xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 20e8ef5c)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/objtool.h>
4 #include <linux/percpu.h>
5 
6 #include <asm/debugreg.h>
7 #include <asm/mmu_context.h>
8 
9 #include "cpuid.h"
10 #include "hyperv.h"
11 #include "mmu.h"
12 #include "nested.h"
13 #include "pmu.h"
14 #include "sgx.h"
15 #include "trace.h"
16 #include "vmx.h"
17 #include "x86.h"
18 
19 static bool __read_mostly enable_shadow_vmcs = 1;
20 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
21 
22 static bool __read_mostly nested_early_check = 0;
23 module_param(nested_early_check, bool, S_IRUGO);
24 
25 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
26 
27 /*
28  * Hyper-V requires all of these, so mark them as supported even though
29  * they are just treated the same as all-context.
30  */
31 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
32 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
33 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
34 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
35 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
36 
37 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
38 
39 enum {
40 	VMX_VMREAD_BITMAP,
41 	VMX_VMWRITE_BITMAP,
42 	VMX_BITMAP_NR
43 };
44 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
45 
46 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
47 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
48 
49 struct shadow_vmcs_field {
50 	u16	encoding;
51 	u16	offset;
52 };
53 static struct shadow_vmcs_field shadow_read_only_fields[] = {
54 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
55 #include "vmcs_shadow_fields.h"
56 };
57 static int max_shadow_read_only_fields =
58 	ARRAY_SIZE(shadow_read_only_fields);
59 
60 static struct shadow_vmcs_field shadow_read_write_fields[] = {
61 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
62 #include "vmcs_shadow_fields.h"
63 };
64 static int max_shadow_read_write_fields =
65 	ARRAY_SIZE(shadow_read_write_fields);
66 
67 static void init_vmcs_shadow_fields(void)
68 {
69 	int i, j;
70 
71 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
72 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
73 
74 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
75 		struct shadow_vmcs_field entry = shadow_read_only_fields[i];
76 		u16 field = entry.encoding;
77 
78 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
79 		    (i + 1 == max_shadow_read_only_fields ||
80 		     shadow_read_only_fields[i + 1].encoding != field + 1))
81 			pr_err("Missing field from shadow_read_only_field %x\n",
82 			       field + 1);
83 
84 		clear_bit(field, vmx_vmread_bitmap);
85 		if (field & 1)
86 #ifdef CONFIG_X86_64
87 			continue;
88 #else
89 			entry.offset += sizeof(u32);
90 #endif
91 		shadow_read_only_fields[j++] = entry;
92 	}
93 	max_shadow_read_only_fields = j;
94 
95 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
96 		struct shadow_vmcs_field entry = shadow_read_write_fields[i];
97 		u16 field = entry.encoding;
98 
99 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
100 		    (i + 1 == max_shadow_read_write_fields ||
101 		     shadow_read_write_fields[i + 1].encoding != field + 1))
102 			pr_err("Missing field from shadow_read_write_field %x\n",
103 			       field + 1);
104 
105 		WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
106 			  field <= GUEST_TR_AR_BYTES,
107 			  "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
108 
109 		/*
110 		 * PML and the preemption timer can be emulated, but the
111 		 * processor cannot vmwrite to fields that don't exist
112 		 * on bare metal.
113 		 */
114 		switch (field) {
115 		case GUEST_PML_INDEX:
116 			if (!cpu_has_vmx_pml())
117 				continue;
118 			break;
119 		case VMX_PREEMPTION_TIMER_VALUE:
120 			if (!cpu_has_vmx_preemption_timer())
121 				continue;
122 			break;
123 		case GUEST_INTR_STATUS:
124 			if (!cpu_has_vmx_apicv())
125 				continue;
126 			break;
127 		default:
128 			break;
129 		}
130 
131 		clear_bit(field, vmx_vmwrite_bitmap);
132 		clear_bit(field, vmx_vmread_bitmap);
133 		if (field & 1)
134 #ifdef CONFIG_X86_64
135 			continue;
136 #else
137 			entry.offset += sizeof(u32);
138 #endif
139 		shadow_read_write_fields[j++] = entry;
140 	}
141 	max_shadow_read_write_fields = j;
142 }
143 
144 /*
145  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
146  * set the success or error code of an emulated VMX instruction (as specified
147  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
148  * instruction.
149  */
150 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
151 {
152 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
153 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
154 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
155 	return kvm_skip_emulated_instruction(vcpu);
156 }
157 
158 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
159 {
160 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
161 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
162 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
163 			| X86_EFLAGS_CF);
164 	return kvm_skip_emulated_instruction(vcpu);
165 }
166 
167 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
168 				u32 vm_instruction_error)
169 {
170 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
171 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
172 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
173 			| X86_EFLAGS_ZF);
174 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
175 	/*
176 	 * We don't need to force sync to shadow VMCS because
177 	 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
178 	 * fields and thus must be synced.
179 	 */
180 	if (to_vmx(vcpu)->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
181 		to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
182 
183 	return kvm_skip_emulated_instruction(vcpu);
184 }
185 
186 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
187 {
188 	struct vcpu_vmx *vmx = to_vmx(vcpu);
189 
190 	/*
191 	 * failValid writes the error number to the current VMCS, which
192 	 * can't be done if there isn't a current VMCS.
193 	 */
194 	if (vmx->nested.current_vmptr == INVALID_GPA &&
195 	    !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
196 		return nested_vmx_failInvalid(vcpu);
197 
198 	return nested_vmx_failValid(vcpu, vm_instruction_error);
199 }
200 
201 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
202 {
203 	/* TODO: not to reset guest simply here. */
204 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
205 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
206 }
207 
208 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
209 {
210 	return fixed_bits_valid(control, low, high);
211 }
212 
213 static inline u64 vmx_control_msr(u32 low, u32 high)
214 {
215 	return low | ((u64)high << 32);
216 }
217 
218 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
219 {
220 	secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
221 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
222 	vmx->nested.need_vmcs12_to_shadow_sync = false;
223 }
224 
225 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
226 {
227 	struct vcpu_vmx *vmx = to_vmx(vcpu);
228 
229 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
230 		kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
231 		vmx->nested.hv_evmcs = NULL;
232 	}
233 
234 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
235 }
236 
237 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
238 				     struct loaded_vmcs *prev)
239 {
240 	struct vmcs_host_state *dest, *src;
241 
242 	if (unlikely(!vmx->guest_state_loaded))
243 		return;
244 
245 	src = &prev->host_state;
246 	dest = &vmx->loaded_vmcs->host_state;
247 
248 	vmx_set_vmcs_host_state(dest, src->cr3, src->fs_sel, src->gs_sel,
249 				src->fs_base, src->gs_base);
250 	dest->ldt_sel = src->ldt_sel;
251 #ifdef CONFIG_X86_64
252 	dest->ds_sel = src->ds_sel;
253 	dest->es_sel = src->es_sel;
254 #endif
255 }
256 
257 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
258 {
259 	struct vcpu_vmx *vmx = to_vmx(vcpu);
260 	struct loaded_vmcs *prev;
261 	int cpu;
262 
263 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
264 		return;
265 
266 	cpu = get_cpu();
267 	prev = vmx->loaded_vmcs;
268 	vmx->loaded_vmcs = vmcs;
269 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
270 	vmx_sync_vmcs_host_state(vmx, prev);
271 	put_cpu();
272 
273 	vcpu->arch.regs_avail = ~VMX_REGS_LAZY_LOAD_SET;
274 
275 	/*
276 	 * All lazily updated registers will be reloaded from VMCS12 on both
277 	 * vmentry and vmexit.
278 	 */
279 	vcpu->arch.regs_dirty = 0;
280 }
281 
282 /*
283  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
284  * just stops using VMX.
285  */
286 static void free_nested(struct kvm_vcpu *vcpu)
287 {
288 	struct vcpu_vmx *vmx = to_vmx(vcpu);
289 
290 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
291 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
292 
293 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
294 		return;
295 
296 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
297 
298 	vmx->nested.vmxon = false;
299 	vmx->nested.smm.vmxon = false;
300 	vmx->nested.vmxon_ptr = INVALID_GPA;
301 	free_vpid(vmx->nested.vpid02);
302 	vmx->nested.posted_intr_nv = -1;
303 	vmx->nested.current_vmptr = INVALID_GPA;
304 	if (enable_shadow_vmcs) {
305 		vmx_disable_shadow_vmcs(vmx);
306 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
307 		free_vmcs(vmx->vmcs01.shadow_vmcs);
308 		vmx->vmcs01.shadow_vmcs = NULL;
309 	}
310 	kfree(vmx->nested.cached_vmcs12);
311 	vmx->nested.cached_vmcs12 = NULL;
312 	kfree(vmx->nested.cached_shadow_vmcs12);
313 	vmx->nested.cached_shadow_vmcs12 = NULL;
314 	/* Unpin physical memory we referred to in the vmcs02 */
315 	if (vmx->nested.apic_access_page) {
316 		kvm_release_page_clean(vmx->nested.apic_access_page);
317 		vmx->nested.apic_access_page = NULL;
318 	}
319 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
320 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
321 	vmx->nested.pi_desc = NULL;
322 
323 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
324 
325 	nested_release_evmcs(vcpu);
326 
327 	free_loaded_vmcs(&vmx->nested.vmcs02);
328 }
329 
330 /*
331  * Ensure that the current vmcs of the logical processor is the
332  * vmcs01 of the vcpu before calling free_nested().
333  */
334 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
335 {
336 	vcpu_load(vcpu);
337 	vmx_leave_nested(vcpu);
338 	vcpu_put(vcpu);
339 }
340 
341 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
342 
343 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
344 {
345 	return VALID_PAGE(root_hpa) &&
346 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
347 }
348 
349 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
350 				       gpa_t addr)
351 {
352 	uint i;
353 	struct kvm_mmu_root_info *cached_root;
354 
355 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
356 
357 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
358 		cached_root = &vcpu->arch.mmu->prev_roots[i];
359 
360 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
361 					    eptp))
362 			vcpu->arch.mmu->invlpg(vcpu, addr, cached_root->hpa);
363 	}
364 }
365 
366 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
367 		struct x86_exception *fault)
368 {
369 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
370 	struct vcpu_vmx *vmx = to_vmx(vcpu);
371 	u32 vm_exit_reason;
372 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
373 
374 	if (vmx->nested.pml_full) {
375 		vm_exit_reason = EXIT_REASON_PML_FULL;
376 		vmx->nested.pml_full = false;
377 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
378 	} else {
379 		if (fault->error_code & PFERR_RSVD_MASK)
380 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
381 		else
382 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
383 
384 		/*
385 		 * Although the caller (kvm_inject_emulated_page_fault) would
386 		 * have already synced the faulting address in the shadow EPT
387 		 * tables for the current EPTP12, we also need to sync it for
388 		 * any other cached EPTP02s based on the same EP4TA, since the
389 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
390 		 */
391 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
392 					   fault->address);
393 	}
394 
395 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
396 	vmcs12->guest_physical_address = fault->address;
397 }
398 
399 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
400 {
401 	struct vcpu_vmx *vmx = to_vmx(vcpu);
402 	bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
403 	int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
404 
405 	kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
406 				nested_ept_ad_enabled(vcpu),
407 				nested_ept_get_eptp(vcpu));
408 }
409 
410 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
411 {
412 	WARN_ON(mmu_is_nested(vcpu));
413 
414 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
415 	nested_ept_new_eptp(vcpu);
416 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
417 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
418 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
419 
420 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
421 }
422 
423 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
424 {
425 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
426 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
427 }
428 
429 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
430 					    u16 error_code)
431 {
432 	bool inequality, bit;
433 
434 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
435 	inequality =
436 		(error_code & vmcs12->page_fault_error_code_mask) !=
437 		 vmcs12->page_fault_error_code_match;
438 	return inequality ^ bit;
439 }
440 
441 
442 /*
443  * KVM wants to inject page-faults which it got to the guest. This function
444  * checks whether in a nested guest, we need to inject them to L1 or L2.
445  */
446 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
447 {
448 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
449 	unsigned int nr = vcpu->arch.exception.nr;
450 	bool has_payload = vcpu->arch.exception.has_payload;
451 	unsigned long payload = vcpu->arch.exception.payload;
452 
453 	if (nr == PF_VECTOR) {
454 		if (vcpu->arch.exception.nested_apf) {
455 			*exit_qual = vcpu->arch.apf.nested_apf_token;
456 			return 1;
457 		}
458 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
459 						    vcpu->arch.exception.error_code)) {
460 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
461 			return 1;
462 		}
463 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
464 		if (nr == DB_VECTOR) {
465 			if (!has_payload) {
466 				payload = vcpu->arch.dr6;
467 				payload &= ~DR6_BT;
468 				payload ^= DR6_ACTIVE_LOW;
469 			}
470 			*exit_qual = payload;
471 		} else
472 			*exit_qual = 0;
473 		return 1;
474 	}
475 
476 	return 0;
477 }
478 
479 
480 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
481 		struct x86_exception *fault)
482 {
483 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
484 
485 	WARN_ON(!is_guest_mode(vcpu));
486 
487 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
488 		!to_vmx(vcpu)->nested.nested_run_pending) {
489 		vmcs12->vm_exit_intr_error_code = fault->error_code;
490 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
491 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
492 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
493 				  fault->address);
494 	} else {
495 		kvm_inject_page_fault(vcpu, fault);
496 	}
497 }
498 
499 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
500 					       struct vmcs12 *vmcs12)
501 {
502 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
503 		return 0;
504 
505 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
506 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
507 		return -EINVAL;
508 
509 	return 0;
510 }
511 
512 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
513 						struct vmcs12 *vmcs12)
514 {
515 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
516 		return 0;
517 
518 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
519 		return -EINVAL;
520 
521 	return 0;
522 }
523 
524 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
525 						struct vmcs12 *vmcs12)
526 {
527 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
528 		return 0;
529 
530 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
531 		return -EINVAL;
532 
533 	return 0;
534 }
535 
536 /*
537  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
538  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
539  * only the "disable intercept" case needs to be handled.
540  */
541 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
542 							unsigned long *msr_bitmap_l0,
543 							u32 msr, int type)
544 {
545 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
546 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
547 
548 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
549 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
550 }
551 
552 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
553 {
554 	int msr;
555 
556 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
557 		unsigned word = msr / BITS_PER_LONG;
558 
559 		msr_bitmap[word] = ~0;
560 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
561 	}
562 }
563 
564 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
565 static inline									\
566 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
567 					 unsigned long *msr_bitmap_l1,		\
568 					 unsigned long *msr_bitmap_l0, u32 msr)	\
569 {										\
570 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
571 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
572 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
573 	else									\
574 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
575 }
576 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
577 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
578 
579 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
580 						    unsigned long *msr_bitmap_l1,
581 						    unsigned long *msr_bitmap_l0,
582 						    u32 msr, int types)
583 {
584 	if (types & MSR_TYPE_R)
585 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
586 						  msr_bitmap_l0, msr);
587 	if (types & MSR_TYPE_W)
588 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
589 						   msr_bitmap_l0, msr);
590 }
591 
592 /*
593  * Merge L0's and L1's MSR bitmap, return false to indicate that
594  * we do not use the hardware.
595  */
596 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
597 						 struct vmcs12 *vmcs12)
598 {
599 	struct vcpu_vmx *vmx = to_vmx(vcpu);
600 	int msr;
601 	unsigned long *msr_bitmap_l1;
602 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
603 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
604 	struct kvm_host_map *map = &vmx->nested.msr_bitmap_map;
605 
606 	/* Nothing to do if the MSR bitmap is not in use.  */
607 	if (!cpu_has_vmx_msr_bitmap() ||
608 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
609 		return false;
610 
611 	/*
612 	 * MSR bitmap update can be skipped when:
613 	 * - MSR bitmap for L1 hasn't changed.
614 	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
615 	 *   before.
616 	 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
617 	 *   and tells KVM (L0) there were no changes in MSR bitmap for L2.
618 	 */
619 	if (!vmx->nested.force_msr_bitmap_recalc && evmcs &&
620 	    evmcs->hv_enlightenments_control.msr_bitmap &&
621 	    evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
622 		return true;
623 
624 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
625 		return false;
626 
627 	msr_bitmap_l1 = (unsigned long *)map->hva;
628 
629 	/*
630 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
631 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
632 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
633 	 */
634 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
635 
636 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
637 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
638 			/*
639 			 * L0 need not intercept reads for MSRs between 0x800
640 			 * and 0x8ff, it just lets the processor take the value
641 			 * from the virtual-APIC page; take those 256 bits
642 			 * directly from the L1 bitmap.
643 			 */
644 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
645 				unsigned word = msr / BITS_PER_LONG;
646 
647 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
648 			}
649 		}
650 
651 		nested_vmx_disable_intercept_for_x2apic_msr(
652 			msr_bitmap_l1, msr_bitmap_l0,
653 			X2APIC_MSR(APIC_TASKPRI),
654 			MSR_TYPE_R | MSR_TYPE_W);
655 
656 		if (nested_cpu_has_vid(vmcs12)) {
657 			nested_vmx_disable_intercept_for_x2apic_msr(
658 				msr_bitmap_l1, msr_bitmap_l0,
659 				X2APIC_MSR(APIC_EOI),
660 				MSR_TYPE_W);
661 			nested_vmx_disable_intercept_for_x2apic_msr(
662 				msr_bitmap_l1, msr_bitmap_l0,
663 				X2APIC_MSR(APIC_SELF_IPI),
664 				MSR_TYPE_W);
665 		}
666 	}
667 
668 	/*
669 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
670 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
671 	 */
672 #ifdef CONFIG_X86_64
673 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
674 					 MSR_FS_BASE, MSR_TYPE_RW);
675 
676 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
677 					 MSR_GS_BASE, MSR_TYPE_RW);
678 
679 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
680 					 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
681 #endif
682 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
683 					 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
684 
685 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
686 					 MSR_IA32_PRED_CMD, MSR_TYPE_W);
687 
688 	kvm_vcpu_unmap(vcpu, &vmx->nested.msr_bitmap_map, false);
689 
690 	vmx->nested.force_msr_bitmap_recalc = false;
691 
692 	return true;
693 }
694 
695 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
696 				       struct vmcs12 *vmcs12)
697 {
698 	struct vcpu_vmx *vmx = to_vmx(vcpu);
699 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
700 
701 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
702 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
703 		return;
704 
705 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
706 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
707 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
708 		return;
709 
710 	kvm_read_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
711 			      VMCS12_SIZE);
712 }
713 
714 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
715 					      struct vmcs12 *vmcs12)
716 {
717 	struct vcpu_vmx *vmx = to_vmx(vcpu);
718 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
719 
720 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
721 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
722 		return;
723 
724 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
725 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
726 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
727 		return;
728 
729 	kvm_write_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
730 			       VMCS12_SIZE);
731 }
732 
733 /*
734  * In nested virtualization, check if L1 has set
735  * VM_EXIT_ACK_INTR_ON_EXIT
736  */
737 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
738 {
739 	return get_vmcs12(vcpu)->vm_exit_controls &
740 		VM_EXIT_ACK_INTR_ON_EXIT;
741 }
742 
743 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
744 					  struct vmcs12 *vmcs12)
745 {
746 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
747 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
748 		return -EINVAL;
749 	else
750 		return 0;
751 }
752 
753 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
754 					   struct vmcs12 *vmcs12)
755 {
756 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
757 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
758 	    !nested_cpu_has_vid(vmcs12) &&
759 	    !nested_cpu_has_posted_intr(vmcs12))
760 		return 0;
761 
762 	/*
763 	 * If virtualize x2apic mode is enabled,
764 	 * virtualize apic access must be disabled.
765 	 */
766 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
767 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
768 		return -EINVAL;
769 
770 	/*
771 	 * If virtual interrupt delivery is enabled,
772 	 * we must exit on external interrupts.
773 	 */
774 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
775 		return -EINVAL;
776 
777 	/*
778 	 * bits 15:8 should be zero in posted_intr_nv,
779 	 * the descriptor address has been already checked
780 	 * in nested_get_vmcs12_pages.
781 	 *
782 	 * bits 5:0 of posted_intr_desc_addr should be zero.
783 	 */
784 	if (nested_cpu_has_posted_intr(vmcs12) &&
785 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
786 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
787 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
788 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
789 		return -EINVAL;
790 
791 	/* tpr shadow is needed by all apicv features. */
792 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
793 		return -EINVAL;
794 
795 	return 0;
796 }
797 
798 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
799 				       u32 count, u64 addr)
800 {
801 	if (count == 0)
802 		return 0;
803 
804 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
805 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
806 		return -EINVAL;
807 
808 	return 0;
809 }
810 
811 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
812 						     struct vmcs12 *vmcs12)
813 {
814 	if (CC(nested_vmx_check_msr_switch(vcpu,
815 					   vmcs12->vm_exit_msr_load_count,
816 					   vmcs12->vm_exit_msr_load_addr)) ||
817 	    CC(nested_vmx_check_msr_switch(vcpu,
818 					   vmcs12->vm_exit_msr_store_count,
819 					   vmcs12->vm_exit_msr_store_addr)))
820 		return -EINVAL;
821 
822 	return 0;
823 }
824 
825 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
826                                                       struct vmcs12 *vmcs12)
827 {
828 	if (CC(nested_vmx_check_msr_switch(vcpu,
829 					   vmcs12->vm_entry_msr_load_count,
830 					   vmcs12->vm_entry_msr_load_addr)))
831                 return -EINVAL;
832 
833 	return 0;
834 }
835 
836 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
837 					 struct vmcs12 *vmcs12)
838 {
839 	if (!nested_cpu_has_pml(vmcs12))
840 		return 0;
841 
842 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
843 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
844 		return -EINVAL;
845 
846 	return 0;
847 }
848 
849 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
850 							struct vmcs12 *vmcs12)
851 {
852 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
853 	       !nested_cpu_has_ept(vmcs12)))
854 		return -EINVAL;
855 	return 0;
856 }
857 
858 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
859 							 struct vmcs12 *vmcs12)
860 {
861 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
862 	       !nested_cpu_has_ept(vmcs12)))
863 		return -EINVAL;
864 	return 0;
865 }
866 
867 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
868 						 struct vmcs12 *vmcs12)
869 {
870 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
871 		return 0;
872 
873 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
874 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
875 		return -EINVAL;
876 
877 	return 0;
878 }
879 
880 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
881 				       struct vmx_msr_entry *e)
882 {
883 	/* x2APIC MSR accesses are not allowed */
884 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
885 		return -EINVAL;
886 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
887 	    CC(e->index == MSR_IA32_UCODE_REV))
888 		return -EINVAL;
889 	if (CC(e->reserved != 0))
890 		return -EINVAL;
891 	return 0;
892 }
893 
894 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
895 				     struct vmx_msr_entry *e)
896 {
897 	if (CC(e->index == MSR_FS_BASE) ||
898 	    CC(e->index == MSR_GS_BASE) ||
899 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
900 	    nested_vmx_msr_check_common(vcpu, e))
901 		return -EINVAL;
902 	return 0;
903 }
904 
905 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
906 				      struct vmx_msr_entry *e)
907 {
908 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
909 	    nested_vmx_msr_check_common(vcpu, e))
910 		return -EINVAL;
911 	return 0;
912 }
913 
914 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
915 {
916 	struct vcpu_vmx *vmx = to_vmx(vcpu);
917 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
918 				       vmx->nested.msrs.misc_high);
919 
920 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
921 }
922 
923 /*
924  * Load guest's/host's msr at nested entry/exit.
