xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 15e3ae36)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/frame.h>
4 #include <linux/percpu.h>
5 
6 #include <asm/debugreg.h>
7 #include <asm/mmu_context.h>
8 
9 #include "cpuid.h"
10 #include "hyperv.h"
11 #include "mmu.h"
12 #include "nested.h"
13 #include "pmu.h"
14 #include "trace.h"
15 #include "x86.h"
16 
17 static bool __read_mostly enable_shadow_vmcs = 1;
18 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
19 
20 static bool __read_mostly nested_early_check = 0;
21 module_param(nested_early_check, bool, S_IRUGO);
22 
23 #define CC(consistency_check)						\
24 ({									\
25 	bool failed = (consistency_check);				\
26 	if (failed)							\
27 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
28 	failed;								\
29 })
30 
31 /*
32  * Hyper-V requires all of these, so mark them as supported even though
33  * they are just treated the same as all-context.
34  */
35 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
36 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
37 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
38 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
39 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
40 
41 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
42 
43 enum {
44 	VMX_VMREAD_BITMAP,
45 	VMX_VMWRITE_BITMAP,
46 	VMX_BITMAP_NR
47 };
48 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
49 
50 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
51 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
52 
53 struct shadow_vmcs_field {
54 	u16	encoding;
55 	u16	offset;
56 };
57 static struct shadow_vmcs_field shadow_read_only_fields[] = {
58 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
59 #include "vmcs_shadow_fields.h"
60 };
61 static int max_shadow_read_only_fields =
62 	ARRAY_SIZE(shadow_read_only_fields);
63 
64 static struct shadow_vmcs_field shadow_read_write_fields[] = {
65 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
66 #include "vmcs_shadow_fields.h"
67 };
68 static int max_shadow_read_write_fields =
69 	ARRAY_SIZE(shadow_read_write_fields);
70 
71 static void init_vmcs_shadow_fields(void)
72 {
73 	int i, j;
74 
75 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
76 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
77 
78 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
79 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
80 		u16 field = entry.encoding;
81 
82 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
83 		    (i + 1 == max_shadow_read_only_fields ||
84 		     shadow_read_only_fields[i + 1].encoding != field + 1))
85 			pr_err("Missing field from shadow_read_only_field %x\n",
86 			       field + 1);
87 
88 		clear_bit(field, vmx_vmread_bitmap);
89 		if (field & 1)
90 #ifdef CONFIG_X86_64
91 			continue;
92 #else
93 			entry.offset += sizeof(u32);
94 #endif
95 		shadow_read_only_fields[j++] = entry;
96 	}
97 	max_shadow_read_only_fields = j;
98 
99 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
100 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
101 		u16 field = entry.encoding;
102 
103 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
104 		    (i + 1 == max_shadow_read_write_fields ||
105 		     shadow_read_write_fields[i + 1].encoding != field + 1))
106 			pr_err("Missing field from shadow_read_write_field %x\n",
107 			       field + 1);
108 
109 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
110 			  field <= GUEST_TR_AR_BYTES,
111 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
112 
113 		/*
114 		 * PML and the preemption timer can be emulated, but the
115 		 * processor cannot vmwrite to fields that don't exist
116 		 * on bare metal.
117 		 */
118 		switch (field) {
119 		case GUEST_PML_INDEX:
120 			if (!cpu_has_vmx_pml())
121 				continue;
122 			break;
123 		case VMX_PREEMPTION_TIMER_VALUE:
124 			if (!cpu_has_vmx_preemption_timer())
125 				continue;
126 			break;
127 		case GUEST_INTR_STATUS:
128 			if (!cpu_has_vmx_apicv())
129 				continue;
130 			break;
131 		default:
132 			break;
133 		}
134 
135 		clear_bit(field, vmx_vmwrite_bitmap);
136 		clear_bit(field, vmx_vmread_bitmap);
137 		if (field & 1)
138 #ifdef CONFIG_X86_64
139 			continue;
140 #else
141 			entry.offset += sizeof(u32);
142 #endif
143 		shadow_read_write_fields[j++] = entry;
144 	}
145 	max_shadow_read_write_fields = j;
146 }
147 
148 /*
149  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
150  * set the success or error code of an emulated VMX instruction (as specified
151  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
152  * instruction.
153  */
154 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
155 {
156 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
157 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
158 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
159 	return kvm_skip_emulated_instruction(vcpu);
160 }
161 
162 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
163 {
164 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
165 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
166 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
167 			| X86_EFLAGS_CF);
168 	return kvm_skip_emulated_instruction(vcpu);
169 }
170 
171 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
172 				u32 vm_instruction_error)
173 {
174 	struct vcpu_vmx *vmx = to_vmx(vcpu);
175 
176 	/*
177 	 * failValid writes the error number to the current VMCS, which
178 	 * can't be done if there isn't a current VMCS.
179 	 */
180 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
181 		return nested_vmx_failInvalid(vcpu);
182 
183 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
184 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
185 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
186 			| X86_EFLAGS_ZF);
187 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
188 	/*
189 	 * We don't need to force a shadow sync because
190 	 * VM_INSTRUCTION_ERROR is not shadowed
191 	 */
192 	return kvm_skip_emulated_instruction(vcpu);
193 }
194 
195 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
196 {
197 	/* TODO: not to reset guest simply here. */
198 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
199 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
200 }
201 
202 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
203 {
204 	return fixed_bits_valid(control, low, high);
205 }
206 
207 static inline u64 vmx_control_msr(u32 low, u32 high)
208 {
209 	return low | ((u64)high << 32);
210 }
211 
212 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
213 {
214 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
215 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
216 	vmx->nested.need_vmcs12_to_shadow_sync = false;
217 }
218 
219 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
220 {
221 	struct vcpu_vmx *vmx = to_vmx(vcpu);
222 
223 	if (!vmx->nested.hv_evmcs)
224 		return;
225 
226 	kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
227 	vmx->nested.hv_evmcs_vmptr = 0;
228 	vmx->nested.hv_evmcs = NULL;
229 }
230 
231 /*
232  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
233  * just stops using VMX.
234  */
235 static void free_nested(struct kvm_vcpu *vcpu)
236 {
237 	struct vcpu_vmx *vmx = to_vmx(vcpu);
238 
239 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
240 		return;
241 
242 	kvm_clear_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
243 
244 	vmx->nested.vmxon = false;
245 	vmx->nested.smm.vmxon = false;
246 	free_vpid(vmx->nested.vpid02);
247 	vmx->nested.posted_intr_nv = -1;
248 	vmx->nested.current_vmptr = -1ull;
249 	if (enable_shadow_vmcs) {
250 		vmx_disable_shadow_vmcs(vmx);
251 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
252 		free_vmcs(vmx->vmcs01.shadow_vmcs);
253 		vmx->vmcs01.shadow_vmcs = NULL;
254 	}
255 	kfree(vmx->nested.cached_vmcs12);
256 	vmx->nested.cached_vmcs12 = NULL;
257 	kfree(vmx->nested.cached_shadow_vmcs12);
258 	vmx->nested.cached_shadow_vmcs12 = NULL;
259 	/* Unpin physical memory we referred to in the vmcs02 */
260 	if (vmx->nested.apic_access_page) {
261 		kvm_release_page_clean(vmx->nested.apic_access_page);
262 		vmx->nested.apic_access_page = NULL;
263 	}
264 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
265 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
266 	vmx->nested.pi_desc = NULL;
267 
268 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
269 
270 	nested_release_evmcs(vcpu);
271 
272 	free_loaded_vmcs(&vmx->nested.vmcs02);
273 }
274 
275 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
276 				     struct loaded_vmcs *prev)
277 {
278 	struct vmcs_host_state *dest, *src;
279 
280 	if (unlikely(!vmx->guest_state_loaded))
281 		return;
282 
283 	src = &prev->host_state;
284 	dest = &vmx->loaded_vmcs->host_state;
285 
286 	vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
287 	dest->ldt_sel = src->ldt_sel;
288 #ifdef CONFIG_X86_64
289 	dest->ds_sel = src->ds_sel;
290 	dest->es_sel = src->es_sel;
291 #endif
292 }
293 
294 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
295 {
296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
297 	struct loaded_vmcs *prev;
298 	int cpu;
299 
300 	if (vmx->loaded_vmcs == vmcs)
301 		return;
302 
303 	cpu = get_cpu();
304 	prev = vmx->loaded_vmcs;
305 	vmx->loaded_vmcs = vmcs;
306 	vmx_vcpu_load_vmcs(vcpu, cpu);
307 	vmx_sync_vmcs_host_state(vmx, prev);
308 	put_cpu();
309 
310 	vmx_segment_cache_clear(vmx);
311 }
312 
313 /*
314  * Ensure that the current vmcs of the logical processor is the
315  * vmcs01 of the vcpu before calling free_nested().
316  */
317 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
318 {
319 	vcpu_load(vcpu);
320 	vmx_leave_nested(vcpu);
321 	vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
322 	free_nested(vcpu);
323 	vcpu_put(vcpu);
324 }
325 
326 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
327 		struct x86_exception *fault)
328 {
329 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
331 	u32 exit_reason;
332 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
333 
334 	if (vmx->nested.pml_full) {
335 		exit_reason = EXIT_REASON_PML_FULL;
336 		vmx->nested.pml_full = false;
337 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
338 	} else if (fault->error_code & PFERR_RSVD_MASK)
339 		exit_reason = EXIT_REASON_EPT_MISCONFIG;
340 	else
341 		exit_reason = EXIT_REASON_EPT_VIOLATION;
342 
343 	nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
344 	vmcs12->guest_physical_address = fault->address;
345 }
346 
347 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
348 {
349 	WARN_ON(mmu_is_nested(vcpu));
350 
351 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
352 	kvm_init_shadow_ept_mmu(vcpu,
353 			to_vmx(vcpu)->nested.msrs.ept_caps &
354 			VMX_EPT_EXECUTE_ONLY_BIT,
355 			nested_ept_ad_enabled(vcpu),
356 			nested_ept_get_eptp(vcpu));
357 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
358 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
359 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
360 
361 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
362 }
363 
364 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
365 {
366 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
367 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
368 }
369 
370 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
371 					    u16 error_code)
372 {
373 	bool inequality, bit;
374 
375 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
376 	inequality =
377 		(error_code & vmcs12->page_fault_error_code_mask) !=
378 		 vmcs12->page_fault_error_code_match;
379 	return inequality ^ bit;
380 }
381 
382 
383 /*
384  * KVM wants to inject page-faults which it got to the guest. This function
385  * checks whether in a nested guest, we need to inject them to L1 or L2.
386  */
387 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
388 {
389 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
390 	unsigned int nr = vcpu->arch.exception.nr;
391 	bool has_payload = vcpu->arch.exception.has_payload;
392 	unsigned long payload = vcpu->arch.exception.payload;
393 
394 	if (nr == PF_VECTOR) {
395 		if (vcpu->arch.exception.nested_apf) {
396 			*exit_qual = vcpu->arch.apf.nested_apf_token;
397 			return 1;
398 		}
399 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
400 						    vcpu->arch.exception.error_code)) {
401 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
402 			return 1;
403 		}
404 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
405 		if (nr == DB_VECTOR) {
406 			if (!has_payload) {
407 				payload = vcpu->arch.dr6;
408 				payload &= ~(DR6_FIXED_1 | DR6_BT);
409 				payload ^= DR6_RTM;
410 			}
411 			*exit_qual = payload;
412 		} else
413 			*exit_qual = 0;
414 		return 1;
415 	}
416 
417 	return 0;
418 }
419 
420 
421 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
422 		struct x86_exception *fault)
423 {
424 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
425 
426 	WARN_ON(!is_guest_mode(vcpu));
427 
428 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
429 		!to_vmx(vcpu)->nested.nested_run_pending) {
430 		vmcs12->vm_exit_intr_error_code = fault->error_code;
431 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
432 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
433 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
434 				  fault->address);
435 	} else {
436 		kvm_inject_page_fault(vcpu, fault);
437 	}
438 }
439 
440 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
441 {
442 	return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
443 }
444 
445 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
446 					       struct vmcs12 *vmcs12)
447 {
448 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
449 		return 0;
450 
451 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
452 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
453 		return -EINVAL;
454 
455 	return 0;
456 }
457 
458 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
459 						struct vmcs12 *vmcs12)
460 {
461 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
462 		return 0;
463 
464 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
465 		return -EINVAL;
466 
467 	return 0;
468 }
469 
470 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
471 						struct vmcs12 *vmcs12)
472 {
473 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
474 		return 0;
475 
476 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
477 		return -EINVAL;
478 
479 	return 0;
480 }
481 
482 /*
483  * Check if MSR is intercepted for L01 MSR bitmap.
484  */
485 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
486 {
487 	unsigned long *msr_bitmap;
488 	int f = sizeof(unsigned long);
489 
490 	if (!cpu_has_vmx_msr_bitmap())
491 		return true;
492 
493 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
494 
495 	if (msr <= 0x1fff) {
496 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
497 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
498 		msr &= 0x1fff;
499 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
500 	}
501 
502 	return true;
503 }
504 
505 /*
506  * If a msr is allowed by L0, we should check whether it is allowed by L1.
507  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
508  */
509 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
510 					       unsigned long *msr_bitmap_nested,
511 					       u32 msr, int type)
512 {
513 	int f = sizeof(unsigned long);
514 
515 	/*
516 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
517 	 * have the write-low and read-high bitmap offsets the wrong way round.
518 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
519 	 */
520 	if (msr <= 0x1fff) {
521 		if (type & MSR_TYPE_R &&
522 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
523 			/* read-low */
524 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
525 
526 		if (type & MSR_TYPE_W &&
527 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
528 			/* write-low */
529 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
530 
531 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
532 		msr &= 0x1fff;
533 		if (type & MSR_TYPE_R &&
534 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
535 			/* read-high */
536 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
537 
538 		if (type & MSR_TYPE_W &&
539 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
540 			/* write-high */
541 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
542 
543 	}
544 }
545 
546 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
547 {
548 	int msr;
549 
550 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
551 		unsigned word = msr / BITS_PER_LONG;
552 
553 		msr_bitmap[word] = ~0;
554 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
555 	}
556 }
557 
558 /*
559  * Merge L0's and L1's MSR bitmap, return false to indicate that
560  * we do not use the hardware.
561  */
562 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
563 						 struct vmcs12 *vmcs12)
564 {
565 	int msr;
566 	unsigned long *msr_bitmap_l1;
567 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
568 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
569 
570 	/* Nothing to do if the MSR bitmap is not in use.  */
571 	if (!cpu_has_vmx_msr_bitmap() ||
572 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
573 		return false;
574 
575 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
576 		return false;
577 
578 	msr_bitmap_l1 = (unsigned long *)map->hva;
579 
580 	/*
581 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
582 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
583 	 * the x2APIC MSR range and selectively disable them below.
584 	 */
585 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
586 
587 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
588 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
589 			/*
590 			 * L0 need not intercept reads for MSRs between 0x800
591 			 * and 0x8ff, it just lets the processor take the value
592 			 * from the virtual-APIC page; take those 256 bits
593 			 * directly from the L1 bitmap.
594 			 */
595 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
596 				unsigned word = msr / BITS_PER_LONG;
597 
598 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
599 			}
600 		}
601 
602 		nested_vmx_disable_intercept_for_msr(
603 			msr_bitmap_l1, msr_bitmap_l0,
604 			X2APIC_MSR(APIC_TASKPRI),
605 			MSR_TYPE_R | MSR_TYPE_W);
606 
607 		if (nested_cpu_has_vid(vmcs12)) {
608 			nested_vmx_disable_intercept_for_msr(
609 				msr_bitmap_l1, msr_bitmap_l0,
610 				X2APIC_MSR(APIC_EOI),
611 				MSR_TYPE_W);
612 			nested_vmx_disable_intercept_for_msr(
613 				msr_bitmap_l1, msr_bitmap_l0,
614 				X2APIC_MSR(APIC_SELF_IPI),
615 				MSR_TYPE_W);
616 		}
617 	}
618 
619 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
620 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
621 					     MSR_FS_BASE, MSR_TYPE_RW);
622 
623 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
624 					     MSR_GS_BASE, MSR_TYPE_RW);
625 
626 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
627 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
628 
629 	/*
630 	 * Checking the L0->L1 bitmap is trying to verify two things:
631 	 *
632 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
633 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
634 	 *    from the L12 MSR bitmap that is too permissive.
635 	 * 2. That L1 or L2s have actually used the MSR. This avoids
636 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
637 	 *    works properly because we only update the L01 MSR bitmap lazily.
638 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
639 	 *    updated to reflect this when L1 (or its L2s) actually write to
640 	 *    the MSR.
641 	 */
642 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
643 		nested_vmx_disable_intercept_for_msr(
644 					msr_bitmap_l1, msr_bitmap_l0,
645 					MSR_IA32_SPEC_CTRL,
646 					MSR_TYPE_R | MSR_TYPE_W);
647 
648 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
649 		nested_vmx_disable_intercept_for_msr(
650 					msr_bitmap_l1, msr_bitmap_l0,
651 					MSR_IA32_PRED_CMD,
652 					MSR_TYPE_W);
653 
654 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
655 
656 	return true;
657 }
658 
659 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
660 				       struct vmcs12 *vmcs12)
661 {
662 	struct kvm_host_map map;
663 	struct vmcs12 *shadow;
664 
665 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
666 	    vmcs12->vmcs_link_pointer == -1ull)
667 		return;
668 
669 	shadow = get_shadow_vmcs12(vcpu);
670 
671 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
672 		return;
673 
674 	memcpy(shadow, map.hva, VMCS12_SIZE);
675 	kvm_vcpu_unmap(vcpu, &map, false);
676 }
677 
678 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
679 					      struct vmcs12 *vmcs12)
680 {
681 	struct vcpu_vmx *vmx = to_vmx(vcpu);
682 
683 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
684 	    vmcs12->vmcs_link_pointer == -1ull)
685 		return;
686 
687 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
688 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
689 }
690 
691 /*
692  * In nested virtualization, check if L1 has set
693  * VM_EXIT_ACK_INTR_ON_EXIT
694  */
695 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
696 {
697 	return get_vmcs12(vcpu)->vm_exit_controls &
698 		VM_EXIT_ACK_INTR_ON_EXIT;
699 }
700 
701 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
702 {
703 	return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
704 }
705 
706 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
707 					  struct vmcs12 *vmcs12)
708 {
709 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
710 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
711 		return -EINVAL;
712 	else
713 		return 0;
714 }
715 
716 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
717 					   struct vmcs12 *vmcs12)
718 {
719 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
720 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
721 	    !nested_cpu_has_vid(vmcs12) &&
722 	    !nested_cpu_has_posted_intr(vmcs12))
723 		return 0;
724 
725 	/*
726 	 * If virtualize x2apic mode is enabled,
727 	 * virtualize apic access must be disabled.
728 	 */
729 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
730 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
731 		return -EINVAL;
732 
733 	/*
734 	 * If virtual interrupt delivery is enabled,
735 	 * we must exit on external interrupts.
736 	 */
737 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
738 		return -EINVAL;
739 
740 	/*
741 	 * bits 15:8 should be zero in posted_intr_nv,
742 	 * the descriptor address has been already checked
743 	 * in nested_get_vmcs12_pages.
744 	 *
745 	 * bits 5:0 of posted_intr_desc_addr should be zero.
746 	 */
747 	if (nested_cpu_has_posted_intr(vmcs12) &&
748 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
749 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
750 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
751 	    CC((vmcs12->posted_intr_desc_addr & 0x3f)) ||
752 	    CC((vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu)))))
753 		return -EINVAL;
754 
755 	/* tpr shadow is needed by all apicv features. */
756 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
757 		return -EINVAL;
758 
759 	return 0;
760 }
761 
762 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
763 				       u32 count, u64 addr)
764 {
765 	int maxphyaddr;
766 
767 	if (count == 0)
768 		return 0;
769 	maxphyaddr = cpuid_maxphyaddr(vcpu);
770 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
771 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
772 		return -EINVAL;
773 
774 	return 0;
775 }
776 
777 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
778 						     struct vmcs12 *vmcs12)
779 {
780 	if (CC(nested_vmx_check_msr_switch(vcpu,
781 					   vmcs12->vm_exit_msr_load_count,
782 					   vmcs12->vm_exit_msr_load_addr)) ||
783 	    CC(nested_vmx_check_msr_switch(vcpu,
784 					   vmcs12->vm_exit_msr_store_count,
785 					   vmcs12->vm_exit_msr_store_addr)))
786 		return -EINVAL;
787 
788 	return 0;
789 }
790 
791 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
792                                                       struct vmcs12 *vmcs12)
793 {
794 	if (CC(nested_vmx_check_msr_switch(vcpu,
795 					   vmcs12->vm_entry_msr_load_count,
796 					   vmcs12->vm_entry_msr_load_addr)))
797                 return -EINVAL;
798 
799 	return 0;
800 }
801 
802 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
803 					 struct vmcs12 *vmcs12)
804 {
805 	if (!nested_cpu_has_pml(vmcs12))
806 		return 0;
807 
808 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
809 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
810 		return -EINVAL;
811 
812 	return 0;
813 }
814 
815 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
816 							struct vmcs12 *vmcs12)
817 {
818 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
819 	       !nested_cpu_has_ept(vmcs12)))
820 		return -EINVAL;
821 	return 0;
822 }
823 
824 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
825 							 struct vmcs12 *vmcs12)
826 {
827 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
828 	       !nested_cpu_has_ept(vmcs12)))
829 		return -EINVAL;
830 	return 0;
831 }
832 
833 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
834 						 struct vmcs12 *vmcs12)
835 {
836 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
837 		return 0;
838 
839 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
840 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
841 		return -EINVAL;
842 
843 	return 0;
844 }
845 
846 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
847 				       struct vmx_msr_entry *e)
848 {
849 	/* x2APIC MSR accesses are not allowed */
850 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
851 		return -EINVAL;
852 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
853 	    CC(e->index == MSR_IA32_UCODE_REV))
854 		return -EINVAL;
855 	if (CC(e->reserved != 0))
856 		return -EINVAL;
857 	return 0;
858 }
859 
860 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
861 				     struct vmx_msr_entry *e)
862 {
863 	if (CC(e->index == MSR_FS_BASE) ||
864 	    CC(e->index == MSR_GS_BASE) ||
865 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
866 	    nested_vmx_msr_check_common(vcpu, e))
867 		return -EINVAL;
868 	return 0;
869 }
870 
871 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
872 				      struct vmx_msr_entry *e)
873 {
874 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
875 	    nested_vmx_msr_check_common(vcpu, e))
876 		return -EINVAL;
877 	return 0;
878 }
879 
880 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
881 {
882 	struct vcpu_vmx *vmx = to_vmx(vcpu);
883 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
884 				       vmx->nested.msrs.misc_high);
885 
886 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
887 }
888 
889 /*
890  * Load guest's/host's msr at nested entry/exit.
