsigned long type, types;
4922 	gva_t gva;
4923 	struct x86_exception e;
4924 	struct {
4925 		u64 vpid;
4926 		u64 gla;
4927 	} operand;
4928 	u16 vpid02;
4929 
4930 	if (!(vmx->nested.msrs.secondary_ctls_high &
4931 	      SECONDARY_EXEC_ENABLE_VPID) ||
4932 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
4933 		kvm_queue_exception(vcpu, UD_VECTOR);
4934 		return 1;
4935 	}
4936 
4937 	if (!nested_vmx_check_permission(vcpu))
4938 		return 1;
4939 
4940 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4941 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4942 
4943 	types = (vmx->nested.msrs.vpid_caps &
4944 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
4945 
4946 	if (type >= 32 || !(types & (1 << type)))
4947 		return nested_vmx_failValid(vcpu,
4948 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4949 
4950 	/* according to the intel vmx instruction reference, the memory
4951 	 * operand is read even if it isn't needed (e.g., for type==global)
4952 	 */
4953 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4954 			vmx_instruction_info, false, sizeof(operand), &gva))
4955 		return 1;
4956 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4957 		kvm_inject_page_fault(vcpu, &e);
4958 		return 1;
4959 	}
4960 	if (operand.vpid >> 16)
4961 		return nested_vmx_failValid(vcpu,
4962 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4963 
4964 	vpid02 = nested_get_vpid02(vcpu);
4965 	switch (type) {
4966 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
4967 		if (!operand.vpid ||
4968 		    is_noncanonical_address(operand.gla, vcpu))
4969 			return nested_vmx_failValid(vcpu,
4970 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4971 		if (cpu_has_vmx_invvpid_individual_addr()) {
4972 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
4973 				vpid02, operand.gla);
4974 		} else
4975 			__vmx_flush_tlb(vcpu, vpid02, false);
4976 		break;
4977 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
4978 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
4979 		if (!operand.vpid)
4980 			return nested_vmx_failValid(vcpu,
4981 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4982 		__vmx_flush_tlb(vcpu, vpid02, false);
4983 		break;
4984 	case VMX_VPID_EXTENT_ALL_CONTEXT:
4985 		__vmx_flush_tlb(vcpu, vpid02, false);
4986 		break;
4987 	default:
4988 		WARN_ON_ONCE(1);
4989 		return kvm_skip_emulated_instruction(vcpu);
4990 	}
4991 
4992 	return nested_vmx_succeed(vcpu);
4993 }
4994 
4995 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
4996 				     struct vmcs12 *vmcs12)
4997 {
4998 	u32 index = kvm_rcx_read(vcpu);
4999 	u64 address;
5000 	bool accessed_dirty;
5001 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
5002 
5003 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
5004 	    !nested_cpu_has_ept(vmcs12))
5005 		return 1;
5006 
5007 	if (index >= VMFUNC_EPTP_ENTRIES)
5008 		return 1;
5009 
5010 
5011 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5012 				     &address, index * 8, 8))
5013 		return 1;
5014 
5015 	accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
5016 
5017 	/*
5018 	 * If the (L2) guest does a vmfunc to the currently
5019 	 * active ept pointer, we don't have to do anything else
5020 	 */
5021 	if (vmcs12->ept_pointer != address) {
5022 		if (!valid_ept_address(vcpu, address))
5023 			return 1;
5024 
5025 		kvm_mmu_unload(vcpu);
5026 		mmu->ept_ad = accessed_dirty;
5027 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
5028 		vmcs12->ept_pointer = address;
5029 		/*
5030 		 * TODO: Check what's the correct approach in case
5031 		 * mmu reload fails. Currently, we just let the next
5032 		 * reload potentially fail
5033 		 */
5034 		kvm_mmu_reload(vcpu);
5035 	}
5036 
5037 	return 0;
5038 }
5039 
5040 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5041 {
5042 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5043 	struct vmcs12 *vmcs12;
5044 	u32 function = kvm_rax_read(vcpu);
5045 
5046 	/*
5047 	 * VMFUNC is only supported for nested guests, but we always enable the
5048 	 * secondary control for simplicity; for non-nested mode, fake that we
5049 	 * didn't by injecting #UD.