925  * return 0 for success, entry index for failure.
926  *
927  * One of the failure modes for MSR load/store is when a list exceeds the
928  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
929  * as possible, process all valid entries before failing rather than precheck
930  * for a capacity violation.
931  */
932 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
933 {
934 	u32 i;
935 	struct vmx_msr_entry e;
936 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
937 
938 	for (i = 0; i < count; i++) {
939 		if (unlikely(i >= max_msr_list_size))
940 			goto fail;
941 
942 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
943 					&e, sizeof(e))) {
944 			pr_debug_ratelimited(
945 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
946 				__func__, i, gpa + i * sizeof(e));
947 			goto fail;
948 		}
949 		if (nested_vmx_load_msr_check(vcpu, &e)) {
950 			pr_debug_ratelimited(
951 				"%s check failed (%u, 0x%x, 0x%x)\n",
952 				__func__, i, e.index, e.reserved);
953 			goto fail;
954 		}
955 		if (kvm_set_msr(vcpu, e.index, e.value)) {
956 			pr_debug_ratelimited(
957 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
958 				__func__, i, e.index, e.value);
959 			goto fail;
960 		}
961 	}
962 	return 0;
963 fail:
964 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
965 	return i + 1;
966 }
967 
968 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
969 					    u32 msr_index,
970 					    u64 *data)
971 {
972 	struct vcpu_vmx *vmx = to_vmx(vcpu);
973 
974 	/*
975 	 * If the L0 hypervisor stored a more accurate value for the TSC that
976 	 * does not include the time taken for emulation of the L2->L1
977 	 * VM-exit in L0, use the more accurate value.
978 	 */
979 	if (msr_index == MSR_IA32_TSC) {
980 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
981 						    MSR_IA32_TSC);
982 
983 		if (i >= 0) {
984 			u64 val = vmx->msr_autostore.guest.val[i].value;
985 
986 			*data = kvm_read_l1_tsc(vcpu, val);
987 			return true;
988 		}
989 	}
990 
991 	if (kvm_get_msr(vcpu, msr_index, data)) {
992 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
993 			msr_index);
994 		return false;
995 	}
996 	return true;
997 }
998 
999 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
1000 				     struct vmx_msr_entry *e)
1001 {
1002 	if (kvm_vcpu_read_guest(vcpu,
1003 				gpa + i * sizeof(*e),
1004 				e, 2 * sizeof(u32))) {
1005 		pr_debug_ratelimited(
1006 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
1007 			__func__, i, gpa + i * sizeof(*e));
1008 		return false;
1009 	}
1010 	if (nested_vmx_store_msr_check(vcpu, e)) {
1011 		pr_debug_ratelimited(
1012 			"%s check failed (%u, 0x%x, 0x%x)\n",
1013 			__func__, i, e->index, e->reserved);
1014 		return false;
1015 	}
1016 	return true;
1017 }
1018 
1019 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1020 {
1021 	u64 data;
1022 	u32 i;
1023 	struct vmx_msr_entry e;
1024 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1025 
1026 	for (i = 0; i < count; i++) {
1027 		if (unlikely(i >= max_msr_list_size))
1028 			return -EINVAL;
1029 
1030 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1031 			return -EINVAL;
1032 
1033 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1034 			return -EINVAL;
1035 
1036 		if (kvm_vcpu_write_guest(vcpu,
1037 					 gpa + i * sizeof(e) +
1038 					     offsetof(struct vmx_msr_entry, value),
1039 					 &data, sizeof(data))) {
1040 			pr_debug_ratelimited(
1041 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1042 				__func__, i, e.index, data);
1043 			return -EINVAL;
1044 		}
1045 	}
1046 	return 0;
1047 }
1048 
1049 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1050 {
1051 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1052 	u32 count = vmcs12->vm_exit_msr_store_count;
1053 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1054 	struct vmx_msr_entry e;
1055 	u32 i;
1056 
1057 	for (i = 0; i < count; i++) {
1058 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1059 			return false;
1060 
1061 		if (e.index == msr_index)
1062 			return true;
1063 	}
1064 	return false;
1065 }
1066 
1067 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1068 					   u32 msr_index)
1069 {
1070 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1071 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1072 	bool in_vmcs12_store_list;
1073 	int msr_autostore_slot;
1074 	bool in_autostore_list;
1075 	int last;
1076 
1077 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1078 	in_autostore_list = msr_autostore_slot >= 0;
1079 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1080 
1081 	if (in_vmcs12_store_list && !in_autostore_list) {
1082 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1083 			/*
1084 			 * Emulated VMEntry does not fail here.  Instead a less
1085 			 * accurate value will be returned by
1086 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1087 			 * instead of reading the value from the vmcs02 VMExit
1088 			 * MSR-store area.
1089 			 */
1090 			pr_warn_ratelimited(
1091 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1092 				msr_index);
1093 			return;
1094 		}
1095 		last = autostore->nr++;
1096 		autostore->val[last].index = msr_index;
1097 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1098 		last = --autostore->nr;
1099 		autostore->val[msr_autostore_slot] = autostore->val[last];
1100 	}
1101 }
1102 
1103 /*
1104  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1105  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1106  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1107  * @entry_failure_code.
1108  */
1109 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1110 			       bool nested_ept, bool reload_pdptrs,
1111 			       enum vm_entry_failure_code *entry_failure_code)
1112 {
1113 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1114 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1115 		return -EINVAL;
1116 	}
1117 
1118 	/*
1119 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1120 	 * must not be dereferenced.
1121 	 */
1122 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1123 	    CC(!load_pdptrs(vcpu, cr3))) {
1124 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1125 		return -EINVAL;
1126 	}
1127 
1128 	if (!nested_ept)
1129 		kvm_mmu_new_pgd(vcpu, cr3);
1130 
1131 	vcpu->arch.cr3 = cr3;
1132 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1133 
1134 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1135 	kvm_init_mmu(vcpu);
1136 
1137 	return 0;
1138 }
1139 
1140 /*
1141  * Returns if KVM is able to config CPU to tag TLB entries
1142  * populated by L2 differently than TLB entries populated
1143  * by L1.
1144  *
1145  * If L0 uses EPT, L1 and L2 run with different EPTP because
1146  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1147  * are tagged with different EPTP.
1148  *
1149  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1150  * with different VPID (L1 entries are tagged with vmx->vpid
1151  * while L2 entries are tagged with vmx->nested.vpid02).
1152  */
1153 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1154 {
1155 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1156 
1157 	return enable_ept ||
1158 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1159 }
1160 
1161 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1162 					    struct vmcs12 *vmcs12,
1163 					    bool is_vmenter)
1164 {
1165 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1166 
1167 	/*
1168 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1169 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1170 	 * full TLB flush from the guest's perspective.  This is required even
1171 	 * if VPID is disabled in the host as KVM may need to synchronize the
1172 	 * MMU in response to the guest TLB flush.
1173 	 *
1174 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1175 	 * EPT is a special snowflake, as guest-physical mappings aren't
1176 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1177 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1178 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1179 	 * those mappings.
1180 	 */
1181 	if (!nested_cpu_has_vpid(vmcs12)) {
1182 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1183 		return;
1184 	}
1185 
1186 	/* L2 should never have a VPID if VPID is disabled. */
1187 	WARN_ON(!enable_vpid);
1188 
1189 	/*
1190 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1191 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1192 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1193 	 * that the new vpid12 has never been used and thus represents a new
1194 	 * guest ASID that cannot have entries in the TLB.
1195 	 */
1196 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1197 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1198 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1199 		return;
1200 	}
1201 
1202 	/*
1203 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1204 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1205 	 * KVM was unable to allocate a VPID for L2, flush the current context
1206 	 * as the effective ASID is common to both L1 and L2.
1207 	 */
1208 	if (!nested_has_guest_tlb_tag(vcpu))
1209 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1210 }
1211 
1212 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1213 {
1214 	superset &= mask;
1215 	subset &= mask;
1216 
1217 	return (superset | subset) == superset;
1218 }
1219 
1220 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1221 {
1222 	const u64 feature_and_reserved =
1223 		/* feature (except bit 48; see below) */
1224 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1225 		/* reserved */
1226 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1227 	u64 vmx_basic = vmx->nested.msrs.basic;
1228 
1229 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1230 		return -EINVAL;
1231 
1232 	/*
1233 	 * KVM does not emulate a version of VMX that constrains physical
1234 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1235 	 */
1236 	if (data & BIT_ULL(48))
1237 		return -EINVAL;
1238 
1239 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1240 	    vmx_basic_vmcs_revision_id(data))
1241 		return -EINVAL;
1242 
1243 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1244 		return -EINVAL;
1245 
1246 	vmx->nested.msrs.basic = data;
1247 	return 0;
1248 }
1249 
1250 static int
1251 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1252 {
1253 	u64 supported;
1254 	u32 *lowp, *highp;
1255 
1256 	switch (msr_index) {
1257 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1258 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1259 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1260 		break;
1261 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1262 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1263 		highp = &vmx->nested.msrs.procbased_ctls_high;
1264 		break;
1265 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1266 		lowp = &vmx->nested.msrs.exit_ctls_low;
1267 		highp = &vmx->nested.msrs.exit_ctls_high;
1268 		break;
1269 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1270 		lowp = &vmx->nested.msrs.entry_ctls_low;
1271 		highp = &vmx->nested.msrs.entry_ctls_high;
1272 		break;
1273 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1274 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1275 		highp = &vmx->nested.msrs.secondary_ctls_high;
1276 		break;
1277 	default:
1278 		BUG();
1279 	}
1280 
1281 	supported = vmx_control_msr(*lowp, *highp);
1282 
1283 	/* Check must-be-1 bits are still 1. */
1284 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1285 		return -EINVAL;
1286 
1287 	/* Check must-be-0 bits are still 0. */
1288 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1289 		return -EINVAL;
1290 
1291 	*lowp = data;
1292 	*highp = data >> 32;
1293 	return 0;
1294 }
1295 
1296 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1297 {
1298 	const u64 feature_and_reserved_bits =
1299 		/* feature */
1300 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1301 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1302 		/* reserved */
1303 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1304 	u64 vmx_misc;
1305 
1306 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1307 				   vmx->nested.msrs.misc_high);
1308 
1309 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1310 		return -EINVAL;
1311 
1312 	if ((vmx->nested.msrs.pinbased_ctls_high &
1313 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1314 	    vmx_misc_preemption_timer_rate(data) !=
1315 	    vmx_misc_preemption_timer_rate(vmx_misc))
1316 		return -EINVAL;
1317 
1318 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1319 		return -EINVAL;
1320 
1321 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1322 		return -EINVAL;
1323 
1324 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1325 		return -EINVAL;
1326 
1327 	vmx->nested.msrs.misc_low = data;
1328 	vmx->nested.msrs.misc_high = data >> 32;
1329 
1330 	return 0;
1331 }
1332 
1333 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1334 {
1335 	u64 vmx_ept_vpid_cap;
1336 
1337 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1338 					   vmx->nested.msrs.vpid_caps);
1339 
1340 	/* Every bit is either reserved or a feature bit. */
1341 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1342 		return -EINVAL;
1343 
1344 	vmx->nested.msrs.ept_caps = data;
1345 	vmx->nested.msrs.vpid_caps = data >> 32;
1346 	return 0;
1347 }
1348 
1349 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1350 {
1351 	u64 *msr;
1352 
1353 	switch (msr_index) {
1354 	case MSR_IA32_VMX_CR0_FIXED0:
1355 		msr = &vmx->nested.msrs.cr0_fixed0;
1356 		break;
1357 	case MSR_IA32_VMX_CR4_FIXED0:
1358 		msr = &vmx->nested.msrs.cr4_fixed0;
1359 		break;
1360 	default:
1361 		BUG();
1362 	}
1363 
1364 	/*
1365 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1366 	 * must be 1 in the restored value.
1367 	 */
1368 	if (!is_bitwise_subset(data, *msr, -1ULL))
1369 		return -EINVAL;
1370 
1371 	*msr = data;
1372 	return 0;
1373 }
1374 
1375 /*
1376  * Called when userspace is restoring VMX MSRs.
1377  *
1378  * Returns 0 on success, non-0 otherwise.
1379  */
1380 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1381 {
1382 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1383 
1384 	/*
1385 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1386 	 * is in VMX operation.
1387 	 */
1388 	if (vmx->nested.vmxon)
1389 		return -EBUSY;
1390 
1391 	switch (msr_index) {
1392 	case MSR_IA32_VMX_BASIC:
1393 		return vmx_restore_vmx_basic(vmx, data);
1394 	case MSR_IA32_VMX_PINBASED_CTLS:
1395 	case MSR_IA32_VMX_PROCBASED_CTLS:
1396 	case MSR_IA32_VMX_EXIT_CTLS:
1397 	case MSR_IA32_VMX_ENTRY_CTLS:
1398 		/*
1399 		 * The "non-true" VMX capability MSRs are generated from the
1400 		 * "true" MSRs, so we do not support restoring them directly.
1401 		 *
1402 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1403 		 * should restore the "true" MSRs with the must-be-1 bits
1404 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1405 		 * DEFAULT SETTINGS".
1406 		 */
1407 		return -EINVAL;
1408 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1409 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1410 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1411 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1412 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1413 		return vmx_restore_control_msr(vmx, msr_index, data);
1414 	case MSR_IA32_VMX_MISC:
1415 		return vmx_restore_vmx_misc(vmx, data);
1416 	case MSR_IA32_VMX_CR0_FIXED0:
1417 	case MSR_IA32_VMX_CR4_FIXED0:
1418 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1419 	case MSR_IA32_VMX_CR0_FIXED1:
1420 	case MSR_IA32_VMX_CR4_FIXED1:
1421 		/*
1422 		 * These MSRs are generated based on the vCPU's CPUID, so we
1423 		 * do not support restoring them directly.
1424 		 */
1425 		return -EINVAL;
1426 	case MSR_IA32_VMX_EPT_VPID_CAP:
1427 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1428 	case MSR_IA32_VMX_VMCS_ENUM:
1429 		vmx->nested.msrs.vmcs_enum = data;
1430 		return 0;
1431 	case MSR_IA32_VMX_VMFUNC:
1432 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1433 			return -EINVAL;
1434 		vmx->nested.msrs.vmfunc_controls = data;
1435 		return 0;
1436 	default:
1437 		/*
1438 		 * The rest of the VMX capability MSRs do not support restore.
1439 		 */
1440 		return -EINVAL;
1441 	}
1442 }
1443 
1444 /* Returns 0 on success, non-0 otherwise. */
1445 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1446 {
1447 	switch (msr_index) {
1448 	case MSR_IA32_VMX_BASIC:
1449 		*pdata = msrs->basic;
1450 		break;
1451 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1452 	case MSR_IA32_VMX_PINBASED_CTLS:
1453 		*pdata = vmx_control_msr(
1454 			msrs->pinbased_ctls_low,
1455 			msrs->pinbased_ctls_high);
1456 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1457 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1458 		break;
1459 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1460 	case MSR_IA32_VMX_PROCBASED_CTLS:
1461 		*pdata = vmx_control_msr(
1462 			msrs->procbased_ctls_low,
1463 			msrs->procbased_ctls_high);
1464 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1465 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1466 		break;
1467 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1468 	case MSR_IA32_VMX_EXIT_CTLS:
1469 		*pdata = vmx_control_msr(
1470 			msrs->exit_ctls_low,
1471 			msrs->exit_ctls_high);
1472 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1473 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1474 		break;
1475 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1476 	case MSR_IA32_VMX_ENTRY_CTLS:
1477 		*pdata = vmx_control_msr(
1478 			msrs->entry_ctls_low,
1479 			msrs->entry_ctls_high);
1480 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1481 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1482 		break;
1483 	case MSR_IA32_VMX_MISC:
1484 		*pdata = vmx_control_msr(
1485 			msrs->misc_low,
1486 			msrs->misc_high);
1487 		break;
1488 	case MSR_IA32_VMX_CR0_FIXED0:
1489 		*pdata = msrs->cr0_fixed0;
1490 		break;
1491 	case MSR_IA32_VMX_CR0_FIXED1:
1492 		*pdata = msrs->cr0_fixed1;
1493 		break;
1494 	case MSR_IA32_VMX_CR4_FIXED0:
1495 		*pdata = msrs->cr4_fixed0;
1496 		break;
1497 	case MSR_IA32_VMX_CR4_FIXED1:
1498 		*pdata = msrs->cr4_fixed1;
1499 		break;
1500 	case MSR_IA32_VMX_VMCS_ENUM:
1501 		*pdata = msrs->vmcs_enum;
1502 		break;
1503 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1504 		*pdata = vmx_control_msr(
1505 			msrs->secondary_ctls_low,
1506 			msrs->secondary_ctls_high);
1507 		break;
1508 	case MSR_IA32_VMX_EPT_VPID_CAP:
1509 		*pdata = msrs->ept_caps |
1510 			((u64)msrs->vpid_caps << 32);
1511 		break;
1512 	case MSR_IA32_VMX_VMFUNC:
1513 		*pdata = msrs->vmfunc_controls;
1514 		break;
1515 	default:
1516 		return 1;
1517 	}
1518 
1519 	return 0;
1520 }
1521 
1522 /*
1523  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1524  * been modified by the L1 guest.  Note, "writable" in this context means
1525  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1526  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1527  * VM-exit information fields (which are actually writable if the vCPU is
1528  * configured to support "VMWRITE to any supported field in the VMCS").
1529  */
1530 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1531 {
1532 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1533 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1534 	struct shadow_vmcs_field field;
1535 	unsigned long val;
1536 	int i;
1537 
1538 	if (WARN_ON(!shadow_vmcs))
1539 		return;
1540 
1541 	preempt_disable();
1542 
1543 	vmcs_load(shadow_vmcs);
1544 
1545 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1546 		field = shadow_read_write_fields[i];
1547 		val = __vmcs_readl(field.encoding);
1548 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1549 	}
1550 
1551 	vmcs_clear(shadow_vmcs);
1552 	vmcs_load(vmx->loaded_vmcs->vmcs);
1553 
1554 	preempt_enable();
1555 }
1556 
1557 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1558 {
1559 	const struct shadow_vmcs_field *fields[] = {
1560 		shadow_read_write_fields,
1561 		shadow_read_only_fields
1562 	};
1563 	const int max_fields[] = {
1564 		max_shadow_read_write_fields,
1565 		max_shadow_read_only_fields
1566 	};
1567 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1568 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1569 	struct shadow_vmcs_field field;
1570 	unsigned long val;
1571 	int i, q;
1572 
1573 	if (WARN_ON(!shadow_vmcs))
1574 		return;
1575 
1576 	vmcs_load(shadow_vmcs);
1577 
1578 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1579 		for (i = 0; i < max_fields[q]; i++) {
1580 			field = fields[q][i];
1581 			val = vmcs12_read_any(vmcs12, field.encoding,
1582 					      field.offset);
1583 			__vmcs_writel(field.encoding, val);
1584 		}
1585 	}
1586 
1587 	vmcs_clear(shadow_vmcs);
1588 	vmcs_load(vmx->loaded_vmcs->vmcs);
1589 }
1590 
1591 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1592 {
1593 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1594 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1595 
1596 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1597 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1598 	vmcs12->guest_rip = evmcs->guest_rip;
1599 
1600 	if (unlikely(!(hv_clean_fields &
1601 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1602 		vmcs12->guest_rsp = evmcs->guest_rsp;
1603 		vmcs12->guest_rflags = evmcs->guest_rflags;
1604 		vmcs12->guest_interruptibility_info =
1605 			evmcs->guest_interruptibility_info;
1606 	}
1607 
1608 	if (unlikely(!(hv_clean_fields &
1609 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1610 		vmcs12->cpu_based_vm_exec_control =
1611 			evmcs->cpu_based_vm_exec_control;
1612 	}
1613 
1614 	if (unlikely(!(hv_clean_fields &
1615 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1616 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1617 	}
1618 
1619 	if (unlikely(!(hv_clean_fields &
1620 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1621 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1622 	}
1623 
1624 	if (unlikely(!(hv_clean_fields &
1625 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1626 		vmcs12->vm_entry_intr_info_field =
1627 			evmcs->vm_entry_intr_info_field;
1628 		vmcs12->vm_entry_exception_error_code =
1629 			evmcs->vm_entry_exception_error_code;
1630 		vmcs12->vm_entry_instruction_len =
1631 			evmcs->vm_entry_instruction_len;
1632 	}
1633 
1634 	if (unlikely(!(hv_clean_fields &
1635 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1636 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1637 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1638 		vmcs12->host_cr0 = evmcs->host_cr0;
1639 		vmcs12->host_cr3 = evmcs->host_cr3;
1640 		vmcs12->host_cr4 = evmcs->host_cr4;
1641 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1642 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1643 		vmcs12->host_rip = evmcs->host_rip;
1644 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1645 		vmcs12->host_es_selector = evmcs->host_es_selector;
1646 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1647 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1648 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1649 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1650 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1651 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1652 	}
1653 
1654 	if (unlikely(!(hv_clean_fields &
1655 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1656 		vmcs12->pin_based_vm_exec_control =
1657 			evmcs->pin_based_vm_exec_control;
1658 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1659 		vmcs12->secondary_vm_exec_control =
1660 			evmcs->secondary_vm_exec_control;
1661 	}
1662 
1663 	if (unlikely(!(hv_clean_fields &
1664 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1665 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1666 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1667 	}
1668 
1669 	if (unlikely(!(hv_clean_fields &
1670 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1671 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1672 	}
1673 
1674 	if (unlikely(!(hv_clean_fields &
1675 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1676 		vmcs12->guest_es_base = evmcs->guest_es_base;
1677 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1678 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1679 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1680 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1681 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1682 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1683 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1684 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1685 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1686 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1687 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1688 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1689 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1690 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1691 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1692 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1693 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1694 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1695 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1696 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1697 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1698 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1699 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1700 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1701 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1702 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1703 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1704 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1705 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1706 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1707 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1708 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1709 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1710 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1711 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1712 	}
1713 
1714 	if (unlikely(!(hv_clean_fields &
1715 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1716 		vmcs12->tsc_offset = evmcs->tsc_offset;
1717 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1718 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1719 	}
1720 
1721 	if (unlikely(!(hv_clean_fields &
1722 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1723 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1724 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1725 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1726 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1727 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1728 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1729 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1730 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1731 	}
1732 
1733 	if (unlikely(!(hv_clean_fields &
1734 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1735 		vmcs12->host_fs_base = evmcs->host_fs_base;
1736 		vmcs12->host_gs_base = evmcs->host_gs_base;
1737 		vmcs12->host_tr_base = evmcs->host_tr_base;
1738 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1739 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1740 		vmcs12->host_rsp = evmcs->host_rsp;
1741 	}
1742 
1743 	if (unlikely(!(hv_clean_fields &
1744 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1745 		vmcs12->ept_pointer = evmcs->ept_pointer;
1746 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1747 	}
1748 
1749 	if (unlikely(!(hv_clean_fields &
1750 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1751 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1752 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1753 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1754 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1755 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1756 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1757 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1758 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1759 		vmcs12->guest_pending_dbg_exceptions =
1760 			evmcs->guest_pending_dbg_exceptions;
1761 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1762 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1763 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1764 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1765 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1766 	}
1767 
1768 	/*
1769 	 * Not used?