891  * return 0 for success, entry index for failure.
892  *
893  * One of the failure modes for MSR load/store is when a list exceeds the
894  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
895  * as possible, process all valid entries before failing rather than precheck
896  * for a capacity violation.
897  */
898 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
899 {
900 	u32 i;
901 	struct vmx_msr_entry e;
902 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
903 
904 	for (i = 0; i < count; i++) {
905 		if (unlikely(i >= max_msr_list_size))
906 			goto fail;
907 
908 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
909 					&e, sizeof(e))) {
910 			pr_debug_ratelimited(
911 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
912 				__func__, i, gpa + i * sizeof(e));
913 			goto fail;
914 		}
915 		if (nested_vmx_load_msr_check(vcpu, &e)) {
916 			pr_debug_ratelimited(
917 				"%s check failed (%u, 0x%x, 0x%x)\n",
918 				__func__, i, e.index, e.reserved);
919 			goto fail;
920 		}
921 		if (kvm_set_msr(vcpu, e.index, e.value)) {
922 			pr_debug_ratelimited(
923 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
924 				__func__, i, e.index, e.value);
925 			goto fail;
926 		}
927 	}
928 	return 0;
929 fail:
930 	return i + 1;
931 }
932 
933 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
934 					    u32 msr_index,
935 					    u64 *data)
936 {
937 	struct vcpu_vmx *vmx = to_vmx(vcpu);
938 
939 	/*
940 	 * If the L0 hypervisor stored a more accurate value for the TSC that
941 	 * does not include the time taken for emulation of the L2->L1
942 	 * VM-exit in L0, use the more accurate value.
943 	 */
944 	if (msr_index == MSR_IA32_TSC) {
945 		int index = vmx_find_msr_index(&vmx->msr_autostore.guest,
946 					       MSR_IA32_TSC);
947 
948 		if (index >= 0) {
949 			u64 val = vmx->msr_autostore.guest.val[index].value;
950 
951 			*data = kvm_read_l1_tsc(vcpu, val);
952 			return true;
953 		}
954 	}
955 
956 	if (kvm_get_msr(vcpu, msr_index, data)) {
957 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
958 			msr_index);
959 		return false;
960 	}
961 	return true;
962 }
963 
964 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
965 				     struct vmx_msr_entry *e)
966 {
967 	if (kvm_vcpu_read_guest(vcpu,
968 				gpa + i * sizeof(*e),
969 				e, 2 * sizeof(u32))) {
970 		pr_debug_ratelimited(
971 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
972 			__func__, i, gpa + i * sizeof(*e));
973 		return false;
974 	}
975 	if (nested_vmx_store_msr_check(vcpu, e)) {
976 		pr_debug_ratelimited(
977 			"%s check failed (%u, 0x%x, 0x%x)\n",
978 			__func__, i, e->index, e->reserved);
979 		return false;
980 	}
981 	return true;
982 }
983 
984 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
985 {
986 	u64 data;
987 	u32 i;
988 	struct vmx_msr_entry e;
989 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
990 
991 	for (i = 0; i < count; i++) {
992 		if (unlikely(i >= max_msr_list_size))
993 			return -EINVAL;
994 
995 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
996 			return -EINVAL;
997 
998 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
999 			return -EINVAL;
1000 
1001 		if (kvm_vcpu_write_guest(vcpu,
1002 					 gpa + i * sizeof(e) +
1003 					     offsetof(struct vmx_msr_entry, value),
1004 					 &data, sizeof(data))) {
1005 			pr_debug_ratelimited(
1006 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1007 				__func__, i, e.index, data);
1008 			return -EINVAL;
1009 		}
1010 	}
1011 	return 0;
1012 }
1013 
1014 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1015 {
1016 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1017 	u32 count = vmcs12->vm_exit_msr_store_count;
1018 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1019 	struct vmx_msr_entry e;
1020 	u32 i;
1021 
1022 	for (i = 0; i < count; i++) {
1023 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1024 			return false;
1025 
1026 		if (e.index == msr_index)
1027 			return true;
1028 	}
1029 	return false;
1030 }
1031 
1032 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1033 					   u32 msr_index)
1034 {
1035 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1036 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1037 	bool in_vmcs12_store_list;
1038 	int msr_autostore_index;
1039 	bool in_autostore_list;
1040 	int last;
1041 
1042 	msr_autostore_index = vmx_find_msr_index(autostore, msr_index);
1043 	in_autostore_list = msr_autostore_index >= 0;
1044 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1045 
1046 	if (in_vmcs12_store_list && !in_autostore_list) {
1047 		if (autostore->nr == NR_LOADSTORE_MSRS) {
1048 			/*
1049 			 * Emulated VMEntry does not fail here.  Instead a less
1050 			 * accurate value will be returned by
1051 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1052 			 * instead of reading the value from the vmcs02 VMExit
1053 			 * MSR-store area.
1054 			 */
1055 			pr_warn_ratelimited(
1056 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1057 				msr_index);
1058 			return;
1059 		}
1060 		last = autostore->nr++;
1061 		autostore->val[last].index = msr_index;
1062 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1063 		last = --autostore->nr;
1064 		autostore->val[msr_autostore_index] = autostore->val[last];
1065 	}
1066 }
1067 
1068 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
1069 {
1070 	unsigned long invalid_mask;
1071 
1072 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1073 	return (val & invalid_mask) == 0;
1074 }
1075 
1076 /*
1077  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1078  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1079  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1080  * @entry_failure_code.
1081  */
1082 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
1083 			       u32 *entry_failure_code)
1084 {
1085 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
1086 		if (CC(!nested_cr3_valid(vcpu, cr3))) {
1087 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
1088 			return -EINVAL;
1089 		}
1090 
1091 		/*
1092 		 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1093 		 * must not be dereferenced.
1094 		 */
1095 		if (is_pae_paging(vcpu) && !nested_ept) {
1096 			if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1097 				*entry_failure_code = ENTRY_FAIL_PDPTE;
1098 				return -EINVAL;
1099 			}
1100 		}
1101 	}
1102 
1103 	if (!nested_ept)
1104 		kvm_mmu_new_cr3(vcpu, cr3, false);
1105 
1106 	vcpu->arch.cr3 = cr3;
1107 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1108 
1109 	kvm_init_mmu(vcpu, false);
1110 
1111 	return 0;
1112 }
1113 
1114 /*
1115  * Returns if KVM is able to config CPU to tag TLB entries
1116  * populated by L2 differently than TLB entries populated
1117  * by L1.
1118  *
1119  * If L0 uses EPT, L1 and L2 run with different EPTP because
1120  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1121  * are tagged with different EPTP.
1122  *
1123  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1124  * with different VPID (L1 entries are tagged with vmx->vpid
1125  * while L2 entries are tagged with vmx->nested.vpid02).
1126  */
1127 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1128 {
1129 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1130 
1131 	return enable_ept ||
1132 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1133 }
1134 
1135 static u16 nested_get_vpid02(struct kvm_vcpu *vcpu)
1136 {
1137 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1138 
1139 	return vmx->nested.vpid02 ? vmx->nested.vpid02 : vmx->vpid;
1140 }
1141 
1142 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1143 {
1144 	superset &= mask;
1145 	subset &= mask;
1146 
1147 	return (superset | subset) == superset;
1148 }
1149 
1150 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1151 {
1152 	const u64 feature_and_reserved =
1153 		/* feature (except bit 48; see below) */
1154 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1155 		/* reserved */
1156 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1157 	u64 vmx_basic = vmx->nested.msrs.basic;
1158 
1159 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1160 		return -EINVAL;
1161 
1162 	/*
1163 	 * KVM does not emulate a version of VMX that constrains physical
1164 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1165 	 */
1166 	if (data & BIT_ULL(48))
1167 		return -EINVAL;
1168 
1169 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1170 	    vmx_basic_vmcs_revision_id(data))
1171 		return -EINVAL;
1172 
1173 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1174 		return -EINVAL;
1175 
1176 	vmx->nested.msrs.basic = data;
1177 	return 0;
1178 }
1179 
1180 static int
1181 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1182 {
1183 	u64 supported;
1184 	u32 *lowp, *highp;
1185 
1186 	switch (msr_index) {
1187 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1188 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1189 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1190 		break;
1191 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1192 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1193 		highp = &vmx->nested.msrs.procbased_ctls_high;
1194 		break;
1195 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1196 		lowp = &vmx->nested.msrs.exit_ctls_low;
1197 		highp = &vmx->nested.msrs.exit_ctls_high;
1198 		break;
1199 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1200 		lowp = &vmx->nested.msrs.entry_ctls_low;
1201 		highp = &vmx->nested.msrs.entry_ctls_high;
1202 		break;
1203 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1204 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1205 		highp = &vmx->nested.msrs.secondary_ctls_high;
1206 		break;
1207 	default:
1208 		BUG();
1209 	}
1210 
1211 	supported = vmx_control_msr(*lowp, *highp);
1212 
1213 	/* Check must-be-1 bits are still 1. */
1214 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1215 		return -EINVAL;
1216 
1217 	/* Check must-be-0 bits are still 0. */
1218 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1219 		return -EINVAL;
1220 
1221 	*lowp = data;
1222 	*highp = data >> 32;
1223 	return 0;
1224 }
1225 
1226 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1227 {
1228 	const u64 feature_and_reserved_bits =
1229 		/* feature */
1230 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1231 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1232 		/* reserved */
1233 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1234 	u64 vmx_misc;
1235 
1236 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1237 				   vmx->nested.msrs.misc_high);
1238 
1239 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1240 		return -EINVAL;
1241 
1242 	if ((vmx->nested.msrs.pinbased_ctls_high &
1243 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1244 	    vmx_misc_preemption_timer_rate(data) !=
1245 	    vmx_misc_preemption_timer_rate(vmx_misc))
1246 		return -EINVAL;
1247 
1248 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1249 		return -EINVAL;
1250 
1251 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1252 		return -EINVAL;
1253 
1254 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1255 		return -EINVAL;
1256 
1257 	vmx->nested.msrs.misc_low = data;
1258 	vmx->nested.msrs.misc_high = data >> 32;
1259 
1260 	return 0;
1261 }
1262 
1263 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1264 {
1265 	u64 vmx_ept_vpid_cap;
1266 
1267 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1268 					   vmx->nested.msrs.vpid_caps);
1269 
1270 	/* Every bit is either reserved or a feature bit. */
1271 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1272 		return -EINVAL;
1273 
1274 	vmx->nested.msrs.ept_caps = data;
1275 	vmx->nested.msrs.vpid_caps = data >> 32;
1276 	return 0;
1277 }
1278 
1279 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1280 {
1281 	u64 *msr;
1282 
1283 	switch (msr_index) {
1284 	case MSR_IA32_VMX_CR0_FIXED0:
1285 		msr = &vmx->nested.msrs.cr0_fixed0;
1286 		break;
1287 	case MSR_IA32_VMX_CR4_FIXED0:
1288 		msr = &vmx->nested.msrs.cr4_fixed0;
1289 		break;
1290 	default:
1291 		BUG();
1292 	}
1293 
1294 	/*
1295 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1296 	 * must be 1 in the restored value.
1297 	 */
1298 	if (!is_bitwise_subset(data, *msr, -1ULL))
1299 		return -EINVAL;
1300 
1301 	*msr = data;
1302 	return 0;
1303 }
1304 
1305 /*
1306  * Called when userspace is restoring VMX MSRs.
1307  *
1308  * Returns 0 on success, non-0 otherwise.
1309  */
1310 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1311 {
1312 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1313 
1314 	/*
1315 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1316 	 * is in VMX operation.
1317 	 */
1318 	if (vmx->nested.vmxon)
1319 		return -EBUSY;
1320 
1321 	switch (msr_index) {
1322 	case MSR_IA32_VMX_BASIC:
1323 		return vmx_restore_vmx_basic(vmx, data);
1324 	case MSR_IA32_VMX_PINBASED_CTLS:
1325 	case MSR_IA32_VMX_PROCBASED_CTLS:
1326 	case MSR_IA32_VMX_EXIT_CTLS:
1327 	case MSR_IA32_VMX_ENTRY_CTLS:
1328 		/*
1329 		 * The "non-true" VMX capability MSRs are generated from the
1330 		 * "true" MSRs, so we do not support restoring them directly.
1331 		 *
1332 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1333 		 * should restore the "true" MSRs with the must-be-1 bits
1334 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1335 		 * DEFAULT SETTINGS".
1336 		 */
1337 		return -EINVAL;
1338 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1339 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1340 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1341 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1342 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1343 		return vmx_restore_control_msr(vmx, msr_index, data);
1344 	case MSR_IA32_VMX_MISC:
1345 		return vmx_restore_vmx_misc(vmx, data);
1346 	case MSR_IA32_VMX_CR0_FIXED0:
1347 	case MSR_IA32_VMX_CR4_FIXED0:
1348 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1349 	case MSR_IA32_VMX_CR0_FIXED1:
1350 	case MSR_IA32_VMX_CR4_FIXED1:
1351 		/*
1352 		 * These MSRs are generated based on the vCPU's CPUID, so we
1353 		 * do not support restoring them directly.
1354 		 */
1355 		return -EINVAL;
1356 	case MSR_IA32_VMX_EPT_VPID_CAP:
1357 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1358 	case MSR_IA32_VMX_VMCS_ENUM:
1359 		vmx->nested.msrs.vmcs_enum = data;
1360 		return 0;
1361 	case MSR_IA32_VMX_VMFUNC:
1362 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1363 			return -EINVAL;
1364 		vmx->nested.msrs.vmfunc_controls = data;
1365 		return 0;
1366 	default:
1367 		/*
1368 		 * The rest of the VMX capability MSRs do not support restore.
1369 		 */
1370 		return -EINVAL;
1371 	}
1372 }
1373 
1374 /* Returns 0 on success, non-0 otherwise. */
1375 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1376 {
1377 	switch (msr_index) {
1378 	case MSR_IA32_VMX_BASIC:
1379 		*pdata = msrs->basic;
1380 		break;
1381 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1382 	case MSR_IA32_VMX_PINBASED_CTLS:
1383 		*pdata = vmx_control_msr(
1384 			msrs->pinbased_ctls_low,
1385 			msrs->pinbased_ctls_high);
1386 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1387 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1388 		break;
1389 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1390 	case MSR_IA32_VMX_PROCBASED_CTLS:
1391 		*pdata = vmx_control_msr(
1392 			msrs->procbased_ctls_low,
1393 			msrs->procbased_ctls_high);
1394 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1395 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1396 		break;
1397 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1398 	case MSR_IA32_VMX_EXIT_CTLS:
1399 		*pdata = vmx_control_msr(
1400 			msrs->exit_ctls_low,
1401 			msrs->exit_ctls_high);
1402 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1403 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1404 		break;
1405 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1406 	case MSR_IA32_VMX_ENTRY_CTLS:
1407 		*pdata = vmx_control_msr(
1408 			msrs->entry_ctls_low,
1409 			msrs->entry_ctls_high);
1410 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1411 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1412 		break;
1413 	case MSR_IA32_VMX_MISC:
1414 		*pdata = vmx_control_msr(
1415 			msrs->misc_low,
1416 			msrs->misc_high);
1417 		break;
1418 	case MSR_IA32_VMX_CR0_FIXED0:
1419 		*pdata = msrs->cr0_fixed0;
1420 		break;
1421 	case MSR_IA32_VMX_CR0_FIXED1:
1422 		*pdata = msrs->cr0_fixed1;
1423 		break;
1424 	case MSR_IA32_VMX_CR4_FIXED0:
1425 		*pdata = msrs->cr4_fixed0;
1426 		break;
1427 	case MSR_IA32_VMX_CR4_FIXED1:
1428 		*pdata = msrs->cr4_fixed1;
1429 		break;
1430 	case MSR_IA32_VMX_VMCS_ENUM:
1431 		*pdata = msrs->vmcs_enum;
1432 		break;
1433 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1434 		*pdata = vmx_control_msr(
1435 			msrs->secondary_ctls_low,
1436 			msrs->secondary_ctls_high);
1437 		break;
1438 	case MSR_IA32_VMX_EPT_VPID_CAP:
1439 		*pdata = msrs->ept_caps |
1440 			((u64)msrs->vpid_caps << 32);
1441 		break;
1442 	case MSR_IA32_VMX_VMFUNC:
1443 		*pdata = msrs->vmfunc_controls;
1444 		break;
1445 	default:
1446 		return 1;
1447 	}
1448 
1449 	return 0;
1450 }
1451 
1452 /*
1453  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1454  * been modified by the L1 guest.  Note, "writable" in this context means
1455  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1456  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1457  * VM-exit information fields (which are actually writable if the vCPU is
1458  * configured to support "VMWRITE to any supported field in the VMCS").
1459  */
1460 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1461 {
1462 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1463 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1464 	struct shadow_vmcs_field field;
1465 	unsigned long val;
1466 	int i;
1467 
1468 	if (WARN_ON(!shadow_vmcs))
1469 		return;
1470 
1471 	preempt_disable();
1472 
1473 	vmcs_load(shadow_vmcs);
1474 
1475 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1476 		field = shadow_read_write_fields[i];
1477 		val = __vmcs_readl(field.encoding);
1478 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1479 	}
1480 
1481 	vmcs_clear(shadow_vmcs);
1482 	vmcs_load(vmx->loaded_vmcs->vmcs);
1483 
1484 	preempt_enable();
1485 }
1486 
1487 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1488 {
1489 	const struct shadow_vmcs_field *fields[] = {
1490 		shadow_read_write_fields,
1491 		shadow_read_only_fields
1492 	};
1493 	const int max_fields[] = {
1494 		max_shadow_read_write_fields,
1495 		max_shadow_read_only_fields
1496 	};
1497 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1498 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1499 	struct shadow_vmcs_field field;
1500 	unsigned long val;
1501 	int i, q;
1502 
1503 	if (WARN_ON(!shadow_vmcs))
1504 		return;
1505 
1506 	vmcs_load(shadow_vmcs);
1507 
1508 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1509 		for (i = 0; i < max_fields[q]; i++) {
1510 			field = fields[q][i];
1511 			val = vmcs12_read_any(vmcs12, field.encoding,
1512 					      field.offset);
1513 			__vmcs_writel(field.encoding, val);
1514 		}
1515 	}
1516 
1517 	vmcs_clear(shadow_vmcs);
1518 	vmcs_load(vmx->loaded_vmcs->vmcs);
1519 }
1520 
1521 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1522 {
1523 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1524 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1525 
1526 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1527 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1528 	vmcs12->guest_rip = evmcs->guest_rip;
1529 
1530 	if (unlikely(!(evmcs->hv_clean_fields &
1531 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1532 		vmcs12->guest_rsp = evmcs->guest_rsp;
1533 		vmcs12->guest_rflags = evmcs->guest_rflags;
1534 		vmcs12->guest_interruptibility_info =
1535 			evmcs->guest_interruptibility_info;
1536 	}
1537 
1538 	if (unlikely(!(evmcs->hv_clean_fields &
1539 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1540 		vmcs12->cpu_based_vm_exec_control =
1541 			evmcs->cpu_based_vm_exec_control;
1542 	}
1543 
1544 	if (unlikely(!(evmcs->hv_clean_fields &
1545 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1546 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1547 	}
1548 
1549 	if (unlikely(!(evmcs->hv_clean_fields &
1550 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1551 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1552 	}
1553 
1554 	if (unlikely(!(evmcs->hv_clean_fields &
1555 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1556 		vmcs12->vm_entry_intr_info_field =
1557 			evmcs->vm_entry_intr_info_field;
1558 		vmcs12->vm_entry_exception_error_code =
1559 			evmcs->vm_entry_exception_error_code;
1560 		vmcs12->vm_entry_instruction_len =
1561 			evmcs->vm_entry_instruction_len;
1562 	}
1563 
1564 	if (unlikely(!(evmcs->hv_clean_fields &
1565 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1566 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1567 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1568 		vmcs12->host_cr0 = evmcs->host_cr0;
1569 		vmcs12->host_cr3 = evmcs->host_cr3;
1570 		vmcs12->host_cr4 = evmcs->host_cr4;
1571 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1572 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1573 		vmcs12->host_rip = evmcs->host_rip;
1574 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1575 		vmcs12->host_es_selector = evmcs->host_es_selector;
1576 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1577 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1578 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1579 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1580 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1581 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1582 	}
1583 
1584 	if (unlikely(!(evmcs->hv_clean_fields &
1585 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1586 		vmcs12->pin_based_vm_exec_control =
1587 			evmcs->pin_based_vm_exec_control;
1588 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1589 		vmcs12->secondary_vm_exec_control =
1590 			evmcs->secondary_vm_exec_control;
1591 	}
1592 
1593 	if (unlikely(!(evmcs->hv_clean_fields &
1594 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1595 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1596 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1597 	}
1598 
1599 	if (unlikely(!(evmcs->hv_clean_fields &
1600 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1601 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1602 	}
1603 
1604 	if (unlikely(!(evmcs->hv_clean_fields &
1605 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1606 		vmcs12->guest_es_base = evmcs->guest_es_base;
1607 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1608 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1609 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1610 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1611 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1612 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1613 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1614 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1615 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1616 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1617 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1618 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1619 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1620 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1621 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1622 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1623 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1624 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1625 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1626 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1627 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1628 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1629 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1630 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1631 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1632 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1633 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1634 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1635 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1636 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1637 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1638 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1639 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1640 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1641 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1642 	}
1643 
1644 	if (unlikely(!(evmcs->hv_clean_fields &
1645 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1646 		vmcs12->tsc_offset = evmcs->tsc_offset;
1647 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1648 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1649 	}
1650 
1651 	if (unlikely(!(evmcs->hv_clean_fields &
1652 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1653 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1654 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1655 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1656 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1657 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1658 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1659 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1660 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1661 	}
1662 
1663 	if (unlikely(!(evmcs->hv_clean_fields &
1664 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1665 		vmcs12->host_fs_base = evmcs->host_fs_base;
1666 		vmcs12->host_gs_base = evmcs->host_gs_base;
1667 		vmcs12->host_tr_base = evmcs->host_tr_base;
1668 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1669 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1670 		vmcs12->host_rsp = evmcs->host_rsp;
1671 	}
1672 
1673 	if (unlikely(!(evmcs->hv_clean_fields &
1674 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1675 		vmcs12->ept_pointer = evmcs->ept_pointer;
1676 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1677 	}
1678 
1679 	if (unlikely(!(evmcs->hv_clean_fields &
1680 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1681 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1682 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1683 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1684 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1685 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1686 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1687 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1688 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1689 		vmcs12->guest_pending_dbg_exceptions =
1690 			evmcs->guest_pending_dbg_exceptions;
1691 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1692 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1693 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1694 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1695 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1696 	}
1697 
1698 	/*
1699 	 * Not used?