5050 	 */
5051 	if (!is_guest_mode(vcpu)) {
5052 		kvm_queue_exception(vcpu, UD_VECTOR);
5053 		return 1;
5054 	}
5055 
5056 	vmcs12 = get_vmcs12(vcpu);
5057 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
5058 		goto fail;
5059 
5060 	switch (function) {
5061 	case 0:
5062 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5063 			goto fail;
5064 		break;
5065 	default:
5066 		goto fail;
5067 	}
5068 	return kvm_skip_emulated_instruction(vcpu);
5069 
5070 fail:
5071 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
5072 			  vmcs_read32(VM_EXIT_INTR_INFO),
5073 			  vmcs_readl(EXIT_QUALIFICATION));
5074 	return 1;
5075 }
5076 
5077 
5078 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5079 				       struct vmcs12 *vmcs12)
5080 {
5081 	unsigned long exit_qualification;
5082 	gpa_t bitmap, last_bitmap;
5083 	unsigned int port;
5084 	int size;
5085 	u8 b;
5086 
5087 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5088 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5089 
5090 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5091 
5092 	port = exit_qualification >> 16;
5093 	size = (exit_qualification & 7) + 1;
5094 
5095 	last_bitmap = (gpa_t)-1;
5096 	b = -1;
5097 
5098 	while (size > 0) {
5099 		if (port < 0x8000)
5100 			bitmap = vmcs12->io_bitmap_a;
5101 		else if (port < 0x10000)
5102 			bitmap = vmcs12->io_bitmap_b;
5103 		else
5104 			return true;
5105 		bitmap += (port & 0x7fff) / 8;
5106 
5107 		if (last_bitmap != bitmap)
5108 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5109 				return true;
5110 		if (b & (1 << (port & 7)))
5111 			return true;
5112 
5113 		port++;
5114 		size--;
5115 		last_bitmap = bitmap;
5116 	}
5117 
5118 	return false;
5119 }
5120 
5121 /*
5122  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5123  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5124  * disinterest in the current event (read or write a specific MSR) by using an
5125  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5126  */
5127 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5128 	struct vmcs12 *vmcs12, u32 exit_reason)
5129 {
5130 	u32 msr_index = kvm_rcx_read(vcpu);
5131 	gpa_t bitmap;
5132 
5133 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5134 		return true;
5135 
5136 	/*
5137 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5138 	 * for the four combinations of read/write and low/high MSR numbers.
5139 	 * First we need to figure out which of the four to use:
5140 	 */
5141 	bitmap = vmcs12->msr_bitmap;
5142 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5143 		bitmap += 2048;
5144 	if (msr_index >= 0xc0000000) {
5145 		msr_index -= 0xc0000000;
5146 		bitmap += 1024;
5147 	}
5148 
5149 	/* Then read the msr_index'th bit from this bitmap: */
5150 	if (msr_index < 1024*8) {
5151 		unsigned char b;
5152 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5153 			return true;
5154 		return 1 & (b >> (msr_index & 7));
5155 	} else
5156 		return true; /* let L1 handle the wrong parameter */
5157 }
5158 
5159 /*
5160  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5161  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5162  * intercept (via guest_host_mask etc.) the current event.
5163  */
5164 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5165 	struct vmcs12 *vmcs12)
5166 {
5167 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5168 	int cr = exit_qualification & 15;
5169 	int reg;
5170 	unsigned long val;
5171 
5172 	switch ((exit_qualification >> 4) & 3) {
5173 	case 0: /* mov to cr */
5174 		reg = (exit_qualification >> 8) & 15;
5175 		val = kvm_register_readl(vcpu, reg);
5176 		switch (cr) {
5177 		case 0:
5178 			if (vmcs12->cr0_guest_host_mask &
5179 			    (val ^ vmcs12->cr0_read_shadow))
5180 				return true;
5181 			break;
5182 		case 3:
5183 			if ((vmcs12->cr3_target_count >= 1 &&
5184 					vmcs12->cr3_target_value0 == val) ||
5185 				(vmcs12->cr3_target_count >= 2 &&
5186 					vmcs12->cr3_target_value1 == val) ||
5187 				(vmcs12->cr3_target_count >= 3 &&
5188 					vmcs12->cr3_target_value2 == val) ||
5189 				(vmcs12->cr3_target_count >= 4 &&
5190 					vmcs12->cr3_target_value3 == val))
5191 				return false;
5192 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5193 				return true;
5194 			break;
5195 		case 4:
5196 			if (vmcs12->cr4_guest_host_mask &
5197 			    (vmcs12->cr4_read_shadow ^ val))
5198 				return true;
5199 			break;
5200 		case 8:
5201 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5202 				return true;
5203 			break;
5204 		}
5205 		break;
5206 	case 2: /* clts */
5207 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5208 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5209 			return true;
5210 		break;
5211 	case 1: /* mov from cr */
5212 		switch (cr) {
5213 		case 3:
5214 			if (vmcs12->cpu_based_vm_exec_control &
5215 			    CPU_BASED_CR3_STORE_EXITING)
5216 				return true;
5217 			break;
5218 		case 8:
5219 			if (vmcs12->cpu_based_vm_exec_control &
5220 			    CPU_BASED_CR8_STORE_EXITING)
5221 				return true;
5222 			break;
5223 		}
5224 		break;
5225 	case 3: /* lmsw */
5226 		/*
5227 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5228 		 * cr0. Other attempted changes are ignored, with no exit.