1770 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1771 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1772 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1773 	 * vmcs12->page_fault_error_code_mask =
1774 	 *		evmcs->page_fault_error_code_mask;
1775 	 * vmcs12->page_fault_error_code_match =
1776 	 *		evmcs->page_fault_error_code_match;
1777 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1778 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1779 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1780 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1781 	 */
1782 
1783 	/*
1784 	 * Read only fields:
1785 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1786 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1787 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1788 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1789 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1790 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1791 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1792 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1793 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1794 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1795 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1796 	 *
1797 	 * Not present in struct vmcs12:
1798 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1799 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1800 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1801 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1802 	 */
1803 
1804 	return;
1805 }
1806 
1807 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1808 {
1809 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1810 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1811 
1812 	/*
1813 	 * Should not be changed by KVM:
1814 	 *
1815 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1816 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1817 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1818 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1819 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1820 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1821 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1822 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1823 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1824 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1825 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1826 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1827 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1828 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1829 	 * evmcs->host_rip = vmcs12->host_rip;
1830 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1831 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1832 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1833 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1834 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1835 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1836 	 * evmcs->host_rsp = vmcs12->host_rsp;
1837 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1838 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1839 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1840 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1841 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1842 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1843 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1844 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1845 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1846 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1847 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1848 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1849 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1850 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1851 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1852 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1853 	 * evmcs->page_fault_error_code_mask =
1854 	 *		vmcs12->page_fault_error_code_mask;
1855 	 * evmcs->page_fault_error_code_match =
1856 	 *		vmcs12->page_fault_error_code_match;
1857 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1858 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1859 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1860 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1861 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1862 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1863 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1864 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1865 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1866 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1867 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1868 	 *
1869 	 * Not present in struct vmcs12:
1870 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1871 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1872 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1873 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1874 	 */
1875 
1876 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1877 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1878 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1879 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1880 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1881 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1882 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1883 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1884 
1885 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1886 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1887 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1888 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1889 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1890 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1891 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1892 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1893 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1894 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1895 
1896 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1897 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1898 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1899 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1900 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1901 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1902 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1903 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1904 
1905 	evmcs->guest_es_base = vmcs12->guest_es_base;
1906 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1907 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1908 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1909 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1910 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1911 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1912 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1913 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1914 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1915 
1916 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1917 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1918 
1919 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1920 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1921 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1922 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1923 
1924 	evmcs->guest_pending_dbg_exceptions =
1925 		vmcs12->guest_pending_dbg_exceptions;
1926 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1927 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1928 
1929 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1930 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1931 
1932 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1933 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1934 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1935 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1936 
1937 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1938 
1939 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1940 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1941 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1942 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1943 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1944 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1945 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1946 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1947 
1948 	evmcs->exit_qualification = vmcs12->exit_qualification;
1949 
1950 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1951 	evmcs->guest_rsp = vmcs12->guest_rsp;
1952 	evmcs->guest_rflags = vmcs12->guest_rflags;
1953 
1954 	evmcs->guest_interruptibility_info =
1955 		vmcs12->guest_interruptibility_info;
1956 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1957 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1958 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1959 	evmcs->vm_entry_exception_error_code =
1960 		vmcs12->vm_entry_exception_error_code;
1961 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1962 
1963 	evmcs->guest_rip = vmcs12->guest_rip;
1964 
1965 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1966 
1967 	return;
1968 }
1969 
1970 /*
1971  * This is an equivalent of the nested hypervisor executing the vmptrld
1972  * instruction.
1973  */
1974 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1975 	struct kvm_vcpu *vcpu, bool from_launch)
1976 {
1977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1978 	bool evmcs_gpa_changed = false;
1979 	u64 evmcs_gpa;
1980 
1981 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1982 		return EVMPTRLD_DISABLED;
1983 
1984 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa)) {
1985 		nested_release_evmcs(vcpu);
1986 		return EVMPTRLD_DISABLED;
1987 	}
1988 
1989 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1990 		vmx->nested.current_vmptr = INVALID_GPA;
1991 
1992 		nested_release_evmcs(vcpu);
1993 
1994 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1995 				 &vmx->nested.hv_evmcs_map))
1996 			return EVMPTRLD_ERROR;
1997 
1998 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1999 
2000 		/*
2001 		 * Currently, KVM only supports eVMCS version 1
2002 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2003 		 * value to first u32 field of eVMCS which should specify eVMCS
2004 		 * VersionNumber.
2005 		 *
2006 		 * Guest should be aware of supported eVMCS versions by host by
2007 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2008 		 * expected to set this CPUID leaf according to the value
2009 		 * returned in vmcs_version from nested_enable_evmcs().
2010 		 *
2011 		 * However, it turns out that Microsoft Hyper-V fails to comply
2012 		 * to their own invented interface: When Hyper-V use eVMCS, it
2013 		 * just sets first u32 field of eVMCS to revision_id specified
2014 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2015 		 * which is one of the supported versions specified in
2016 		 * CPUID.0x4000000A.EAX[0:15].
2017 		 *
2018 		 * To overcome Hyper-V bug, we accept here either a supported
2019 		 * eVMCS version or VMCS12 revision_id as valid values for first
2020 		 * u32 field of eVMCS.
2021 		 */
2022 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2023 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2024 			nested_release_evmcs(vcpu);
2025 			return EVMPTRLD_VMFAIL;
2026 		}
2027 
2028 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2029 
2030 		evmcs_gpa_changed = true;
2031 		/*
2032 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2033 		 * reloaded from guest's memory (read only fields, fields not
2034 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2035 		 * are no leftovers.
2036 		 */
2037 		if (from_launch) {
2038 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2039 			memset(vmcs12, 0, sizeof(*vmcs12));
2040 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2041 		}
2042 
2043 	}
2044 
2045 	/*
2046 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2047 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2048 	 */
2049 	if (from_launch || evmcs_gpa_changed) {
2050 		vmx->nested.hv_evmcs->hv_clean_fields &=
2051 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2052 
2053 		vmx->nested.force_msr_bitmap_recalc = true;
2054 	}
2055 
2056 	return EVMPTRLD_SUCCEEDED;
2057 }
2058 
2059 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2060 {
2061 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2062 
2063 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2064 		copy_vmcs12_to_enlightened(vmx);
2065 	else
2066 		copy_vmcs12_to_shadow(vmx);
2067 
2068 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2069 }
2070 
2071 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2072 {
2073 	struct vcpu_vmx *vmx =
2074 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2075 
2076 	vmx->nested.preemption_timer_expired = true;
2077 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2078 	kvm_vcpu_kick(&vmx->vcpu);
2079 
2080 	return HRTIMER_NORESTART;
2081 }
2082 
2083 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2084 {
2085 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2086 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2087 
2088 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2089 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2090 
2091 	if (!vmx->nested.has_preemption_timer_deadline) {
2092 		vmx->nested.preemption_timer_deadline =
2093 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2094 		vmx->nested.has_preemption_timer_deadline = true;
2095 	}
2096 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2097 }
2098 
2099 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2100 					u64 preemption_timeout)
2101 {
2102 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2103 
2104 	/*
2105 	 * A timer value of zero is architecturally guaranteed to cause
2106 	 * a VMExit prior to executing any instructions in the guest.
2107 	 */
2108 	if (preemption_timeout == 0) {
2109 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2110 		return;
2111 	}
2112 
2113 	if (vcpu->arch.virtual_tsc_khz == 0)
2114 		return;
2115 
2116 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2117 	preemption_timeout *= 1000000;
2118 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2119 	hrtimer_start(&vmx->nested.preemption_timer,
2120 		      ktime_add_ns(ktime_get(), preemption_timeout),
2121 		      HRTIMER_MODE_ABS_PINNED);
2122 }
2123 
2124 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2125 {
2126 	if (vmx->nested.nested_run_pending &&
2127 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2128 		return vmcs12->guest_ia32_efer;
2129 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2130 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2131 	else
2132 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2133 }
2134 
2135 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2136 {
2137 	/*
2138 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2139 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2140 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2141 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2142 	 */
2143 	if (vmx->nested.vmcs02_initialized)
2144 		return;
2145 	vmx->nested.vmcs02_initialized = true;
2146 
2147 	/*
2148 	 * We don't care what the EPTP value is we just need to guarantee
2149 	 * it's valid so we don't get a false positive when doing early
2150 	 * consistency checks.
2151 	 */
2152 	if (enable_ept && nested_early_check)
2153 		vmcs_write64(EPT_POINTER,
2154 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2155 
2156 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2157 	if (cpu_has_vmx_vmfunc())
2158 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2159 
2160 	if (cpu_has_vmx_posted_intr())
2161 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2162 
2163 	if (cpu_has_vmx_msr_bitmap())
2164 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2165 
2166 	/*
2167 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2168 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2169 	 * deal with filtering out L2 GPAs from the buffer.
2170 	 */
2171 	if (enable_pml) {
2172 		vmcs_write64(PML_ADDRESS, 0);
2173 		vmcs_write16(GUEST_PML_INDEX, -1);
2174 	}
2175 
2176 	if (cpu_has_vmx_encls_vmexit())
2177 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2178 
2179 	/*
2180 	 * Set the MSR load/store lists to match L0's settings.  Only the
2181 	 * addresses are constant (for vmcs02), the counts can change based
2182 	 * on L2's behavior, e.g. switching to/from long mode.
2183 	 */
2184 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2185 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2186 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2187 
2188 	vmx_set_constant_host_state(vmx);
2189 }
2190 
2191 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2192 				      struct vmcs12 *vmcs12)
2193 {
2194 	prepare_vmcs02_constant_state(vmx);
2195 
2196 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2197 
2198 	if (enable_vpid) {
2199 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2200 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2201 		else
2202 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2203 	}
2204 }
2205 
2206 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2207 				 struct vmcs12 *vmcs12)
2208 {
2209 	u32 exec_control;
2210 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2211 
2212 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2213 		prepare_vmcs02_early_rare(vmx, vmcs12);
2214 
2215 	/*
2216 	 * PIN CONTROLS
2217 	 */
2218 	exec_control = __pin_controls_get(vmcs01);
2219 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2220 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2221 
2222 	/* Posted interrupts setting is only taken from vmcs12.  */
2223 	vmx->nested.pi_pending = false;
2224 	if (nested_cpu_has_posted_intr(vmcs12))
2225 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2226 	else
2227 		exec_control &= ~PIN_BASED_POSTED_INTR;
2228 	pin_controls_set(vmx, exec_control);
2229 
2230 	/*
2231 	 * EXEC CONTROLS
2232 	 */
2233 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2234 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2235 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2236 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2237 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2238 
2239 	vmx->nested.l1_tpr_threshold = -1;
2240 	if (exec_control & CPU_BASED_TPR_SHADOW)
2241 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2242 #ifdef CONFIG_X86_64
2243 	else
2244 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2245 				CPU_BASED_CR8_STORE_EXITING;
2246 #endif
2247 
2248 	/*
2249 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2250 	 * for I/O port accesses.
2251 	 */
2252 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2253 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2254 
2255 	/*
2256 	 * This bit will be computed in nested_get_vmcs12_pages, because
2257 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2258 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2259 	 * only set/clear this bit.
2260 	 */
2261 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2262 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2263 
2264 	exec_controls_set(vmx, exec_control);
2265 
2266 	/*
2267 	 * SECONDARY EXEC CONTROLS
2268 	 */
2269 	if (cpu_has_secondary_exec_ctrls()) {
2270 		exec_control = __secondary_exec_controls_get(vmcs01);
2271 
2272 		/* Take the following fields only from vmcs12 */
2273 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2274 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2275 				  SECONDARY_EXEC_ENABLE_INVPCID |
2276 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2277 				  SECONDARY_EXEC_XSAVES |
2278 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2279 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2280 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2281 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2282 				  SECONDARY_EXEC_TSC_SCALING |
2283 				  SECONDARY_EXEC_DESC);
2284 
2285 		if (nested_cpu_has(vmcs12,
2286 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2287 			exec_control |= vmcs12->secondary_vm_exec_control;
2288 
2289 		/* PML is emulated and never enabled in hardware for L2. */
2290 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2291 
2292 		/* VMCS shadowing for L2 is emulated for now */
2293 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2294 
2295 		/*
2296 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2297 		 * will not have to rewrite the controls just for this bit.
2298 		 */
2299 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2300 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2301 			exec_control |= SECONDARY_EXEC_DESC;
2302 
2303 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2304 			vmcs_write16(GUEST_INTR_STATUS,
2305 				vmcs12->guest_intr_status);
2306 
2307 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2308 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2309 
2310 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2311 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2312 
2313 		secondary_exec_controls_set(vmx, exec_control);
2314 	}
2315 
2316 	/*
2317 	 * ENTRY CONTROLS
2318 	 *
2319 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2320 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2321 	 * on the related bits (if supported by the CPU) in the hope that
2322 	 * we can avoid VMWrites during vmx_set_efer().
2323 	 */
2324 	exec_control = __vm_entry_controls_get(vmcs01);
2325 	exec_control |= vmcs12->vm_entry_controls;
2326 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2327 	if (cpu_has_load_ia32_efer()) {
2328 		if (guest_efer & EFER_LMA)
2329 			exec_control |= VM_ENTRY_IA32E_MODE;
2330 		if (guest_efer != host_efer)
2331 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2332 	}
2333 	vm_entry_controls_set(vmx, exec_control);
2334 
2335 	/*
2336 	 * EXIT CONTROLS
2337 	 *
2338 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2339 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2340 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2341 	 */
2342 	exec_control = __vm_exit_controls_get(vmcs01);
2343 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2344 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2345 	else
2346 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2347 	vm_exit_controls_set(vmx, exec_control);
2348 
2349 	/*
2350 	 * Interrupt/Exception Fields
2351 	 */
2352 	if (vmx->nested.nested_run_pending) {
2353 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2354 			     vmcs12->vm_entry_intr_info_field);
2355 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2356 			     vmcs12->vm_entry_exception_error_code);
2357 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2358 			     vmcs12->vm_entry_instruction_len);
2359 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2360 			     vmcs12->guest_interruptibility_info);
2361 		vmx->loaded_vmcs->nmi_known_unmasked =
2362 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2363 	} else {
2364 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2365 	}
2366 }
2367 
2368 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2369 {
2370 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2371 
2372 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2373 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2374 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2375 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2376 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2377 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2378 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2379 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2380 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2381 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2382 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2383 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2384 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2385 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2386 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2387 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2388 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2389 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2390 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2391 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2392 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2393 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2394 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2395 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2396 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2397 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2398 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2399 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2400 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2401 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2402 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2403 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2404 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2405 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2406 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2407 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2408 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2409 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2410 
2411 		vmx->segment_cache.bitmask = 0;
2412 	}
2413 
2414 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2415 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2416 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2417 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2418 			    vmcs12->guest_pending_dbg_exceptions);
2419 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2420 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2421 
2422 		/*
2423 		 * L1 may access the L2's PDPTR, so save them to construct
2424 		 * vmcs12
2425 		 */
2426 		if (enable_ept) {
2427 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2428 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2429 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2430 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2431 		}
2432 
2433 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2434 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2435 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2436 	}
2437 
2438 	if (nested_cpu_has_xsaves(vmcs12))
2439 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2440 
2441 	/*
2442 	 * Whether page-faults are trapped is determined by a combination of
2443 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2444 	 * doesn't care about page faults then we should set all of these to
2445 	 * L1's desires. However, if L0 does care about (some) page faults, it
2446 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2447 	 * simply ask to exit on each and every L2 page fault. This is done by
2448 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2449 	 * Note that below we don't need special code to set EB.PF beyond the
2450 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2451 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2452 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2453 	 */
2454 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2455 		/*
2456 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2457 		 * go ahead and use it?
2458 		 */
2459 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2460 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2461 	} else {
2462 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2463 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2464 	}
2465 
2466 	if (cpu_has_vmx_apicv()) {
2467 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2468 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2469 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2470 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2471 	}
2472 
2473 	/*
2474 	 * Make sure the msr_autostore list is up to date before we set the
2475 	 * count in the vmcs02.
2476 	 */
2477 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2478 
2479 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2480 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2481 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2482 
2483 	set_cr4_guest_host_mask(vmx);
2484 }
2485 
2486 /*
2487  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2488  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2489  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2490  * guest in a way that will both be appropriate to L1's requests, and our
2491  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2492  * function also has additional necessary side-effects, like setting various
2493  * vcpu->arch fields.
2494  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2495  * is assigned to entry_failure_code on failure.
2496  */
2497 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2498 			  bool from_vmentry,
2499 			  enum vm_entry_failure_code *entry_failure_code)
2500 {
2501 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2502 	bool load_guest_pdptrs_vmcs12 = false;
2503 
2504 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2505 		prepare_vmcs02_rare(vmx, vmcs12);
2506 		vmx->nested.dirty_vmcs12 = false;
2507 
2508 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2509 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2510 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2511 	}
2512 
2513 	if (vmx->nested.nested_run_pending &&
2514 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2515 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2516 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2517 	} else {
2518 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2519 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2520 	}
2521 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2522 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2523 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2524 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2525 
2526 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2527 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2528 	 * trap. Note that CR0.TS also needs updating - we do this later.
2529 	 */
2530 	vmx_update_exception_bitmap(vcpu);
2531 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2532 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2533 
2534 	if (vmx->nested.nested_run_pending &&
2535 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2536 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2537 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2538 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2539 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2540 	}
2541 
2542 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2543 			vcpu->arch.l1_tsc_offset,
2544 			vmx_get_l2_tsc_offset(vcpu),
2545 			vmx_get_l2_tsc_multiplier(vcpu));
2546 
2547 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2548 			vcpu->arch.l1_tsc_scaling_ratio,
2549 			vmx_get_l2_tsc_multiplier(vcpu));
2550 
2551 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2552 	if (kvm_has_tsc_control)
2553 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2554 
2555 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2556 
2557 	if (nested_cpu_has_ept(vmcs12))
2558 		nested_ept_init_mmu_context(vcpu);
2559 
2560 	/*
2561 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2562 	 * bits which we consider mandatory enabled.
2563 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2564 	 * the specifications by L1; It's not enough to take
2565 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2566 	 * have more bits than L1 expected.
2567 	 */
2568 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2569 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2570 
2571 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2572 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2573 
2574 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2575 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2576 	vmx_set_efer(vcpu, vcpu->arch.efer);
2577 
2578 	/*
2579 	 * Guest state is invalid and unrestricted guest is disabled,
2580 	 * which means L1 attempted VMEntry to L2 with invalid state.
2581 	 * Fail the VMEntry.
2582 	 *
2583 	 * However when force loading the guest state (SMM exit or
2584 	 * loading nested state after migration, it is possible to
2585 	 * have invalid guest state now, which will be later fixed by
2586 	 * restoring L2 register state
2587 	 */
2588 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2589 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2590 		return -EINVAL;
2591 	}
2592 
2593 	/* Shadow page tables on either EPT or shadow page tables. */
2594 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2595 				from_vmentry, entry_failure_code))
2596 		return -EINVAL;
2597 
2598 	/*
2599 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2600 	 * on nested VM-Exit, which can occur without actually running L2 and
2601 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2602 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2603 	 * transition to HLT instead of running L2.
2604 	 */
2605 	if (enable_ept)
2606 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2607 
2608 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2609 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2610 	    is_pae_paging(vcpu)) {
2611 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2612 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2613 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2614 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2615 	}
2616 
2617 	if (!enable_ept)
2618 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2619 
2620 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2621 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2622 				     vmcs12->guest_ia32_perf_global_ctrl))) {
2623 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2624 		return -EINVAL;
2625 	}
2626 
2627 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2628 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2629 
2630 	/*
2631 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2632 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2633 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2634 	 * here.