1700 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1701 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1702 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1703 	 * vmcs12->cr3_target_value0 = evmcs->cr3_target_value0;
1704 	 * vmcs12->cr3_target_value1 = evmcs->cr3_target_value1;
1705 	 * vmcs12->cr3_target_value2 = evmcs->cr3_target_value2;
1706 	 * vmcs12->cr3_target_value3 = evmcs->cr3_target_value3;
1707 	 * vmcs12->page_fault_error_code_mask =
1708 	 *		evmcs->page_fault_error_code_mask;
1709 	 * vmcs12->page_fault_error_code_match =
1710 	 *		evmcs->page_fault_error_code_match;
1711 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1712 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1713 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1714 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1715 	 */
1716 
1717 	/*
1718 	 * Read only fields:
1719 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1720 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1721 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1722 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1723 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1724 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1725 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1726 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1727 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1728 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1729 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1730 	 *
1731 	 * Not present in struct vmcs12:
1732 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1733 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1734 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1735 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1736 	 */
1737 
1738 	return 0;
1739 }
1740 
1741 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1742 {
1743 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1744 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1745 
1746 	/*
1747 	 * Should not be changed by KVM:
1748 	 *
1749 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1750 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1751 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1752 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1753 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1754 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1755 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1756 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1757 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1758 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1759 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1760 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1761 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1762 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1763 	 * evmcs->host_rip = vmcs12->host_rip;
1764 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1765 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1766 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1767 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1768 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1769 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1770 	 * evmcs->host_rsp = vmcs12->host_rsp;
1771 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1772 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1773 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1774 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1775 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1776 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1777 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1778 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1779 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1780 	 * evmcs->cr3_target_value0 = vmcs12->cr3_target_value0;
1781 	 * evmcs->cr3_target_value1 = vmcs12->cr3_target_value1;
1782 	 * evmcs->cr3_target_value2 = vmcs12->cr3_target_value2;
1783 	 * evmcs->cr3_target_value3 = vmcs12->cr3_target_value3;
1784 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1785 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1786 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1787 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1788 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1789 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1790 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1791 	 * evmcs->page_fault_error_code_mask =
1792 	 *		vmcs12->page_fault_error_code_mask;
1793 	 * evmcs->page_fault_error_code_match =
1794 	 *		vmcs12->page_fault_error_code_match;
1795 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1796 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1797 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1798 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1799 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1800 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1801 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1802 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1803 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1804 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1805 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1806 	 *
1807 	 * Not present in struct vmcs12:
1808 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1809 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1810 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1811 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1812 	 */
1813 
1814 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1815 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1816 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1817 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1818 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1819 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1820 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1821 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1822 
1823 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1824 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1825 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1826 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1827 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1828 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1829 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1830 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1831 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1832 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1833 
1834 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1835 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1836 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1837 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1838 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1839 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1840 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1841 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1842 
1843 	evmcs->guest_es_base = vmcs12->guest_es_base;
1844 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1845 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1846 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1847 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1848 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1849 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1850 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1851 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1852 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1853 
1854 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1855 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1856 
1857 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1858 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1859 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1860 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1861 
1862 	evmcs->guest_pending_dbg_exceptions =
1863 		vmcs12->guest_pending_dbg_exceptions;
1864 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1865 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1866 
1867 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1868 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1869 
1870 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1871 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1872 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1873 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1874 
1875 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1876 
1877 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1878 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1879 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1880 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1881 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1882 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1883 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1884 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1885 
1886 	evmcs->exit_qualification = vmcs12->exit_qualification;
1887 
1888 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1889 	evmcs->guest_rsp = vmcs12->guest_rsp;
1890 	evmcs->guest_rflags = vmcs12->guest_rflags;
1891 
1892 	evmcs->guest_interruptibility_info =
1893 		vmcs12->guest_interruptibility_info;
1894 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1895 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1896 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1897 	evmcs->vm_entry_exception_error_code =
1898 		vmcs12->vm_entry_exception_error_code;
1899 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1900 
1901 	evmcs->guest_rip = vmcs12->guest_rip;
1902 
1903 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1904 
1905 	return 0;
1906 }
1907 
1908 /*
1909  * This is an equivalent of the nested hypervisor executing the vmptrld
1910  * instruction.
1911  */
1912 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1913 	struct kvm_vcpu *vcpu, bool from_launch)
1914 {
1915 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1916 	bool evmcs_gpa_changed = false;
1917 	u64 evmcs_gpa;
1918 
1919 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1920 		return EVMPTRLD_DISABLED;
1921 
1922 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa))
1923 		return EVMPTRLD_DISABLED;
1924 
1925 	if (unlikely(!vmx->nested.hv_evmcs ||
1926 		     evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1927 		if (!vmx->nested.hv_evmcs)
1928 			vmx->nested.current_vmptr = -1ull;
1929 
1930 		nested_release_evmcs(vcpu);
1931 
1932 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1933 				 &vmx->nested.hv_evmcs_map))
1934 			return EVMPTRLD_ERROR;
1935 
1936 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1937 
1938 		/*
1939 		 * Currently, KVM only supports eVMCS version 1
1940 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1941 		 * value to first u32 field of eVMCS which should specify eVMCS
1942 		 * VersionNumber.
1943 		 *
1944 		 * Guest should be aware of supported eVMCS versions by host by
1945 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1946 		 * expected to set this CPUID leaf according to the value
1947 		 * returned in vmcs_version from nested_enable_evmcs().
1948 		 *
1949 		 * However, it turns out that Microsoft Hyper-V fails to comply
1950 		 * to their own invented interface: When Hyper-V use eVMCS, it
1951 		 * just sets first u32 field of eVMCS to revision_id specified
1952 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1953 		 * which is one of the supported versions specified in
1954 		 * CPUID.0x4000000A.EAX[0:15].
1955 		 *
1956 		 * To overcome Hyper-V bug, we accept here either a supported
1957 		 * eVMCS version or VMCS12 revision_id as valid values for first
1958 		 * u32 field of eVMCS.
1959 		 */
1960 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1961 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1962 			nested_release_evmcs(vcpu);
1963 			return EVMPTRLD_VMFAIL;
1964 		}
1965 
1966 		vmx->nested.dirty_vmcs12 = true;
1967 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
1968 
1969 		evmcs_gpa_changed = true;
1970 		/*
1971 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
1972 		 * reloaded from guest's memory (read only fields, fields not
1973 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
1974 		 * are no leftovers.
1975 		 */
1976 		if (from_launch) {
1977 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1978 			memset(vmcs12, 0, sizeof(*vmcs12));
1979 			vmcs12->hdr.revision_id = VMCS12_REVISION;
1980 		}
1981 
1982 	}
1983 
1984 	/*
1985 	 * Clean fields data can't be used on VMLAUNCH and when we switch
1986 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
1987 	 */
1988 	if (from_launch || evmcs_gpa_changed)
1989 		vmx->nested.hv_evmcs->hv_clean_fields &=
1990 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1991 
1992 	return EVMPTRLD_SUCCEEDED;
1993 }
1994 
1995 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
1996 {
1997 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1998 
1999 	if (vmx->nested.hv_evmcs) {
2000 		copy_vmcs12_to_enlightened(vmx);
2001 		/* All fields are clean */
2002 		vmx->nested.hv_evmcs->hv_clean_fields |=
2003 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2004 	} else {
2005 		copy_vmcs12_to_shadow(vmx);
2006 	}
2007 
2008 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2009 }
2010 
2011 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2012 {
2013 	struct vcpu_vmx *vmx =
2014 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2015 
2016 	vmx->nested.preemption_timer_expired = true;
2017 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2018 	kvm_vcpu_kick(&vmx->vcpu);
2019 
2020 	return HRTIMER_NORESTART;
2021 }
2022 
2023 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
2024 {
2025 	u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
2026 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2027 
2028 	/*
2029 	 * A timer value of zero is architecturally guaranteed to cause
2030 	 * a VMExit prior to executing any instructions in the guest.
2031 	 */
2032 	if (preemption_timeout == 0) {
2033 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2034 		return;
2035 	}
2036 
2037 	if (vcpu->arch.virtual_tsc_khz == 0)
2038 		return;
2039 
2040 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2041 	preemption_timeout *= 1000000;
2042 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2043 	hrtimer_start(&vmx->nested.preemption_timer,
2044 		      ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
2045 }
2046 
2047 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2048 {
2049 	if (vmx->nested.nested_run_pending &&
2050 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2051 		return vmcs12->guest_ia32_efer;
2052 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2053 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2054 	else
2055 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2056 }
2057 
2058 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2059 {
2060 	/*
2061 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2062 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2063 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2064 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2065 	 */
2066 	if (vmx->nested.vmcs02_initialized)
2067 		return;
2068 	vmx->nested.vmcs02_initialized = true;
2069 
2070 	/*
2071 	 * We don't care what the EPTP value is we just need to guarantee
2072 	 * it's valid so we don't get a false positive when doing early
2073 	 * consistency checks.
2074 	 */
2075 	if (enable_ept && nested_early_check)
2076 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
2077 
2078 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2079 	if (cpu_has_vmx_vmfunc())
2080 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2081 
2082 	if (cpu_has_vmx_posted_intr())
2083 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2084 
2085 	if (cpu_has_vmx_msr_bitmap())
2086 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2087 
2088 	/*
2089 	 * The PML address never changes, so it is constant in vmcs02.
2090 	 * Conceptually we want to copy the PML index from vmcs01 here,
2091 	 * and then back to vmcs01 on nested vmexit.  But since we flush
2092 	 * the log and reset GUEST_PML_INDEX on each vmexit, the PML
2093 	 * index is also effectively constant in vmcs02.
2094 	 */
2095 	if (enable_pml) {
2096 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
2097 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2098 	}
2099 
2100 	if (cpu_has_vmx_encls_vmexit())
2101 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2102 
2103 	/*
2104 	 * Set the MSR load/store lists to match L0's settings.  Only the
2105 	 * addresses are constant (for vmcs02), the counts can change based
2106 	 * on L2's behavior, e.g. switching to/from long mode.
2107 	 */
2108 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2109 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2110 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2111 
2112 	vmx_set_constant_host_state(vmx);
2113 }
2114 
2115 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2116 				      struct vmcs12 *vmcs12)
2117 {
2118 	prepare_vmcs02_constant_state(vmx);
2119 
2120 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2121 
2122 	if (enable_vpid) {
2123 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2124 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2125 		else
2126 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2127 	}
2128 }
2129 
2130 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2131 {
2132 	u32 exec_control, vmcs12_exec_ctrl;
2133 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2134 
2135 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
2136 		prepare_vmcs02_early_rare(vmx, vmcs12);
2137 
2138 	/*
2139 	 * PIN CONTROLS
2140 	 */
2141 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2142 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2143 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2144 
2145 	/* Posted interrupts setting is only taken from vmcs12.  */
2146 	if (nested_cpu_has_posted_intr(vmcs12)) {
2147 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2148 		vmx->nested.pi_pending = false;
2149 	} else {
2150 		exec_control &= ~PIN_BASED_POSTED_INTR;
2151 	}
2152 	pin_controls_set(vmx, exec_control);
2153 
2154 	/*
2155 	 * EXEC CONTROLS
2156 	 */
2157 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2158 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2159 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2160 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2161 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2162 
2163 	vmx->nested.l1_tpr_threshold = -1;
2164 	if (exec_control & CPU_BASED_TPR_SHADOW)
2165 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2166 #ifdef CONFIG_X86_64
2167 	else
2168 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2169 				CPU_BASED_CR8_STORE_EXITING;
2170 #endif
2171 
2172 	/*
2173 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2174 	 * for I/O port accesses.
2175 	 */
2176 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2177 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2178 
2179 	/*
2180 	 * This bit will be computed in nested_get_vmcs12_pages, because
2181 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2182 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2183 	 * only set/clear this bit.
2184 	 */
2185 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2186 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2187 
2188 	exec_controls_set(vmx, exec_control);
2189 
2190 	/*
2191 	 * SECONDARY EXEC CONTROLS
2192 	 */
2193 	if (cpu_has_secondary_exec_ctrls()) {
2194 		exec_control = vmx->secondary_exec_control;
2195 
2196 		/* Take the following fields only from vmcs12 */
2197 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2198 				  SECONDARY_EXEC_ENABLE_INVPCID |
2199 				  SECONDARY_EXEC_RDTSCP |
2200 				  SECONDARY_EXEC_XSAVES |
2201 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2202 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2203 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2204 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2205 		if (nested_cpu_has(vmcs12,
2206 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2207 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2208 				~SECONDARY_EXEC_ENABLE_PML;
2209 			exec_control |= vmcs12_exec_ctrl;
2210 		}
2211 
2212 		/* VMCS shadowing for L2 is emulated for now */
2213 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2214 
2215 		/*
2216 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2217 		 * will not have to rewrite the controls just for this bit.
2218 		 */
2219 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2220 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2221 			exec_control |= SECONDARY_EXEC_DESC;
2222 
2223 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2224 			vmcs_write16(GUEST_INTR_STATUS,
2225 				vmcs12->guest_intr_status);
2226 
2227 		secondary_exec_controls_set(vmx, exec_control);
2228 	}
2229 
2230 	/*
2231 	 * ENTRY CONTROLS
2232 	 *
2233 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2234 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2235 	 * on the related bits (if supported by the CPU) in the hope that
2236 	 * we can avoid VMWrites during vmx_set_efer().
2237 	 */
2238 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2239 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2240 	if (cpu_has_load_ia32_efer()) {
2241 		if (guest_efer & EFER_LMA)
2242 			exec_control |= VM_ENTRY_IA32E_MODE;
2243 		if (guest_efer != host_efer)
2244 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2245 	}
2246 	vm_entry_controls_set(vmx, exec_control);
2247 
2248 	/*
2249 	 * EXIT CONTROLS
2250 	 *
2251 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2252 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2253 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2254 	 */
2255 	exec_control = vmx_vmexit_ctrl();
2256 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2257 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2258 	vm_exit_controls_set(vmx, exec_control);
2259 
2260 	/*
2261 	 * Interrupt/Exception Fields
2262 	 */
2263 	if (vmx->nested.nested_run_pending) {
2264 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2265 			     vmcs12->vm_entry_intr_info_field);
2266 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2267 			     vmcs12->vm_entry_exception_error_code);
2268 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2269 			     vmcs12->vm_entry_instruction_len);
2270 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2271 			     vmcs12->guest_interruptibility_info);
2272 		vmx->loaded_vmcs->nmi_known_unmasked =
2273 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2274 	} else {
2275 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2276 	}
2277 }
2278 
2279 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2280 {
2281 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2282 
2283 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2284 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2285 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2286 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2287 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2288 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2289 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2290 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2291 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2292 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2293 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2294 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2295 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2296 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2297 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2298 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2299 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2300 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2301 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2302 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2303 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2304 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2305 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2306 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2307 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2308 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2309 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2310 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2311 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2312 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2313 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2314 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2315 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2316 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2317 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2318 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2319 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2320 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2321 	}
2322 
2323 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2324 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2325 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2326 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2327 			    vmcs12->guest_pending_dbg_exceptions);
2328 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2329 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2330 
2331 		/*
2332 		 * L1 may access the L2's PDPTR, so save them to construct
2333 		 * vmcs12
2334 		 */
2335 		if (enable_ept) {
2336 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2337 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2338 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2339 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2340 		}
2341 
2342 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2343 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2344 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2345 	}
2346 
2347 	if (nested_cpu_has_xsaves(vmcs12))
2348 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2349 
2350 	/*
2351 	 * Whether page-faults are trapped is determined by a combination of
2352 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2353 	 * If enable_ept, L0 doesn't care about page faults and we should
2354 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2355 	 * care about (at least some) page faults, and because it is not easy
2356 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2357 	 * to exit on each and every L2 page fault. This is done by setting
2358 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2359 	 * Note that below we don't need special code to set EB.PF beyond the
2360 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2361 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2362 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2363 	 */
2364 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2365 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2366 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2367 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2368 
2369 	if (cpu_has_vmx_apicv()) {
2370 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2371 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2372 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2373 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2374 	}
2375 
2376 	/*
2377 	 * Make sure the msr_autostore list is up to date before we set the
2378 	 * count in the vmcs02.
2379 	 */
2380 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2381 
2382 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2383 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2384 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2385 
2386 	set_cr4_guest_host_mask(vmx);
2387 }
2388 
2389 /*
2390  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2391  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2392  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2393  * guest in a way that will both be appropriate to L1's requests, and our
2394  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2395  * function also has additional necessary side-effects, like setting various
2396  * vcpu->arch fields.
2397  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2398  * is assigned to entry_failure_code on failure.
2399  */
2400 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2401 			  u32 *entry_failure_code)
2402 {
2403 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2404 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2405 	bool load_guest_pdptrs_vmcs12 = false;
2406 
2407 	if (vmx->nested.dirty_vmcs12 || hv_evmcs) {
2408 		prepare_vmcs02_rare(vmx, vmcs12);
2409 		vmx->nested.dirty_vmcs12 = false;
2410 
2411 		load_guest_pdptrs_vmcs12 = !hv_evmcs ||
2412 			!(hv_evmcs->hv_clean_fields &
2413 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2414 	}
2415 
2416 	if (vmx->nested.nested_run_pending &&
2417 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2418 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2419 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2420 	} else {
2421 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2422 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2423 	}
2424 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2425 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2426 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2427 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2428 
2429 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2430 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2431 	 * trap. Note that CR0.TS also needs updating - we do this later.
2432 	 */
2433 	update_exception_bitmap(vcpu);
2434 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2435 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2436 
2437 	if (vmx->nested.nested_run_pending &&
2438 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2439 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2440 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2441 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2442 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2443 	}
2444 
2445 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2446 
2447 	if (kvm_has_tsc_control)
2448 		decache_tsc_multiplier(vmx);
2449 
2450 	if (enable_vpid) {
2451 		/*
2452 		 * There is no direct mapping between vpid02 and vpid12, the
2453 		 * vpid02 is per-vCPU for L0 and reused while the value of
2454 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2455 		 * The vpid12 is allocated by L1 for L2, so it will not
2456 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2457 		 * even if spawn a lot of nested vCPUs.
2458 		 */
2459 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2460 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2461 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2462 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2463 			}
2464 		} else {
2465 			/*
2466 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2467 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2468 			 * flush until vmcs02->eptp is fully updated by
2469 			 * KVM_REQ_LOAD_MMU_PGD. Note that this assumes
2470 			 * KVM_REQ_TLB_FLUSH is evaluated after
2471 			 * KVM_REQ_LOAD_MMU_PGD in vcpu_enter_guest().