5229 		 */
5230 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5231 		if (vmcs12->cr0_guest_host_mask & 0xe &
5232 		    (val ^ vmcs12->cr0_read_shadow))
5233 			return true;
5234 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5235 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5236 		    (val & 0x1))
5237 			return true;
5238 		break;
5239 	}
5240 	return false;
5241 }
5242 
5243 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5244 	struct vmcs12 *vmcs12, gpa_t bitmap)
5245 {
5246 	u32 vmx_instruction_info;
5247 	unsigned long field;
5248 	u8 b;
5249 
5250 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5251 		return true;
5252 
5253 	/* Decode instruction info and find the field to access */
5254 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5255 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5256 
5257 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5258 	if (field >> 15)
5259 		return true;
5260 
5261 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5262 		return true;
5263 
5264 	return 1 & (b >> (field & 7));
5265 }
5266 
5267 /*
5268  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5269  * should handle it ourselves in L0 (and then continue L2). Only call this
5270  * when in is_guest_mode (L2).
5271  */
5272 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5273 {
5274 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5275 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5276 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5277 
5278 	if (vmx->nested.nested_run_pending)
5279 		return false;
5280 
5281 	if (unlikely(vmx->fail)) {
5282 		trace_kvm_nested_vmenter_failed(
5283 			"hardware VM-instruction error: ",
5284 			vmcs_read32(VM_INSTRUCTION_ERROR));
5285 		return true;
5286 	}
5287 
5288 	/*
5289 	 * The host physical addresses of some pages of guest memory
5290 	 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5291 	 * Page). The CPU may write to these pages via their host
5292 	 * physical address while L2 is running, bypassing any
5293 	 * address-translation-based dirty tracking (e.g. EPT write
5294 	 * protection).
5295 	 *
5296 	 * Mark them dirty on every exit from L2 to prevent them from
5297 	 * getting out of sync with dirty tracking.
5298 	 */
5299 	nested_mark_vmcs12_pages_dirty(vcpu);
5300 
5301 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5302 				vmcs_readl(EXIT_QUALIFICATION),
5303 				vmx->idt_vectoring_info,
5304 				intr_info,
5305 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5306 				KVM_ISA_VMX);
5307 
5308 	switch (exit_reason) {
5309 	case EXIT_REASON_EXCEPTION_NMI:
5310 		if (is_nmi(intr_info))
5311 			return false;
5312 		else if (is_page_fault(intr_info))
5313 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5314 		else if (is_debug(intr_info) &&
5315 			 vcpu->guest_debug &
5316 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5317 			return false;
5318 		else if (is_breakpoint(intr_info) &&
5319 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5320 			return false;
5321 		return vmcs12->exception_bitmap &
5322 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5323 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5324 		return false;
5325 	case EXIT_REASON_TRIPLE_FAULT:
5326 		return true;
5327 	case EXIT_REASON_PENDING_INTERRUPT:
5328 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
5329 	case EXIT_REASON_NMI_WINDOW:
5330 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
5331 	case EXIT_REASON_TASK_SWITCH:
5332 		return true;
5333 	case EXIT_REASON_CPUID:
5334 		return true;
5335 	case EXIT_REASON_HLT:
5336 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5337 	case EXIT_REASON_INVD:
5338 		return true;
5339 	case EXIT_REASON_INVLPG:
5340 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5341 	case EXIT_REASON_RDPMC:
5342 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5343 	case EXIT_REASON_RDRAND:
5344 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5345 	case EXIT_REASON_RDSEED:
5346 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5347 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5348 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5349 	case EXIT_REASON_VMREAD:
5350 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5351 			vmcs12->vmread_bitmap);
5352 	case EXIT_REASON_VMWRITE:
5353 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5354 			vmcs12->vmwrite_bitmap);