2635 	 */
2636 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2637 		vmx->nested.hv_evmcs->hv_clean_fields |=
2638 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2639 
2640 	return 0;
2641 }
2642 
2643 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2644 {
2645 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2646 	       nested_cpu_has_virtual_nmis(vmcs12)))
2647 		return -EINVAL;
2648 
2649 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2650 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2651 		return -EINVAL;
2652 
2653 	return 0;
2654 }
2655 
2656 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2657 {
2658 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2659 
2660 	/* Check for memory type validity */
2661 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2662 	case VMX_EPTP_MT_UC:
2663 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2664 			return false;
2665 		break;
2666 	case VMX_EPTP_MT_WB:
2667 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2668 			return false;
2669 		break;
2670 	default:
2671 		return false;
2672 	}
2673 
2674 	/* Page-walk levels validity. */
2675 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2676 	case VMX_EPTP_PWL_5:
2677 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2678 			return false;
2679 		break;
2680 	case VMX_EPTP_PWL_4:
2681 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2682 			return false;
2683 		break;
2684 	default:
2685 		return false;
2686 	}
2687 
2688 	/* Reserved bits should not be set */
2689 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2690 		return false;
2691 
2692 	/* AD, if set, should be supported */
2693 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2694 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2695 			return false;
2696 	}
2697 
2698 	return true;
2699 }
2700 
2701 /*
2702  * Checks related to VM-Execution Control Fields
2703  */
2704 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2705                                               struct vmcs12 *vmcs12)
2706 {
2707 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2708 
2709 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2710 				   vmx->nested.msrs.pinbased_ctls_low,
2711 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2712 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2713 				   vmx->nested.msrs.procbased_ctls_low,
2714 				   vmx->nested.msrs.procbased_ctls_high)))
2715 		return -EINVAL;
2716 
2717 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2718 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2719 				   vmx->nested.msrs.secondary_ctls_low,
2720 				   vmx->nested.msrs.secondary_ctls_high)))
2721 		return -EINVAL;
2722 
2723 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2724 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2725 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2726 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2727 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2728 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2729 	    nested_vmx_check_nmi_controls(vmcs12) ||
2730 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2731 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2732 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2733 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2734 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2735 		return -EINVAL;
2736 
2737 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2738 	    nested_cpu_has_save_preemption_timer(vmcs12))
2739 		return -EINVAL;
2740 
2741 	if (nested_cpu_has_ept(vmcs12) &&
2742 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2743 		return -EINVAL;
2744 
2745 	if (nested_cpu_has_vmfunc(vmcs12)) {
2746 		if (CC(vmcs12->vm_function_control &
2747 		       ~vmx->nested.msrs.vmfunc_controls))
2748 			return -EINVAL;
2749 
2750 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2751 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2752 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2753 				return -EINVAL;
2754 		}
2755 	}
2756 
2757 	return 0;
2758 }
2759 
2760 /*
2761  * Checks related to VM-Exit Control Fields
2762  */
2763 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2764                                          struct vmcs12 *vmcs12)
2765 {
2766 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2767 
2768 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2769 				    vmx->nested.msrs.exit_ctls_low,
2770 				    vmx->nested.msrs.exit_ctls_high)) ||
2771 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2772 		return -EINVAL;
2773 
2774 	return 0;
2775 }
2776 
2777 /*
2778  * Checks related to VM-Entry Control Fields
2779  */
2780 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2781 					  struct vmcs12 *vmcs12)
2782 {
2783 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2784 
2785 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2786 				    vmx->nested.msrs.entry_ctls_low,
2787 				    vmx->nested.msrs.entry_ctls_high)))
2788 		return -EINVAL;
2789 
2790 	/*
2791 	 * From the Intel SDM, volume 3:
2792 	 * Fields relevant to VM-entry event injection must be set properly.
2793 	 * These fields are the VM-entry interruption-information field, the
2794 	 * VM-entry exception error code, and the VM-entry instruction length.
2795 	 */
2796 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2797 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2798 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2799 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2800 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2801 		bool should_have_error_code;
2802 		bool urg = nested_cpu_has2(vmcs12,
2803 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2804 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2805 
2806 		/* VM-entry interruption-info field: interruption type */
2807 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2808 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2809 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2810 			return -EINVAL;
2811 
2812 		/* VM-entry interruption-info field: vector */
2813 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2814 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2815 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2816 			return -EINVAL;
2817 
2818 		/* VM-entry interruption-info field: deliver error code */
2819 		should_have_error_code =
2820 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2821 			x86_exception_has_error_code(vector);
2822 		if (CC(has_error_code != should_have_error_code))
2823 			return -EINVAL;
2824 
2825 		/* VM-entry exception error code */
2826 		if (CC(has_error_code &&
2827 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2828 			return -EINVAL;
2829 
2830 		/* VM-entry interruption-info field: reserved bits */
2831 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2832 			return -EINVAL;
2833 
2834 		/* VM-entry instruction length */
2835 		switch (intr_type) {
2836 		case INTR_TYPE_SOFT_EXCEPTION:
2837 		case INTR_TYPE_SOFT_INTR:
2838 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2839 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2840 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2841 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2842 				return -EINVAL;
2843 		}
2844 	}
2845 
2846 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2847 		return -EINVAL;
2848 
2849 	return 0;
2850 }
2851 
2852 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2853 				     struct vmcs12 *vmcs12)
2854 {
2855 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2856 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2857 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2858 		return -EINVAL;
2859 
2860 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2861 		return nested_evmcs_check_controls(vmcs12);
2862 
2863 	return 0;
2864 }
2865 
2866 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
2867 				       struct vmcs12 *vmcs12)
2868 {
2869 #ifdef CONFIG_X86_64
2870 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
2871 		!!(vcpu->arch.efer & EFER_LMA)))
2872 		return -EINVAL;
2873 #endif
2874 	return 0;
2875 }
2876 
2877 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2878 				       struct vmcs12 *vmcs12)
2879 {
2880 	bool ia32e;
2881 
2882 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2883 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2884 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2885 		return -EINVAL;
2886 
2887 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2888 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2889 		return -EINVAL;
2890 
2891 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2892 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2893 		return -EINVAL;
2894 
2895 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2896 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2897 					   vmcs12->host_ia32_perf_global_ctrl)))
2898 		return -EINVAL;
2899 
2900 #ifdef CONFIG_X86_64
2901 	ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
2902 #else
2903 	ia32e = false;
2904 #endif
2905 
2906 	if (ia32e) {
2907 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2908 			return -EINVAL;
2909 	} else {
2910 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2911 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2912 		    CC((vmcs12->host_rip) >> 32))
2913 			return -EINVAL;
2914 	}
2915 
2916 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2917 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2918 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2919 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2920 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2921 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2922 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2923 	    CC(vmcs12->host_cs_selector == 0) ||
2924 	    CC(vmcs12->host_tr_selector == 0) ||
2925 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2926 		return -EINVAL;
2927 
2928 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2929 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2930 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2931 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2932 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2933 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2934 		return -EINVAL;
2935 
2936 	/*
2937 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2938 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2939 	 * the values of the LMA and LME bits in the field must each be that of
2940 	 * the host address-space size VM-exit control.
2941 	 */
2942 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2943 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2944 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2945 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2946 			return -EINVAL;
2947 	}
2948 
2949 	return 0;
2950 }
2951 
2952 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2953 					  struct vmcs12 *vmcs12)
2954 {
2955 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2956 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
2957 	struct vmcs_hdr hdr;
2958 
2959 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
2960 		return 0;
2961 
2962 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2963 		return -EINVAL;
2964 
2965 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
2966 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
2967 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
2968                 return -EINVAL;
2969 
2970 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
2971 					    offsetof(struct vmcs12, hdr),
2972 					    sizeof(hdr))))
2973 		return -EINVAL;
2974 
2975 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
2976 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2977 		return -EINVAL;
2978 
2979 	return 0;
2980 }
2981 
2982 /*
2983  * Checks related to Guest Non-register State
2984  */
2985 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2986 {
2987 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2988 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2989 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2990 		return -EINVAL;
2991 
2992 	return 0;
2993 }
2994 
2995 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2996 					struct vmcs12 *vmcs12,
2997 					enum vm_entry_failure_code *entry_failure_code)
2998 {
2999 	bool ia32e;
3000 
3001 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
3002 
3003 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3004 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3005 		return -EINVAL;
3006 
3007 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
3008 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
3009 		return -EINVAL;
3010 
3011 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3012 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3013 		return -EINVAL;
3014 
3015 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3016 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3017 		return -EINVAL;
3018 	}
3019 
3020 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3021 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3022 					   vmcs12->guest_ia32_perf_global_ctrl)))
3023 		return -EINVAL;
3024 
3025 	/*
3026 	 * If the load IA32_EFER VM-entry control is 1, the following checks
3027 	 * are performed on the field for the IA32_EFER MSR:
3028 	 * - Bits reserved in the IA32_EFER MSR must be 0.
3029 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3030 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3031 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3032 	 *   CR0.PG) is 1.
3033 	 */
3034 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3035 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3036 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
3037 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3038 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3039 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3040 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3041 			return -EINVAL;
3042 	}
3043 
3044 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3045 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3046 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3047 		return -EINVAL;
3048 
3049 	if (nested_check_guest_non_reg_state(vmcs12))
3050 		return -EINVAL;
3051 
3052 	return 0;
3053 }
3054 
3055 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3056 {
3057 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3058 	unsigned long cr4;
3059 	bool vm_fail;
3060 
3061 	if (!nested_early_check)
3062 		return 0;
3063 
3064 	if (vmx->msr_autoload.host.nr)
3065 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3066 	if (vmx->msr_autoload.guest.nr)
3067 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3068 
3069 	preempt_disable();
3070 
3071 	vmx_prepare_switch_to_guest(vcpu);
3072 
3073 	/*
3074 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3075 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3076 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3077 	 * there is no need to preserve other bits or save/restore the field.
3078 	 */
3079 	vmcs_writel(GUEST_RFLAGS, 0);
3080 
3081 	cr4 = cr4_read_shadow();
3082 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3083 		vmcs_writel(HOST_CR4, cr4);
3084 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3085 	}
3086 
3087 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3088 				 vmx->loaded_vmcs->launched);
3089 
3090 	if (vmx->msr_autoload.host.nr)
3091 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3092 	if (vmx->msr_autoload.guest.nr)
3093 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3094 
3095 	if (vm_fail) {
3096 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3097 
3098 		preempt_enable();
3099 
3100 		trace_kvm_nested_vmenter_failed(
3101 			"early hardware check VM-instruction error: ", error);
3102 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3103 		return 1;
3104 	}
3105 
3106 	/*
3107 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3108 	 */
3109 	if (hw_breakpoint_active())
3110 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3111 	local_irq_enable();
3112 	preempt_enable();
3113 
3114 	/*
3115 	 * A non-failing VMEntry means we somehow entered guest mode with
3116 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3117 	 * is no telling what memory has been modified or what state has
3118 	 * been exposed to unknown code.  Hitting this all but guarantees
3119 	 * a (very critical) hardware issue.
3120 	 */
3121 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3122 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3123 
3124 	return 0;
3125 }
3126 
3127 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3128 {
3129 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3130 
3131 	/*
3132 	 * hv_evmcs may end up being not mapped after migration (when
3133 	 * L2 was running), map it here to make sure vmcs12 changes are
3134 	 * properly reflected.
3135 	 */
3136 	if (vmx->nested.enlightened_vmcs_enabled &&
3137 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3138 		enum nested_evmptrld_status evmptrld_status =
3139 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3140 
3141 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3142 		    evmptrld_status == EVMPTRLD_ERROR)
3143 			return false;
3144 
3145 		/*
3146 		 * Post migration VMCS12 always provides the most actual
3147 		 * information, copy it to eVMCS upon entry.
3148 		 */
3149 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3150 	}
3151 
3152 	return true;
3153 }
3154 
3155 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3156 {
3157 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3158 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3159 	struct kvm_host_map *map;
3160 	struct page *page;
3161 	u64 hpa;
3162 
3163 	if (!vcpu->arch.pdptrs_from_userspace &&
3164 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3165 		/*
3166 		 * Reload the guest's PDPTRs since after a migration
3167 		 * the guest CR3 might be restored prior to setting the nested
3168 		 * state which can lead to a load of wrong PDPTRs.
3169 		 */
3170 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3171 			return false;
3172 	}
3173 
3174 
3175 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3176 		/*
3177 		 * Translate L1 physical address to host physical
3178 		 * address for vmcs02. Keep the page pinned, so this
3179 		 * physical address remains valid. We keep a reference
3180 		 * to it so we can release it later.
3181 		 */
3182 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3183 			kvm_release_page_clean(vmx->nested.apic_access_page);
3184 			vmx->nested.apic_access_page = NULL;
3185 		}
3186 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3187 		if (!is_error_page(page)) {
3188 			vmx->nested.apic_access_page = page;
3189 			hpa = page_to_phys(vmx->nested.apic_access_page);
3190 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3191 		} else {
3192 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3193 					     __func__);
3194 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3195 			vcpu->run->internal.suberror =
3196 				KVM_INTERNAL_ERROR_EMULATION;
3197 			vcpu->run->internal.ndata = 0;
3198 			return false;
3199 		}
3200 	}
3201 
3202 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3203 		map = &vmx->nested.virtual_apic_map;
3204 
3205 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3206 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3207 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3208 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3209 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3210 			/*
3211 			 * The processor will never use the TPR shadow, simply
3212 			 * clear the bit from the execution control.  Such a
3213 			 * configuration is useless, but it happens in tests.
3214 			 * For any other configuration, failing the vm entry is
3215 			 * _not_ what the processor does but it's basically the
3216 			 * only possibility we have.
3217 			 */
3218 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3219 		} else {
3220 			/*
3221 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3222 			 * force VM-Entry to fail.
3223 			 */
3224 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3225 		}
3226 	}
3227 
3228 	if (nested_cpu_has_posted_intr(vmcs12)) {
3229 		map = &vmx->nested.pi_desc_map;
3230 
3231 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3232 			vmx->nested.pi_desc =
3233 				(struct pi_desc *)(((void *)map->hva) +
3234 				offset_in_page(vmcs12->posted_intr_desc_addr));
3235 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3236 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3237 		} else {
3238 			/*
3239 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3240 			 * access the contents of the VMCS12 posted interrupt
3241 			 * descriptor. (Note that KVM may do this when it
3242 			 * should not, per the architectural specification.)
3243 			 */
3244 			vmx->nested.pi_desc = NULL;
3245 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3246 		}
3247 	}
3248 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3249 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3250 	else
3251 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3252 
3253 	return true;
3254 }
3255 
3256 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3257 {
3258 	if (!nested_get_evmcs_page(vcpu)) {
3259 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3260 				     __func__);
3261 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3262 		vcpu->run->internal.suberror =
3263 			KVM_INTERNAL_ERROR_EMULATION;
3264 		vcpu->run->internal.ndata = 0;
3265 
3266 		return false;
3267 	}
3268 
3269 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3270 		return false;
3271 
3272 	return true;
3273 }
3274 
3275 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3276 {
3277 	struct vmcs12 *vmcs12;
3278 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3279 	gpa_t dst;
3280 
3281 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3282 		return 0;
3283 
3284 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3285 		return 1;
3286 
3287 	/*
3288 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3289 	 * set is already checked as part of A/D emulation.
3290 	 */
3291 	vmcs12 = get_vmcs12(vcpu);
3292 	if (!nested_cpu_has_pml(vmcs12))
3293 		return 0;
3294 
3295 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3296 		vmx->nested.pml_full = true;
3297 		return 1;
3298 	}
3299 
3300 	gpa &= ~0xFFFull;
3301 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3302 
3303 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3304 				 offset_in_page(dst), sizeof(gpa)))
3305 		return 0;
3306 
3307 	vmcs12->guest_pml_index--;
3308 
3309 	return 0;
3310 }
3311 
3312 /*
3313  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3314  * for running VMX instructions (except VMXON, whose prerequisites are
3315  * slightly different). It also specifies what exception to inject otherwise.
3316  * Note that many of these exceptions have priority over VM exits, so they
3317  * don't have to be checked again here.
3318  */
3319 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3320 {
3321 	if (!to_vmx(vcpu)->nested.vmxon) {
3322 		kvm_queue_exception(vcpu, UD_VECTOR);
3323 		return 0;
3324 	}
3325 
3326 	if (vmx_get_cpl(vcpu)) {
3327 		kvm_inject_gp(vcpu, 0);
3328 		return 0;
3329 	}
3330 
3331 	return 1;
3332 }
3333 
3334 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3335 {
3336 	u8 rvi = vmx_get_rvi();
3337 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3338 
3339 	return ((rvi & 0xf0) > (vppr & 0xf0));
3340 }
3341 
3342 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3343 				   struct vmcs12 *vmcs12);
3344 
3345 /*
3346  * If from_vmentry is false, this is being called from state restore (either RSM
3347  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3348  *
3349  * Returns:
3350  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3351  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3352  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3353  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3354  */
3355 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3356 							bool from_vmentry)
3357 {
3358 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3359 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3360 	enum vm_entry_failure_code entry_failure_code;
3361 	bool evaluate_pending_interrupts;
3362 	union vmx_exit_reason exit_reason = {
3363 		.basic = EXIT_REASON_INVALID_STATE,
3364 		.failed_vmentry = 1,
3365 	};
3366 	u32 failed_index;
3367 
3368 	kvm_service_local_tlb_flush_requests(vcpu);
3369 
3370 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3371 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3372 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3373 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3374 
3375 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3376 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3377 	if (kvm_mpx_supported() &&
3378 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3379 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3380 
3381 	/*
3382 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3383 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3384 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3385 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3386 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3387 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3388 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3389 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3390 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3391 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3392 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3393 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3394 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3395 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3396 	 */
3397 	if (!enable_ept && !nested_early_check)
3398 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3399 
3400 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3401 
3402 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3403 
3404 	if (from_vmentry) {
3405 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3406 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3407 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3408 		}
3409 
3410 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3411 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3412 			return NVMX_VMENTRY_VMFAIL;
3413 		}
3414 
3415 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3416 						 &entry_failure_code)) {
3417 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3418 			vmcs12->exit_qualification = entry_failure_code;
3419 			goto vmentry_fail_vmexit;
3420 		}
3421 	}
3422 
3423 	enter_guest_mode(vcpu);
3424 
3425 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3426 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3427 		vmcs12->exit_qualification = entry_failure_code;
3428 		goto vmentry_fail_vmexit_guest_mode;
3429 	}
3430 
3431 	if (from_vmentry) {
3432 		failed_index = nested_vmx_load_msr(vcpu,
3433 						   vmcs12->vm_entry_msr_load_addr,
3434 						   vmcs12->vm_entry_msr_load_count);
3435 		if (failed_index) {
3436 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3437 			vmcs12->exit_qualification = failed_index;
3438 			goto vmentry_fail_vmexit_guest_mode;
3439 		}
3440 	} else {
3441 		/*
3442 		 * The MMU is not initialized to point at the right entities yet and
3443 		 * "get pages" would need to read data from the guest (i.e. we will
3444 		 * need to perform gpa to hpa translation). Request a call
3445 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3446 		 * have already been set at vmentry time and should not be reset.
3447 		 */
3448 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3449 	}
3450 
3451 	/*
3452 	 * If L1 had a pending IRQ/NMI until it executed
3453 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3454 	 * disallowed (e.g. interrupts disabled), L0 needs to
3455 	 * evaluate if this pending event should cause an exit from L2
3456 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3457 	 * intercept EXTERNAL_INTERRUPT).
3458 	 *
3459 	 * Usually this would be handled by the processor noticing an
3460 	 * IRQ/NMI window request, or checking RVI during evaluation of
3461 	 * pending virtual interrupts.  However, this setting was done
3462 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3463 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3464 	 */
3465 	if (unlikely(evaluate_pending_interrupts))
3466 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3467 
3468 	/*
3469 	 * Do not start the preemption timer hrtimer until after we know
3470 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3471 	 * the timer.
3472 	 */
3473 	vmx->nested.preemption_timer_expired = false;
3474 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3475 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3476 		vmx_start_preemption_timer(vcpu, timer_value);
3477 	}
3478 
3479 	/*
3480 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3481 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3482 	 * returned as far as L1 is concerned. It will only return (and set
3483 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3484 	 */
3485 	return NVMX_VMENTRY_SUCCESS;
3486 
3487 	/*
3488 	 * A failed consistency check that leads to a VMExit during L1's
3489 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3490 	 * 26.7 "VM-entry failures during or after loading guest state".
3491 	 */
3492 vmentry_fail_vmexit_guest_mode:
3493 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3494 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3495 	leave_guest_mode(vcpu);
3496 
3497 vmentry_fail_vmexit:
3498 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3499 
3500 	if (!from_vmentry)
3501 		return NVMX_VMENTRY_VMEXIT;
3502 
3503 	load_vmcs12_host_state(vcpu, vmcs12);
3504 	vmcs12->vm_exit_reason = exit_reason.full;
3505 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3506 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3507 	return NVMX_VMENTRY_VMEXIT;
3508 }
3509 
3510 /*
3511  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3512  * for running an L2 nested guest.
3513  */
3514 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3515 {
3516 	struct vmcs12 *vmcs12;
3517 	enum nvmx_vmentry_status status;
3518 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3519 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3520 	enum nested_evmptrld_status evmptrld_status;
3521 
3522 	if (!nested_vmx_check_permission(vcpu))
3523 		return 1;
3524 
3525 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3526 	if (evmptrld_status == EVMPTRLD_ERROR) {
3527 		kvm_queue_exception(vcpu, UD_VECTOR);
3528 		return 1;
3529 	}
3530 
3531 	kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
3532 
3533 	if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3534 		return nested_vmx_failInvalid(vcpu);
3535 
3536 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3537 	       vmx->nested.current_vmptr == INVALID_GPA))
3538 		return nested_vmx_failInvalid(vcpu);
3539 
3540 	vmcs12 = get_vmcs12(vcpu);
3541 
3542 	/*
3543 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3544 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3545 	 * rather than RFLAGS.ZF, and no error number is stored to the
3546 	 * VM-instruction error field.
3547 	 */
3548 	if (CC(vmcs12->hdr.shadow_vmcs))
3549 		return nested_vmx_failInvalid(vcpu);
3550 
3551 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3552 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3553 		/* Enlightened VMCS doesn't have launch state */
3554 		vmcs12->launch_state = !launch;
3555 	} else if (enable_shadow_vmcs) {
3556 		copy_shadow_to_vmcs12(vmx);
3557 	}
3558 
3559 	/*
3560 	 * The nested entry process starts with enforcing various prerequisites
3561 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3562 	 * they fail: As the SDM explains, some conditions should cause the
3563 	 * instruction to fail, while others will cause the instruction to seem
3564 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3565 	 * To speed up the normal (success) code path, we should avoid checking
3566 	 * for misconfigurations which will anyway be caught by the processor
3567 	 * when using the merged vmcs02.