2472 			 */
2473 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2474 		}
2475 	}
2476 
2477 	if (nested_cpu_has_ept(vmcs12))
2478 		nested_ept_init_mmu_context(vcpu);
2479 
2480 	/*
2481 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2482 	 * bits which we consider mandatory enabled.
2483 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2484 	 * the specifications by L1; It's not enough to take
2485 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2486 	 * have more bits than L1 expected.
2487 	 */
2488 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2489 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2490 
2491 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2492 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2493 
2494 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2495 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2496 	vmx_set_efer(vcpu, vcpu->arch.efer);
2497 
2498 	/*
2499 	 * Guest state is invalid and unrestricted guest is disabled,
2500 	 * which means L1 attempted VMEntry to L2 with invalid state.
2501 	 * Fail the VMEntry.
2502 	 */
2503 	if (vmx->emulation_required) {
2504 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2505 		return -EINVAL;
2506 	}
2507 
2508 	/* Shadow page tables on either EPT or shadow page tables. */
2509 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2510 				entry_failure_code))
2511 		return -EINVAL;
2512 
2513 	/*
2514 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2515 	 * on nested VM-Exit, which can occur without actually running L2 and
2516 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2517 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2518 	 * transition to HLT instead of running L2.
2519 	 */
2520 	if (enable_ept)
2521 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2522 
2523 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2524 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2525 	    is_pae_paging(vcpu)) {
2526 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2527 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2528 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2529 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2530 	}
2531 
2532 	if (!enable_ept)
2533 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2534 
2535 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2536 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2537 				     vmcs12->guest_ia32_perf_global_ctrl)))
2538 		return -EINVAL;
2539 
2540 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2541 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2542 	return 0;
2543 }
2544 
2545 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2546 {
2547 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2548 	       nested_cpu_has_virtual_nmis(vmcs12)))
2549 		return -EINVAL;
2550 
2551 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2552 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2553 		return -EINVAL;
2554 
2555 	return 0;
2556 }
2557 
2558 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2559 {
2560 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2561 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2562 
2563 	/* Check for memory type validity */
2564 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2565 	case VMX_EPTP_MT_UC:
2566 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2567 			return false;
2568 		break;
2569 	case VMX_EPTP_MT_WB:
2570 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2571 			return false;
2572 		break;
2573 	default:
2574 		return false;
2575 	}
2576 
2577 	/* Page-walk levels validity. */
2578 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2579 	case VMX_EPTP_PWL_5:
2580 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2581 			return false;
2582 		break;
2583 	case VMX_EPTP_PWL_4:
2584 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2585 			return false;
2586 		break;
2587 	default:
2588 		return false;
2589 	}
2590 
2591 	/* Reserved bits should not be set */
2592 	if (CC(new_eptp >> maxphyaddr || ((new_eptp >> 7) & 0x1f)))
2593 		return false;
2594 
2595 	/* AD, if set, should be supported */
2596 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2597 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2598 			return false;
2599 	}
2600 
2601 	return true;
2602 }
2603 
2604 /*
2605  * Checks related to VM-Execution Control Fields
2606  */
2607 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2608                                               struct vmcs12 *vmcs12)
2609 {
2610 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2611 
2612 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2613 				   vmx->nested.msrs.pinbased_ctls_low,
2614 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2615 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2616 				   vmx->nested.msrs.procbased_ctls_low,
2617 				   vmx->nested.msrs.procbased_ctls_high)))
2618 		return -EINVAL;
2619 
2620 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2621 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2622 				   vmx->nested.msrs.secondary_ctls_low,
2623 				   vmx->nested.msrs.secondary_ctls_high)))
2624 		return -EINVAL;
2625 
2626 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2627 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2628 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2629 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2630 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2631 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2632 	    nested_vmx_check_nmi_controls(vmcs12) ||
2633 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2634 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2635 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2636 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2637 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2638 		return -EINVAL;
2639 
2640 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2641 	    nested_cpu_has_save_preemption_timer(vmcs12))
2642 		return -EINVAL;
2643 
2644 	if (nested_cpu_has_ept(vmcs12) &&
2645 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2646 		return -EINVAL;
2647 
2648 	if (nested_cpu_has_vmfunc(vmcs12)) {
2649 		if (CC(vmcs12->vm_function_control &
2650 		       ~vmx->nested.msrs.vmfunc_controls))
2651 			return -EINVAL;
2652 
2653 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2654 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2655 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2656 				return -EINVAL;
2657 		}
2658 	}
2659 
2660 	return 0;
2661 }
2662 
2663 /*
2664  * Checks related to VM-Exit Control Fields
2665  */
2666 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2667                                          struct vmcs12 *vmcs12)
2668 {
2669 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2670 
2671 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2672 				    vmx->nested.msrs.exit_ctls_low,
2673 				    vmx->nested.msrs.exit_ctls_high)) ||
2674 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2675 		return -EINVAL;
2676 
2677 	return 0;
2678 }
2679 
2680 /*
2681  * Checks related to VM-Entry Control Fields
2682  */
2683 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2684 					  struct vmcs12 *vmcs12)
2685 {
2686 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2687 
2688 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2689 				    vmx->nested.msrs.entry_ctls_low,
2690 				    vmx->nested.msrs.entry_ctls_high)))
2691 		return -EINVAL;
2692 
2693 	/*
2694 	 * From the Intel SDM, volume 3:
2695 	 * Fields relevant to VM-entry event injection must be set properly.
2696 	 * These fields are the VM-entry interruption-information field, the
2697 	 * VM-entry exception error code, and the VM-entry instruction length.
2698 	 */
2699 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2700 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2701 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2702 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2703 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2704 		bool should_have_error_code;
2705 		bool urg = nested_cpu_has2(vmcs12,
2706 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2707 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2708 
2709 		/* VM-entry interruption-info field: interruption type */
2710 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2711 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2712 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2713 			return -EINVAL;
2714 
2715 		/* VM-entry interruption-info field: vector */
2716 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2717 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2718 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2719 			return -EINVAL;
2720 
2721 		/* VM-entry interruption-info field: deliver error code */
2722 		should_have_error_code =
2723 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2724 			x86_exception_has_error_code(vector);
2725 		if (CC(has_error_code != should_have_error_code))
2726 			return -EINVAL;
2727 
2728 		/* VM-entry exception error code */
2729 		if (CC(has_error_code &&
2730 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2731 			return -EINVAL;
2732 
2733 		/* VM-entry interruption-info field: reserved bits */
2734 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2735 			return -EINVAL;
2736 
2737 		/* VM-entry instruction length */
2738 		switch (intr_type) {
2739 		case INTR_TYPE_SOFT_EXCEPTION:
2740 		case INTR_TYPE_SOFT_INTR:
2741 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2742 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2743 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2744 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2745 				return -EINVAL;
2746 		}
2747 	}
2748 
2749 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2750 		return -EINVAL;
2751 
2752 	return 0;
2753 }
2754 
2755 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2756 				     struct vmcs12 *vmcs12)
2757 {
2758 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2759 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2760 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2761 		return -EINVAL;
2762 
2763 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2764 		return nested_evmcs_check_controls(vmcs12);
2765 
2766 	return 0;
2767 }
2768 
2769 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2770 				       struct vmcs12 *vmcs12)
2771 {
2772 	bool ia32e;
2773 
2774 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2775 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2776 	    CC(!nested_cr3_valid(vcpu, vmcs12->host_cr3)))
2777 		return -EINVAL;
2778 
2779 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2780 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2781 		return -EINVAL;
2782 
2783 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2784 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2785 		return -EINVAL;
2786 
2787 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2788 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2789 					   vmcs12->host_ia32_perf_global_ctrl)))
2790 		return -EINVAL;
2791 
2792 #ifdef CONFIG_X86_64
2793 	ia32e = !!(vcpu->arch.efer & EFER_LMA);
2794 #else
2795 	ia32e = false;
2796 #endif
2797 
2798 	if (ia32e) {
2799 		if (CC(!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)) ||
2800 		    CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2801 			return -EINVAL;
2802 	} else {
2803 		if (CC(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) ||
2804 		    CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2805 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2806 		    CC((vmcs12->host_rip) >> 32))
2807 			return -EINVAL;
2808 	}
2809 
2810 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2811 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2812 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2813 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2814 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2815 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2816 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2817 	    CC(vmcs12->host_cs_selector == 0) ||
2818 	    CC(vmcs12->host_tr_selector == 0) ||
2819 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2820 		return -EINVAL;
2821 
2822 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2823 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2824 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2825 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2826 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2827 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2828 		return -EINVAL;
2829 
2830 	/*
2831 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2832 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2833 	 * the values of the LMA and LME bits in the field must each be that of
2834 	 * the host address-space size VM-exit control.
2835 	 */
2836 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2837 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2838 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2839 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2840 			return -EINVAL;
2841 	}
2842 
2843 	return 0;
2844 }
2845 
2846 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2847 					  struct vmcs12 *vmcs12)
2848 {
2849 	int r = 0;
2850 	struct vmcs12 *shadow;
2851 	struct kvm_host_map map;
2852 
2853 	if (vmcs12->vmcs_link_pointer == -1ull)
2854 		return 0;
2855 
2856 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2857 		return -EINVAL;
2858 
2859 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2860 		return -EINVAL;
2861 
2862 	shadow = map.hva;
2863 
2864 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2865 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2866 		r = -EINVAL;
2867 
2868 	kvm_vcpu_unmap(vcpu, &map, false);
2869 	return r;
2870 }
2871 
2872 /*
2873  * Checks related to Guest Non-register State
2874  */
2875 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2876 {
2877 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2878 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT))
2879 		return -EINVAL;
2880 
2881 	return 0;
2882 }
2883 
2884 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2885 					struct vmcs12 *vmcs12,
2886 					u32 *exit_qual)
2887 {
2888 	bool ia32e;
2889 
2890 	*exit_qual = ENTRY_FAIL_DEFAULT;
2891 
2892 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2893 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2894 		return -EINVAL;
2895 
2896 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2897 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2898 		return -EINVAL;
2899 
2900 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2901 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2902 		return -EINVAL;
2903 
2904 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2905 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2906 		return -EINVAL;
2907 	}
2908 
2909 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2910 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2911 					   vmcs12->guest_ia32_perf_global_ctrl)))
2912 		return -EINVAL;
2913 
2914 	/*
2915 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2916 	 * are performed on the field for the IA32_EFER MSR:
2917 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2918 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2919 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2920 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2921 	 *   CR0.PG) is 1.
2922 	 */
2923 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2924 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2925 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2926 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
2927 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
2928 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
2929 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
2930 			return -EINVAL;
2931 	}
2932 
2933 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2934 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
2935 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
2936 		return -EINVAL;
2937 
2938 	if (nested_check_guest_non_reg_state(vmcs12))
2939 		return -EINVAL;
2940 
2941 	return 0;
2942 }
2943 
2944 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2945 {
2946 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2947 	unsigned long cr3, cr4;
2948 	bool vm_fail;
2949 
2950 	if (!nested_early_check)
2951 		return 0;
2952 
2953 	if (vmx->msr_autoload.host.nr)
2954 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2955 	if (vmx->msr_autoload.guest.nr)
2956 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2957 
2958 	preempt_disable();
2959 
2960 	vmx_prepare_switch_to_guest(vcpu);
2961 
2962 	/*
2963 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2964 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2965 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
2966 	 * there is no need to preserve other bits or save/restore the field.
2967 	 */
2968 	vmcs_writel(GUEST_RFLAGS, 0);
2969 
2970 	cr3 = __get_current_cr3_fast();
2971 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2972 		vmcs_writel(HOST_CR3, cr3);
2973 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2974 	}
2975 
2976 	cr4 = cr4_read_shadow();
2977 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2978 		vmcs_writel(HOST_CR4, cr4);
2979 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2980 	}
2981 
2982 	asm(
2983 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2984 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2985 		"je 1f \n\t"
2986 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
2987 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2988 		"1: \n\t"
2989 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2990 
2991 		/* Check if vmlaunch or vmresume is needed */
2992 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
2993 
2994 		/*
2995 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
2996 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
2997 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
2998 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
2999 		 */
3000 		"call vmx_vmenter\n\t"
3001 
3002 		CC_SET(be)
3003 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
3004 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
3005 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
3006 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
3007 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
3008 		[wordsize]"i"(sizeof(ulong))
3009 	      : "memory"
3010 	);
3011 
3012 	if (vmx->msr_autoload.host.nr)
3013 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3014 	if (vmx->msr_autoload.guest.nr)
3015 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3016 
3017 	if (vm_fail) {
3018 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3019 
3020 		preempt_enable();
3021 
3022 		trace_kvm_nested_vmenter_failed(
3023 			"early hardware check VM-instruction error: ", error);
3024 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3025 		return 1;
3026 	}
3027 
3028 	/*
3029 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3030 	 */
3031 	local_irq_enable();
3032 	if (hw_breakpoint_active())
3033 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3034 	preempt_enable();
3035 
3036 	/*
3037 	 * A non-failing VMEntry means we somehow entered guest mode with
3038 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3039 	 * is no telling what memory has been modified or what state has
3040 	 * been exposed to unknown code.  Hitting this all but guarantees
3041 	 * a (very critical) hardware issue.
3042 	 */
3043 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3044 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3045 
3046 	return 0;
3047 }
3048 
3049 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3050 {
3051 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3052 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3053 	struct kvm_host_map *map;
3054 	struct page *page;
3055 	u64 hpa;
3056 
3057 	/*
3058 	 * hv_evmcs may end up being not mapped after migration (when
3059 	 * L2 was running), map it here to make sure vmcs12 changes are
3060 	 * properly reflected.
3061 	 */
3062 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs) {
3063 		enum nested_evmptrld_status evmptrld_status =
3064 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3065 
3066 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3067 		    evmptrld_status == EVMPTRLD_ERROR) {
3068 			pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3069 					     __func__);
3070 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3071 			vcpu->run->internal.suberror =
3072 				KVM_INTERNAL_ERROR_EMULATION;
3073 			vcpu->run->internal.ndata = 0;
3074 			return false;
3075 		}
3076 	}
3077 
3078 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3079 		/*
3080 		 * Translate L1 physical address to host physical
3081 		 * address for vmcs02. Keep the page pinned, so this
3082 		 * physical address remains valid. We keep a reference
3083 		 * to it so we can release it later.
3084 		 */
3085 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3086 			kvm_release_page_clean(vmx->nested.apic_access_page);
3087 			vmx->nested.apic_access_page = NULL;
3088 		}
3089 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3090 		if (!is_error_page(page)) {
3091 			vmx->nested.apic_access_page = page;
3092 			hpa = page_to_phys(vmx->nested.apic_access_page);
3093 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3094 		} else {
3095 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3096 					     __func__);
3097 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3098 			vcpu->run->internal.suberror =
3099 				KVM_INTERNAL_ERROR_EMULATION;
3100 			vcpu->run->internal.ndata = 0;
3101 			return false;
3102 		}
3103 	}
3104 
3105 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3106 		map = &vmx->nested.virtual_apic_map;
3107 
3108 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3109 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3110 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3111 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3112 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3113 			/*
3114 			 * The processor will never use the TPR shadow, simply
3115 			 * clear the bit from the execution control.  Such a
3116 			 * configuration is useless, but it happens in tests.
3117 			 * For any other configuration, failing the vm entry is
3118 			 * _not_ what the processor does but it's basically the
3119 			 * only possibility we have.
3120 			 */
3121 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3122 		} else {
3123 			/*
3124 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3125 			 * force VM-Entry to fail.
3126 			 */
3127 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
3128 		}
3129 	}
3130 
3131 	if (nested_cpu_has_posted_intr(vmcs12)) {
3132 		map = &vmx->nested.pi_desc_map;
3133 
3134 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3135 			vmx->nested.pi_desc =
3136 				(struct pi_desc *)(((void *)map->hva) +
3137 				offset_in_page(vmcs12->posted_intr_desc_addr));
3138 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3139 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3140 		}
3141 	}
3142 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3143 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3144 	else
3145 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3146 	return true;
3147 }
3148 
3149 /*
3150  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3151  * for running VMX instructions (except VMXON, whose prerequisites are
3152  * slightly different). It also specifies what exception to inject otherwise.
3153  * Note that many of these exceptions have priority over VM exits, so they
3154  * don't have to be checked again here.
3155  */
3156 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3157 {
3158 	if (!to_vmx(vcpu)->nested.vmxon) {
3159 		kvm_queue_exception(vcpu, UD_VECTOR);
3160 		return 0;
3161 	}
3162 
3163 	if (vmx_get_cpl(vcpu)) {
3164 		kvm_inject_gp(vcpu, 0);
3165 		return 0;
3166 	}
3167 
3168 	return 1;
3169 }
3170 
3171 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3172 {
3173 	u8 rvi = vmx_get_rvi();
3174 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3175 
3176 	return ((rvi & 0xf0) > (vppr & 0xf0));
3177 }
3178 
3179 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3180 				   struct vmcs12 *vmcs12);
3181 
3182 /*
3183  * If from_vmentry is false, this is being called from state restore (either RSM
3184  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3185  *
3186  * Returns:
3187  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3188  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3189  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3190  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3191  */
3192 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3193 							bool from_vmentry)
3194 {
3195 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3196 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3197 	bool evaluate_pending_interrupts;
3198 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
3199 	u32 exit_qual;
3200 
3201 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3202 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3203 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3204 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3205 
3206 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3207 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3208 	if (kvm_mpx_supported() &&
3209 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3210 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3211 
3212 	/*
3213 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3214 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3215 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3216 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3217 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3218 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3219 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3220 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3221 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3222 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3223 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3224 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3225 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3226 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3227 	 */
3228 	if (!enable_ept && !nested_early_check)
3229 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3230 
3231 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3232 
3233 	prepare_vmcs02_early(vmx, vmcs12);
3234 
3235 	if (from_vmentry) {
3236 		if (unlikely(!nested_get_vmcs12_pages(vcpu)))
3237 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3238 
3239 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3240 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3241 			return NVMX_VMENTRY_VMFAIL;
3242 		}
3243 
3244 		if (nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
3245 			goto vmentry_fail_vmexit;
3246 	}
3247 
3248 	enter_guest_mode(vcpu);
3249 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3250 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3251 
3252 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
3253 		goto vmentry_fail_vmexit_guest_mode;
3254 
3255 	if (from_vmentry) {
3256 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3257 		exit_qual = nested_vmx_load_msr(vcpu,
3258 						vmcs12->vm_entry_msr_load_addr,
3259 						vmcs12->vm_entry_msr_load_count);
3260 		if (exit_qual)
3261 			goto vmentry_fail_vmexit_guest_mode;
3262 	} else {
3263 		/*
3264 		 * The MMU is not initialized to point at the right entities yet and
3265 		 * "get pages" would need to read data from the guest (i.e. we will
3266 		 * need to perform gpa to hpa translation). Request a call
3267 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3268 		 * have already been set at vmentry time and should not be reset.
3269 		 */
3270 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3271 	}
3272 
3273 	/*
3274 	 * If L1 had a pending IRQ/NMI until it executed
3275 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3276 	 * disallowed (e.g. interrupts disabled), L0 needs to
3277 	 * evaluate if this pending event should cause an exit from L2
3278 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3279 	 * intercept EXTERNAL_INTERRUPT).
3280 	 *
3281 	 * Usually this would be handled by the processor noticing an
3282 	 * IRQ/NMI window request, or checking RVI during evaluation of
3283 	 * pending virtual interrupts.  However, this setting was done
3284 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3285 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3286 	 */
3287 	if (unlikely(evaluate_pending_interrupts))
3288 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3289 
3290 	/*
3291 	 * Do not start the preemption timer hrtimer until after we know
3292 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3293 	 * the timer.
3294 	 */
3295 	vmx->nested.preemption_timer_expired = false;
3296 	if (nested_cpu_has_preemption_timer(vmcs12))
3297 		vmx_start_preemption_timer(vcpu);
3298 
3299 	/*
3300 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3301 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3302 	 * returned as far as L1 is concerned. It will only return (and set
3303 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3304 	 */
3305 	return NVMX_VMENTRY_SUCCESS;
3306 
3307 	/*
3308 	 * A failed consistency check that leads to a VMExit during L1's
3309 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3310 	 * 26.7 "VM-entry failures during or after loading guest state".
3311 	 */
3312 vmentry_fail_vmexit_guest_mode:
3313 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3314 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3315 	leave_guest_mode(vcpu);
3316 
3317 vmentry_fail_vmexit:
3318 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3319 
3320 	if (!from_vmentry)
3321 		return NVMX_VMENTRY_VMEXIT;
3322 
3323 	load_vmcs12_host_state(vcpu, vmcs12);
3324 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3325 	vmcs12->exit_qualification = exit_qual;
3326 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3327 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3328 	return NVMX_VMENTRY_VMEXIT;
3329 }
3330 
3331 /*
3332  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3333  * for running an L2 nested guest.