5355 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5356 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5357 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5358 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5359 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5360 		/*
5361 		 * VMX instructions trap unconditionally. This allows L1 to
5362 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5363 		 */
5364 		return true;
5365 	case EXIT_REASON_CR_ACCESS:
5366 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5367 	case EXIT_REASON_DR_ACCESS:
5368 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5369 	case EXIT_REASON_IO_INSTRUCTION:
5370 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5371 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5372 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5373 	case EXIT_REASON_MSR_READ:
5374 	case EXIT_REASON_MSR_WRITE:
5375 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5376 	case EXIT_REASON_INVALID_STATE:
5377 		return true;
5378 	case EXIT_REASON_MWAIT_INSTRUCTION:
5379 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5380 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5381 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
5382 	case EXIT_REASON_MONITOR_INSTRUCTION:
5383 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5384 	case EXIT_REASON_PAUSE_INSTRUCTION:
5385 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5386 			nested_cpu_has2(vmcs12,
5387 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5388 	case EXIT_REASON_MCE_DURING_VMENTRY:
5389 		return false;
5390 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5391 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5392 	case EXIT_REASON_APIC_ACCESS:
5393 	case EXIT_REASON_APIC_WRITE:
5394 	case EXIT_REASON_EOI_INDUCED:
5395 		/*
5396 		 * The controls for "virtualize APIC accesses," "APIC-
5397 		 * register virtualization," and "virtual-interrupt
5398 		 * delivery" only come from vmcs12.
5399 		 */
5400 		return true;
5401 	case EXIT_REASON_EPT_VIOLATION:
5402 		/*
5403 		 * L0 always deals with the EPT violation. If nested EPT is
5404 		 * used, and the nested mmu code discovers that the address is
5405 		 * missing in the guest EPT table (EPT12), the EPT violation
5406 		 * will be injected with nested_ept_inject_page_fault()
5407 		 */
5408 		return false;
5409 	case EXIT_REASON_EPT_MISCONFIG:
5410 		/*
5411 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5412 		 * table (shadow on EPT) or a merged EPT table that L0 built
5413 		 * (EPT on EPT). So any problems with the structure of the
5414 		 * table is L0's fault.
5415 		 */
5416 		return false;
5417 	case EXIT_REASON_INVPCID:
5418 		return
5419 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5420 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5421 	case EXIT_REASON_WBINVD:
5422 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5423 	case EXIT_REASON_XSETBV:
5424 		return true;
5425 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5426 		/*
5427 		 * This should never happen, since it is not possible to
5428 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5429 		 * If if it were, XSS would have to be checked against
5430 		 * the XSS exit bitmap in vmcs12.
5431 		 */
5432 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5433 	case EXIT_REASON_PREEMPTION_TIMER:
5434 		return false;
5435 	case EXIT_REASON_PML_FULL:
5436 		/* We emulate PML support to L1. */
5437 		return false;
5438 	case EXIT_REASON_VMFUNC:
5439 		/* VM functions are emulated through L2->L0 vmexits. */
5440 		return false;
5441 	case EXIT_REASON_ENCLS:
5442 		/* SGX is never exposed to L1 */
5443 		return false;
5444 	default:
5445 		return true;
5446 	}
5447 }
5448 
5449 
5450 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5451 				struct kvm_nested_state __user *user_kvm_nested_state,
5452 				u32 user_data_size)
5453 {
5454 	struct vcpu_vmx *vmx;
5455 	struct vmcs12 *vmcs12;
5456 	struct kvm_nested_state kvm_state = {
5457 		.flags = 0,
5458 		.format = KVM_STATE_NESTED_FORMAT_VMX,
5459 		.size = sizeof(kvm_state),
5460 		.hdr.vmx.vmxon_pa = -1ull,
5461 		.hdr.vmx.vmcs12_pa = -1ull,
5462 	};
5463 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5464 		&user_kvm_nested_state->data.vmx[0];
5465 
5466 	if (!vcpu)
5467 		return kvm_state.size + sizeof(*user_vmx_nested_state);
5468 
5469 	vmx = to_vmx(vcpu);
5470 	vmcs12 = get_vmcs12(vcpu);
5471 
5472 	if (nested_vmx_allowed(vcpu) &&
5473 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5474 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5475 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