3568 	 */
3569 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3570 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3571 
3572 	if (CC(vmcs12->launch_state == launch))
3573 		return nested_vmx_fail(vcpu,
3574 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3575 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3576 
3577 	if (nested_vmx_check_controls(vcpu, vmcs12))
3578 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3579 
3580 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3581 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3582 
3583 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3584 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3585 
3586 	/*
3587 	 * We're finally done with prerequisite checking, and can start with
3588 	 * the nested entry.
3589 	 */
3590 	vmx->nested.nested_run_pending = 1;
3591 	vmx->nested.has_preemption_timer_deadline = false;
3592 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3593 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3594 		goto vmentry_failed;
3595 
3596 	/* Emulate processing of posted interrupts on VM-Enter. */
3597 	if (nested_cpu_has_posted_intr(vmcs12) &&
3598 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3599 		vmx->nested.pi_pending = true;
3600 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3601 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3602 	}
3603 
3604 	/* Hide L1D cache contents from the nested guest.  */
3605 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3606 
3607 	/*
3608 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3609 	 * also be used as part of restoring nVMX state for
3610 	 * snapshot restore (migration).
3611 	 *
3612 	 * In this flow, it is assumed that vmcs12 cache was
3613 	 * transferred as part of captured nVMX state and should
3614 	 * therefore not be read from guest memory (which may not
3615 	 * exist on destination host yet).
3616 	 */
3617 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3618 
3619 	switch (vmcs12->guest_activity_state) {
3620 	case GUEST_ACTIVITY_HLT:
3621 		/*
3622 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3623 		 * awakened by event injection or by an NMI-window VM-exit or
3624 		 * by an interrupt-window VM-exit, halt the vcpu.
3625 		 */
3626 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3627 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3628 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3629 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3630 			vmx->nested.nested_run_pending = 0;
3631 			return kvm_emulate_halt_noskip(vcpu);
3632 		}
3633 		break;
3634 	case GUEST_ACTIVITY_WAIT_SIPI:
3635 		vmx->nested.nested_run_pending = 0;
3636 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3637 		break;
3638 	default:
3639 		break;
3640 	}
3641 
3642 	return 1;
3643 
3644 vmentry_failed:
3645 	vmx->nested.nested_run_pending = 0;
3646 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3647 		return 0;
3648 	if (status == NVMX_VMENTRY_VMEXIT)
3649 		return 1;
3650 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3651 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3652 }
3653 
3654 /*
3655  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3656  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3657  * This function returns the new value we should put in vmcs12.guest_cr0.
3658  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3659  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3660  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3661  *     didn't trap the bit, because if L1 did, so would L0).
3662  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3663  *     been modified by L2, and L1 knows it. So just leave the old value of
3664  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3665  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3666  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3667  *     changed these bits, and therefore they need to be updated, but L0
3668  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3669  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3670  */
3671 static inline unsigned long
3672 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3673 {
3674 	return
3675 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3676 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3677 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3678 			vcpu->arch.cr0_guest_owned_bits));
3679 }
3680 
3681 static inline unsigned long
3682 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3683 {
3684 	return
3685 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3686 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3687 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3688 			vcpu->arch.cr4_guest_owned_bits));
3689 }
3690 
3691 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3692 				      struct vmcs12 *vmcs12)
3693 {
3694 	u32 idt_vectoring;
3695 	unsigned int nr;
3696 
3697 	if (vcpu->arch.exception.injected) {
3698 		nr = vcpu->arch.exception.nr;
3699 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3700 
3701 		if (kvm_exception_is_soft(nr)) {
3702 			vmcs12->vm_exit_instruction_len =
3703 				vcpu->arch.event_exit_inst_len;
3704 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3705 		} else
3706 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3707 
3708 		if (vcpu->arch.exception.has_error_code) {
3709 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3710 			vmcs12->idt_vectoring_error_code =
3711 				vcpu->arch.exception.error_code;
3712 		}
3713 
3714 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3715 	} else if (vcpu->arch.nmi_injected) {
3716 		vmcs12->idt_vectoring_info_field =
3717 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3718 	} else if (vcpu->arch.interrupt.injected) {
3719 		nr = vcpu->arch.interrupt.nr;
3720 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3721 
3722 		if (vcpu->arch.interrupt.soft) {
3723 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3724 			vmcs12->vm_entry_instruction_len =
3725 				vcpu->arch.event_exit_inst_len;
3726 		} else
3727 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3728 
3729 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3730 	}
3731 }
3732 
3733 
3734 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3735 {
3736 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3737 	gfn_t gfn;
3738 
3739 	/*
3740 	 * Don't need to mark the APIC access page dirty; it is never
3741 	 * written to by the CPU during APIC virtualization.
3742 	 */
3743 
3744 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3745 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3746 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3747 	}
3748 
3749 	if (nested_cpu_has_posted_intr(vmcs12)) {
3750 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3751 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3752 	}
3753 }
3754 
3755 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3756 {
3757 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3758 	int max_irr;
3759 	void *vapic_page;
3760 	u16 status;
3761 
3762 	if (!vmx->nested.pi_pending)
3763 		return 0;
3764 
3765 	if (!vmx->nested.pi_desc)
3766 		goto mmio_needed;
3767 
3768 	vmx->nested.pi_pending = false;
3769 
3770 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3771 		return 0;
3772 
3773 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3774 	if (max_irr != 256) {
3775 		vapic_page = vmx->nested.virtual_apic_map.hva;
3776 		if (!vapic_page)
3777 			goto mmio_needed;
3778 
3779 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3780 			vapic_page, &max_irr);
3781 		status = vmcs_read16(GUEST_INTR_STATUS);
3782 		if ((u8)max_irr > ((u8)status & 0xff)) {
3783 			status &= ~0xff;
3784 			status |= (u8)max_irr;
3785 			vmcs_write16(GUEST_INTR_STATUS, status);
3786 		}
3787 	}
3788 
3789 	nested_mark_vmcs12_pages_dirty(vcpu);
3790 	return 0;
3791 
3792 mmio_needed:
3793 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3794 	return -ENXIO;
3795 }
3796 
3797 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3798 					       unsigned long exit_qual)
3799 {
3800 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3801 	unsigned int nr = vcpu->arch.exception.nr;
3802 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3803 
3804 	if (vcpu->arch.exception.has_error_code) {
3805 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3806 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3807 	}
3808 
3809 	if (kvm_exception_is_soft(nr))
3810 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3811 	else
3812 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3813 
3814 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3815 	    vmx_get_nmi_mask(vcpu))
3816 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3817 
3818 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3819 }
3820 
3821 /*
3822  * Returns true if a debug trap is pending delivery.
3823  *
3824  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3825  * exception may be inferred from the presence of an exception payload.
3826  */
3827 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3828 {
3829 	return vcpu->arch.exception.pending &&
3830 			vcpu->arch.exception.nr == DB_VECTOR &&
3831 			vcpu->arch.exception.payload;
3832 }
3833 
3834 /*
3835  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3836  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3837  * represents these debug traps with a payload that is said to be compatible
3838  * with the 'pending debug exceptions' field, write the payload to the VMCS
3839  * field if a VM-exit is delivered before the debug trap.
3840  */
3841 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3842 {
3843 	if (vmx_pending_dbg_trap(vcpu))
3844 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3845 			    vcpu->arch.exception.payload);
3846 }
3847 
3848 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3849 {
3850 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3851 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3852 }
3853 
3854 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3855 {
3856 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3857 	unsigned long exit_qual;
3858 	bool block_nested_events =
3859 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3860 	bool mtf_pending = vmx->nested.mtf_pending;
3861 	struct kvm_lapic *apic = vcpu->arch.apic;
3862 
3863 	/*
3864 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3865 	 * this state is discarded.
3866 	 */
3867 	if (!block_nested_events)
3868 		vmx->nested.mtf_pending = false;
3869 
3870 	if (lapic_in_kernel(vcpu) &&
3871 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3872 		if (block_nested_events)
3873 			return -EBUSY;
3874 		nested_vmx_update_pending_dbg(vcpu);
3875 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3876 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3877 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3878 		return 0;
3879 	}
3880 
3881 	if (lapic_in_kernel(vcpu) &&
3882 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3883 		if (block_nested_events)
3884 			return -EBUSY;
3885 
3886 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3887 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3888 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3889 						apic->sipi_vector & 0xFFUL);
3890 		return 0;
3891 	}
3892 
3893 	/*
3894 	 * Process any exceptions that are not debug traps before MTF.
3895 	 *
3896 	 * Note that only a pending nested run can block a pending exception.
3897 	 * Otherwise an injected NMI/interrupt should either be
3898 	 * lost or delivered to the nested hypervisor in the IDT_VECTORING_INFO,
3899 	 * while delivering the pending exception.
3900 	 */
3901 
3902 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3903 		if (vmx->nested.nested_run_pending)
3904 			return -EBUSY;
3905 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3906 			goto no_vmexit;
3907 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3908 		return 0;
3909 	}
3910 
3911 	if (mtf_pending) {
3912 		if (block_nested_events)
3913 			return -EBUSY;
3914 		nested_vmx_update_pending_dbg(vcpu);
3915 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3916 		return 0;
3917 	}
3918 
3919 	if (vcpu->arch.exception.pending) {
3920 		if (vmx->nested.nested_run_pending)
3921 			return -EBUSY;
3922 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3923 			goto no_vmexit;
3924 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3925 		return 0;
3926 	}
3927 
3928 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3929 		if (block_nested_events)
3930 			return -EBUSY;
3931 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3932 		return 0;
3933 	}
3934 
3935 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3936 		if (block_nested_events)
3937 			return -EBUSY;
3938 		goto no_vmexit;
3939 	}
3940 
3941 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3942 		if (block_nested_events)
3943 			return -EBUSY;
3944 		if (!nested_exit_on_nmi(vcpu))
3945 			goto no_vmexit;
3946 
3947 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3948 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3949 				  INTR_INFO_VALID_MASK, 0);
3950 		/*
3951 		 * The NMI-triggered VM exit counts as injection:
3952 		 * clear this one and block further NMIs.
3953 		 */
3954 		vcpu->arch.nmi_pending = 0;
3955 		vmx_set_nmi_mask(vcpu, true);
3956 		return 0;
3957 	}
3958 
3959 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3960 		if (block_nested_events)
3961 			return -EBUSY;
3962 		if (!nested_exit_on_intr(vcpu))
3963 			goto no_vmexit;
3964 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3965 		return 0;
3966 	}
3967 
3968 no_vmexit:
3969 	return vmx_complete_nested_posted_interrupt(vcpu);
3970 }
3971 
3972 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3973 {
3974 	ktime_t remaining =
3975 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3976 	u64 value;
3977 
3978 	if (ktime_to_ns(remaining) <= 0)
3979 		return 0;
3980 
3981 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3982 	do_div(value, 1000000);
3983 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3984 }
3985 
3986 static bool is_vmcs12_ext_field(unsigned long field)
3987 {
3988 	switch (field) {
3989 	case GUEST_ES_SELECTOR:
3990 	case GUEST_CS_SELECTOR:
3991 	case GUEST_SS_SELECTOR:
3992 	case GUEST_DS_SELECTOR:
3993 	case GUEST_FS_SELECTOR:
3994 	case GUEST_GS_SELECTOR:
3995 	case GUEST_LDTR_SELECTOR:
3996 	case GUEST_TR_SELECTOR:
3997 	case GUEST_ES_LIMIT:
3998 	case GUEST_CS_LIMIT:
3999 	case GUEST_SS_LIMIT:
4000 	case GUEST_DS_LIMIT:
4001 	case GUEST_FS_LIMIT:
4002 	case GUEST_GS_LIMIT:
4003 	case GUEST_LDTR_LIMIT:
4004 	case GUEST_TR_LIMIT:
4005 	case GUEST_GDTR_LIMIT:
4006 	case GUEST_IDTR_LIMIT:
4007 	case GUEST_ES_AR_BYTES:
4008 	case GUEST_DS_AR_BYTES:
4009 	case GUEST_FS_AR_BYTES:
4010 	case GUEST_GS_AR_BYTES:
4011 	case GUEST_LDTR_AR_BYTES:
4012 	case GUEST_TR_AR_BYTES:
4013 	case GUEST_ES_BASE:
4014 	case GUEST_CS_BASE:
4015 	case GUEST_SS_BASE:
4016 	case GUEST_DS_BASE:
4017 	case GUEST_FS_BASE:
4018 	case GUEST_GS_BASE:
4019 	case GUEST_LDTR_BASE:
4020 	case GUEST_TR_BASE:
4021 	case GUEST_GDTR_BASE:
4022 	case GUEST_IDTR_BASE:
4023 	case GUEST_PENDING_DBG_EXCEPTIONS:
4024 	case GUEST_BNDCFGS:
4025 		return true;
4026 	default:
4027 		break;
4028 	}
4029 
4030 	return false;
4031 }
4032 
4033 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4034 				       struct vmcs12 *vmcs12)
4035 {
4036 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4037 
4038 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4039 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4040 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4041 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4042 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4043 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4044 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4045 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4046 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4047 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4048 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4049 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4050 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4051 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4052 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4053 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4054 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4055 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4056 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4057 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4058 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4059 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4060 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4061 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4062 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4063 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4064 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4065 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4066 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4067 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4068 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4069 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4070 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4071 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4072 	vmcs12->guest_pending_dbg_exceptions =
4073 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4074 	if (kvm_mpx_supported())
4075 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
4076 
4077 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4078 }
4079 
4080 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4081 				       struct vmcs12 *vmcs12)
4082 {
4083 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4084 	int cpu;
4085 
4086 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4087 		return;
4088 
4089 
4090 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4091 
4092 	cpu = get_cpu();
4093 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4094 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4095 
4096 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4097 
4098 	vmx->loaded_vmcs = &vmx->vmcs01;
4099 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4100 	put_cpu();
4101 }
4102 
4103 /*
4104  * Update the guest state fields of vmcs12 to reflect changes that
4105  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4106  * VM-entry controls is also updated, since this is really a guest
4107  * state bit.)
4108  */
4109 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4110 {
4111 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4112 
4113 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4114 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4115 
4116 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4117 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4118 
4119 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4120 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4121 
4122 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4123 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4124 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4125 
4126 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4127 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4128 
4129 	vmcs12->guest_interruptibility_info =
4130 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4131 
4132 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4133 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4134 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4135 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4136 	else
4137 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4138 
4139 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4140 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4141 	    !vmx->nested.nested_run_pending)
4142 		vmcs12->vmx_preemption_timer_value =
4143 			vmx_get_preemption_timer_value(vcpu);
4144 
4145 	/*
4146 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4147 	 * own CR3 without exiting. If it has changed it, we must keep it.
4148 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4149 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4150 	 *
4151 	 * Additionally, restore L2's PDPTR to vmcs12.
4152 	 */
4153 	if (enable_ept) {
4154 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4155 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4156 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4157 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4158 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4159 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4160 		}
4161 	}
4162 
4163 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4164 
4165 	if (nested_cpu_has_vid(vmcs12))
4166 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4167 
4168 	vmcs12->vm_entry_controls =
4169 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4170 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4171 
4172 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4173 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4174 
4175 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4176 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4177 }
4178 
4179 /*
4180  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4181  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4182  * and this function updates it to reflect the changes to the guest state while
4183  * L2 was running (and perhaps made some exits which were handled directly by L0
4184  * without going back to L1), and to reflect the exit reason.
4185  * Note that we do not have to copy here all VMCS fields, just those that
4186  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4187  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4188  * which already writes to vmcs12 directly.
4189  */
4190 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4191 			   u32 vm_exit_reason, u32 exit_intr_info,
4192 			   unsigned long exit_qualification)
4193 {
4194 	/* update exit information fields: */
4195 	vmcs12->vm_exit_reason = vm_exit_reason;
4196 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4197 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4198 	vmcs12->exit_qualification = exit_qualification;
4199 	vmcs12->vm_exit_intr_info = exit_intr_info;
4200 
4201 	vmcs12->idt_vectoring_info_field = 0;
4202 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4203 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4204 
4205 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4206 		vmcs12->launch_state = 1;
4207 
4208 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4209 		 * instead of reading the real value. */
4210 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4211 
4212 		/*
4213 		 * Transfer the event that L0 or L1 may wanted to inject into
4214 		 * L2 to IDT_VECTORING_INFO_FIELD.
4215 		 */
4216 		vmcs12_save_pending_event(vcpu, vmcs12);
4217 
4218 		/*
4219 		 * According to spec, there's no need to store the guest's
4220 		 * MSRs if the exit is due to a VM-entry failure that occurs
4221 		 * during or after loading the guest state. Since this exit
4222 		 * does not fall in that category, we need to save the MSRs.
4223 		 */
4224 		if (nested_vmx_store_msr(vcpu,
4225 					 vmcs12->vm_exit_msr_store_addr,
4226 					 vmcs12->vm_exit_msr_store_count))
4227 			nested_vmx_abort(vcpu,
4228 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4229 	}
4230 
4231 	/*
4232 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4233 	 * preserved above and would only end up incorrectly in L1.
4234 	 */
4235 	vcpu->arch.nmi_injected = false;
4236 	kvm_clear_exception_queue(vcpu);
4237 	kvm_clear_interrupt_queue(vcpu);
4238 }
4239 
4240 /*
4241  * A part of what we need to when the nested L2 guest exits and we want to
4242  * run its L1 parent, is to reset L1's guest state to the host state specified
4243  * in vmcs12.
4244  * This function is to be called not only on normal nested exit, but also on
4245  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4246  * Failures During or After Loading Guest State").
4247  * This function should be called when the active VMCS is L1's (vmcs01).
4248  */
4249 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4250 				   struct vmcs12 *vmcs12)
4251 {
4252 	enum vm_entry_failure_code ignored;
4253 	struct kvm_segment seg;
4254 
4255 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4256 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4257 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4258 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4259 	else
4260 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4261 	vmx_set_efer(vcpu, vcpu->arch.efer);
4262 
4263 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4264 	kvm_rip_write(vcpu, vmcs12->host_rip);
4265 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4266 	vmx_set_interrupt_shadow(vcpu, 0);
4267 
4268 	/*
4269 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4270 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4271 	 *
4272 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4273 	 * (KVM doesn't change it);
4274 	 */
4275 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4276 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4277 
4278 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4279 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4280 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4281 
4282 	nested_ept_uninit_mmu_context(vcpu);
4283 
4284 	/*
4285 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4286 	 * couldn't have changed.
4287 	 */
4288 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4289 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4290 
4291 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4292 
4293 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4294 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4295 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4296 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4297 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4298 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4299 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4300 
4301 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4302 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4303 		vmcs_write64(GUEST_BNDCFGS, 0);
4304 
4305 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4306 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4307 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4308 	}
4309 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4310 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4311 					 vmcs12->host_ia32_perf_global_ctrl));
4312 
4313 	/* Set L1 segment info according to Intel SDM
4314 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4315 	seg = (struct kvm_segment) {
4316 		.base = 0,
4317 		.limit = 0xFFFFFFFF,
4318 		.selector = vmcs12->host_cs_selector,
4319 		.type = 11,
4320 		.present = 1,
4321 		.s = 1,
4322 		.g = 1
4323 	};
4324 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4325 		seg.l = 1;
4326 	else
4327 		seg.db = 1;
4328 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4329 	seg = (struct kvm_segment) {
4330 		.base = 0,
4331 		.limit = 0xFFFFFFFF,
4332 		.type = 3,
4333 		.present = 1,
4334 		.s = 1,
4335 		.db = 1,
4336 		.g = 1
4337 	};
4338 	seg.selector = vmcs12->host_ds_selector;
4339 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4340 	seg.selector = vmcs12->host_es_selector;
4341 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4342 	seg.selector = vmcs12->host_ss_selector;
4343 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4344 	seg.selector = vmcs12->host_fs_selector;
4345 	seg.base = vmcs12->host_fs_base;
4346 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4347 	seg.selector = vmcs12->host_gs_selector;
4348 	seg.base = vmcs12->host_gs_base;
4349 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4350 	seg = (struct kvm_segment) {
4351 		.base = vmcs12->host_tr_base,
4352 		.limit = 0x67,
4353 		.selector = vmcs12->host_tr_selector,
4354 		.type = 11,
4355 		.present = 1
4356 	};
4357 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4358 
4359 	memset(&seg, 0, sizeof(seg));
4360 	seg.unusable = 1;
4361 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4362 
4363 	kvm_set_dr(vcpu, 7, 0x400);
4364 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4365 
4366 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4367 				vmcs12->vm_exit_msr_load_count))
4368 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4369 
4370 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4371 }
4372 
4373 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4374 {
4375 	struct vmx_uret_msr *efer_msr;
4376 	unsigned int i;
4377 
4378 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4379 		return vmcs_read64(GUEST_IA32_EFER);
4380 
4381 	if (cpu_has_load_ia32_efer())
4382 		return host_efer;
4383 
4384 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4385 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4386 			return vmx->msr_autoload.guest.val[i].value;
4387 	}
4388 
4389 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4390 	if (efer_msr)
4391 		return efer_msr->data;
4392 
4393 	return host_efer;
4394 }
4395 
4396 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4397 {
4398 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4399 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4400 	struct vmx_msr_entry g, h;
4401 	gpa_t gpa;
4402 	u32 i, j;
4403 
4404 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4405 
4406 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4407 		/*
4408 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4409 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4410 		 * and vcpu->arch.dr7 is not squirreled away before the
4411 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4412 		 */
4413 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4414 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4415 		else
4416 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4417 	}
4418 
4419 	/*
4420 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4421 	 * handle a variety of side effects to KVM's software model.
4422 	 */
4423 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4424 
4425 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4426 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4427 
4428 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4429 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4430 
4431 	nested_ept_uninit_mmu_context(vcpu);
4432 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4433 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4434 
4435 	/*
4436 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4437 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4438 	 * VMFail, like everything else we just need to ensure our
4439 	 * software model is up-to-date.
4440 	 */
4441 	if (enable_ept && is_pae_paging(vcpu))
4442 		ept_save_pdptrs(vcpu);
4443 
4444 	kvm_mmu_reset_context(vcpu);
4445 
4446 	/*
4447 	 * This nasty bit of open coding is a compromise between blindly
4448 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4449 	 * of VMFail), leaving the nested VM's MSRs in the software model
4450 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4451 	 * expensive since the lists are unbound by hardware).  For each
4452 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4453 	 * list, reload it from the exit load list if it exists and differs
4454 	 * from the guest value.  The intent is to stuff host state as
4455 	 * silently as possible, not to fully process the exit load list.