3334  */
3335 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3336 {
3337 	struct vmcs12 *vmcs12;
3338 	enum nvmx_vmentry_status status;
3339 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3340 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3341 	enum nested_evmptrld_status evmptrld_status;
3342 
3343 	if (!nested_vmx_check_permission(vcpu))
3344 		return 1;
3345 
3346 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3347 	if (evmptrld_status == EVMPTRLD_ERROR) {
3348 		kvm_queue_exception(vcpu, UD_VECTOR);
3349 		return 1;
3350 	} else if (evmptrld_status == EVMPTRLD_VMFAIL) {
3351 		return nested_vmx_failInvalid(vcpu);
3352 	}
3353 
3354 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3355 		return nested_vmx_failInvalid(vcpu);
3356 
3357 	vmcs12 = get_vmcs12(vcpu);
3358 
3359 	/*
3360 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3361 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3362 	 * rather than RFLAGS.ZF, and no error number is stored to the
3363 	 * VM-instruction error field.
3364 	 */
3365 	if (vmcs12->hdr.shadow_vmcs)
3366 		return nested_vmx_failInvalid(vcpu);
3367 
3368 	if (vmx->nested.hv_evmcs) {
3369 		copy_enlightened_to_vmcs12(vmx);
3370 		/* Enlightened VMCS doesn't have launch state */
3371 		vmcs12->launch_state = !launch;
3372 	} else if (enable_shadow_vmcs) {
3373 		copy_shadow_to_vmcs12(vmx);
3374 	}
3375 
3376 	/*
3377 	 * The nested entry process starts with enforcing various prerequisites
3378 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3379 	 * they fail: As the SDM explains, some conditions should cause the
3380 	 * instruction to fail, while others will cause the instruction to seem
3381 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3382 	 * To speed up the normal (success) code path, we should avoid checking
3383 	 * for misconfigurations which will anyway be caught by the processor
3384 	 * when using the merged vmcs02.
3385 	 */
3386 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3387 		return nested_vmx_failValid(vcpu,
3388 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3389 
3390 	if (vmcs12->launch_state == launch)
3391 		return nested_vmx_failValid(vcpu,
3392 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3393 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3394 
3395 	if (nested_vmx_check_controls(vcpu, vmcs12))
3396 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3397 
3398 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3399 		return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3400 
3401 	/*
3402 	 * We're finally done with prerequisite checking, and can start with
3403 	 * the nested entry.
3404 	 */
3405 	vmx->nested.nested_run_pending = 1;
3406 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3407 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3408 		goto vmentry_failed;
3409 
3410 	/* Hide L1D cache contents from the nested guest.  */
3411 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3412 
3413 	/*
3414 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3415 	 * also be used as part of restoring nVMX state for
3416 	 * snapshot restore (migration).
3417 	 *
3418 	 * In this flow, it is assumed that vmcs12 cache was
3419 	 * trasferred as part of captured nVMX state and should
3420 	 * therefore not be read from guest memory (which may not
3421 	 * exist on destination host yet).
3422 	 */
3423 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3424 
3425 	/*
3426 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3427 	 * awakened by event injection or by an NMI-window VM-exit or
3428 	 * by an interrupt-window VM-exit, halt the vcpu.
3429 	 */
3430 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3431 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3432 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_NMI_WINDOW_EXITING) &&
3433 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_INTR_WINDOW_EXITING) &&
3434 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3435 		vmx->nested.nested_run_pending = 0;
3436 		return kvm_vcpu_halt(vcpu);
3437 	}
3438 	return 1;
3439 
3440 vmentry_failed:
3441 	vmx->nested.nested_run_pending = 0;
3442 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3443 		return 0;
3444 	if (status == NVMX_VMENTRY_VMEXIT)
3445 		return 1;
3446 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3447 	return nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3448 }
3449 
3450 /*
3451  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3452  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3453  * This function returns the new value we should put in vmcs12.guest_cr0.
3454  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3455  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3456  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3457  *     didn't trap the bit, because if L1 did, so would L0).
3458  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3459  *     been modified by L2, and L1 knows it. So just leave the old value of
3460  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3461  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3462  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3463  *     changed these bits, and therefore they need to be updated, but L0
3464  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3465  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3466  */
3467 static inline unsigned long
3468 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3469 {
3470 	return
3471 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3472 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3473 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3474 			vcpu->arch.cr0_guest_owned_bits));
3475 }
3476 
3477 static inline unsigned long
3478 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3479 {
3480 	return
3481 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3482 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3483 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3484 			vcpu->arch.cr4_guest_owned_bits));
3485 }
3486 
3487 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3488 				      struct vmcs12 *vmcs12)
3489 {
3490 	u32 idt_vectoring;
3491 	unsigned int nr;
3492 
3493 	if (vcpu->arch.exception.injected) {
3494 		nr = vcpu->arch.exception.nr;
3495 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3496 
3497 		if (kvm_exception_is_soft(nr)) {
3498 			vmcs12->vm_exit_instruction_len =
3499 				vcpu->arch.event_exit_inst_len;
3500 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3501 		} else
3502 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3503 
3504 		if (vcpu->arch.exception.has_error_code) {
3505 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3506 			vmcs12->idt_vectoring_error_code =
3507 				vcpu->arch.exception.error_code;
3508 		}
3509 
3510 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3511 	} else if (vcpu->arch.nmi_injected) {
3512 		vmcs12->idt_vectoring_info_field =
3513 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3514 	} else if (vcpu->arch.interrupt.injected) {
3515 		nr = vcpu->arch.interrupt.nr;
3516 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3517 
3518 		if (vcpu->arch.interrupt.soft) {
3519 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3520 			vmcs12->vm_entry_instruction_len =
3521 				vcpu->arch.event_exit_inst_len;
3522 		} else
3523 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3524 
3525 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3526 	}
3527 }
3528 
3529 
3530 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3531 {
3532 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3533 	gfn_t gfn;
3534 
3535 	/*
3536 	 * Don't need to mark the APIC access page dirty; it is never
3537 	 * written to by the CPU during APIC virtualization.
3538 	 */
3539 
3540 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3541 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3542 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3543 	}
3544 
3545 	if (nested_cpu_has_posted_intr(vmcs12)) {
3546 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3547 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3548 	}
3549 }
3550 
3551 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3552 {
3553 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3554 	int max_irr;
3555 	void *vapic_page;
3556 	u16 status;
3557 
3558 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3559 		return;
3560 
3561 	vmx->nested.pi_pending = false;
3562 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3563 		return;
3564 
3565 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3566 	if (max_irr != 256) {
3567 		vapic_page = vmx->nested.virtual_apic_map.hva;
3568 		if (!vapic_page)
3569 			return;
3570 
3571 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3572 			vapic_page, &max_irr);
3573 		status = vmcs_read16(GUEST_INTR_STATUS);
3574 		if ((u8)max_irr > ((u8)status & 0xff)) {
3575 			status &= ~0xff;
3576 			status |= (u8)max_irr;
3577 			vmcs_write16(GUEST_INTR_STATUS, status);
3578 		}
3579 	}
3580 
3581 	nested_mark_vmcs12_pages_dirty(vcpu);
3582 }
3583 
3584 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3585 					       unsigned long exit_qual)
3586 {
3587 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3588 	unsigned int nr = vcpu->arch.exception.nr;
3589 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3590 
3591 	if (vcpu->arch.exception.has_error_code) {
3592 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3593 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3594 	}
3595 
3596 	if (kvm_exception_is_soft(nr))
3597 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3598 	else
3599 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3600 
3601 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3602 	    vmx_get_nmi_mask(vcpu))
3603 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3604 
3605 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3606 }
3607 
3608 /*
3609  * Returns true if a debug trap is pending delivery.
3610  *
3611  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3612  * exception may be inferred from the presence of an exception payload.
3613  */
3614 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3615 {
3616 	return vcpu->arch.exception.pending &&
3617 			vcpu->arch.exception.nr == DB_VECTOR &&
3618 			vcpu->arch.exception.payload;
3619 }
3620 
3621 /*
3622  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3623  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3624  * represents these debug traps with a payload that is said to be compatible
3625  * with the 'pending debug exceptions' field, write the payload to the VMCS
3626  * field if a VM-exit is delivered before the debug trap.
3627  */
3628 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3629 {
3630 	if (vmx_pending_dbg_trap(vcpu))
3631 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3632 			    vcpu->arch.exception.payload);
3633 }
3634 
3635 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3636 {
3637 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3638 	unsigned long exit_qual;
3639 	bool block_nested_events =
3640 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3641 	bool mtf_pending = vmx->nested.mtf_pending;
3642 	struct kvm_lapic *apic = vcpu->arch.apic;
3643 
3644 	/*
3645 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3646 	 * this state is discarded.
3647 	 */
3648 	if (!block_nested_events)
3649 		vmx->nested.mtf_pending = false;
3650 
3651 	if (lapic_in_kernel(vcpu) &&
3652 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3653 		if (block_nested_events)
3654 			return -EBUSY;
3655 		nested_vmx_update_pending_dbg(vcpu);
3656 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3657 		nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3658 		return 0;
3659 	}
3660 
3661 	/*
3662 	 * Process any exceptions that are not debug traps before MTF.
3663 	 */
3664 	if (vcpu->arch.exception.pending &&
3665 	    !vmx_pending_dbg_trap(vcpu) &&
3666 	    nested_vmx_check_exception(vcpu, &exit_qual)) {
3667 		if (block_nested_events)
3668 			return -EBUSY;
3669 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3670 		return 0;
3671 	}
3672 
3673 	if (mtf_pending) {
3674 		if (block_nested_events)
3675 			return -EBUSY;
3676 		nested_vmx_update_pending_dbg(vcpu);
3677 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3678 		return 0;
3679 	}
3680 
3681 	if (vcpu->arch.exception.pending &&
3682 	    nested_vmx_check_exception(vcpu, &exit_qual)) {
3683 		if (block_nested_events)
3684 			return -EBUSY;
3685 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3686 		return 0;
3687 	}
3688 
3689 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3690 	    vmx->nested.preemption_timer_expired) {
3691 		if (block_nested_events)
3692 			return -EBUSY;
3693 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3694 		return 0;
3695 	}
3696 
3697 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3698 		if (block_nested_events)
3699 			return -EBUSY;
3700 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3701 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3702 				  INTR_INFO_VALID_MASK, 0);
3703 		/*
3704 		 * The NMI-triggered VM exit counts as injection:
3705 		 * clear this one and block further NMIs.
3706 		 */
3707 		vcpu->arch.nmi_pending = 0;
3708 		vmx_set_nmi_mask(vcpu, true);
3709 		return 0;
3710 	}
3711 
3712 	if (kvm_cpu_has_interrupt(vcpu) && nested_exit_on_intr(vcpu)) {
3713 		if (block_nested_events)
3714 			return -EBUSY;
3715 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3716 		return 0;
3717 	}
3718 
3719 	vmx_complete_nested_posted_interrupt(vcpu);
3720 	return 0;
3721 }
3722 
3723 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3724 {
3725 	ktime_t remaining =
3726 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3727 	u64 value;
3728 
3729 	if (ktime_to_ns(remaining) <= 0)
3730 		return 0;
3731 
3732 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3733 	do_div(value, 1000000);
3734 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3735 }
3736 
3737 static bool is_vmcs12_ext_field(unsigned long field)
3738 {
3739 	switch (field) {
3740 	case GUEST_ES_SELECTOR:
3741 	case GUEST_CS_SELECTOR:
3742 	case GUEST_SS_SELECTOR:
3743 	case GUEST_DS_SELECTOR:
3744 	case GUEST_FS_SELECTOR:
3745 	case GUEST_GS_SELECTOR:
3746 	case GUEST_LDTR_SELECTOR:
3747 	case GUEST_TR_SELECTOR:
3748 	case GUEST_ES_LIMIT:
3749 	case GUEST_CS_LIMIT:
3750 	case GUEST_SS_LIMIT:
3751 	case GUEST_DS_LIMIT:
3752 	case GUEST_FS_LIMIT:
3753 	case GUEST_GS_LIMIT:
3754 	case GUEST_LDTR_LIMIT:
3755 	case GUEST_TR_LIMIT:
3756 	case GUEST_GDTR_LIMIT:
3757 	case GUEST_IDTR_LIMIT:
3758 	case GUEST_ES_AR_BYTES:
3759 	case GUEST_DS_AR_BYTES:
3760 	case GUEST_FS_AR_BYTES:
3761 	case GUEST_GS_AR_BYTES:
3762 	case GUEST_LDTR_AR_BYTES:
3763 	case GUEST_TR_AR_BYTES:
3764 	case GUEST_ES_BASE:
3765 	case GUEST_CS_BASE:
3766 	case GUEST_SS_BASE:
3767 	case GUEST_DS_BASE:
3768 	case GUEST_FS_BASE:
3769 	case GUEST_GS_BASE:
3770 	case GUEST_LDTR_BASE:
3771 	case GUEST_TR_BASE:
3772 	case GUEST_GDTR_BASE:
3773 	case GUEST_IDTR_BASE:
3774 	case GUEST_PENDING_DBG_EXCEPTIONS:
3775 	case GUEST_BNDCFGS:
3776 		return true;
3777 	default:
3778 		break;
3779 	}
3780 
3781 	return false;
3782 }
3783 
3784 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3785 				       struct vmcs12 *vmcs12)
3786 {
3787 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3788 
3789 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3790 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3791 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3792 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3793 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3794 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3795 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3796 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3797 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3798 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3799 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3800 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3801 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3802 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3803 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3804 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3805 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3806 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3807 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3808 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3809 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3810 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3811 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3812 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3813 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3814 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3815 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3816 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3817 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3818 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3819 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3820 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3821 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3822 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3823 	vmcs12->guest_pending_dbg_exceptions =
3824 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3825 	if (kvm_mpx_supported())
3826 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3827 
3828 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
3829 }
3830 
3831 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3832 				       struct vmcs12 *vmcs12)
3833 {
3834 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3835 	int cpu;
3836 
3837 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
3838 		return;
3839 
3840 
3841 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
3842 
3843 	cpu = get_cpu();
3844 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
3845 	vmx_vcpu_load(&vmx->vcpu, cpu);
3846 
3847 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3848 
3849 	vmx->loaded_vmcs = &vmx->vmcs01;
3850 	vmx_vcpu_load(&vmx->vcpu, cpu);
3851 	put_cpu();
3852 }
3853 
3854 /*
3855  * Update the guest state fields of vmcs12 to reflect changes that
3856  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3857  * VM-entry controls is also updated, since this is really a guest
3858  * state bit.)
3859  */
3860 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3861 {
3862 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3863 
3864 	if (vmx->nested.hv_evmcs)
3865 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
3866 
3867 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = !vmx->nested.hv_evmcs;
3868 
3869 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3870 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3871 
3872 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
3873 	vmcs12->guest_rip = kvm_rip_read(vcpu);
3874 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3875 
3876 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3877 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3878 
3879 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3880 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3881 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3882 
3883 	vmcs12->guest_interruptibility_info =
3884 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3885 
3886 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3887 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3888 	else
3889 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3890 
3891 	if (nested_cpu_has_preemption_timer(vmcs12) &&
3892 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3893 			vmcs12->vmx_preemption_timer_value =
3894 				vmx_get_preemption_timer_value(vcpu);
3895 
3896 	/*
3897 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3898 	 * own CR3 without exiting. If it has changed it, we must keep it.
3899 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3900 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3901 	 *
3902 	 * Additionally, restore L2's PDPTR to vmcs12.
3903 	 */
3904 	if (enable_ept) {
3905 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3906 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3907 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3908 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3909 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3910 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3911 		}
3912 	}
3913 
3914 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3915 
3916 	if (nested_cpu_has_vid(vmcs12))
3917 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3918 
3919 	vmcs12->vm_entry_controls =
3920 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3921 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3922 
3923 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
3924 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3925 
3926 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3927 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3928 }
3929 
3930 /*
3931  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3932  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3933  * and this function updates it to reflect the changes to the guest state while
3934  * L2 was running (and perhaps made some exits which were handled directly by L0
3935  * without going back to L1), and to reflect the exit reason.
3936  * Note that we do not have to copy here all VMCS fields, just those that
3937  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3938  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3939  * which already writes to vmcs12 directly.
3940  */
3941 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3942 			   u32 exit_reason, u32 exit_intr_info,
3943 			   unsigned long exit_qualification)
3944 {
3945 	/* update exit information fields: */
3946 	vmcs12->vm_exit_reason = exit_reason;
3947 	vmcs12->exit_qualification = exit_qualification;
3948 	vmcs12->vm_exit_intr_info = exit_intr_info;
3949 
3950 	vmcs12->idt_vectoring_info_field = 0;
3951 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3952 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3953 
3954 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3955 		vmcs12->launch_state = 1;
3956 
3957 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3958 		 * instead of reading the real value. */
3959 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3960 
3961 		/*
3962 		 * Transfer the event that L0 or L1 may wanted to inject into
3963 		 * L2 to IDT_VECTORING_INFO_FIELD.
3964 		 */
3965 		vmcs12_save_pending_event(vcpu, vmcs12);
3966 
3967 		/*
3968 		 * According to spec, there's no need to store the guest's
3969 		 * MSRs if the exit is due to a VM-entry failure that occurs
3970 		 * during or after loading the guest state. Since this exit
3971 		 * does not fall in that category, we need to save the MSRs.
3972 		 */
3973 		if (nested_vmx_store_msr(vcpu,
3974 					 vmcs12->vm_exit_msr_store_addr,
3975 					 vmcs12->vm_exit_msr_store_count))
3976 			nested_vmx_abort(vcpu,
3977 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3978 	}
3979 
3980 	/*
3981 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3982 	 * preserved above and would only end up incorrectly in L1.
3983 	 */
3984 	vcpu->arch.nmi_injected = false;
3985 	kvm_clear_exception_queue(vcpu);
3986 	kvm_clear_interrupt_queue(vcpu);
3987 }
3988 
3989 /*
3990  * A part of what we need to when the nested L2 guest exits and we want to
3991  * run its L1 parent, is to reset L1's guest state to the host state specified
3992  * in vmcs12.
3993  * This function is to be called not only on normal nested exit, but also on
3994  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3995  * Failures During or After Loading Guest State").
3996  * This function should be called when the active VMCS is L1's (vmcs01).
3997  */
3998 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3999 				   struct vmcs12 *vmcs12)
4000 {
4001 	struct kvm_segment seg;
4002 	u32 entry_failure_code;
4003 
4004 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4005 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4006 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4007 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4008 	else
4009 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4010 	vmx_set_efer(vcpu, vcpu->arch.efer);
4011 
4012 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4013 	kvm_rip_write(vcpu, vmcs12->host_rip);
4014 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4015 	vmx_set_interrupt_shadow(vcpu, 0);
4016 
4017 	/*
4018 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4019 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4020 	 *
4021 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4022 	 * (KVM doesn't change it);
4023 	 */
4024 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
4025 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4026 
4027 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4028 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4029 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4030 
4031 	nested_ept_uninit_mmu_context(vcpu);
4032 
4033 	/*
4034 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4035 	 * couldn't have changed.
4036 	 */
4037 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
4038 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4039 
4040 	if (!enable_ept)
4041 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
4042 
4043 	/*
4044 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
4045 	 * VMEntry/VMExit. Thus, no need to flush TLB.
4046 	 *
4047 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
4048 	 * flushed on every VMEntry/VMExit.
4049 	 *
4050 	 * Otherwise, we can preserve TLB entries as long as we are
4051 	 * able to tag L1 TLB entries differently than L2 TLB entries.
4052 	 *
4053 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
4054 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
4055 	 * has been set by KVM_REQ_LOAD_MMU_PGD.
4056 	 */
4057 	if (enable_vpid &&
4058 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
4059 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4060 	}
4061 
4062 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4063 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4064 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4065 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4066 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4067 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4068 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4069 
4070 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4071 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4072 		vmcs_write64(GUEST_BNDCFGS, 0);
4073 
4074 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4075 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4076 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4077 	}
4078 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4079 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4080 					 vmcs12->host_ia32_perf_global_ctrl));
4081 
4082 	/* Set L1 segment info according to Intel SDM
4083 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4084 	seg = (struct kvm_segment) {
4085 		.base = 0,
4086 		.limit = 0xFFFFFFFF,
4087 		.selector = vmcs12->host_cs_selector,
4088 		.type = 11,
4089 		.present = 1,
4090 		.s = 1,
4091 		.g = 1
4092 	};
4093 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4094 		seg.l = 1;
4095 	else
4096 		seg.db = 1;
4097 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4098 	seg = (struct kvm_segment) {
4099 		.base = 0,
4100 		.limit = 0xFFFFFFFF,
4101 		.type = 3,
4102 		.present = 1,
4103 		.s = 1,
4104 		.db = 1,
4105 		.g = 1
4106 	};
4107 	seg.selector = vmcs12->host_ds_selector;
4108 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4109 	seg.selector = vmcs12->host_es_selector;
4110 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4111 	seg.selector = vmcs12->host_ss_selector;
4112 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4113 	seg.selector = vmcs12->host_fs_selector;
4114 	seg.base = vmcs12->host_fs_base;
4115 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4116 	seg.selector = vmcs12->host_gs_selector;
4117 	seg.base = vmcs12->host_gs_base;
4118 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4119 	seg = (struct kvm_segment) {
4120 		.base = vmcs12->host_tr_base,
4121 		.limit = 0x67,
4122 		.selector = vmcs12->host_tr_selector,
4123 		.type = 11,
4124 		.present = 1
4125 	};
4126 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4127 
4128 	kvm_set_dr(vcpu, 7, 0x400);
4129 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4130 
4131 	if (cpu_has_vmx_msr_bitmap())
4132 		vmx_update_msr_bitmap(vcpu);
4133 
4134 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4135 				vmcs12->vm_exit_msr_load_count))
4136 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4137 }
4138 
4139 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4140 {
4141 	struct shared_msr_entry *efer_msr;
4142 	unsigned int i;
4143 
4144 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4145 		return vmcs_read64(GUEST_IA32_EFER);
4146 
4147 	if (cpu_has_load_ia32_efer())
4148 		return host_efer;
4149 
4150 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4151 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4152 			return vmx->msr_autoload.guest.val[i].value;
4153 	}
4154 
4155 	efer_msr = find_msr_entry(vmx, MSR_EFER);
4156 	if (efer_msr)
4157 		return efer_msr->data;
4158 
4159 	return host_efer;
4160 }
4161 
4162 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4163 {
4164 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4165 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4166 	struct vmx_msr_entry g, h;
4167 	gpa_t gpa;
4168 	u32 i, j;
4169 
4170 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4171 
4172 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4173 		/*
4174 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4175 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4176 		 * and vcpu->arch.dr7 is not squirreled away before the
4177 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4178 		 */
4179 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4180 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4181 		else
4182 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4183 	}
4184 
4185 	/*
4186 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4187 	 * handle a variety of side effects to KVM's software model.