5476 
5477 		if (vmx_has_valid_vmcs12(vcpu)) {
5478 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
5479 
5480 			if (vmx->nested.hv_evmcs)
5481 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5482 
5483 			if (is_guest_mode(vcpu) &&
5484 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5485 			    vmcs12->vmcs_link_pointer != -1ull)
5486 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
5487 		}
5488 
5489 		if (vmx->nested.smm.vmxon)
5490 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5491 
5492 		if (vmx->nested.smm.guest_mode)
5493 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5494 
5495 		if (is_guest_mode(vcpu)) {
5496 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5497 
5498 			if (vmx->nested.nested_run_pending)
5499 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5500 		}
5501 	}
5502 
5503 	if (user_data_size < kvm_state.size)
5504 		goto out;
5505 
5506 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5507 		return -EFAULT;
5508 
5509 	if (!vmx_has_valid_vmcs12(vcpu))
5510 		goto out;
5511 
5512 	/*
5513 	 * When running L2, the authoritative vmcs12 state is in the
5514 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5515 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5516 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
5517 	 * vmcs12 state is in the vmcs12 already.
5518 	 */
5519 	if (is_guest_mode(vcpu)) {
5520 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
5521 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5522 	} else if (!vmx->nested.need_vmcs12_to_shadow_sync) {
5523 		if (vmx->nested.hv_evmcs)
5524 			copy_enlightened_to_vmcs12(vmx);
5525 		else if (enable_shadow_vmcs)
5526 			copy_shadow_to_vmcs12(vmx);
5527 	}
5528 
5529 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
5530 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
5531 
5532 	/*
5533 	 * Copy over the full allocated size of vmcs12 rather than just the size
5534 	 * of the struct.
5535 	 */
5536 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
5537 		return -EFAULT;
5538 
5539 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5540 	    vmcs12->vmcs_link_pointer != -1ull) {
5541 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
5542 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
5543 			return -EFAULT;
5544 	}
5545 
5546 out:
5547 	return kvm_state.size;
5548 }
5549 
5550 /*
5551  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5552  */
5553 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5554 {
5555 	if (is_guest_mode(vcpu)) {
5556 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5557 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5558 	}
5559 	free_nested(vcpu);
5560 }
5561 
5562 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5563 				struct kvm_nested_state __user *user_kvm_nested_state,
5564 				struct kvm_nested_state *kvm_state)
5565 {
5566 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5567 	struct vmcs12 *vmcs12;
5568 	u32 exit_qual;
5569 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5570 		&user_kvm_nested_state->data.vmx[0];
5571 	int ret;
5572 
5573 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
5574 		return -EINVAL;
5575 
5576 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
5577 		if (kvm_state->hdr.vmx.smm.flags)
5578 			return -EINVAL;
5579 
5580 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
5581 			return -EINVAL;
5582 
5583 		/*
5584 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
5585 		 * enable eVMCS capability on vCPU. However, since then
5586 		 * code was changed such that flag signals vmcs12 should
5587 		 * be copied into eVMCS in guest memory.
5588 		 *
5589 		 * To preserve backwards compatability, allow user
5590 		 * to set this flag even when there is no VMXON region.
5591 		 */
5592 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
5593 			return -EINVAL;
5594 	} else {
5595 		if (!nested_vmx_allowed(vcpu))
5596 			return -EINVAL;
5597 
5598 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
5599 			return -EINVAL;
5600 	}
5601 
5602 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5603 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5604 		return -EINVAL;
5605 
5606 	if (kvm_state->hdr.vmx.smm.flags &
5607 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5608 		return -EINVAL;
5609 
5610 	/*
5611 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5612 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5613 	 * must be zero.
5614 	 */
5615 	if (is_smm(vcpu) ?