4456 	 */
4457 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4458 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4459 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4460 			pr_debug_ratelimited(
4461 				"%s read MSR index failed (%u, 0x%08llx)\n",
4462 				__func__, i, gpa);
4463 			goto vmabort;
4464 		}
4465 
4466 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4467 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4468 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4469 				pr_debug_ratelimited(
4470 					"%s read MSR failed (%u, 0x%08llx)\n",
4471 					__func__, j, gpa);
4472 				goto vmabort;
4473 			}
4474 			if (h.index != g.index)
4475 				continue;
4476 			if (h.value == g.value)
4477 				break;
4478 
4479 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4480 				pr_debug_ratelimited(
4481 					"%s check failed (%u, 0x%x, 0x%x)\n",
4482 					__func__, j, h.index, h.reserved);
4483 				goto vmabort;
4484 			}
4485 
4486 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4487 				pr_debug_ratelimited(
4488 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4489 					__func__, j, h.index, h.value);
4490 				goto vmabort;
4491 			}
4492 		}
4493 	}
4494 
4495 	return;
4496 
4497 vmabort:
4498 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4499 }
4500 
4501 /*
4502  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4503  * and modify vmcs12 to make it see what it would expect to see there if
4504  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4505  */
4506 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4507 		       u32 exit_intr_info, unsigned long exit_qualification)
4508 {
4509 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4510 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4511 
4512 	/* trying to cancel vmlaunch/vmresume is a bug */
4513 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4514 
4515 	/* Similarly, triple faults in L2 should never escape. */
4516 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
4517 
4518 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4519 		/*
4520 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4521 		 * Enlightened VMCS after migration and we still need to
4522 		 * do that when something is forcing L2->L1 exit prior to
4523 		 * the first L2 run.
4524 		 */
4525 		(void)nested_get_evmcs_page(vcpu);
4526 	}
4527 
4528 	/* Service pending TLB flush requests for L2 before switching to L1. */
4529 	kvm_service_local_tlb_flush_requests(vcpu);
4530 
4531 	/*
4532 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4533 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4534 	 * up-to-date before switching to L1.
4535 	 */
4536 	if (enable_ept && is_pae_paging(vcpu))
4537 		vmx_ept_load_pdptrs(vcpu);
4538 
4539 	leave_guest_mode(vcpu);
4540 
4541 	if (nested_cpu_has_preemption_timer(vmcs12))
4542 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4543 
4544 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4545 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4546 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4547 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4548 	}
4549 
4550 	if (likely(!vmx->fail)) {
4551 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4552 
4553 		if (vm_exit_reason != -1)
4554 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4555 				       exit_intr_info, exit_qualification);
4556 
4557 		/*
4558 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4559 		 * also be used to capture vmcs12 cache as part of
4560 		 * capturing nVMX state for snapshot (migration).
4561 		 *
4562 		 * Otherwise, this flush will dirty guest memory at a
4563 		 * point it is already assumed by user-space to be
4564 		 * immutable.
4565 		 */
4566 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4567 	} else {
4568 		/*
4569 		 * The only expected VM-instruction error is "VM entry with
4570 		 * invalid control field(s)." Anything else indicates a
4571 		 * problem with L0.  And we should never get here with a
4572 		 * VMFail of any type if early consistency checks are enabled.
4573 		 */
4574 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4575 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4576 		WARN_ON_ONCE(nested_early_check);
4577 	}
4578 
4579 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4580 
4581 	/* Update any VMCS fields that might have changed while L2 ran */
4582 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4583 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4584 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4585 	if (kvm_has_tsc_control)
4586 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4587 
4588 	if (vmx->nested.l1_tpr_threshold != -1)
4589 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4590 
4591 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4592 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4593 		vmx_set_virtual_apic_mode(vcpu);
4594 	}
4595 
4596 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4597 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4598 		vmx_update_cpu_dirty_logging(vcpu);
4599 	}
4600 
4601 	/* Unpin physical memory we referred to in vmcs02 */
4602 	if (vmx->nested.apic_access_page) {
4603 		kvm_release_page_clean(vmx->nested.apic_access_page);
4604 		vmx->nested.apic_access_page = NULL;
4605 	}
4606 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4607 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4608 	vmx->nested.pi_desc = NULL;
4609 
4610 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4611 		vmx->nested.reload_vmcs01_apic_access_page = false;
4612 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4613 	}
4614 
4615 	if ((vm_exit_reason != -1) &&
4616 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4617 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4618 
4619 	/* in case we halted in L2 */
4620 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4621 
4622 	if (likely(!vmx->fail)) {
4623 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4624 		    nested_exit_intr_ack_set(vcpu)) {
4625 			int irq = kvm_cpu_get_interrupt(vcpu);
4626 			WARN_ON(irq < 0);
4627 			vmcs12->vm_exit_intr_info = irq |
4628 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4629 		}
4630 
4631 		if (vm_exit_reason != -1)
4632 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4633 						       vmcs12->exit_qualification,
4634 						       vmcs12->idt_vectoring_info_field,
4635 						       vmcs12->vm_exit_intr_info,
4636 						       vmcs12->vm_exit_intr_error_code,
4637 						       KVM_ISA_VMX);
4638 
4639 		load_vmcs12_host_state(vcpu, vmcs12);
4640 
4641 		return;
4642 	}
4643 
4644 	/*
4645 	 * After an early L2 VM-entry failure, we're now back
4646 	 * in L1 which thinks it just finished a VMLAUNCH or
4647 	 * VMRESUME instruction, so we need to set the failure
4648 	 * flag and the VM-instruction error field of the VMCS
4649 	 * accordingly, and skip the emulated instruction.
4650 	 */
4651 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4652 
4653 	/*
4654 	 * Restore L1's host state to KVM's software model.  We're here
4655 	 * because a consistency check was caught by hardware, which
4656 	 * means some amount of guest state has been propagated to KVM's
4657 	 * model and needs to be unwound to the host's state.
4658 	 */
4659 	nested_vmx_restore_host_state(vcpu);
4660 
4661 	vmx->fail = 0;
4662 }
4663 
4664 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4665 {
4666 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4667 }
4668 
4669 /*
4670  * Decode the memory-address operand of a vmx instruction, as recorded on an
4671  * exit caused by such an instruction (run by a guest hypervisor).
4672  * On success, returns 0. When the operand is invalid, returns 1 and throws
4673  * #UD, #GP, or #SS.
4674  */
4675 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4676 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4677 {
4678 	gva_t off;
4679 	bool exn;
4680 	struct kvm_segment s;
4681 
4682 	/*
4683 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4684 	 * Execution", on an exit, vmx_instruction_info holds most of the
4685 	 * addressing components of the operand. Only the displacement part
4686 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4687 	 * For how an actual address is calculated from all these components,
4688 	 * refer to Vol. 1, "Operand Addressing".
4689 	 */
4690 	int  scaling = vmx_instruction_info & 3;
4691 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4692 	bool is_reg = vmx_instruction_info & (1u << 10);
4693 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4694 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4695 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4696 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4697 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4698 
4699 	if (is_reg) {
4700 		kvm_queue_exception(vcpu, UD_VECTOR);
4701 		return 1;
4702 	}
4703 
4704 	/* Addr = segment_base + offset */
4705 	/* offset = base + [index * scale] + displacement */
4706 	off = exit_qualification; /* holds the displacement */
4707 	if (addr_size == 1)
4708 		off = (gva_t)sign_extend64(off, 31);
4709 	else if (addr_size == 0)
4710 		off = (gva_t)sign_extend64(off, 15);
4711 	if (base_is_valid)
4712 		off += kvm_register_read(vcpu, base_reg);
4713 	if (index_is_valid)
4714 		off += kvm_register_read(vcpu, index_reg) << scaling;
4715 	vmx_get_segment(vcpu, &s, seg_reg);
4716 
4717 	/*
4718 	 * The effective address, i.e. @off, of a memory operand is truncated
4719 	 * based on the address size of the instruction.  Note that this is
4720 	 * the *effective address*, i.e. the address prior to accounting for
4721 	 * the segment's base.
4722 	 */
4723 	if (addr_size == 1) /* 32 bit */
4724 		off &= 0xffffffff;
4725 	else if (addr_size == 0) /* 16 bit */
4726 		off &= 0xffff;
4727 
4728 	/* Checks for #GP/#SS exceptions. */
4729 	exn = false;
4730 	if (is_long_mode(vcpu)) {
4731 		/*
4732 		 * The virtual/linear address is never truncated in 64-bit
4733 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4734 		 * address when using FS/GS with a non-zero base.
4735 		 */
4736 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4737 			*ret = s.base + off;
4738 		else
4739 			*ret = off;
4740 
4741 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4742 		 * non-canonical form. This is the only check on the memory
4743 		 * destination for long mode!
4744 		 */
4745 		exn = is_noncanonical_address(*ret, vcpu);
4746 	} else {
4747 		/*
4748 		 * When not in long mode, the virtual/linear address is
4749 		 * unconditionally truncated to 32 bits regardless of the
4750 		 * address size.
4751 		 */
4752 		*ret = (s.base + off) & 0xffffffff;
4753 
4754 		/* Protected mode: apply checks for segment validity in the
4755 		 * following order:
4756 		 * - segment type check (#GP(0) may be thrown)
4757 		 * - usability check (#GP(0)/#SS(0))
4758 		 * - limit check (#GP(0)/#SS(0))
4759 		 */
4760 		if (wr)
4761 			/* #GP(0) if the destination operand is located in a
4762 			 * read-only data segment or any code segment.
4763 			 */
4764 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4765 		else
4766 			/* #GP(0) if the source operand is located in an
4767 			 * execute-only code segment
4768 			 */
4769 			exn = ((s.type & 0xa) == 8);
4770 		if (exn) {
4771 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4772 			return 1;
4773 		}
4774 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4775 		 */
4776 		exn = (s.unusable != 0);
4777 
4778 		/*
4779 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4780 		 * outside the segment limit.  All CPUs that support VMX ignore
4781 		 * limit checks for flat segments, i.e. segments with base==0,
4782 		 * limit==0xffffffff and of type expand-up data or code.
4783 		 */
4784 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4785 		     ((s.type & 8) || !(s.type & 4))))
4786 			exn = exn || ((u64)off + len - 1 > s.limit);
4787 	}
4788 	if (exn) {
4789 		kvm_queue_exception_e(vcpu,
4790 				      seg_reg == VCPU_SREG_SS ?
4791 						SS_VECTOR : GP_VECTOR,
4792 				      0);
4793 		return 1;
4794 	}
4795 
4796 	return 0;
4797 }
4798 
4799 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4800 {
4801 	struct vcpu_vmx *vmx;
4802 
4803 	if (!nested_vmx_allowed(vcpu))
4804 		return;
4805 
4806 	vmx = to_vmx(vcpu);
4807 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4808 		vmx->nested.msrs.entry_ctls_high |=
4809 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4810 		vmx->nested.msrs.exit_ctls_high |=
4811 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4812 	} else {
4813 		vmx->nested.msrs.entry_ctls_high &=
4814 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4815 		vmx->nested.msrs.exit_ctls_high &=
4816 				~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4817 	}
4818 }
4819 
4820 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4821 				int *ret)
4822 {
4823 	gva_t gva;
4824 	struct x86_exception e;
4825 	int r;
4826 
4827 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4828 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4829 				sizeof(*vmpointer), &gva)) {
4830 		*ret = 1;
4831 		return -EINVAL;
4832 	}
4833 
4834 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4835 	if (r != X86EMUL_CONTINUE) {
4836 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4837 		return -EINVAL;
4838 	}
4839 
4840 	return 0;
4841 }
4842 
4843 /*
4844  * Allocate a shadow VMCS and associate it with the currently loaded
4845  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4846  * VMCS is also VMCLEARed, so that it is ready for use.
4847  */
4848 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4849 {
4850 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4851 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4852 
4853 	/*
4854 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4855 	 * executes VMXON and free it when L1 executes VMXOFF.
4856 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4857 	 * here when vmcs01 already have an allocated shadow vmcs.
4858 	 */
4859 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4860 
4861 	if (!loaded_vmcs->shadow_vmcs) {
4862 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4863 		if (loaded_vmcs->shadow_vmcs)
4864 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4865 	}
4866 	return loaded_vmcs->shadow_vmcs;
4867 }
4868 
4869 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4870 {
4871 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4872 	int r;
4873 
4874 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4875 	if (r < 0)
4876 		goto out_vmcs02;
4877 
4878 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4879 	if (!vmx->nested.cached_vmcs12)
4880 		goto out_cached_vmcs12;
4881 
4882 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
4883 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4884 	if (!vmx->nested.cached_shadow_vmcs12)
4885 		goto out_cached_shadow_vmcs12;
4886 
4887 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4888 		goto out_shadow_vmcs;
4889 
4890 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4891 		     HRTIMER_MODE_ABS_PINNED);
4892 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4893 
4894 	vmx->nested.vpid02 = allocate_vpid();
4895 
4896 	vmx->nested.vmcs02_initialized = false;
4897 	vmx->nested.vmxon = true;
4898 
4899 	if (vmx_pt_mode_is_host_guest()) {
4900 		vmx->pt_desc.guest.ctl = 0;
4901 		pt_update_intercept_for_msr(vcpu);
4902 	}
4903 
4904 	return 0;
4905 
4906 out_shadow_vmcs:
4907 	kfree(vmx->nested.cached_shadow_vmcs12);
4908 
4909 out_cached_shadow_vmcs12:
4910 	kfree(vmx->nested.cached_vmcs12);
4911 
4912 out_cached_vmcs12:
4913 	free_loaded_vmcs(&vmx->nested.vmcs02);
4914 
4915 out_vmcs02:
4916 	return -ENOMEM;
4917 }
4918 
4919 /* Emulate the VMXON instruction. */
4920 static int handle_vmon(struct kvm_vcpu *vcpu)
4921 {
4922 	int ret;
4923 	gpa_t vmptr;
4924 	uint32_t revision;
4925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4926 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4927 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4928 
4929 	/*
4930 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4931 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4932 	 * 1 (see vmx_is_valid_cr4() for when we allow the guest to set this).
4933 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4934 	 * have already been checked by hardware, prior to the VM-exit for
4935 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4936 	 * that bit set to 1 in non-root mode.
4937 	 */
4938 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4939 		kvm_queue_exception(vcpu, UD_VECTOR);
4940 		return 1;
4941 	}
4942 
4943 	/* CPL=0 must be checked manually. */
4944 	if (vmx_get_cpl(vcpu)) {
4945 		kvm_inject_gp(vcpu, 0);
4946 		return 1;
4947 	}
4948 
4949 	if (vmx->nested.vmxon)
4950 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4951 
4952 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4953 			!= VMXON_NEEDED_FEATURES) {
4954 		kvm_inject_gp(vcpu, 0);
4955 		return 1;
4956 	}
4957 
4958 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4959 		return ret;
4960 
4961 	/*
4962 	 * SDM 3: 24.11.5
4963 	 * The first 4 bytes of VMXON region contain the supported
4964 	 * VMCS revision identifier
4965 	 *
4966 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4967 	 * which replaces physical address width with 32
4968 	 */
4969 	if (!page_address_valid(vcpu, vmptr))
4970 		return nested_vmx_failInvalid(vcpu);
4971 
4972 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4973 	    revision != VMCS12_REVISION)
4974 		return nested_vmx_failInvalid(vcpu);
4975 
4976 	vmx->nested.vmxon_ptr = vmptr;
4977 	ret = enter_vmx_operation(vcpu);
4978 	if (ret)
4979 		return ret;
4980 
4981 	return nested_vmx_succeed(vcpu);
4982 }
4983 
4984 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4985 {
4986 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4987 
4988 	if (vmx->nested.current_vmptr == INVALID_GPA)
4989 		return;
4990 
4991 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4992 
4993 	if (enable_shadow_vmcs) {
4994 		/* copy to memory all shadowed fields in case
4995 		   they were modified */
4996 		copy_shadow_to_vmcs12(vmx);
4997 		vmx_disable_shadow_vmcs(vmx);
4998 	}
4999 	vmx->nested.posted_intr_nv = -1;
5000 
5001 	/* Flush VMCS12 to guest memory */
5002 	kvm_vcpu_write_guest_page(vcpu,
5003 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
5004 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5005 
5006 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5007 
5008 	vmx->nested.current_vmptr = INVALID_GPA;
5009 }
5010 
5011 /* Emulate the VMXOFF instruction */
5012 static int handle_vmoff(struct kvm_vcpu *vcpu)
5013 {
5014 	if (!nested_vmx_check_permission(vcpu))
5015 		return 1;
5016 
5017 	free_nested(vcpu);
5018 
5019 	/* Process a latched INIT during time CPU was in VMX operation */
5020 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5021 
5022 	return nested_vmx_succeed(vcpu);
5023 }
5024 
5025 /* Emulate the VMCLEAR instruction */
5026 static int handle_vmclear(struct kvm_vcpu *vcpu)
5027 {
5028 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5029 	u32 zero = 0;
5030 	gpa_t vmptr;
5031 	u64 evmcs_gpa;
5032 	int r;
5033 
5034 	if (!nested_vmx_check_permission(vcpu))
5035 		return 1;
5036 
5037 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5038 		return r;
5039 
5040 	if (!page_address_valid(vcpu, vmptr))
5041 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5042 
5043 	if (vmptr == vmx->nested.vmxon_ptr)
5044 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5045 
5046 	/*
5047 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
5048 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
5049 	 * way to distinguish it from VMCS12) and we must not corrupt it by
5050 	 * writing to the non-existent 'launch_state' field. The area doesn't
5051 	 * have to be the currently active EVMCS on the calling CPU and there's
5052 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5053 	 * state. It is possible that the area will stay mapped as
5054 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5055 	 */
5056 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
5057 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
5058 		if (vmptr == vmx->nested.current_vmptr)
5059 			nested_release_vmcs12(vcpu);
5060 
5061 		kvm_vcpu_write_guest(vcpu,
5062 				     vmptr + offsetof(struct vmcs12,
5063 						      launch_state),
5064 				     &zero, sizeof(zero));
5065 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5066 		nested_release_evmcs(vcpu);
5067 	}
5068 
5069 	return nested_vmx_succeed(vcpu);
5070 }
5071 
5072 /* Emulate the VMLAUNCH instruction */
5073 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5074 {
5075 	return nested_vmx_run(vcpu, true);
5076 }
5077 
5078 /* Emulate the VMRESUME instruction */
5079 static int handle_vmresume(struct kvm_vcpu *vcpu)
5080 {
5081 
5082 	return nested_vmx_run(vcpu, false);
5083 }
5084 
5085 static int handle_vmread(struct kvm_vcpu *vcpu)
5086 {
5087 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5088 						    : get_vmcs12(vcpu);
5089 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5090 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5091 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5092 	struct x86_exception e;
5093 	unsigned long field;
5094 	u64 value;
5095 	gva_t gva = 0;
5096 	short offset;
5097 	int len, r;
5098 
5099 	if (!nested_vmx_check_permission(vcpu))
5100 		return 1;
5101 
5102 	/*
5103 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5104 	 * any VMREAD sets the ALU flags for VMfailInvalid.
5105 	 */
5106 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5107 	    (is_guest_mode(vcpu) &&
5108 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5109 		return nested_vmx_failInvalid(vcpu);
5110 
5111 	/* Decode instruction info and find the field to read */
5112 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5113 
5114 	offset = vmcs_field_to_offset(field);
5115 	if (offset < 0)
5116 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5117 
5118 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5119 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5120 
5121 	/* Read the field, zero-extended to a u64 value */
5122 	value = vmcs12_read_any(vmcs12, field, offset);
5123 
5124 	/*
5125 	 * Now copy part of this value to register or memory, as requested.
5126 	 * Note that the number of bits actually copied is 32 or 64 depending
5127 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5128 	 */
5129 	if (instr_info & BIT(10)) {
5130 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5131 	} else {
5132 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5133 		if (get_vmx_mem_address(vcpu, exit_qualification,
5134 					instr_info, true, len, &gva))
5135 			return 1;
5136 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5137 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5138 		if (r != X86EMUL_CONTINUE)
5139 			return kvm_handle_memory_failure(vcpu, r, &e);
5140 	}
5141 
5142 	return nested_vmx_succeed(vcpu);
5143 }
5144 
5145 static bool is_shadow_field_rw(unsigned long field)
5146 {
5147 	switch (field) {
5148 #define SHADOW_FIELD_RW(x, y) case x:
5149 #include "vmcs_shadow_fields.h"
5150 		return true;
5151 	default:
5152 		break;
5153 	}
5154 	return false;
5155 }
5156 
5157 static bool is_shadow_field_ro(unsigned long field)
5158 {
5159 	switch (field) {
5160 #define SHADOW_FIELD_RO(x, y) case x:
5161 #include "vmcs_shadow_fields.h"
5162 		return true;
5163 	default:
5164 		break;
5165 	}
5166 	return false;
5167 }
5168 
5169 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5170 {
5171 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5172 						    : get_vmcs12(vcpu);
5173 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5174 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5175 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5176 	struct x86_exception e;
5177 	unsigned long field;
5178 	short offset;
5179 	gva_t gva;
5180 	int len, r;
5181 
5182 	/*
5183 	 * The value to write might be 32 or 64 bits, depending on L1's long
5184 	 * mode, and eventually we need to write that into a field of several
5185 	 * possible lengths. The code below first zero-extends the value to 64
5186 	 * bit (value), and then copies only the appropriate number of
5187 	 * bits into the vmcs12 field.
5188 	 */
5189 	u64 value = 0;
5190 
5191 	if (!nested_vmx_check_permission(vcpu))
5192 		return 1;
5193 
5194 	/*
5195 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5196 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5197 	 */
5198 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5199 	    (is_guest_mode(vcpu) &&
5200 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5201 		return nested_vmx_failInvalid(vcpu);
5202 
5203 	if (instr_info & BIT(10))
5204 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5205 	else {
5206 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5207 		if (get_vmx_mem_address(vcpu, exit_qualification,
5208 					instr_info, false, len, &gva))
5209 			return 1;
5210 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5211 		if (r != X86EMUL_CONTINUE)
5212 			return kvm_handle_memory_failure(vcpu, r, &e);
5213 	}
5214 
5215 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5216 
5217 	offset = vmcs_field_to_offset(field);
5218 	if (offset < 0)
5219 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5220 
5221 	/*
5222 	 * If the vCPU supports "VMWRITE to any supported field in the
5223 	 * VMCS," then the "read-only" fields are actually read/write.