4188 	 */
4189 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4190 
4191 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
4192 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4193 
4194 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4195 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4196 
4197 	nested_ept_uninit_mmu_context(vcpu);
4198 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4199 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4200 
4201 	/*
4202 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4203 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4204 	 * VMFail, like everything else we just need to ensure our
4205 	 * software model is up-to-date.
4206 	 */
4207 	if (enable_ept)
4208 		ept_save_pdptrs(vcpu);
4209 
4210 	kvm_mmu_reset_context(vcpu);
4211 
4212 	if (cpu_has_vmx_msr_bitmap())
4213 		vmx_update_msr_bitmap(vcpu);
4214 
4215 	/*
4216 	 * This nasty bit of open coding is a compromise between blindly
4217 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4218 	 * of VMFail), leaving the nested VM's MSRs in the software model
4219 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4220 	 * expensive since the lists are unbound by hardware).  For each
4221 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4222 	 * list, reload it from the exit load list if it exists and differs
4223 	 * from the guest value.  The intent is to stuff host state as
4224 	 * silently as possible, not to fully process the exit load list.
4225 	 */
4226 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4227 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4228 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4229 			pr_debug_ratelimited(
4230 				"%s read MSR index failed (%u, 0x%08llx)\n",
4231 				__func__, i, gpa);
4232 			goto vmabort;
4233 		}
4234 
4235 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4236 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4237 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4238 				pr_debug_ratelimited(
4239 					"%s read MSR failed (%u, 0x%08llx)\n",
4240 					__func__, j, gpa);
4241 				goto vmabort;
4242 			}
4243 			if (h.index != g.index)
4244 				continue;
4245 			if (h.value == g.value)
4246 				break;
4247 
4248 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4249 				pr_debug_ratelimited(
4250 					"%s check failed (%u, 0x%x, 0x%x)\n",
4251 					__func__, j, h.index, h.reserved);
4252 				goto vmabort;
4253 			}
4254 
4255 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4256 				pr_debug_ratelimited(
4257 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4258 					__func__, j, h.index, h.value);
4259 				goto vmabort;
4260 			}
4261 		}
4262 	}
4263 
4264 	return;
4265 
4266 vmabort:
4267 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4268 }
4269 
4270 /*
4271  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4272  * and modify vmcs12 to make it see what it would expect to see there if
4273  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4274  */
4275 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
4276 		       u32 exit_intr_info, unsigned long exit_qualification)
4277 {
4278 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4279 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4280 
4281 	/* trying to cancel vmlaunch/vmresume is a bug */
4282 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4283 
4284 	leave_guest_mode(vcpu);
4285 
4286 	if (nested_cpu_has_preemption_timer(vmcs12))
4287 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4288 
4289 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
4290 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
4291 
4292 	if (likely(!vmx->fail)) {
4293 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4294 
4295 		if (exit_reason != -1)
4296 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
4297 				       exit_qualification);
4298 
4299 		/*
4300 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4301 		 * also be used to capture vmcs12 cache as part of
4302 		 * capturing nVMX state for snapshot (migration).
4303 		 *
4304 		 * Otherwise, this flush will dirty guest memory at a
4305 		 * point it is already assumed by user-space to be
4306 		 * immutable.
4307 		 */
4308 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4309 	} else {
4310 		/*
4311 		 * The only expected VM-instruction error is "VM entry with
4312 		 * invalid control field(s)." Anything else indicates a
4313 		 * problem with L0.  And we should never get here with a
4314 		 * VMFail of any type if early consistency checks are enabled.
4315 		 */
4316 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4317 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4318 		WARN_ON_ONCE(nested_early_check);
4319 	}
4320 
4321 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4322 
4323 	/* Update any VMCS fields that might have changed while L2 ran */
4324 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4325 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4326 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4327 	if (vmx->nested.l1_tpr_threshold != -1)
4328 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4329 
4330 	if (kvm_has_tsc_control)
4331 		decache_tsc_multiplier(vmx);
4332 
4333 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4334 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4335 		vmx_set_virtual_apic_mode(vcpu);
4336 	}
4337 
4338 	/* Unpin physical memory we referred to in vmcs02 */
4339 	if (vmx->nested.apic_access_page) {
4340 		kvm_release_page_clean(vmx->nested.apic_access_page);
4341 		vmx->nested.apic_access_page = NULL;
4342 	}
4343 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4344 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4345 	vmx->nested.pi_desc = NULL;
4346 
4347 	/*
4348 	 * We are now running in L2, mmu_notifier will force to reload the
4349 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
4350 	 */
4351 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4352 
4353 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
4354 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4355 
4356 	/* in case we halted in L2 */
4357 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4358 
4359 	if (likely(!vmx->fail)) {
4360 		if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4361 		    nested_exit_intr_ack_set(vcpu)) {
4362 			int irq = kvm_cpu_get_interrupt(vcpu);
4363 			WARN_ON(irq < 0);
4364 			vmcs12->vm_exit_intr_info = irq |
4365 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4366 		}
4367 
4368 		if (exit_reason != -1)
4369 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4370 						       vmcs12->exit_qualification,
4371 						       vmcs12->idt_vectoring_info_field,
4372 						       vmcs12->vm_exit_intr_info,
4373 						       vmcs12->vm_exit_intr_error_code,
4374 						       KVM_ISA_VMX);
4375 
4376 		load_vmcs12_host_state(vcpu, vmcs12);
4377 
4378 		return;
4379 	}
4380 
4381 	/*
4382 	 * After an early L2 VM-entry failure, we're now back
4383 	 * in L1 which thinks it just finished a VMLAUNCH or
4384 	 * VMRESUME instruction, so we need to set the failure
4385 	 * flag and the VM-instruction error field of the VMCS
4386 	 * accordingly, and skip the emulated instruction.
4387 	 */
4388 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4389 
4390 	/*
4391 	 * Restore L1's host state to KVM's software model.  We're here
4392 	 * because a consistency check was caught by hardware, which
4393 	 * means some amount of guest state has been propagated to KVM's
4394 	 * model and needs to be unwound to the host's state.
4395 	 */
4396 	nested_vmx_restore_host_state(vcpu);
4397 
4398 	vmx->fail = 0;
4399 }
4400 
4401 /*
4402  * Decode the memory-address operand of a vmx instruction, as recorded on an
4403  * exit caused by such an instruction (run by a guest hypervisor).
4404  * On success, returns 0. When the operand is invalid, returns 1 and throws
4405  * #UD, #GP, or #SS.
4406  */
4407 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4408 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4409 {
4410 	gva_t off;
4411 	bool exn;
4412 	struct kvm_segment s;
4413 
4414 	/*
4415 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4416 	 * Execution", on an exit, vmx_instruction_info holds most of the
4417 	 * addressing components of the operand. Only the displacement part
4418 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4419 	 * For how an actual address is calculated from all these components,
4420 	 * refer to Vol. 1, "Operand Addressing".
4421 	 */
4422 	int  scaling = vmx_instruction_info & 3;
4423 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4424 	bool is_reg = vmx_instruction_info & (1u << 10);
4425 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4426 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4427 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4428 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4429 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4430 
4431 	if (is_reg) {
4432 		kvm_queue_exception(vcpu, UD_VECTOR);
4433 		return 1;
4434 	}
4435 
4436 	/* Addr = segment_base + offset */
4437 	/* offset = base + [index * scale] + displacement */
4438 	off = exit_qualification; /* holds the displacement */
4439 	if (addr_size == 1)
4440 		off = (gva_t)sign_extend64(off, 31);
4441 	else if (addr_size == 0)
4442 		off = (gva_t)sign_extend64(off, 15);
4443 	if (base_is_valid)
4444 		off += kvm_register_read(vcpu, base_reg);
4445 	if (index_is_valid)
4446 		off += kvm_register_read(vcpu, index_reg) << scaling;
4447 	vmx_get_segment(vcpu, &s, seg_reg);
4448 
4449 	/*
4450 	 * The effective address, i.e. @off, of a memory operand is truncated
4451 	 * based on the address size of the instruction.  Note that this is
4452 	 * the *effective address*, i.e. the address prior to accounting for
4453 	 * the segment's base.
4454 	 */
4455 	if (addr_size == 1) /* 32 bit */
4456 		off &= 0xffffffff;
4457 	else if (addr_size == 0) /* 16 bit */
4458 		off &= 0xffff;
4459 
4460 	/* Checks for #GP/#SS exceptions. */
4461 	exn = false;
4462 	if (is_long_mode(vcpu)) {
4463 		/*
4464 		 * The virtual/linear address is never truncated in 64-bit
4465 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4466 		 * address when using FS/GS with a non-zero base.
4467 		 */
4468 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4469 			*ret = s.base + off;
4470 		else
4471 			*ret = off;
4472 
4473 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4474 		 * non-canonical form. This is the only check on the memory
4475 		 * destination for long mode!
4476 		 */
4477 		exn = is_noncanonical_address(*ret, vcpu);
4478 	} else {
4479 		/*
4480 		 * When not in long mode, the virtual/linear address is
4481 		 * unconditionally truncated to 32 bits regardless of the
4482 		 * address size.
4483 		 */
4484 		*ret = (s.base + off) & 0xffffffff;
4485 
4486 		/* Protected mode: apply checks for segment validity in the
4487 		 * following order:
4488 		 * - segment type check (#GP(0) may be thrown)
4489 		 * - usability check (#GP(0)/#SS(0))
4490 		 * - limit check (#GP(0)/#SS(0))
4491 		 */
4492 		if (wr)
4493 			/* #GP(0) if the destination operand is located in a
4494 			 * read-only data segment or any code segment.
4495 			 */
4496 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4497 		else
4498 			/* #GP(0) if the source operand is located in an
4499 			 * execute-only code segment
4500 			 */
4501 			exn = ((s.type & 0xa) == 8);
4502 		if (exn) {
4503 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4504 			return 1;
4505 		}
4506 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4507 		 */
4508 		exn = (s.unusable != 0);
4509 
4510 		/*
4511 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4512 		 * outside the segment limit.  All CPUs that support VMX ignore
4513 		 * limit checks for flat segments, i.e. segments with base==0,
4514 		 * limit==0xffffffff and of type expand-up data or code.
4515 		 */
4516 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4517 		     ((s.type & 8) || !(s.type & 4))))
4518 			exn = exn || ((u64)off + len - 1 > s.limit);
4519 	}
4520 	if (exn) {
4521 		kvm_queue_exception_e(vcpu,
4522 				      seg_reg == VCPU_SREG_SS ?
4523 						SS_VECTOR : GP_VECTOR,
4524 				      0);
4525 		return 1;
4526 	}
4527 
4528 	return 0;
4529 }
4530 
4531 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4532 {
4533 	struct vcpu_vmx *vmx;
4534 
4535 	if (!nested_vmx_allowed(vcpu))
4536 		return;
4537 
4538 	vmx = to_vmx(vcpu);
4539 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4540 		vmx->nested.msrs.entry_ctls_high |=
4541 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4542 		vmx->nested.msrs.exit_ctls_high |=
4543 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4544 	} else {
4545 		vmx->nested.msrs.entry_ctls_high &=
4546 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4547 		vmx->nested.msrs.exit_ctls_high &=
4548 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4549 	}
4550 }
4551 
4552 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4553 {
4554 	gva_t gva;
4555 	struct x86_exception e;
4556 
4557 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4558 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4559 				sizeof(*vmpointer), &gva))
4560 		return 1;
4561 
4562 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4563 		kvm_inject_page_fault(vcpu, &e);
4564 		return 1;
4565 	}
4566 
4567 	return 0;
4568 }
4569 
4570 /*
4571  * Allocate a shadow VMCS and associate it with the currently loaded
4572  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4573  * VMCS is also VMCLEARed, so that it is ready for use.
4574  */
4575 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4576 {
4577 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4578 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4579 
4580 	/*
4581 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4582 	 * executes VMXON and free it when L1 executes VMXOFF.
4583 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4584 	 * here when vmcs01 already have an allocated shadow vmcs.
4585 	 */
4586 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4587 
4588 	if (!loaded_vmcs->shadow_vmcs) {
4589 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4590 		if (loaded_vmcs->shadow_vmcs)
4591 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4592 	}
4593 	return loaded_vmcs->shadow_vmcs;
4594 }
4595 
4596 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4597 {
4598 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4599 	int r;
4600 
4601 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4602 	if (r < 0)
4603 		goto out_vmcs02;
4604 
4605 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4606 	if (!vmx->nested.cached_vmcs12)
4607 		goto out_cached_vmcs12;
4608 
4609 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4610 	if (!vmx->nested.cached_shadow_vmcs12)
4611 		goto out_cached_shadow_vmcs12;
4612 
4613 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4614 		goto out_shadow_vmcs;
4615 
4616 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4617 		     HRTIMER_MODE_REL_PINNED);
4618 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4619 
4620 	vmx->nested.vpid02 = allocate_vpid();
4621 
4622 	vmx->nested.vmcs02_initialized = false;
4623 	vmx->nested.vmxon = true;
4624 
4625 	if (vmx_pt_mode_is_host_guest()) {
4626 		vmx->pt_desc.guest.ctl = 0;
4627 		pt_update_intercept_for_msr(vmx);
4628 	}
4629 
4630 	return 0;
4631 
4632 out_shadow_vmcs:
4633 	kfree(vmx->nested.cached_shadow_vmcs12);
4634 
4635 out_cached_shadow_vmcs12:
4636 	kfree(vmx->nested.cached_vmcs12);
4637 
4638 out_cached_vmcs12:
4639 	free_loaded_vmcs(&vmx->nested.vmcs02);
4640 
4641 out_vmcs02:
4642 	return -ENOMEM;
4643 }
4644 
4645 /*
4646  * Emulate the VMXON instruction.
4647  * Currently, we just remember that VMX is active, and do not save or even
4648  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4649  * do not currently need to store anything in that guest-allocated memory
4650  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4651  * argument is different from the VMXON pointer (which the spec says they do).
4652  */
4653 static int handle_vmon(struct kvm_vcpu *vcpu)
4654 {
4655 	int ret;
4656 	gpa_t vmptr;
4657 	uint32_t revision;
4658 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4659 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4660 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4661 
4662 	/*
4663 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4664 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4665 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4666 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4667 	 * have already been checked by hardware, prior to the VM-exit for
4668 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4669 	 * that bit set to 1 in non-root mode.
4670 	 */
4671 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4672 		kvm_queue_exception(vcpu, UD_VECTOR);
4673 		return 1;
4674 	}
4675 
4676 	/* CPL=0 must be checked manually. */
4677 	if (vmx_get_cpl(vcpu)) {
4678 		kvm_inject_gp(vcpu, 0);
4679 		return 1;
4680 	}
4681 
4682 	if (vmx->nested.vmxon)
4683 		return nested_vmx_failValid(vcpu,
4684 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4685 
4686 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4687 			!= VMXON_NEEDED_FEATURES) {
4688 		kvm_inject_gp(vcpu, 0);
4689 		return 1;
4690 	}
4691 
4692 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4693 		return 1;
4694 
4695 	/*
4696 	 * SDM 3: 24.11.5
4697 	 * The first 4 bytes of VMXON region contain the supported
4698 	 * VMCS revision identifier
4699 	 *
4700 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4701 	 * which replaces physical address width with 32
4702 	 */
4703 	if (!page_address_valid(vcpu, vmptr))
4704 		return nested_vmx_failInvalid(vcpu);
4705 
4706 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4707 	    revision != VMCS12_REVISION)
4708 		return nested_vmx_failInvalid(vcpu);
4709 
4710 	vmx->nested.vmxon_ptr = vmptr;
4711 	ret = enter_vmx_operation(vcpu);
4712 	if (ret)
4713 		return ret;
4714 
4715 	return nested_vmx_succeed(vcpu);
4716 }
4717 
4718 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4719 {
4720 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4721 
4722 	if (vmx->nested.current_vmptr == -1ull)
4723 		return;
4724 
4725 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4726 
4727 	if (enable_shadow_vmcs) {
4728 		/* copy to memory all shadowed fields in case
4729 		   they were modified */
4730 		copy_shadow_to_vmcs12(vmx);
4731 		vmx_disable_shadow_vmcs(vmx);
4732 	}
4733 	vmx->nested.posted_intr_nv = -1;
4734 
4735 	/* Flush VMCS12 to guest memory */
4736 	kvm_vcpu_write_guest_page(vcpu,
4737 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4738 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4739 
4740 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4741 
4742 	vmx->nested.current_vmptr = -1ull;
4743 }
4744 
4745 /* Emulate the VMXOFF instruction */
4746 static int handle_vmoff(struct kvm_vcpu *vcpu)
4747 {
4748 	if (!nested_vmx_check_permission(vcpu))
4749 		return 1;
4750 
4751 	free_nested(vcpu);
4752 
4753 	/* Process a latched INIT during time CPU was in VMX operation */
4754 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4755 
4756 	return nested_vmx_succeed(vcpu);
4757 }
4758 
4759 /* Emulate the VMCLEAR instruction */
4760 static int handle_vmclear(struct kvm_vcpu *vcpu)
4761 {
4762 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4763 	u32 zero = 0;
4764 	gpa_t vmptr;
4765 	u64 evmcs_gpa;
4766 
4767 	if (!nested_vmx_check_permission(vcpu))
4768 		return 1;
4769 
4770 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4771 		return 1;
4772 
4773 	if (!page_address_valid(vcpu, vmptr))
4774 		return nested_vmx_failValid(vcpu,
4775 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4776 
4777 	if (vmptr == vmx->nested.vmxon_ptr)
4778 		return nested_vmx_failValid(vcpu,
4779 			VMXERR_VMCLEAR_VMXON_POINTER);
4780 
4781 	/*
4782 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4783 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4784 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4785 	 * writing to the non-existent 'launch_state' field. The area doesn't
4786 	 * have to be the currently active EVMCS on the calling CPU and there's
4787 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
4788 	 * state. It is possible that the area will stay mapped as
4789 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
4790 	 */
4791 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
4792 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
4793 		if (vmptr == vmx->nested.current_vmptr)
4794 			nested_release_vmcs12(vcpu);
4795 
4796 		kvm_vcpu_write_guest(vcpu,
4797 				     vmptr + offsetof(struct vmcs12,
4798 						      launch_state),
4799 				     &zero, sizeof(zero));
4800 	}
4801 
4802 	return nested_vmx_succeed(vcpu);
4803 }
4804 
4805 /* Emulate the VMLAUNCH instruction */
4806 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4807 {
4808 	return nested_vmx_run(vcpu, true);
4809 }
4810 
4811 /* Emulate the VMRESUME instruction */
4812 static int handle_vmresume(struct kvm_vcpu *vcpu)
4813 {
4814 
4815 	return nested_vmx_run(vcpu, false);
4816 }
4817 
4818 static int handle_vmread(struct kvm_vcpu *vcpu)
4819 {
4820 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4821 						    : get_vmcs12(vcpu);
4822 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4823 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4824 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4825 	struct x86_exception e;
4826 	unsigned long field;
4827 	u64 value;
4828 	gva_t gva = 0;
4829 	short offset;
4830 	int len;
4831 
4832 	if (!nested_vmx_check_permission(vcpu))
4833 		return 1;
4834 
4835 	/*
4836 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
4837 	 * any VMREAD sets the ALU flags for VMfailInvalid.
4838 	 */
4839 	if (vmx->nested.current_vmptr == -1ull ||
4840 	    (is_guest_mode(vcpu) &&
4841 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
4842 		return nested_vmx_failInvalid(vcpu);
4843 
4844 	/* Decode instruction info and find the field to read */
4845 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
4846 
4847 	offset = vmcs_field_to_offset(field);
4848 	if (offset < 0)
4849 		return nested_vmx_failValid(vcpu,
4850 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4851 
4852 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
4853 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4854 
4855 	/* Read the field, zero-extended to a u64 value */
4856 	value = vmcs12_read_any(vmcs12, field, offset);
4857 
4858 	/*
4859 	 * Now copy part of this value to register or memory, as requested.