5616 		(kvm_state->flags &
5617 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
5618 		: kvm_state->hdr.vmx.smm.flags)
5619 		return -EINVAL;
5620 
5621 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5622 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5623 		return -EINVAL;
5624 
5625 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
5626 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
5627 			return -EINVAL;
5628 
5629 	vmx_leave_nested(vcpu);
5630 
5631 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
5632 		return 0;
5633 
5634 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
5635 	ret = enter_vmx_operation(vcpu);
5636 	if (ret)
5637 		return ret;
5638 
5639 	/* Empty 'VMXON' state is permitted */
5640 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12))
5641 		return 0;
5642 
5643 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
5644 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
5645 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
5646 			return -EINVAL;
5647 
5648 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
5649 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5650 		/*
5651 		 * Sync eVMCS upon entry as we may not have
5652 		 * HV_X64_MSR_VP_ASSIST_PAGE set up yet.
5653 		 */
5654 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5655 	} else {
5656 		return -EINVAL;
5657 	}
5658 
5659 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5660 		vmx->nested.smm.vmxon = true;
5661 		vmx->nested.vmxon = false;
5662 
5663 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5664 			vmx->nested.smm.guest_mode = true;
5665 	}
5666 
5667 	vmcs12 = get_vmcs12(vcpu);
5668 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
5669 		return -EFAULT;
5670 
5671 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5672 		return -EINVAL;
5673 
5674 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5675 		return 0;
5676 
5677 	vmx->nested.nested_run_pending =
5678 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5679 
5680 	ret = -EINVAL;
5681 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5682 	    vmcs12->vmcs_link_pointer != -1ull) {
5683 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5684 
5685 		if (kvm_state->size <
5686 		    sizeof(*kvm_state) +
5687 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
5688 			goto error_guest_mode;
5689 
5690 		if (copy_from_user(shadow_vmcs12,
5691 				   user_vmx_nested_state->shadow_vmcs12,
5692 				   sizeof(*shadow_vmcs12))) {
5693 			ret = -EFAULT;
5694 			goto error_guest_mode;
5695 		}
5696 
5697 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5698 		    !shadow_vmcs12->hdr.shadow_vmcs)
5699 			goto error_guest_mode;
5700 	}
5701 
5702 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
5703 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
5704 	    nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
5705 		goto error_guest_mode;
5706 
5707 	vmx->nested.dirty_vmcs12 = true;
5708 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5709 	if (ret)
5710 		goto error_guest_mode;
5711 
5712 	return 0;
5713 
5714 error_guest_mode:
5715 	vmx->nested.nested_run_pending = 0;
5716 	return ret;
5717 }
5718 
5719 void nested_vmx_vcpu_setup(void)
5720 {
5721 	if (enable_shadow_vmcs) {
5722 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5723 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5724 	}
5725 }
5726 
5727 /*
5728  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5729  * returned for the various VMX controls MSRs when nested VMX is enabled.
5730  * The same values should also be used to verify that vmcs12 control fields are
5731  * valid during nested entry from L1 to L2.
5732  * Each of these control msrs has a low and high 32-bit half: A low bit is on
5733  * if the corresponding bit in the (32-bit) control field *must* be on, and a
5734  * bit in the high half is on if the corresponding bit in the control field
5735  * may be on. See also vmx_control_verify().
5736  */
5737 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps,
5738 				bool apicv)
5739 {
5740 	/*
5741 	 * Note that as a general rule, the high half of the MSRs (bits in
5742 	 * the control fields which may be 1) should be initialized by the
5743 	 * intersection of the underlying hardware's MSR (i.e., features which
5744 	 * can be supported) and the list of features we want to expose -
5745 	 * because they are known to be properly supported in our code.
5746 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
5747 	 * be set to 0, meaning that L1 may turn off any of these bits. The
5748 	 * reason is that if one of these bits is necessary, it will appear
5749 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
5750 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
5751 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
5752 	 * These rules have exceptions below.