5224 	 */
5225 	if (vmcs_field_readonly(field) &&
5226 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5227 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5228 
5229 	/*
5230 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5231 	 * vmcs12, else we may crush a field or consume a stale value.
5232 	 */
5233 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5234 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5235 
5236 	/*
5237 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5238 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5239 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5240 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5241 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5242 	 * the stripped down value, L2 sees the full value as stored by KVM).
5243 	 */
5244 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5245 		value &= 0x1f0ff;
5246 
5247 	vmcs12_write_any(vmcs12, field, offset, value);
5248 
5249 	/*
5250 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5251 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5252 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5253 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5254 	 */
5255 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5256 		/*
5257 		 * L1 can read these fields without exiting, ensure the
5258 		 * shadow VMCS is up-to-date.
5259 		 */
5260 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5261 			preempt_disable();
5262 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5263 
5264 			__vmcs_writel(field, value);
5265 
5266 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5267 			vmcs_load(vmx->loaded_vmcs->vmcs);
5268 			preempt_enable();
5269 		}
5270 		vmx->nested.dirty_vmcs12 = true;
5271 	}
5272 
5273 	return nested_vmx_succeed(vcpu);
5274 }
5275 
5276 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5277 {
5278 	vmx->nested.current_vmptr = vmptr;
5279 	if (enable_shadow_vmcs) {
5280 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5281 		vmcs_write64(VMCS_LINK_POINTER,
5282 			     __pa(vmx->vmcs01.shadow_vmcs));
5283 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5284 	}
5285 	vmx->nested.dirty_vmcs12 = true;
5286 	vmx->nested.force_msr_bitmap_recalc = true;
5287 }
5288 
5289 /* Emulate the VMPTRLD instruction */
5290 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5291 {
5292 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5293 	gpa_t vmptr;
5294 	int r;
5295 
5296 	if (!nested_vmx_check_permission(vcpu))
5297 		return 1;
5298 
5299 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5300 		return r;
5301 
5302 	if (!page_address_valid(vcpu, vmptr))
5303 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5304 
5305 	if (vmptr == vmx->nested.vmxon_ptr)
5306 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5307 
5308 	/* Forbid normal VMPTRLD if Enlightened version was used */
5309 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5310 		return 1;
5311 
5312 	if (vmx->nested.current_vmptr != vmptr) {
5313 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5314 		struct vmcs_hdr hdr;
5315 
5316 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5317 			/*
5318 			 * Reads from an unbacked page return all 1s,
5319 			 * which means that the 32 bits located at the
5320 			 * given physical address won't match the required
5321 			 * VMCS12_REVISION identifier.
5322 			 */
5323 			return nested_vmx_fail(vcpu,
5324 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5325 		}
5326 
5327 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5328 						 offsetof(struct vmcs12, hdr),
5329 						 sizeof(hdr))) {
5330 			return nested_vmx_fail(vcpu,
5331 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5332 		}
5333 
5334 		if (hdr.revision_id != VMCS12_REVISION ||
5335 		    (hdr.shadow_vmcs &&
5336 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5337 			return nested_vmx_fail(vcpu,
5338 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5339 		}
5340 
5341 		nested_release_vmcs12(vcpu);
5342 
5343 		/*
5344 		 * Load VMCS12 from guest memory since it is not already
5345 		 * cached.
5346 		 */
5347 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5348 					  VMCS12_SIZE)) {
5349 			return nested_vmx_fail(vcpu,
5350 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5351 		}
5352 
5353 		set_current_vmptr(vmx, vmptr);
5354 	}
5355 
5356 	return nested_vmx_succeed(vcpu);
5357 }
5358 
5359 /* Emulate the VMPTRST instruction */
5360 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5361 {
5362 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5363 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5364 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5365 	struct x86_exception e;
5366 	gva_t gva;
5367 	int r;
5368 
5369 	if (!nested_vmx_check_permission(vcpu))
5370 		return 1;
5371 
5372 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5373 		return 1;
5374 
5375 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5376 				true, sizeof(gpa_t), &gva))
5377 		return 1;
5378 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5379 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5380 					sizeof(gpa_t), &e);
5381 	if (r != X86EMUL_CONTINUE)
5382 		return kvm_handle_memory_failure(vcpu, r, &e);
5383 
5384 	return nested_vmx_succeed(vcpu);
5385 }
5386 
5387 /* Emulate the INVEPT instruction */
5388 static int handle_invept(struct kvm_vcpu *vcpu)
5389 {
5390 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5391 	u32 vmx_instruction_info, types;
5392 	unsigned long type, roots_to_free;
5393 	struct kvm_mmu *mmu;
5394 	gva_t gva;
5395 	struct x86_exception e;
5396 	struct {
5397 		u64 eptp, gpa;
5398 	} operand;
5399 	int i, r, gpr_index;
5400 
5401 	if (!(vmx->nested.msrs.secondary_ctls_high &
5402 	      SECONDARY_EXEC_ENABLE_EPT) ||
5403 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5404 		kvm_queue_exception(vcpu, UD_VECTOR);
5405 		return 1;
5406 	}
5407 
5408 	if (!nested_vmx_check_permission(vcpu))
5409 		return 1;
5410 
5411 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5412 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5413 	type = kvm_register_read(vcpu, gpr_index);
5414 
5415 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5416 
5417 	if (type >= 32 || !(types & (1 << type)))
5418 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5419 
5420 	/* According to the Intel VMX instruction reference, the memory
5421 	 * operand is read even if it isn't needed (e.g., for type==global)
5422 	 */
5423 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5424 			vmx_instruction_info, false, sizeof(operand), &gva))
5425 		return 1;
5426 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5427 	if (r != X86EMUL_CONTINUE)
5428 		return kvm_handle_memory_failure(vcpu, r, &e);
5429 
5430 	/*
5431 	 * Nested EPT roots are always held through guest_mmu,
5432 	 * not root_mmu.
5433 	 */
5434 	mmu = &vcpu->arch.guest_mmu;
5435 
5436 	switch (type) {
5437 	case VMX_EPT_EXTENT_CONTEXT:
5438 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5439 			return nested_vmx_fail(vcpu,
5440 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5441 
5442 		roots_to_free = 0;
5443 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
5444 					    operand.eptp))
5445 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5446 
5447 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5448 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5449 						    mmu->prev_roots[i].pgd,
5450 						    operand.eptp))
5451 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5452 		}
5453 		break;
5454 	case VMX_EPT_EXTENT_GLOBAL:
5455 		roots_to_free = KVM_MMU_ROOTS_ALL;
5456 		break;
5457 	default:
5458 		BUG();
5459 		break;
5460 	}
5461 
5462 	if (roots_to_free)
5463 		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
5464 
5465 	return nested_vmx_succeed(vcpu);
5466 }
5467 
5468 static int handle_invvpid(struct kvm_vcpu *vcpu)
5469 {
5470 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5471 	u32 vmx_instruction_info;
5472 	unsigned long type, types;
5473 	gva_t gva;
5474 	struct x86_exception e;
5475 	struct {
5476 		u64 vpid;
5477 		u64 gla;
5478 	} operand;
5479 	u16 vpid02;
5480 	int r, gpr_index;
5481 
5482 	if (!(vmx->nested.msrs.secondary_ctls_high &
5483 	      SECONDARY_EXEC_ENABLE_VPID) ||
5484 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5485 		kvm_queue_exception(vcpu, UD_VECTOR);
5486 		return 1;
5487 	}
5488 
5489 	if (!nested_vmx_check_permission(vcpu))
5490 		return 1;
5491 
5492 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5493 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5494 	type = kvm_register_read(vcpu, gpr_index);
5495 
5496 	types = (vmx->nested.msrs.vpid_caps &
5497 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5498 
5499 	if (type >= 32 || !(types & (1 << type)))
5500 		return nested_vmx_fail(vcpu,
5501 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5502 
5503 	/* according to the intel vmx instruction reference, the memory
5504 	 * operand is read even if it isn't needed (e.g., for type==global)
5505 	 */
5506 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5507 			vmx_instruction_info, false, sizeof(operand), &gva))
5508 		return 1;
5509 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5510 	if (r != X86EMUL_CONTINUE)
5511 		return kvm_handle_memory_failure(vcpu, r, &e);
5512 
5513 	if (operand.vpid >> 16)
5514 		return nested_vmx_fail(vcpu,
5515 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5516 
5517 	vpid02 = nested_get_vpid02(vcpu);
5518 	switch (type) {
5519 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5520 		if (!operand.vpid ||
5521 		    is_noncanonical_address(operand.gla, vcpu))
5522 			return nested_vmx_fail(vcpu,
5523 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5524 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5525 		break;
5526 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5527 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5528 		if (!operand.vpid)
5529 			return nested_vmx_fail(vcpu,
5530 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5531 		vpid_sync_context(vpid02);
5532 		break;
5533 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5534 		vpid_sync_context(vpid02);
5535 		break;
5536 	default:
5537 		WARN_ON_ONCE(1);
5538 		return kvm_skip_emulated_instruction(vcpu);
5539 	}
5540 
5541 	/*
5542 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5543 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5544 	 * roots as VPIDs are not tracked in the MMU role.
5545 	 *
5546 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5547 	 * an MMU when EPT is disabled.
5548 	 *
5549 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5550 	 */
5551 	if (!enable_ept)
5552 		kvm_mmu_free_guest_mode_roots(vcpu, &vcpu->arch.root_mmu);
5553 
5554 	return nested_vmx_succeed(vcpu);
5555 }
5556 
5557 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5558 				     struct vmcs12 *vmcs12)
5559 {
5560 	u32 index = kvm_rcx_read(vcpu);
5561 	u64 new_eptp;
5562 
5563 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5564 		return 1;
5565 	if (index >= VMFUNC_EPTP_ENTRIES)
5566 		return 1;
5567 
5568 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5569 				     &new_eptp, index * 8, 8))
5570 		return 1;
5571 
5572 	/*
5573 	 * If the (L2) guest does a vmfunc to the currently
5574 	 * active ept pointer, we don't have to do anything else
5575 	 */
5576 	if (vmcs12->ept_pointer != new_eptp) {
5577 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5578 			return 1;
5579 
5580 		vmcs12->ept_pointer = new_eptp;
5581 		nested_ept_new_eptp(vcpu);
5582 
5583 		if (!nested_cpu_has_vpid(vmcs12))
5584 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5585 	}
5586 
5587 	return 0;
5588 }
5589 
5590 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5591 {
5592 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5593 	struct vmcs12 *vmcs12;
5594 	u32 function = kvm_rax_read(vcpu);
5595 
5596 	/*
5597 	 * VMFUNC is only supported for nested guests, but we always enable the
5598 	 * secondary control for simplicity; for non-nested mode, fake that we
5599 	 * didn't by injecting #UD.
5600 	 */
5601 	if (!is_guest_mode(vcpu)) {
5602 		kvm_queue_exception(vcpu, UD_VECTOR);
5603 		return 1;
5604 	}
5605 
5606 	vmcs12 = get_vmcs12(vcpu);
5607 
5608 	/*
5609 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5610 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5611 	 */
5612 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5613 		kvm_queue_exception(vcpu, UD_VECTOR);
5614 		return 1;
5615 	}
5616 
5617 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5618 		goto fail;
5619 
5620 	switch (function) {
5621 	case 0:
5622 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5623 			goto fail;
5624 		break;
5625 	default:
5626 		goto fail;
5627 	}
5628 	return kvm_skip_emulated_instruction(vcpu);
5629 
5630 fail:
5631 	/*
5632 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5633 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5634 	 * EXIT_REASON_VMFUNC as the exit reason.
5635 	 */
5636 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5637 			  vmx_get_intr_info(vcpu),
5638 			  vmx_get_exit_qual(vcpu));
5639 	return 1;
5640 }
5641 
5642 /*
5643  * Return true if an IO instruction with the specified port and size should cause
5644  * a VM-exit into L1.
5645  */
5646 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5647 				 int size)
5648 {
5649 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5650 	gpa_t bitmap, last_bitmap;
5651 	u8 b;
5652 
5653 	last_bitmap = INVALID_GPA;
5654 	b = -1;
5655 
5656 	while (size > 0) {
5657 		if (port < 0x8000)
5658 			bitmap = vmcs12->io_bitmap_a;
5659 		else if (port < 0x10000)
5660 			bitmap = vmcs12->io_bitmap_b;
5661 		else
5662 			return true;
5663 		bitmap += (port & 0x7fff) / 8;
5664 
5665 		if (last_bitmap != bitmap)
5666 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5667 				return true;
5668 		if (b & (1 << (port & 7)))
5669 			return true;
5670 
5671 		port++;
5672 		size--;
5673 		last_bitmap = bitmap;
5674 	}
5675 
5676 	return false;
5677 }
5678 
5679 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5680 				       struct vmcs12 *vmcs12)
5681 {
5682 	unsigned long exit_qualification;
5683 	unsigned short port;
5684 	int size;
5685 
5686 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5687 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5688 
5689 	exit_qualification = vmx_get_exit_qual(vcpu);
5690 
5691 	port = exit_qualification >> 16;
5692 	size = (exit_qualification & 7) + 1;
5693 
5694 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5695 }
5696 
5697 /*
5698  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5699  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5700  * disinterest in the current event (read or write a specific MSR) by using an
5701  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5702  */
5703 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5704 					struct vmcs12 *vmcs12,
5705 					union vmx_exit_reason exit_reason)
5706 {
5707 	u32 msr_index = kvm_rcx_read(vcpu);
5708 	gpa_t bitmap;
5709 
5710 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5711 		return true;
5712 
5713 	/*
5714 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5715 	 * for the four combinations of read/write and low/high MSR numbers.
5716 	 * First we need to figure out which of the four to use:
5717 	 */
5718 	bitmap = vmcs12->msr_bitmap;
5719 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5720 		bitmap += 2048;
5721 	if (msr_index >= 0xc0000000) {
5722 		msr_index -= 0xc0000000;
5723 		bitmap += 1024;
5724 	}
5725 
5726 	/* Then read the msr_index'th bit from this bitmap: */
5727 	if (msr_index < 1024*8) {
5728 		unsigned char b;
5729 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5730 			return true;
5731 		return 1 & (b >> (msr_index & 7));
5732 	} else
5733 		return true; /* let L1 handle the wrong parameter */
5734 }
5735 
5736 /*
5737  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5738  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5739  * intercept (via guest_host_mask etc.) the current event.
5740  */
5741 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5742 	struct vmcs12 *vmcs12)
5743 {
5744 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5745 	int cr = exit_qualification & 15;
5746 	int reg;
5747 	unsigned long val;
5748 
5749 	switch ((exit_qualification >> 4) & 3) {
5750 	case 0: /* mov to cr */
5751 		reg = (exit_qualification >> 8) & 15;
5752 		val = kvm_register_read(vcpu, reg);
5753 		switch (cr) {
5754 		case 0:
5755 			if (vmcs12->cr0_guest_host_mask &
5756 			    (val ^ vmcs12->cr0_read_shadow))
5757 				return true;
5758 			break;
5759 		case 3:
5760 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5761 				return true;
5762 			break;
5763 		case 4:
5764 			if (vmcs12->cr4_guest_host_mask &
5765 			    (vmcs12->cr4_read_shadow ^ val))
5766 				return true;
5767 			break;
5768 		case 8:
5769 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5770 				return true;
5771 			break;
5772 		}
5773 		break;
5774 	case 2: /* clts */
5775 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5776 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5777 			return true;
5778 		break;
5779 	case 1: /* mov from cr */
5780 		switch (cr) {
5781 		case 3:
5782 			if (vmcs12->cpu_based_vm_exec_control &
5783 			    CPU_BASED_CR3_STORE_EXITING)
5784 				return true;
5785 			break;
5786 		case 8:
5787 			if (vmcs12->cpu_based_vm_exec_control &
5788 			    CPU_BASED_CR8_STORE_EXITING)
5789 				return true;
5790 			break;
5791 		}
5792 		break;
5793 	case 3: /* lmsw */
5794 		/*
5795 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5796 		 * cr0. Other attempted changes are ignored, with no exit.
5797 		 */
5798 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5799 		if (vmcs12->cr0_guest_host_mask & 0xe &
5800 		    (val ^ vmcs12->cr0_read_shadow))
5801 			return true;
5802 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5803 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5804 		    (val & 0x1))
5805 			return true;
5806 		break;
5807 	}
5808 	return false;
5809 }
5810 
5811 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
5812 					  struct vmcs12 *vmcs12)
5813 {
5814 	u32 encls_leaf;
5815 
5816 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
5817 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
5818 		return false;
5819 
5820 	encls_leaf = kvm_rax_read(vcpu);
5821 	if (encls_leaf > 62)
5822 		encls_leaf = 63;
5823 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
5824 }
5825 
5826 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5827 	struct vmcs12 *vmcs12, gpa_t bitmap)
5828 {
5829 	u32 vmx_instruction_info;
5830 	unsigned long field;
5831 	u8 b;
5832 
5833 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5834 		return true;
5835 
5836 	/* Decode instruction info and find the field to access */
5837 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5838 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5839 
5840 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5841 	if (field >> 15)
5842 		return true;
5843 
5844 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5845 		return true;
5846 
5847 	return 1 & (b >> (field & 7));
5848 }
5849 
5850 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5851 {
5852 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5853 
5854 	if (nested_cpu_has_mtf(vmcs12))
5855 		return true;
5856 
5857 	/*
5858 	 * An MTF VM-exit may be injected into the guest by setting the
5859 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5860 	 * is the case regardless of the 'monitor trap flag' VM-execution
5861 	 * control.
5862 	 */
5863 	return entry_intr_info == (INTR_INFO_VALID_MASK
5864 				   | INTR_TYPE_OTHER_EVENT);
5865 }
5866 
5867 /*
5868  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5869  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5870  */
5871 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
5872 				     union vmx_exit_reason exit_reason)
5873 {
5874 	u32 intr_info;
5875 
5876 	switch ((u16)exit_reason.basic) {
5877 	case EXIT_REASON_EXCEPTION_NMI:
5878 		intr_info = vmx_get_intr_info(vcpu);
5879 		if (is_nmi(intr_info))
5880 			return true;
5881 		else if (is_page_fault(intr_info))
5882 			return vcpu->arch.apf.host_apf_flags ||
5883 			       vmx_need_pf_intercept(vcpu);
5884 		else if (is_debug(intr_info) &&
5885 			 vcpu->guest_debug &
5886 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5887 			return true;
5888 		else if (is_breakpoint(intr_info) &&
5889 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5890 			return true;
5891 		else if (is_alignment_check(intr_info) &&
5892 			 !vmx_guest_inject_ac(vcpu))
5893 			return true;
5894 		return false;
5895 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5896 		return true;
5897 	case EXIT_REASON_MCE_DURING_VMENTRY:
5898 		return true;
5899 	case EXIT_REASON_EPT_VIOLATION:
5900 		/*
5901 		 * L0 always deals with the EPT violation. If nested EPT is
5902 		 * used, and the nested mmu code discovers that the address is
5903 		 * missing in the guest EPT table (EPT12), the EPT violation
5904 		 * will be injected with nested_ept_inject_page_fault()
5905 		 */
5906 		return true;
5907 	case EXIT_REASON_EPT_MISCONFIG:
5908 		/*
5909 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5910 		 * table (shadow on EPT) or a merged EPT table that L0 built
5911 		 * (EPT on EPT). So any problems with the structure of the
5912 		 * table is L0's fault.
5913 		 */
5914 		return true;
5915 	case EXIT_REASON_PREEMPTION_TIMER:
5916 		return true;
5917 	case EXIT_REASON_PML_FULL:
5918 		/*
5919 		 * PML is emulated for an L1 VMM and should never be enabled in
5920 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
5921 		 */
5922 		return true;
5923 	case EXIT_REASON_VMFUNC:
5924 		/* VM functions are emulated through L2->L0 vmexits. */
5925 		return true;
5926 	case EXIT_REASON_BUS_LOCK:
5927 		/*
5928 		 * At present, bus lock VM exit is never exposed to L1.
5929 		 * Handle L2's bus locks in L0 directly.
5930 		 */
5931 		return true;
5932 	default:
5933 		break;
5934 	}
5935 	return false;
5936 }
5937 
5938 /*
5939  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5940  * is_guest_mode (L2).
5941  */
5942 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
5943 				     union vmx_exit_reason exit_reason)
5944 {
5945 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5946 	u32 intr_info;
5947 
5948 	switch ((u16)exit_reason.basic) {
5949 	case EXIT_REASON_EXCEPTION_NMI:
5950 		intr_info = vmx_get_intr_info(vcpu);
5951 		if (is_nmi(intr_info))
5952 			return true;
5953 		else if (is_page_fault(intr_info))
5954 			return true;
5955 		return vmcs12->exception_bitmap &
5956 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5957 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5958 		return nested_exit_on_intr(vcpu);
5959 	case EXIT_REASON_TRIPLE_FAULT:
5960 		return true;
5961 	case EXIT_REASON_INTERRUPT_WINDOW:
5962 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5963 	case EXIT_REASON_NMI_WINDOW:
5964 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5965 	case EXIT_REASON_TASK_SWITCH:
5966 		return true;
5967 	case EXIT_REASON_CPUID:
5968 		return true;
5969 	case EXIT_REASON_HLT:
5970 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5971 	case EXIT_REASON_INVD:
5972 		return true;
5973 	case EXIT_REASON_INVLPG:
5974 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5975 	case EXIT_REASON_RDPMC:
5976 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5977 	case EXIT_REASON_RDRAND:
5978 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5979 	case EXIT_REASON_RDSEED:
5980 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5981 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5982 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5983 	case EXIT_REASON_VMREAD:
5984 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5985 			vmcs12->vmread_bitmap);
5986 	case EXIT_REASON_VMWRITE:
5987 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5988 			vmcs12->vmwrite_bitmap);
5989 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5990 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5991 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5992 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5993 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5994 		/*
5995 		 * VMX instructions trap unconditionally. This allows L1 to
5996 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5997 		 */
5998 		return true;
5999 	case EXIT_REASON_CR_ACCESS:
6000 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6001 	case EXIT_REASON_DR_ACCESS:
6002 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6003 	case EXIT_REASON_IO_INSTRUCTION:
6004 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
6005 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6006 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6007 	case EXIT_REASON_MSR_READ:
6008 	case EXIT_REASON_MSR_WRITE:
6009 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6010 	case EXIT_REASON_INVALID_STATE:
6011 		return true;
6012 	case EXIT_REASON_MWAIT_INSTRUCTION:
6013 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6014 	case EXIT_REASON_MONITOR_TRAP_FLAG:
6015 		return nested_vmx_exit_handled_mtf(vmcs12);
6016 	case EXIT_REASON_MONITOR_INSTRUCTION:
6017 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6018 	case EXIT_REASON_PAUSE_INSTRUCTION:
6019 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6020 			nested_cpu_has2(vmcs12,
6021 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6022 	case EXIT_REASON_MCE_DURING_VMENTRY:
6023 		return true;
6024 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
6025 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6026 	case EXIT_REASON_APIC_ACCESS:
6027 	case EXIT_REASON_APIC_WRITE:
6028 	case EXIT_REASON_EOI_INDUCED:
6029 		/*
6030 		 * The controls for "virtualize APIC accesses," "APIC-
6031 		 * register virtualization," and "virtual-interrupt
6032 		 * delivery" only come from vmcs12.