4860 	 * Note that the number of bits actually copied is 32 or 64 depending
4861 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4862 	 */
4863 	if (instr_info & BIT(10)) {
4864 		kvm_register_writel(vcpu, (((instr_info) >> 3) & 0xf), value);
4865 	} else {
4866 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4867 		if (get_vmx_mem_address(vcpu, exit_qualification,
4868 					instr_info, true, len, &gva))
4869 			return 1;
4870 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4871 		if (kvm_write_guest_virt_system(vcpu, gva, &value, len, &e)) {
4872 			kvm_inject_page_fault(vcpu, &e);
4873 			return 1;
4874 		}
4875 	}
4876 
4877 	return nested_vmx_succeed(vcpu);
4878 }
4879 
4880 static bool is_shadow_field_rw(unsigned long field)
4881 {
4882 	switch (field) {
4883 #define SHADOW_FIELD_RW(x, y) case x:
4884 #include "vmcs_shadow_fields.h"
4885 		return true;
4886 	default:
4887 		break;
4888 	}
4889 	return false;
4890 }
4891 
4892 static bool is_shadow_field_ro(unsigned long field)
4893 {
4894 	switch (field) {
4895 #define SHADOW_FIELD_RO(x, y) case x:
4896 #include "vmcs_shadow_fields.h"
4897 		return true;
4898 	default:
4899 		break;
4900 	}
4901 	return false;
4902 }
4903 
4904 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4905 {
4906 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
4907 						    : get_vmcs12(vcpu);
4908 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4909 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4910 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4911 	struct x86_exception e;
4912 	unsigned long field;
4913 	short offset;
4914 	gva_t gva;
4915 	int len;
4916 
4917 	/*
4918 	 * The value to write might be 32 or 64 bits, depending on L1's long
4919 	 * mode, and eventually we need to write that into a field of several
4920 	 * possible lengths. The code below first zero-extends the value to 64
4921 	 * bit (value), and then copies only the appropriate number of
4922 	 * bits into the vmcs12 field.
4923 	 */
4924 	u64 value = 0;
4925 
4926 	if (!nested_vmx_check_permission(vcpu))
4927 		return 1;
4928 
4929 	/*
4930 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
4931 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
4932 	 */
4933 	if (vmx->nested.current_vmptr == -1ull ||
4934 	    (is_guest_mode(vcpu) &&
4935 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
4936 		return nested_vmx_failInvalid(vcpu);
4937 
4938 	if (instr_info & BIT(10))
4939 		value = kvm_register_readl(vcpu, (((instr_info) >> 3) & 0xf));
4940 	else {
4941 		len = is_64_bit_mode(vcpu) ? 8 : 4;
4942 		if (get_vmx_mem_address(vcpu, exit_qualification,
4943 					instr_info, false, len, &gva))
4944 			return 1;
4945 		if (kvm_read_guest_virt(vcpu, gva, &value, len, &e)) {
4946 			kvm_inject_page_fault(vcpu, &e);
4947 			return 1;
4948 		}
4949 	}
4950 
4951 	field = kvm_register_readl(vcpu, (((instr_info) >> 28) & 0xf));
4952 
4953 	offset = vmcs_field_to_offset(field);
4954 	if (offset < 0)
4955 		return nested_vmx_failValid(vcpu,
4956 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4957 
4958 	/*
4959 	 * If the vCPU supports "VMWRITE to any supported field in the
4960 	 * VMCS," then the "read-only" fields are actually read/write.
4961 	 */
4962 	if (vmcs_field_readonly(field) &&
4963 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4964 		return nested_vmx_failValid(vcpu,
4965 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4966 
4967 	/*
4968 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
4969 	 * vmcs12, else we may crush a field or consume a stale value.
4970 	 */
4971 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
4972 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4973 
4974 	/*
4975 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
4976 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
4977 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
4978 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
4979 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
4980 	 * the stripped down value, L2 sees the full value as stored by KVM).
4981 	 */
4982 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
4983 		value &= 0x1f0ff;
4984 
4985 	vmcs12_write_any(vmcs12, field, offset, value);
4986 
4987 	/*
4988 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
4989 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
4990 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
4991 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
4992 	 */
4993 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
4994 		/*
4995 		 * L1 can read these fields without exiting, ensure the
4996 		 * shadow VMCS is up-to-date.
4997 		 */
4998 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
4999 			preempt_disable();
5000 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5001 
5002 			__vmcs_writel(field, value);
5003 
5004 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5005 			vmcs_load(vmx->loaded_vmcs->vmcs);
5006 			preempt_enable();
5007 		}
5008 		vmx->nested.dirty_vmcs12 = true;
5009 	}
5010 
5011 	return nested_vmx_succeed(vcpu);
5012 }
5013 
5014 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5015 {
5016 	vmx->nested.current_vmptr = vmptr;
5017 	if (enable_shadow_vmcs) {
5018 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5019 		vmcs_write64(VMCS_LINK_POINTER,
5020 			     __pa(vmx->vmcs01.shadow_vmcs));
5021 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5022 	}
5023 	vmx->nested.dirty_vmcs12 = true;
5024 }
5025 
5026 /* Emulate the VMPTRLD instruction */
5027 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5028 {
5029 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5030 	gpa_t vmptr;
5031 
5032 	if (!nested_vmx_check_permission(vcpu))
5033 		return 1;
5034 
5035 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
5036 		return 1;
5037 
5038 	if (!page_address_valid(vcpu, vmptr))
5039 		return nested_vmx_failValid(vcpu,
5040 			VMXERR_VMPTRLD_INVALID_ADDRESS);
5041 
5042 	if (vmptr == vmx->nested.vmxon_ptr)
5043 		return nested_vmx_failValid(vcpu,
5044 			VMXERR_VMPTRLD_VMXON_POINTER);
5045 
5046 	/* Forbid normal VMPTRLD if Enlightened version was used */
5047 	if (vmx->nested.hv_evmcs)
5048 		return 1;
5049 
5050 	if (vmx->nested.current_vmptr != vmptr) {
5051 		struct kvm_host_map map;
5052 		struct vmcs12 *new_vmcs12;
5053 
5054 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
5055 			/*
5056 			 * Reads from an unbacked page return all 1s,
5057 			 * which means that the 32 bits located at the
5058 			 * given physical address won't match the required
5059 			 * VMCS12_REVISION identifier.
5060 			 */
5061 			return nested_vmx_failValid(vcpu,
5062 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5063 		}
5064 
5065 		new_vmcs12 = map.hva;
5066 
5067 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5068 		    (new_vmcs12->hdr.shadow_vmcs &&
5069 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5070 			kvm_vcpu_unmap(vcpu, &map, false);
5071 			return nested_vmx_failValid(vcpu,
5072 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5073 		}
5074 
5075 		nested_release_vmcs12(vcpu);
5076 
5077 		/*
5078 		 * Load VMCS12 from guest memory since it is not already
5079 		 * cached.
5080 		 */
5081 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
5082 		kvm_vcpu_unmap(vcpu, &map, false);
5083 
5084 		set_current_vmptr(vmx, vmptr);
5085 	}
5086 
5087 	return nested_vmx_succeed(vcpu);
5088 }
5089 
5090 /* Emulate the VMPTRST instruction */
5091 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5092 {
5093 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
5094 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5095 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5096 	struct x86_exception e;
5097 	gva_t gva;
5098 
5099 	if (!nested_vmx_check_permission(vcpu))
5100 		return 1;
5101 
5102 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
5103 		return 1;
5104 
5105 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5106 				true, sizeof(gpa_t), &gva))
5107 		return 1;
5108 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5109 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5110 					sizeof(gpa_t), &e)) {
5111 		kvm_inject_page_fault(vcpu, &e);
5112 		return 1;
5113 	}
5114 	return nested_vmx_succeed(vcpu);
5115 }
5116 
5117 /* Emulate the INVEPT instruction */
5118 static int handle_invept(struct kvm_vcpu *vcpu)
5119 {
5120 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5121 	u32 vmx_instruction_info, types;
5122 	unsigned long type;
5123 	gva_t gva;
5124 	struct x86_exception e;
5125 	struct {
5126 		u64 eptp, gpa;
5127 	} operand;
5128 
5129 	if (!(vmx->nested.msrs.secondary_ctls_high &
5130 	      SECONDARY_EXEC_ENABLE_EPT) ||
5131 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5132 		kvm_queue_exception(vcpu, UD_VECTOR);
5133 		return 1;
5134 	}
5135 
5136 	if (!nested_vmx_check_permission(vcpu))
5137 		return 1;
5138 
5139 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5140 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5141 
5142 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5143 
5144 	if (type >= 32 || !(types & (1 << type)))
5145 		return nested_vmx_failValid(vcpu,
5146 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5147 
5148 	/* According to the Intel VMX instruction reference, the memory
5149 	 * operand is read even if it isn't needed (e.g., for type==global)
5150 	 */
5151 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5152 			vmx_instruction_info, false, sizeof(operand), &gva))
5153 		return 1;
5154 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5155 		kvm_inject_page_fault(vcpu, &e);
5156 		return 1;
5157 	}
5158 
5159 	switch (type) {
5160 	case VMX_EPT_EXTENT_GLOBAL:
5161 	case VMX_EPT_EXTENT_CONTEXT:
5162 	/*
5163 	 * TODO: Sync the necessary shadow EPT roots here, rather than
5164 	 * at the next emulated VM-entry.
5165 	 */
5166 		break;
5167 	default:
5168 		BUG_ON(1);
5169 		break;
5170 	}
5171 
5172 	return nested_vmx_succeed(vcpu);
5173 }
5174 
5175 static int handle_invvpid(struct kvm_vcpu *vcpu)
5176 {
5177 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5178 	u32 vmx_instruction_info;
5179 	unsigned long type, types;
5180 	gva_t gva;
5181 	struct x86_exception e;
5182 	struct {
5183 		u64 vpid;
5184 		u64 gla;
5185 	} operand;
5186 	u16 vpid02;
5187 
5188 	if (!(vmx->nested.msrs.secondary_ctls_high &
5189 	      SECONDARY_EXEC_ENABLE_VPID) ||
5190 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5191 		kvm_queue_exception(vcpu, UD_VECTOR);
5192 		return 1;
5193 	}
5194 
5195 	if (!nested_vmx_check_permission(vcpu))
5196 		return 1;
5197 
5198 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5199 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5200 
5201 	types = (vmx->nested.msrs.vpid_caps &
5202 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5203 
5204 	if (type >= 32 || !(types & (1 << type)))
5205 		return nested_vmx_failValid(vcpu,
5206 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5207 
5208 	/* according to the intel vmx instruction reference, the memory
5209 	 * operand is read even if it isn't needed (e.g., for type==global)
5210 	 */
5211 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5212 			vmx_instruction_info, false, sizeof(operand), &gva))
5213 		return 1;
5214 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5215 		kvm_inject_page_fault(vcpu, &e);
5216 		return 1;
5217 	}
5218 	if (operand.vpid >> 16)
5219 		return nested_vmx_failValid(vcpu,
5220 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5221 
5222 	vpid02 = nested_get_vpid02(vcpu);
5223 	switch (type) {
5224 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5225 		if (!operand.vpid ||
5226 		    is_noncanonical_address(operand.gla, vcpu))
5227 			return nested_vmx_failValid(vcpu,
5228 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5229 		if (cpu_has_vmx_invvpid_individual_addr()) {
5230 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
5231 				vpid02, operand.gla);
5232 		} else
5233 			__vmx_flush_tlb(vcpu, vpid02, false);
5234 		break;
5235 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5236 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5237 		if (!operand.vpid)
5238 			return nested_vmx_failValid(vcpu,
5239 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5240 		__vmx_flush_tlb(vcpu, vpid02, false);
5241 		break;
5242 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5243 		__vmx_flush_tlb(vcpu, vpid02, false);
5244 		break;
5245 	default:
5246 		WARN_ON_ONCE(1);
5247 		return kvm_skip_emulated_instruction(vcpu);
5248 	}
5249 
5250 	return nested_vmx_succeed(vcpu);
5251 }
5252 
5253 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5254 				     struct vmcs12 *vmcs12)
5255 {
5256 	u32 index = kvm_rcx_read(vcpu);
5257 	u64 new_eptp;
5258 	bool accessed_dirty;
5259 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
5260 
5261 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
5262 	    !nested_cpu_has_ept(vmcs12))
5263 		return 1;
5264 
5265 	if (index >= VMFUNC_EPTP_ENTRIES)
5266 		return 1;
5267 
5268 
5269 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5270 				     &new_eptp, index * 8, 8))
5271 		return 1;
5272 
5273 	accessed_dirty = !!(new_eptp & VMX_EPTP_AD_ENABLE_BIT);
5274 
5275 	/*
5276 	 * If the (L2) guest does a vmfunc to the currently
5277 	 * active ept pointer, we don't have to do anything else
5278 	 */
5279 	if (vmcs12->ept_pointer != new_eptp) {
5280 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5281 			return 1;
5282 
5283 		kvm_mmu_unload(vcpu);
5284 		mmu->ept_ad = accessed_dirty;
5285 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
5286 		vmcs12->ept_pointer = new_eptp;
5287 		/*
5288 		 * TODO: Check what's the correct approach in case
5289 		 * mmu reload fails. Currently, we just let the next
5290 		 * reload potentially fail
5291 		 */
5292 		kvm_mmu_reload(vcpu);
5293 	}
5294 
5295 	return 0;
5296 }
5297 
5298 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5299 {
5300 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5301 	struct vmcs12 *vmcs12;
5302 	u32 function = kvm_rax_read(vcpu);
5303 
5304 	/*
5305 	 * VMFUNC is only supported for nested guests, but we always enable the
5306 	 * secondary control for simplicity; for non-nested mode, fake that we
5307 	 * didn't by injecting #UD.
5308 	 */
5309 	if (!is_guest_mode(vcpu)) {
5310 		kvm_queue_exception(vcpu, UD_VECTOR);
5311 		return 1;
5312 	}
5313 
5314 	vmcs12 = get_vmcs12(vcpu);
5315 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
5316 		goto fail;
5317 
5318 	switch (function) {
5319 	case 0:
5320 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5321 			goto fail;
5322 		break;
5323 	default:
5324 		goto fail;
5325 	}
5326 	return kvm_skip_emulated_instruction(vcpu);
5327 
5328 fail:
5329 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
5330 			  vmcs_read32(VM_EXIT_INTR_INFO),
5331 			  vmcs_readl(EXIT_QUALIFICATION));
5332 	return 1;
5333 }
5334 
5335 /*
5336  * Return true if an IO instruction with the specified port and size should cause
5337  * a VM-exit into L1.
5338  */
5339 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5340 				 int size)
5341 {
5342 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5343 	gpa_t bitmap, last_bitmap;
5344 	u8 b;
5345 
5346 	last_bitmap = (gpa_t)-1;
5347 	b = -1;
5348 
5349 	while (size > 0) {
5350 		if (port < 0x8000)
5351 			bitmap = vmcs12->io_bitmap_a;
5352 		else if (port < 0x10000)
5353 			bitmap = vmcs12->io_bitmap_b;
5354 		else
5355 			return true;
5356 		bitmap += (port & 0x7fff) / 8;
5357 
5358 		if (last_bitmap != bitmap)
5359 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5360 				return true;
5361 		if (b & (1 << (port & 7)))
5362 			return true;
5363 
5364 		port++;
5365 		size--;
5366 		last_bitmap = bitmap;
5367 	}
5368 
5369 	return false;
5370 }
5371 
5372 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5373 				       struct vmcs12 *vmcs12)
5374 {
5375 	unsigned long exit_qualification;
5376 	unsigned short port;
5377 	int size;
5378 
5379 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5380 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5381 
5382 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5383 
5384 	port = exit_qualification >> 16;
5385 	size = (exit_qualification & 7) + 1;
5386 
5387 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5388 }
5389 
5390 /*
5391  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5392  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5393  * disinterest in the current event (read or write a specific MSR) by using an
5394  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5395  */
5396 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5397 	struct vmcs12 *vmcs12, u32 exit_reason)
5398 {
5399 	u32 msr_index = kvm_rcx_read(vcpu);
5400 	gpa_t bitmap;
5401 
5402 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5403 		return true;
5404 
5405 	/*
5406 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5407 	 * for the four combinations of read/write and low/high MSR numbers.
5408 	 * First we need to figure out which of the four to use:
5409 	 */
5410 	bitmap = vmcs12->msr_bitmap;
5411 	if (exit_reason == EXIT_REASON_MSR_WRITE)
5412 		bitmap += 2048;
5413 	if (msr_index >= 0xc0000000) {
5414 		msr_index -= 0xc0000000;
5415 		bitmap += 1024;
5416 	}
5417 
5418 	/* Then read the msr_index'th bit from this bitmap: */
5419 	if (msr_index < 1024*8) {
5420 		unsigned char b;
5421 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5422 			return true;
5423 		return 1 & (b >> (msr_index & 7));
5424 	} else
5425 		return true; /* let L1 handle the wrong parameter */
5426 }
5427 
5428 /*
5429  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5430  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5431  * intercept (via guest_host_mask etc.) the current event.
5432  */
5433 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5434 	struct vmcs12 *vmcs12)
5435 {
5436 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5437 	int cr = exit_qualification & 15;
5438 	int reg;
5439 	unsigned long val;
5440 
5441 	switch ((exit_qualification >> 4) & 3) {
5442 	case 0: /* mov to cr */
5443 		reg = (exit_qualification >> 8) & 15;
5444 		val = kvm_register_readl(vcpu, reg);
5445 		switch (cr) {
5446 		case 0:
5447 			if (vmcs12->cr0_guest_host_mask &
5448 			    (val ^ vmcs12->cr0_read_shadow))
5449 				return true;
5450 			break;
5451 		case 3:
5452 			if ((vmcs12->cr3_target_count >= 1 &&
5453 					vmcs12->cr3_target_value0 == val) ||
5454 				(vmcs12->cr3_target_count >= 2 &&
5455 					vmcs12->cr3_target_value1 == val) ||
5456 				(vmcs12->cr3_target_count >= 3 &&
5457 					vmcs12->cr3_target_value2 == val) ||
5458 				(vmcs12->cr3_target_count >= 4 &&
5459 					vmcs12->cr3_target_value3 == val))
5460 				return false;
5461 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5462 				return true;
5463 			break;
5464 		case 4:
5465 			if (vmcs12->cr4_guest_host_mask &
5466 			    (vmcs12->cr4_read_shadow ^ val))
5467 				return true;
5468 			break;
5469 		case 8:
5470 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5471 				return true;
5472 			break;
5473 		}
5474 		break;
5475 	case 2: /* clts */
5476 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5477 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5478 			return true;
5479 		break;
5480 	case 1: /* mov from cr */
5481 		switch (cr) {
5482 		case 3:
5483 			if (vmcs12->cpu_based_vm_exec_control &
5484 			    CPU_BASED_CR3_STORE_EXITING)
5485 				return true;
5486 			break;
5487 		case 8:
5488 			if (vmcs12->cpu_based_vm_exec_control &
5489 			    CPU_BASED_CR8_STORE_EXITING)
5490 				return true;
5491 			break;
5492 		}
5493 		break;
5494 	case 3: /* lmsw */
5495 		/*
5496 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5497 		 * cr0. Other attempted changes are ignored, with no exit.
5498 		 */
5499 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5500 		if (vmcs12->cr0_guest_host_mask & 0xe &
5501 		    (val ^ vmcs12->cr0_read_shadow))
5502 			return true;
5503 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5504 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5505 		    (val & 0x1))
5506 			return true;
5507 		break;
5508 	}
5509 	return false;
5510 }
5511 
5512 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5513 	struct vmcs12 *vmcs12, gpa_t bitmap)
5514 {
5515 	u32 vmx_instruction_info;
5516 	unsigned long field;
5517 	u8 b;
5518 
5519 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5520 		return true;
5521 
5522 	/* Decode instruction info and find the field to access */
5523 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5524 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5525 
5526 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5527 	if (field >> 15)
5528 		return true;
5529 
5530 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5531 		return true;
5532 
5533 	return 1 & (b >> (field & 7));
5534 }
5535 
5536 /*
5537  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5538  * should handle it ourselves in L0 (and then continue L2). Only call this
5539  * when in is_guest_mode (L2).