5753 	 */
5754 
5755 	/* pin-based controls */
5756 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
5757 		msrs->pinbased_ctls_low,
5758 		msrs->pinbased_ctls_high);
5759 	msrs->pinbased_ctls_low |=
5760 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5761 	msrs->pinbased_ctls_high &=
5762 		PIN_BASED_EXT_INTR_MASK |
5763 		PIN_BASED_NMI_EXITING |
5764 		PIN_BASED_VIRTUAL_NMIS |
5765 		(apicv ? PIN_BASED_POSTED_INTR : 0);
5766 	msrs->pinbased_ctls_high |=
5767 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5768 		PIN_BASED_VMX_PREEMPTION_TIMER;
5769 
5770 	/* exit controls */
5771 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
5772 		msrs->exit_ctls_low,
5773 		msrs->exit_ctls_high);
5774 	msrs->exit_ctls_low =
5775 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
5776 
5777 	msrs->exit_ctls_high &=
5778 #ifdef CONFIG_X86_64
5779 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
5780 #endif
5781 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
5782 	msrs->exit_ctls_high |=
5783 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
5784 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
5785 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
5786 
5787 	/* We support free control of debug control saving. */
5788 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
5789 
5790 	/* entry controls */
5791 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
5792 		msrs->entry_ctls_low,
5793 		msrs->entry_ctls_high);
5794 	msrs->entry_ctls_low =
5795 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
5796 	msrs->entry_ctls_high &=
5797 #ifdef CONFIG_X86_64
5798 		VM_ENTRY_IA32E_MODE |
5799 #endif
5800 		VM_ENTRY_LOAD_IA32_PAT;
5801 	msrs->entry_ctls_high |=
5802 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
5803 
5804 	/* We support free control of debug control loading. */
5805 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
5806 
5807 	/* cpu-based controls */
5808 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
5809 		msrs->procbased_ctls_low,
5810 		msrs->procbased_ctls_high);
5811 	msrs->procbased_ctls_low =
5812 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5813 	msrs->procbased_ctls_high &=
5814 		CPU_BASED_VIRTUAL_INTR_PENDING |
5815 		CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
5816 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
5817 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
5818 		CPU_BASED_CR3_STORE_EXITING |
5819 #ifdef CONFIG_X86_64
5820 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
5821 #endif
5822 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
5823 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
5824 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
5825 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
5826 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
5827 	/*
5828 	 * We can allow some features even when not supported by the
5829 	 * hardware. For example, L1 can specify an MSR bitmap - and we
5830 	 * can use it to avoid exits to L1 - even when L0 runs L2
5831 	 * without MSR bitmaps.
5832 	 */
5833 	msrs->procbased_ctls_high |=
5834 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5835 		CPU_BASED_USE_MSR_BITMAPS;
5836 
5837 	/* We support free control of CR3 access interception. */
5838 	msrs->procbased_ctls_low &=
5839 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
5840 
5841 	/*
5842 	 * secondary cpu-based controls.  Do not include those that
5843 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
5844 	 */
5845 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
5846 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
5847 		      msrs->secondary_ctls_low,
5848 		      msrs->secondary_ctls_high);
5849 
5850 	msrs->secondary_ctls_low = 0;
5851 	msrs->secondary_ctls_high &=
5852 		SECONDARY_EXEC_DESC |
5853 		SECONDARY_EXEC_RDTSCP |
5854 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
5855 		SECONDARY_EXEC_WBINVD_EXITING |
5856 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
5857 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
5858 		SECONDARY_EXEC_RDRAND_EXITING |
5859 		SECONDARY_EXEC_ENABLE_INVPCID |
5860 		SECONDARY_EXEC_RDSEED_EXITING |
5861 		SECONDARY_EXEC_XSAVES;
5862 
5863 	/*
5864 	 * We can emulate "VMCS shadowing," even if the hardware
5865 	 * doesn't support it.
5866 	 */
5867 	msrs->secondary_ctls_high |=
5868 		SECONDARY_EXEC_SHADOW_VMCS;
5869 
5870 	if (enable_ept) {
5871 		/* nested EPT: emulate EPT also to L1 */
5872 		msrs->secondary_ctls_high |=
5873 			SECONDARY_EXEC_ENABLE_EPT;
5874 		msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
5875 			 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
5876 		if (cpu_has_vmx_ept_execute_only())
5877 			msrs->ept_caps |=
5878 				VMX_EPT_EXECUTE_ONLY_BIT;
5879 		msrs->ept_caps &= ept_caps;
5880 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
5881 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
5882 			VMX_EPT_1GB_PAGE_BIT;
5883 		if (enable_ept_ad_bits) {
5884 			msrs->secondary_ctls_high |=
5885 				SECONDARY_EXEC_ENABLE_PML;
5886 			msrs->ept_caps |= VMX_EPT_AD_BIT;
5887 		}
5888 	}
5889 
5890 	if (cpu_has_vmx_vmfunc()) {
5891 		msrs->secondary_ctls_high |=
5892 			SECONDARY_EXEC_ENABLE_VMFUNC;
5893 		/*
5894 		 * Advertise EPTP switching unconditionally
5895 		 * since we emulate it
5896 		 */
5897 		if (enable_ept)
5898 			msrs->vmfunc_controls =
5899 				VMX_VMFUNC_EPTP_SWITCHING;
5900 	}
5901 
5902 	/*
5903 	 * Old versions of KVM use the single-context version without
5904 	 * checking for support, so declare that it is supported even
5905 	 * though it is treated as global context.  The alternative is
5906 	 * not failing the single-context invvpid, and it is worse.