6033 		 */
6034 		return true;
6035 	case EXIT_REASON_INVPCID:
6036 		return
6037 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6038 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6039 	case EXIT_REASON_WBINVD:
6040 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6041 	case EXIT_REASON_XSETBV:
6042 		return true;
6043 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6044 		/*
6045 		 * This should never happen, since it is not possible to
6046 		 * set XSS to a non-zero value---neither in L1 nor in L2.
6047 		 * If if it were, XSS would have to be checked against
6048 		 * the XSS exit bitmap in vmcs12.
6049 		 */
6050 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
6051 	case EXIT_REASON_UMWAIT:
6052 	case EXIT_REASON_TPAUSE:
6053 		return nested_cpu_has2(vmcs12,
6054 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6055 	case EXIT_REASON_ENCLS:
6056 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6057 	default:
6058 		return true;
6059 	}
6060 }
6061 
6062 /*
6063  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6064  * reflected into L1.
6065  */
6066 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6067 {
6068 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6069 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6070 	unsigned long exit_qual;
6071 	u32 exit_intr_info;
6072 
6073 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6074 
6075 	/*
6076 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6077 	 * has already loaded L2's state.
6078 	 */
6079 	if (unlikely(vmx->fail)) {
6080 		trace_kvm_nested_vmenter_failed(
6081 			"hardware VM-instruction error: ",
6082 			vmcs_read32(VM_INSTRUCTION_ERROR));
6083 		exit_intr_info = 0;
6084 		exit_qual = 0;
6085 		goto reflect_vmexit;
6086 	}
6087 
6088 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6089 
6090 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6091 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6092 		return false;
6093 
6094 	/* If L1 doesn't want the exit, handle it in L0. */
6095 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6096 		return false;
6097 
6098 	/*
6099 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6100 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6101 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6102 	 * interrupts are never reflected to L1 so it's a non-issue.
6103 	 */
6104 	exit_intr_info = vmx_get_intr_info(vcpu);
6105 	if (is_exception_with_error_code(exit_intr_info)) {
6106 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6107 
6108 		vmcs12->vm_exit_intr_error_code =
6109 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6110 	}
6111 	exit_qual = vmx_get_exit_qual(vcpu);
6112 
6113 reflect_vmexit:
6114 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6115 	return true;
6116 }
6117 
6118 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6119 				struct kvm_nested_state __user *user_kvm_nested_state,
6120 				u32 user_data_size)
6121 {
6122 	struct vcpu_vmx *vmx;
6123 	struct vmcs12 *vmcs12;
6124 	struct kvm_nested_state kvm_state = {
6125 		.flags = 0,
6126 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6127 		.size = sizeof(kvm_state),
6128 		.hdr.vmx.flags = 0,
6129 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6130 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6131 		.hdr.vmx.preemption_timer_deadline = 0,
6132 	};
6133 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6134 		&user_kvm_nested_state->data.vmx[0];
6135 
6136 	if (!vcpu)
6137 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6138 
6139 	vmx = to_vmx(vcpu);
6140 	vmcs12 = get_vmcs12(vcpu);
6141 
6142 	if (nested_vmx_allowed(vcpu) &&
6143 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6144 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6145 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6146 
6147 		if (vmx_has_valid_vmcs12(vcpu)) {
6148 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6149 
6150 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6151 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6152 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6153 
6154 			if (is_guest_mode(vcpu) &&
6155 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6156 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6157 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6158 		}
6159 
6160 		if (vmx->nested.smm.vmxon)
6161 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6162 
6163 		if (vmx->nested.smm.guest_mode)
6164 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6165 
6166 		if (is_guest_mode(vcpu)) {
6167 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6168 
6169 			if (vmx->nested.nested_run_pending)
6170 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6171 
6172 			if (vmx->nested.mtf_pending)
6173 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6174 
6175 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6176 			    vmx->nested.has_preemption_timer_deadline) {
6177 				kvm_state.hdr.vmx.flags |=
6178 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6179 				kvm_state.hdr.vmx.preemption_timer_deadline =
6180 					vmx->nested.preemption_timer_deadline;
6181 			}
6182 		}
6183 	}
6184 
6185 	if (user_data_size < kvm_state.size)
6186 		goto out;
6187 
6188 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6189 		return -EFAULT;
6190 
6191 	if (!vmx_has_valid_vmcs12(vcpu))
6192 		goto out;
6193 
6194 	/*
6195 	 * When running L2, the authoritative vmcs12 state is in the
6196 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6197 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6198 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6199 	 * vmcs12 state is in the vmcs12 already.
6200 	 */
6201 	if (is_guest_mode(vcpu)) {
6202 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6203 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6204 	} else  {
6205 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6206 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6207 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6208 				/*
6209 				 * L1 hypervisor is not obliged to keep eVMCS
6210 				 * clean fields data always up-to-date while
6211 				 * not in guest mode, 'hv_clean_fields' is only
6212 				 * supposed to be actual upon vmentry so we need
6213 				 * to ignore it here and do full copy.
6214 				 */
6215 				copy_enlightened_to_vmcs12(vmx, 0);
6216 			else if (enable_shadow_vmcs)
6217 				copy_shadow_to_vmcs12(vmx);
6218 		}
6219 	}
6220 
6221 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6222 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6223 
6224 	/*
6225 	 * Copy over the full allocated size of vmcs12 rather than just the size
6226 	 * of the struct.
6227 	 */
6228 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6229 		return -EFAULT;
6230 
6231 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6232 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6233 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6234 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6235 			return -EFAULT;
6236 	}
6237 out:
6238 	return kvm_state.size;
6239 }
6240 
6241 /*
6242  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6243  */
6244 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6245 {
6246 	if (is_guest_mode(vcpu)) {
6247 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6248 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6249 	}
6250 	free_nested(vcpu);
6251 }
6252 
6253 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6254 				struct kvm_nested_state __user *user_kvm_nested_state,
6255 				struct kvm_nested_state *kvm_state)
6256 {
6257 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6258 	struct vmcs12 *vmcs12;
6259 	enum vm_entry_failure_code ignored;
6260 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6261 		&user_kvm_nested_state->data.vmx[0];
6262 	int ret;
6263 
6264 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6265 		return -EINVAL;
6266 
6267 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6268 		if (kvm_state->hdr.vmx.smm.flags)
6269 			return -EINVAL;
6270 
6271 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6272 			return -EINVAL;
6273 
6274 		/*
6275 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6276 		 * enable eVMCS capability on vCPU. However, since then
6277 		 * code was changed such that flag signals vmcs12 should
6278 		 * be copied into eVMCS in guest memory.
6279 		 *
6280 		 * To preserve backwards compatability, allow user
6281 		 * to set this flag even when there is no VMXON region.
6282 		 */
6283 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6284 			return -EINVAL;
6285 	} else {
6286 		if (!nested_vmx_allowed(vcpu))
6287 			return -EINVAL;
6288 
6289 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6290 			return -EINVAL;
6291 	}
6292 
6293 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6294 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6295 		return -EINVAL;
6296 
6297 	if (kvm_state->hdr.vmx.smm.flags &
6298 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6299 		return -EINVAL;
6300 
6301 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6302 		return -EINVAL;
6303 
6304 	/*
6305 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6306 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6307 	 * must be zero.
6308 	 */
6309 	if (is_smm(vcpu) ?
6310 		(kvm_state->flags &
6311 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6312 		: kvm_state->hdr.vmx.smm.flags)
6313 		return -EINVAL;
6314 
6315 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6316 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6317 		return -EINVAL;
6318 
6319 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6320 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6321 			return -EINVAL;
6322 
6323 	vmx_leave_nested(vcpu);
6324 
6325 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6326 		return 0;
6327 
6328 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6329 	ret = enter_vmx_operation(vcpu);
6330 	if (ret)
6331 		return ret;
6332 
6333 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6334 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6335 		/* See vmx_has_valid_vmcs12.  */
6336 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6337 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6338 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6339 			return -EINVAL;
6340 		else
6341 			return 0;
6342 	}
6343 
6344 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6345 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6346 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6347 			return -EINVAL;
6348 
6349 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6350 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6351 		/*
6352 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6353 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6354 		 * restored yet. EVMCS will be mapped from
6355 		 * nested_get_vmcs12_pages().
6356 		 */
6357 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6358 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6359 	} else {
6360 		return -EINVAL;
6361 	}
6362 
6363 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6364 		vmx->nested.smm.vmxon = true;
6365 		vmx->nested.vmxon = false;
6366 
6367 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6368 			vmx->nested.smm.guest_mode = true;
6369 	}
6370 
6371 	vmcs12 = get_vmcs12(vcpu);
6372 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6373 		return -EFAULT;
6374 
6375 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6376 		return -EINVAL;
6377 
6378 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6379 		return 0;
6380 
6381 	vmx->nested.nested_run_pending =
6382 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6383 
6384 	vmx->nested.mtf_pending =
6385 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6386 
6387 	ret = -EINVAL;
6388 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6389 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6390 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6391 
6392 		if (kvm_state->size <
6393 		    sizeof(*kvm_state) +
6394 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6395 			goto error_guest_mode;
6396 
6397 		if (copy_from_user(shadow_vmcs12,
6398 				   user_vmx_nested_state->shadow_vmcs12,
6399 				   sizeof(*shadow_vmcs12))) {
6400 			ret = -EFAULT;
6401 			goto error_guest_mode;
6402 		}
6403 
6404 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6405 		    !shadow_vmcs12->hdr.shadow_vmcs)
6406 			goto error_guest_mode;
6407 	}
6408 
6409 	vmx->nested.has_preemption_timer_deadline = false;
6410 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6411 		vmx->nested.has_preemption_timer_deadline = true;
6412 		vmx->nested.preemption_timer_deadline =
6413 			kvm_state->hdr.vmx.preemption_timer_deadline;
6414 	}
6415 
6416 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6417 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6418 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6419 		goto error_guest_mode;
6420 
6421 	vmx->nested.dirty_vmcs12 = true;
6422 	vmx->nested.force_msr_bitmap_recalc = true;
6423 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6424 	if (ret)
6425 		goto error_guest_mode;
6426 
6427 	return 0;
6428 
6429 error_guest_mode:
6430 	vmx->nested.nested_run_pending = 0;
6431 	return ret;
6432 }
6433 
6434 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6435 {
6436 	if (enable_shadow_vmcs) {
6437 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6438 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6439 	}
6440 }
6441 
6442 /*
6443  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6444  * that madness to get the encoding for comparison.
6445  */
6446 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6447 
6448 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6449 {
6450 	/*
6451 	 * Note these are the so called "index" of the VMCS field encoding, not
6452 	 * the index into vmcs12.
6453 	 */
6454 	unsigned int max_idx, idx;
6455 	int i;
6456 
6457 	/*
6458 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6459 	 * vmcs12, regardless of whether or not the associated feature is
6460 	 * exposed to L1.  Simply find the field with the highest index.
6461 	 */
6462 	max_idx = 0;
6463 	for (i = 0; i < nr_vmcs12_fields; i++) {
6464 		/* The vmcs12 table is very, very sparsely populated. */
6465 		if (!vmcs_field_to_offset_table[i])
6466 			continue;
6467 
6468 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6469 		if (idx > max_idx)
6470 			max_idx = idx;
6471 	}
6472 
6473 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6474 }
6475 
6476 /*
6477  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6478  * returned for the various VMX controls MSRs when nested VMX is enabled.
6479  * The same values should also be used to verify that vmcs12 control fields are
6480  * valid during nested entry from L1 to L2.
6481  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6482  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6483  * bit in the high half is on if the corresponding bit in the control field
6484  * may be on. See also vmx_control_verify().
6485  */
6486 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6487 {
6488 	/*
6489 	 * Note that as a general rule, the high half of the MSRs (bits in
6490 	 * the control fields which may be 1) should be initialized by the
6491 	 * intersection of the underlying hardware's MSR (i.e., features which
6492 	 * can be supported) and the list of features we want to expose -
6493 	 * because they are known to be properly supported in our code.
6494 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6495 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6496 	 * reason is that if one of these bits is necessary, it will appear
6497 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6498 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6499 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6500 	 * These rules have exceptions below.
6501 	 */
6502 
6503 	/* pin-based controls */
6504 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6505 		msrs->pinbased_ctls_low,
6506 		msrs->pinbased_ctls_high);
6507 	msrs->pinbased_ctls_low |=
6508 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6509 	msrs->pinbased_ctls_high &=
6510 		PIN_BASED_EXT_INTR_MASK |
6511 		PIN_BASED_NMI_EXITING |
6512 		PIN_BASED_VIRTUAL_NMIS |
6513 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6514 	msrs->pinbased_ctls_high |=
6515 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6516 		PIN_BASED_VMX_PREEMPTION_TIMER;
6517 
6518 	/* exit controls */
6519 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6520 		msrs->exit_ctls_low,
6521 		msrs->exit_ctls_high);
6522 	msrs->exit_ctls_low =
6523 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6524 
6525 	msrs->exit_ctls_high &=
6526 #ifdef CONFIG_X86_64
6527 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6528 #endif
6529 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6530 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6531 	msrs->exit_ctls_high |=
6532 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6533 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6534 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6535 
6536 	/* We support free control of debug control saving. */
6537 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6538 
6539 	/* entry controls */
6540 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6541 		msrs->entry_ctls_low,
6542 		msrs->entry_ctls_high);
6543 	msrs->entry_ctls_low =
6544 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6545 	msrs->entry_ctls_high &=
6546 #ifdef CONFIG_X86_64
6547 		VM_ENTRY_IA32E_MODE |
6548 #endif
6549 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6550 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6551 	msrs->entry_ctls_high |=
6552 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6553 
6554 	/* We support free control of debug control loading. */
6555 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6556 
6557 	/* cpu-based controls */
6558 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6559 		msrs->procbased_ctls_low,
6560 		msrs->procbased_ctls_high);
6561 	msrs->procbased_ctls_low =
6562 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6563 	msrs->procbased_ctls_high &=
6564 		CPU_BASED_INTR_WINDOW_EXITING |
6565 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6566 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6567 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6568 		CPU_BASED_CR3_STORE_EXITING |
6569 #ifdef CONFIG_X86_64
6570 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6571 #endif
6572 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6573 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6574 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6575 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6576 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6577 	/*
6578 	 * We can allow some features even when not supported by the
6579 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6580 	 * can use it to avoid exits to L1 - even when L0 runs L2
6581 	 * without MSR bitmaps.
6582 	 */
6583 	msrs->procbased_ctls_high |=
6584 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6585 		CPU_BASED_USE_MSR_BITMAPS;
6586 
6587 	/* We support free control of CR3 access interception. */
6588 	msrs->procbased_ctls_low &=
6589 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6590 
6591 	/*
6592 	 * secondary cpu-based controls.  Do not include those that
6593 	 * depend on CPUID bits, they are added later by
6594 	 * vmx_vcpu_after_set_cpuid.
6595 	 */
6596 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6597 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6598 		      msrs->secondary_ctls_low,
6599 		      msrs->secondary_ctls_high);
6600 
6601 	msrs->secondary_ctls_low = 0;
6602 	msrs->secondary_ctls_high &=
6603 		SECONDARY_EXEC_DESC |
6604 		SECONDARY_EXEC_ENABLE_RDTSCP |
6605 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6606 		SECONDARY_EXEC_WBINVD_EXITING |
6607 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6608 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6609 		SECONDARY_EXEC_RDRAND_EXITING |
6610 		SECONDARY_EXEC_ENABLE_INVPCID |
6611 		SECONDARY_EXEC_RDSEED_EXITING |
6612 		SECONDARY_EXEC_XSAVES |
6613 		SECONDARY_EXEC_TSC_SCALING;
6614 
6615 	/*
6616 	 * We can emulate "VMCS shadowing," even if the hardware
6617 	 * doesn't support it.
6618 	 */
6619 	msrs->secondary_ctls_high |=
6620 		SECONDARY_EXEC_SHADOW_VMCS;
6621 
6622 	if (enable_ept) {
6623 		/* nested EPT: emulate EPT also to L1 */
6624 		msrs->secondary_ctls_high |=
6625 			SECONDARY_EXEC_ENABLE_EPT;
6626 		msrs->ept_caps =
6627 			VMX_EPT_PAGE_WALK_4_BIT |
6628 			VMX_EPT_PAGE_WALK_5_BIT |
6629 			VMX_EPTP_WB_BIT |
6630 			VMX_EPT_INVEPT_BIT |
6631 			VMX_EPT_EXECUTE_ONLY_BIT;
6632 
6633 		msrs->ept_caps &= ept_caps;
6634 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6635 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6636 			VMX_EPT_1GB_PAGE_BIT;
6637 		if (enable_ept_ad_bits) {
6638 			msrs->secondary_ctls_high |=
6639 				SECONDARY_EXEC_ENABLE_PML;
6640 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6641 		}
6642 	}
6643 
6644 	if (cpu_has_vmx_vmfunc()) {
6645 		msrs->secondary_ctls_high |=
6646 			SECONDARY_EXEC_ENABLE_VMFUNC;
6647 		/*
6648 		 * Advertise EPTP switching unconditionally
6649 		 * since we emulate it
6650 		 */
6651 		if (enable_ept)
6652 			msrs->vmfunc_controls =
6653 				VMX_VMFUNC_EPTP_SWITCHING;
6654 	}
6655 
6656 	/*
6657 	 * Old versions of KVM use the single-context version without
6658 	 * checking for support, so declare that it is supported even
6659 	 * though it is treated as global context.  The alternative is
6660 	 * not failing the single-context invvpid, and it is worse.
6661 	 */
6662 	if (enable_vpid) {
6663 		msrs->secondary_ctls_high |=
6664 			SECONDARY_EXEC_ENABLE_VPID;
6665 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6666 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6667 	}
6668 
6669 	if (enable_unrestricted_guest)
6670 		msrs->secondary_ctls_high |=
6671 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6672 
6673 	if (flexpriority_enabled)
6674 		msrs->secondary_ctls_high |=
6675 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6676 
6677 	if (enable_sgx)
6678 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6679 
6680 	/* miscellaneous data */
6681 	rdmsr(MSR_IA32_VMX_MISC,
6682 		msrs->misc_low,
6683 		msrs->misc_high);
6684 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6685 	msrs->misc_low |=
6686 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6687 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6688 		VMX_MISC_ACTIVITY_HLT |
6689 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6690 	msrs->misc_high = 0;
6691 
6692 	/*
6693 	 * This MSR reports some information about VMX support. We
6694 	 * should return information about the VMX we emulate for the
6695 	 * guest, and the VMCS structure we give it - not about the
6696 	 * VMX support of the underlying hardware.
6697 	 */
6698 	msrs->basic =
6699 		VMCS12_REVISION |
6700 		VMX_BASIC_TRUE_CTLS |
6701 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6702 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6703 
6704 	if (cpu_has_vmx_basic_inout())
6705 		msrs->basic |= VMX_BASIC_INOUT;
6706 
6707 	/*
6708 	 * These MSRs specify bits which the guest must keep fixed on
6709 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6710 	 * We picked the standard core2 setting.
6711 	 */
6712 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6713 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6714 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6715 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6716 
6717 	/* These MSRs specify bits which the guest must keep fixed off. */
6718 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6719 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6720 
6721 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
6722 }
6723 
6724 void nested_vmx_hardware_unsetup(void)
6725 {
6726 	int i;
6727 
6728 	if (enable_shadow_vmcs) {
6729 		for (i = 0; i < VMX_BITMAP_NR; i++)
6730 			free_page((unsigned long)vmx_bitmap[i]);
6731 	}
6732 }
6733 
6734 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6735 {
6736 	int i;
6737 
6738 	if (!cpu_has_vmx_shadow_vmcs())
6739 		enable_shadow_vmcs = 0;
6740 	if (enable_shadow_vmcs) {
6741 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6742 			/*
6743 			 * The vmx_bitmap is not tied to a VM and so should
6744 			 * not be charged to a memcg.
6745 			 */
6746 			vmx_bitmap[i] = (unsigned long *)
6747 				__get_free_page(GFP_KERNEL);
6748 			if (!vmx_bitmap[i]) {
6749 				nested_vmx_hardware_unsetup();
6750 				return -ENOMEM;
6751 			}
6752 		}
6753 
6754 		init_vmcs_shadow_fields();
6755 	}
6756 
6757 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6758 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6759 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6760 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6761 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6762 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6763 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6764 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6765 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6766 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6767 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6768 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6769 
6770 	return 0;
6771 }
6772 
6773 struct kvm_x86_nested_ops vmx_nested_ops = {
6774 	.check_events = vmx_check_nested_events,
6775 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6776 	.triple_fault = nested_vmx_triple_fault,
6777 	.get_state = vmx_get_nested_state,
6778 	.set_state = vmx_set_nested_state,
6779 	.get_nested_state_pages = vmx_get_nested_state_pages,
6780 	.write_log_dirty = nested_vmx_write_pml_buffer,
6781 	.enable_evmcs = nested_enable_evmcs,
6782 	.get_evmcs_version = nested_get_evmcs_version,
6783 };
6784