5540  */
5541 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5542 {
5543 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5544 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5545 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5546 
5547 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
5548 
5549 	if (unlikely(vmx->fail)) {
5550 		trace_kvm_nested_vmenter_failed(
5551 			"hardware VM-instruction error: ",
5552 			vmcs_read32(VM_INSTRUCTION_ERROR));
5553 		return true;
5554 	}
5555 
5556 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5557 				vmcs_readl(EXIT_QUALIFICATION),
5558 				vmx->idt_vectoring_info,
5559 				intr_info,
5560 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5561 				KVM_ISA_VMX);
5562 
5563 	switch (exit_reason) {
5564 	case EXIT_REASON_EXCEPTION_NMI:
5565 		if (is_nmi(intr_info))
5566 			return false;
5567 		else if (is_page_fault(intr_info))
5568 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5569 		else if (is_debug(intr_info) &&
5570 			 vcpu->guest_debug &
5571 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5572 			return false;
5573 		else if (is_breakpoint(intr_info) &&
5574 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5575 			return false;
5576 		return vmcs12->exception_bitmap &
5577 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5578 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5579 		return false;
5580 	case EXIT_REASON_TRIPLE_FAULT:
5581 		return true;
5582 	case EXIT_REASON_INTERRUPT_WINDOW:
5583 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5584 	case EXIT_REASON_NMI_WINDOW:
5585 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5586 	case EXIT_REASON_TASK_SWITCH:
5587 		return true;
5588 	case EXIT_REASON_CPUID:
5589 		return true;
5590 	case EXIT_REASON_HLT:
5591 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5592 	case EXIT_REASON_INVD:
5593 		return true;
5594 	case EXIT_REASON_INVLPG:
5595 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5596 	case EXIT_REASON_RDPMC:
5597 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5598 	case EXIT_REASON_RDRAND:
5599 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5600 	case EXIT_REASON_RDSEED:
5601 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5602 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5603 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5604 	case EXIT_REASON_VMREAD:
5605 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5606 			vmcs12->vmread_bitmap);
5607 	case EXIT_REASON_VMWRITE:
5608 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5609 			vmcs12->vmwrite_bitmap);
5610 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5611 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5612 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5613 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5614 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5615 		/*
5616 		 * VMX instructions trap unconditionally. This allows L1 to
5617 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5618 		 */
5619 		return true;
5620 	case EXIT_REASON_CR_ACCESS:
5621 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5622 	case EXIT_REASON_DR_ACCESS:
5623 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5624 	case EXIT_REASON_IO_INSTRUCTION:
5625 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5626 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5627 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5628 	case EXIT_REASON_MSR_READ:
5629 	case EXIT_REASON_MSR_WRITE:
5630 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5631 	case EXIT_REASON_INVALID_STATE:
5632 		return true;
5633 	case EXIT_REASON_MWAIT_INSTRUCTION:
5634 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5635 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5636 		return nested_cpu_has_mtf(vmcs12);
5637 	case EXIT_REASON_MONITOR_INSTRUCTION:
5638 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5639 	case EXIT_REASON_PAUSE_INSTRUCTION:
5640 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5641 			nested_cpu_has2(vmcs12,
5642 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5643 	case EXIT_REASON_MCE_DURING_VMENTRY:
5644 		return false;
5645 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5646 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5647 	case EXIT_REASON_APIC_ACCESS:
5648 	case EXIT_REASON_APIC_WRITE:
5649 	case EXIT_REASON_EOI_INDUCED:
5650 		/*
5651 		 * The controls for "virtualize APIC accesses," "APIC-
5652 		 * register virtualization," and "virtual-interrupt
5653 		 * delivery" only come from vmcs12.
5654 		 */
5655 		return true;
5656 	case EXIT_REASON_EPT_VIOLATION:
5657 		/*
5658 		 * L0 always deals with the EPT violation. If nested EPT is
5659 		 * used, and the nested mmu code discovers that the address is
5660 		 * missing in the guest EPT table (EPT12), the EPT violation
5661 		 * will be injected with nested_ept_inject_page_fault()
5662 		 */
5663 		return false;
5664 	case EXIT_REASON_EPT_MISCONFIG:
5665 		/*
5666 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5667 		 * table (shadow on EPT) or a merged EPT table that L0 built
5668 		 * (EPT on EPT). So any problems with the structure of the
5669 		 * table is L0's fault.
5670 		 */
5671 		return false;
5672 	case EXIT_REASON_INVPCID:
5673 		return
5674 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5675 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5676 	case EXIT_REASON_WBINVD:
5677 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5678 	case EXIT_REASON_XSETBV:
5679 		return true;
5680 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5681 		/*
5682 		 * This should never happen, since it is not possible to
5683 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5684 		 * If if it were, XSS would have to be checked against
5685 		 * the XSS exit bitmap in vmcs12.
5686 		 */
5687 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5688 	case EXIT_REASON_PREEMPTION_TIMER:
5689 		return false;
5690 	case EXIT_REASON_PML_FULL:
5691 		/* We emulate PML support to L1. */
5692 		return false;
5693 	case EXIT_REASON_VMFUNC:
5694 		/* VM functions are emulated through L2->L0 vmexits. */
5695 		return false;
5696 	case EXIT_REASON_ENCLS:
5697 		/* SGX is never exposed to L1 */
5698 		return false;
5699 	case EXIT_REASON_UMWAIT:
5700 	case EXIT_REASON_TPAUSE:
5701 		return nested_cpu_has2(vmcs12,
5702 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
5703 	default:
5704 		return true;
5705 	}
5706 }
5707 
5708 
5709 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5710 				struct kvm_nested_state __user *user_kvm_nested_state,
5711 				u32 user_data_size)
5712 {
5713 	struct vcpu_vmx *vmx;
5714 	struct vmcs12 *vmcs12;
5715 	struct kvm_nested_state kvm_state = {
5716 		.flags = 0,
5717 		.format = KVM_STATE_NESTED_FORMAT_VMX,
5718 		.size = sizeof(kvm_state),
5719 		.hdr.vmx.vmxon_pa = -1ull,
5720 		.hdr.vmx.vmcs12_pa = -1ull,
5721 	};
5722 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5723 		&user_kvm_nested_state->data.vmx[0];
5724 
5725 	if (!vcpu)
5726 		return kvm_state.size + sizeof(*user_vmx_nested_state);
5727 
5728 	vmx = to_vmx(vcpu);
5729 	vmcs12 = get_vmcs12(vcpu);
5730 
5731 	if (nested_vmx_allowed(vcpu) &&
5732 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5733 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5734 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
5735 
5736 		if (vmx_has_valid_vmcs12(vcpu)) {
5737 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
5738 
5739 			if (vmx->nested.hv_evmcs)
5740 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5741 
5742 			if (is_guest_mode(vcpu) &&
5743 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5744 			    vmcs12->vmcs_link_pointer != -1ull)
5745 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
5746 		}
5747 
5748 		if (vmx->nested.smm.vmxon)
5749 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5750 
5751 		if (vmx->nested.smm.guest_mode)
5752 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5753 
5754 		if (is_guest_mode(vcpu)) {
5755 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5756 
5757 			if (vmx->nested.nested_run_pending)
5758 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5759 
5760 			if (vmx->nested.mtf_pending)
5761 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
5762 		}
5763 	}
5764 
5765 	if (user_data_size < kvm_state.size)
5766 		goto out;
5767 
5768 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5769 		return -EFAULT;
5770 
5771 	if (!vmx_has_valid_vmcs12(vcpu))
5772 		goto out;
5773 
5774 	/*
5775 	 * When running L2, the authoritative vmcs12 state is in the
5776 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5777 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5778 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
5779 	 * vmcs12 state is in the vmcs12 already.
5780 	 */
5781 	if (is_guest_mode(vcpu)) {
5782 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
5783 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5784 	} else if (!vmx->nested.need_vmcs12_to_shadow_sync) {
5785 		if (vmx->nested.hv_evmcs)
5786 			copy_enlightened_to_vmcs12(vmx);
5787 		else if (enable_shadow_vmcs)
5788 			copy_shadow_to_vmcs12(vmx);
5789 	}
5790 
5791 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
5792 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
5793 
5794 	/*
5795 	 * Copy over the full allocated size of vmcs12 rather than just the size
5796 	 * of the struct.
5797 	 */
5798 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
5799 		return -EFAULT;
5800 
5801 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5802 	    vmcs12->vmcs_link_pointer != -1ull) {
5803 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
5804 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
5805 			return -EFAULT;
5806 	}
5807 
5808 out:
5809 	return kvm_state.size;
5810 }
5811 
5812 /*
5813  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5814  */
5815 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5816 {
5817 	if (is_guest_mode(vcpu)) {
5818 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5819 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5820 	}
5821 	free_nested(vcpu);
5822 }
5823 
5824 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5825 				struct kvm_nested_state __user *user_kvm_nested_state,
5826 				struct kvm_nested_state *kvm_state)
5827 {
5828 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5829 	struct vmcs12 *vmcs12;
5830 	u32 exit_qual;
5831 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
5832 		&user_kvm_nested_state->data.vmx[0];
5833 	int ret;
5834 
5835 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
5836 		return -EINVAL;
5837 
5838 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
5839 		if (kvm_state->hdr.vmx.smm.flags)
5840 			return -EINVAL;
5841 
5842 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
5843 			return -EINVAL;
5844 
5845 		/*
5846 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
5847 		 * enable eVMCS capability on vCPU. However, since then
5848 		 * code was changed such that flag signals vmcs12 should
5849 		 * be copied into eVMCS in guest memory.
5850 		 *
5851 		 * To preserve backwards compatability, allow user
5852 		 * to set this flag even when there is no VMXON region.
5853 		 */
5854 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
5855 			return -EINVAL;
5856 	} else {
5857 		if (!nested_vmx_allowed(vcpu))
5858 			return -EINVAL;
5859 
5860 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
5861 			return -EINVAL;
5862 	}
5863 
5864 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5865 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5866 		return -EINVAL;
5867 
5868 	if (kvm_state->hdr.vmx.smm.flags &
5869 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5870 		return -EINVAL;
5871 
5872 	/*
5873 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5874 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5875 	 * must be zero.
5876 	 */
5877 	if (is_smm(vcpu) ?
5878 		(kvm_state->flags &
5879 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
5880 		: kvm_state->hdr.vmx.smm.flags)
5881 		return -EINVAL;
5882 
5883 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5884 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5885 		return -EINVAL;
5886 
5887 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
5888 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
5889 			return -EINVAL;
5890 
5891 	vmx_leave_nested(vcpu);
5892 
5893 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
5894 		return 0;
5895 
5896 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
5897 	ret = enter_vmx_operation(vcpu);
5898 	if (ret)
5899 		return ret;
5900 
5901 	/* Empty 'VMXON' state is permitted */
5902 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12))
5903 		return 0;
5904 
5905 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
5906 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
5907 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
5908 			return -EINVAL;
5909 
5910 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
5911 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5912 		/*
5913 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
5914 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
5915 		 * restored yet. EVMCS will be mapped from
5916 		 * nested_get_vmcs12_pages().
5917 		 */
5918 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
5919 	} else {
5920 		return -EINVAL;
5921 	}
5922 
5923 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5924 		vmx->nested.smm.vmxon = true;
5925 		vmx->nested.vmxon = false;
5926 
5927 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5928 			vmx->nested.smm.guest_mode = true;
5929 	}
5930 
5931 	vmcs12 = get_vmcs12(vcpu);
5932 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
5933 		return -EFAULT;
5934 
5935 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5936 		return -EINVAL;
5937 
5938 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5939 		return 0;
5940 
5941 	vmx->nested.nested_run_pending =
5942 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5943 
5944 	vmx->nested.mtf_pending =
5945 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
5946 
5947 	ret = -EINVAL;
5948 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5949 	    vmcs12->vmcs_link_pointer != -1ull) {
5950 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5951 
5952 		if (kvm_state->size <
5953 		    sizeof(*kvm_state) +
5954 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
5955 			goto error_guest_mode;
5956 
5957 		if (copy_from_user(shadow_vmcs12,
5958 				   user_vmx_nested_state->shadow_vmcs12,
5959 				   sizeof(*shadow_vmcs12))) {
5960 			ret = -EFAULT;
5961 			goto error_guest_mode;
5962 		}
5963 
5964 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5965 		    !shadow_vmcs12->hdr.shadow_vmcs)
5966 			goto error_guest_mode;
5967 	}
5968 
5969 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
5970 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
5971 	    nested_vmx_check_guest_state(vcpu, vmcs12, &exit_qual))
5972 		goto error_guest_mode;
5973 
5974 	vmx->nested.dirty_vmcs12 = true;
5975 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5976 	if (ret)
5977 		goto error_guest_mode;
5978 
5979 	return 0;
5980 
5981 error_guest_mode:
5982 	vmx->nested.nested_run_pending = 0;
5983 	return ret;
5984 }
5985 
5986 void nested_vmx_set_vmcs_shadowing_bitmap(void)
5987 {
5988 	if (enable_shadow_vmcs) {
5989 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5990 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5991 	}
5992 }
5993 
5994 /*
5995  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5996  * returned for the various VMX controls MSRs when nested VMX is enabled.
5997  * The same values should also be used to verify that vmcs12 control fields are
5998  * valid during nested entry from L1 to L2.
5999  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6000  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6001  * bit in the high half is on if the corresponding bit in the control field
6002  * may be on. See also vmx_control_verify().
6003  */
6004 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6005 {
6006 	/*
6007 	 * Note that as a general rule, the high half of the MSRs (bits in
6008 	 * the control fields which may be 1) should be initialized by the
6009 	 * intersection of the underlying hardware's MSR (i.e., features which
6010 	 * can be supported) and the list of features we want to expose -
6011 	 * because they are known to be properly supported in our code.
6012 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6013 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6014 	 * reason is that if one of these bits is necessary, it will appear
6015 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6016 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6017 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
6018 	 * These rules have exceptions below.
6019 	 */
6020 
6021 	/* pin-based controls */
6022 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6023 		msrs->pinbased_ctls_low,
6024 		msrs->pinbased_ctls_high);
6025 	msrs->pinbased_ctls_low |=
6026 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6027 	msrs->pinbased_ctls_high &=
6028 		PIN_BASED_EXT_INTR_MASK |
6029 		PIN_BASED_NMI_EXITING |
6030 		PIN_BASED_VIRTUAL_NMIS |
6031 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6032 	msrs->pinbased_ctls_high |=
6033 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6034 		PIN_BASED_VMX_PREEMPTION_TIMER;
6035 
6036 	/* exit controls */
6037 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6038 		msrs->exit_ctls_low,
6039 		msrs->exit_ctls_high);
6040 	msrs->exit_ctls_low =
6041 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6042 
6043 	msrs->exit_ctls_high &=
6044 #ifdef CONFIG_X86_64
6045 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6046 #endif
6047 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
6048 	msrs->exit_ctls_high |=
6049 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6050 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6051 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6052 
6053 	/* We support free control of debug control saving. */
6054 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6055 
6056 	/* entry controls */
6057 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6058 		msrs->entry_ctls_low,
6059 		msrs->entry_ctls_high);
6060 	msrs->entry_ctls_low =
6061 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6062 	msrs->entry_ctls_high &=
6063 #ifdef CONFIG_X86_64
6064 		VM_ENTRY_IA32E_MODE |
6065 #endif
6066 		VM_ENTRY_LOAD_IA32_PAT;
6067 	msrs->entry_ctls_high |=
6068 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6069 
6070 	/* We support free control of debug control loading. */
6071 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6072 
6073 	/* cpu-based controls */
6074 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6075 		msrs->procbased_ctls_low,
6076 		msrs->procbased_ctls_high);
6077 	msrs->procbased_ctls_low =
6078 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6079 	msrs->procbased_ctls_high &=
6080 		CPU_BASED_INTR_WINDOW_EXITING |
6081 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6082 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6083 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6084 		CPU_BASED_CR3_STORE_EXITING |
6085 #ifdef CONFIG_X86_64
6086 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6087 #endif
6088 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6089 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6090 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6091 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6092 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6093 	/*
6094 	 * We can allow some features even when not supported by the
6095 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6096 	 * can use it to avoid exits to L1 - even when L0 runs L2
6097 	 * without MSR bitmaps.
6098 	 */
6099 	msrs->procbased_ctls_high |=
6100 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6101 		CPU_BASED_USE_MSR_BITMAPS;
6102 
6103 	/* We support free control of CR3 access interception. */
6104 	msrs->procbased_ctls_low &=
6105 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6106 
6107 	/*
6108 	 * secondary cpu-based controls.  Do not include those that
6109 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
6110 	 */
6111 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6112 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6113 		      msrs->secondary_ctls_low,
6114 		      msrs->secondary_ctls_high);
6115 
6116 	msrs->secondary_ctls_low = 0;
6117 	msrs->secondary_ctls_high &=
6118 		SECONDARY_EXEC_DESC |
6119 		SECONDARY_EXEC_RDTSCP |
6120 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6121 		SECONDARY_EXEC_WBINVD_EXITING |
6122 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6123 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6124 		SECONDARY_EXEC_RDRAND_EXITING |
6125 		SECONDARY_EXEC_ENABLE_INVPCID |
6126 		SECONDARY_EXEC_RDSEED_EXITING |
6127 		SECONDARY_EXEC_XSAVES;
6128 
6129 	/*
6130 	 * We can emulate "VMCS shadowing," even if the hardware
6131 	 * doesn't support it.
6132 	 */
6133 	msrs->secondary_ctls_high |=
6134 		SECONDARY_EXEC_SHADOW_VMCS;
6135 
6136 	if (enable_ept) {
6137 		/* nested EPT: emulate EPT also to L1 */
6138 		msrs->secondary_ctls_high |=
6139 			SECONDARY_EXEC_ENABLE_EPT;
6140 		msrs->ept_caps =
6141 			VMX_EPT_PAGE_WALK_4_BIT |
6142 			VMX_EPT_PAGE_WALK_5_BIT |
6143 			VMX_EPTP_WB_BIT |
6144 			VMX_EPT_INVEPT_BIT |
6145 			VMX_EPT_EXECUTE_ONLY_BIT;
6146 
6147 		msrs->ept_caps &= ept_caps;
6148 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6149 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6150 			VMX_EPT_1GB_PAGE_BIT;
6151 		if (enable_ept_ad_bits) {
6152 			msrs->secondary_ctls_high |=
6153 				SECONDARY_EXEC_ENABLE_PML;
6154 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6155 		}
6156 	}
6157 
6158 	if (cpu_has_vmx_vmfunc()) {
6159 		msrs->secondary_ctls_high |=
6160 			SECONDARY_EXEC_ENABLE_VMFUNC;
6161 		/*
6162 		 * Advertise EPTP switching unconditionally
6163 		 * since we emulate it
6164 		 */
6165 		if (enable_ept)
6166 			msrs->vmfunc_controls =
6167 				VMX_VMFUNC_EPTP_SWITCHING;
6168 	}
6169 
6170 	/*
6171 	 * Old versions of KVM use the single-context version without
6172 	 * checking for support, so declare that it is supported even
6173 	 * though it is treated as global context.  The alternative is
6174 	 * not failing the single-context invvpid, and it is worse.
6175 	 */
6176 	if (enable_vpid) {
6177 		msrs->secondary_ctls_high |=
6178 			SECONDARY_EXEC_ENABLE_VPID;
6179 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6180 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6181 	}
6182 
6183 	if (enable_unrestricted_guest)
6184 		msrs->secondary_ctls_high |=
6185 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6186 
6187 	if (flexpriority_enabled)
6188 		msrs->secondary_ctls_high |=
6189 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6190 
6191 	/* miscellaneous data */
6192 	rdmsr(MSR_IA32_VMX_MISC,
6193 		msrs->misc_low,
6194 		msrs->misc_high);
6195 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6196 	msrs->misc_low |=
6197 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6198 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6199 		VMX_MISC_ACTIVITY_HLT;
6200 	msrs->misc_high = 0;
6201 
6202 	/*
6203 	 * This MSR reports some information about VMX support. We
6204 	 * should return information about the VMX we emulate for the
6205 	 * guest, and the VMCS structure we give it - not about the
6206 	 * VMX support of the underlying hardware.
6207 	 */
6208 	msrs->basic =
6209 		VMCS12_REVISION |
6210 		VMX_BASIC_TRUE_CTLS |
6211 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6212 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6213 
6214 	if (cpu_has_vmx_basic_inout())
6215 		msrs->basic |= VMX_BASIC_INOUT;
6216 
6217 	/*
6218 	 * These MSRs specify bits which the guest must keep fixed on
6219 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6220 	 * We picked the standard core2 setting.
6221 	 */
6222 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6223 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6224 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6225 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6226 
6227 	/* These MSRs specify bits which the guest must keep fixed off. */
6228 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6229 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6230 
6231 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
6232 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
6233 }
6234 
6235 void nested_vmx_hardware_unsetup(void)
6236 {
6237 	int i;
6238 
6239 	if (enable_shadow_vmcs) {
6240 		for (i = 0; i < VMX_BITMAP_NR; i++)
6241 			free_page((unsigned long)vmx_bitmap[i]);
6242 	}
6243 }
6244 
6245 __init int nested_vmx_hardware_setup(struct kvm_x86_ops *ops,
6246 				     int (*exit_handlers[])(struct kvm_vcpu *))
6247 {
6248 	int i;
6249 
6250 	if (!cpu_has_vmx_shadow_vmcs())
6251 		enable_shadow_vmcs = 0;
6252 	if (enable_shadow_vmcs) {
6253 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6254 			/*
6255 			 * The vmx_bitmap is not tied to a VM and so should
6256 			 * not be charged to a memcg.
6257 			 */
6258 			vmx_bitmap[i] = (unsigned long *)
6259 				__get_free_page(GFP_KERNEL);
6260 			if (!vmx_bitmap[i]) {
6261 				nested_vmx_hardware_unsetup();
6262 				return -ENOMEM;
6263 			}
6264 		}
6265 
6266 		init_vmcs_shadow_fields();
6267 	}
6268 
6269 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6270 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6271 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6272 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6273 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6274 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6275 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6276 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6277 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6278 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6279 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6280 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6281 
6282 	ops->check_nested_events = vmx_check_nested_events;
6283 	ops->get_nested_state = vmx_get_nested_state;
6284 	ops->set_nested_state = vmx_set_nested_state;
6285 	ops->get_vmcs12_pages = nested_get_vmcs12_pages;
6286 	ops->nested_enable_evmcs = nested_enable_evmcs;
6287 	ops->nested_get_evmcs_version = nested_get_evmcs_version;
6288 
6289 	return 0;
6290 }
6291