5907 	 */
5908 	if (enable_vpid) {
5909 		msrs->secondary_ctls_high |=
5910 			SECONDARY_EXEC_ENABLE_VPID;
5911 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
5912 			VMX_VPID_EXTENT_SUPPORTED_MASK;
5913 	}
5914 
5915 	if (enable_unrestricted_guest)
5916 		msrs->secondary_ctls_high |=
5917 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
5918 
5919 	if (flexpriority_enabled)
5920 		msrs->secondary_ctls_high |=
5921 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5922 
5923 	/* miscellaneous data */
5924 	rdmsr(MSR_IA32_VMX_MISC,
5925 		msrs->misc_low,
5926 		msrs->misc_high);
5927 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
5928 	msrs->misc_low |=
5929 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
5930 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
5931 		VMX_MISC_ACTIVITY_HLT;
5932 	msrs->misc_high = 0;
5933 
5934 	/*
5935 	 * This MSR reports some information about VMX support. We
5936 	 * should return information about the VMX we emulate for the
5937 	 * guest, and the VMCS structure we give it - not about the
5938 	 * VMX support of the underlying hardware.
5939 	 */
5940 	msrs->basic =
5941 		VMCS12_REVISION |
5942 		VMX_BASIC_TRUE_CTLS |
5943 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
5944 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
5945 
5946 	if (cpu_has_vmx_basic_inout())
5947 		msrs->basic |= VMX_BASIC_INOUT;
5948 
5949 	/*
5950 	 * These MSRs specify bits which the guest must keep fixed on
5951 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
5952 	 * We picked the standard core2 setting.
5953 	 */
5954 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
5955 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
5956 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
5957 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
5958 
5959 	/* These MSRs specify bits which the guest must keep fixed off. */
5960 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
5961 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
5962 
5963 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
5964 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
5965 }
5966 
5967 void nested_vmx_hardware_unsetup(void)
5968 {
5969 	int i;
5970 
5971 	if (enable_shadow_vmcs) {
5972 		for (i = 0; i < VMX_BITMAP_NR; i++)
5973 			free_page((unsigned long)vmx_bitmap[i]);
5974 	}
5975 }
5976 
5977 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
5978 {
5979 	int i;
5980 
5981 	if (!cpu_has_vmx_shadow_vmcs())
5982 		enable_shadow_vmcs = 0;
5983 	if (enable_shadow_vmcs) {
5984 		for (i = 0; i < VMX_BITMAP_NR; i++) {
5985 			/*
5986 			 * The vmx_bitmap is not tied to a VM and so should
5987 			 * not be charged to a memcg.
5988 			 */
5989 			vmx_bitmap[i] = (unsigned long *)
5990 				__get_free_page(GFP_KERNEL);
5991 			if (!vmx_bitmap[i]) {
5992 				nested_vmx_hardware_unsetup();
5993 				return -ENOMEM;
5994 			}
5995 		}
5996 
5997 		init_vmcs_shadow_fields();
5998 	}
5999 
6000 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear,
6001 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch,
6002 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld,
6003 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst,
6004 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread,
6005 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume,
6006 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite,
6007 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff,
6008 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon,
6009 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept,
6010 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid,
6011 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc,
6012 
6013 	kvm_x86_ops->check_nested_events = vmx_check_nested_events;
6014 	kvm_x86_ops->get_nested_state = vmx_get_nested_state;
6015 	kvm_x86_ops->set_nested_state = vmx_set_nested_state;
6016 	kvm_x86_ops->get_vmcs12_pages = nested_get_vmcs12_pages,
6017 	kvm_x86_ops->nested_enable_evmcs = nested_enable_evmcs;
6018 	kvm_x86_ops->nested_get_evmcs_version = nested_get_evmcs_version;
6019 
6020 	return 0;
6021 }
6022