xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision ada066b2)
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_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
249 	dest->ldt_sel = src->ldt_sel;
250 #ifdef CONFIG_X86_64
251 	dest->ds_sel = src->ds_sel;
252 	dest->es_sel = src->es_sel;
253 #endif
254 }
255 
256 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
257 {
258 	struct vcpu_vmx *vmx = to_vmx(vcpu);
259 	struct loaded_vmcs *prev;
260 	int cpu;
261 
262 	if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
263 		return;
264 
265 	cpu = get_cpu();
266 	prev = vmx->loaded_vmcs;
267 	vmx->loaded_vmcs = vmcs;
268 	vmx_vcpu_load_vmcs(vcpu, cpu, prev);
269 	vmx_sync_vmcs_host_state(vmx, prev);
270 	put_cpu();
271 
272 	vmx_register_cache_reset(vcpu);
273 }
274 
275 /*
276  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
277  * just stops using VMX.
278  */
279 static void free_nested(struct kvm_vcpu *vcpu)
280 {
281 	struct vcpu_vmx *vmx = to_vmx(vcpu);
282 
283 	if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
284 		vmx_switch_vmcs(vcpu, &vmx->vmcs01);
285 
286 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
287 		return;
288 
289 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
290 
291 	vmx->nested.vmxon = false;
292 	vmx->nested.smm.vmxon = false;
293 	vmx->nested.vmxon_ptr = INVALID_GPA;
294 	free_vpid(vmx->nested.vpid02);
295 	vmx->nested.posted_intr_nv = -1;
296 	vmx->nested.current_vmptr = INVALID_GPA;
297 	if (enable_shadow_vmcs) {
298 		vmx_disable_shadow_vmcs(vmx);
299 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
300 		free_vmcs(vmx->vmcs01.shadow_vmcs);
301 		vmx->vmcs01.shadow_vmcs = NULL;
302 	}
303 	kfree(vmx->nested.cached_vmcs12);
304 	vmx->nested.cached_vmcs12 = NULL;
305 	kfree(vmx->nested.cached_shadow_vmcs12);
306 	vmx->nested.cached_shadow_vmcs12 = NULL;
307 	/* Unpin physical memory we referred to in the vmcs02 */
308 	if (vmx->nested.apic_access_page) {
309 		kvm_release_page_clean(vmx->nested.apic_access_page);
310 		vmx->nested.apic_access_page = NULL;
311 	}
312 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
313 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
314 	vmx->nested.pi_desc = NULL;
315 
316 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
317 
318 	nested_release_evmcs(vcpu);
319 
320 	free_loaded_vmcs(&vmx->nested.vmcs02);
321 }
322 
323 /*
324  * Ensure that the current vmcs of the logical processor is the
325  * vmcs01 of the vcpu before calling free_nested().
326  */
327 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
328 {
329 	vcpu_load(vcpu);
330 	vmx_leave_nested(vcpu);
331 	vcpu_put(vcpu);
332 }
333 
334 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
335 
336 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
337 {
338 	return VALID_PAGE(root_hpa) &&
339 	       ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
340 }
341 
342 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
343 				       gpa_t addr)
344 {
345 	uint i;
346 	struct kvm_mmu_root_info *cached_root;
347 
348 	WARN_ON_ONCE(!mmu_is_nested(vcpu));
349 
350 	for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
351 		cached_root = &vcpu->arch.mmu->prev_roots[i];
352 
353 		if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
354 					    eptp))
355 			vcpu->arch.mmu->invlpg(vcpu, addr, cached_root->hpa);
356 	}
357 }
358 
359 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
360 		struct x86_exception *fault)
361 {
362 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
363 	struct vcpu_vmx *vmx = to_vmx(vcpu);
364 	u32 vm_exit_reason;
365 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
366 
367 	if (vmx->nested.pml_full) {
368 		vm_exit_reason = EXIT_REASON_PML_FULL;
369 		vmx->nested.pml_full = false;
370 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
371 	} else {
372 		if (fault->error_code & PFERR_RSVD_MASK)
373 			vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
374 		else
375 			vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
376 
377 		/*
378 		 * Although the caller (kvm_inject_emulated_page_fault) would
379 		 * have already synced the faulting address in the shadow EPT
380 		 * tables for the current EPTP12, we also need to sync it for
381 		 * any other cached EPTP02s based on the same EP4TA, since the
382 		 * TLB associates mappings to the EP4TA rather than the full EPTP.
383 		 */
384 		nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
385 					   fault->address);
386 	}
387 
388 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
389 	vmcs12->guest_physical_address = fault->address;
390 }
391 
392 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
393 {
394 	kvm_init_shadow_ept_mmu(vcpu,
395 				to_vmx(vcpu)->nested.msrs.ept_caps &
396 				VMX_EPT_EXECUTE_ONLY_BIT,
397 				nested_ept_ad_enabled(vcpu),
398 				nested_ept_get_eptp(vcpu));
399 }
400 
401 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
402 {
403 	WARN_ON(mmu_is_nested(vcpu));
404 
405 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
406 	nested_ept_new_eptp(vcpu);
407 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
408 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
409 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
410 
411 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
412 }
413 
414 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
415 {
416 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
417 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
418 }
419 
420 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
421 					    u16 error_code)
422 {
423 	bool inequality, bit;
424 
425 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
426 	inequality =
427 		(error_code & vmcs12->page_fault_error_code_mask) !=
428 		 vmcs12->page_fault_error_code_match;
429 	return inequality ^ bit;
430 }
431 
432 
433 /*
434  * KVM wants to inject page-faults which it got to the guest. This function
435  * checks whether in a nested guest, we need to inject them to L1 or L2.
436  */
437 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
438 {
439 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
440 	unsigned int nr = vcpu->arch.exception.nr;
441 	bool has_payload = vcpu->arch.exception.has_payload;
442 	unsigned long payload = vcpu->arch.exception.payload;
443 
444 	if (nr == PF_VECTOR) {
445 		if (vcpu->arch.exception.nested_apf) {
446 			*exit_qual = vcpu->arch.apf.nested_apf_token;
447 			return 1;
448 		}
449 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
450 						    vcpu->arch.exception.error_code)) {
451 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
452 			return 1;
453 		}
454 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
455 		if (nr == DB_VECTOR) {
456 			if (!has_payload) {
457 				payload = vcpu->arch.dr6;
458 				payload &= ~DR6_BT;
459 				payload ^= DR6_ACTIVE_LOW;
460 			}
461 			*exit_qual = payload;
462 		} else
463 			*exit_qual = 0;
464 		return 1;
465 	}
466 
467 	return 0;
468 }
469 
470 
471 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
472 		struct x86_exception *fault)
473 {
474 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
475 
476 	WARN_ON(!is_guest_mode(vcpu));
477 
478 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
479 		!to_vmx(vcpu)->nested.nested_run_pending) {
480 		vmcs12->vm_exit_intr_error_code = fault->error_code;
481 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
482 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
483 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
484 				  fault->address);
485 	} else {
486 		kvm_inject_page_fault(vcpu, fault);
487 	}
488 }
489 
490 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
491 					       struct vmcs12 *vmcs12)
492 {
493 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
494 		return 0;
495 
496 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
497 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
498 		return -EINVAL;
499 
500 	return 0;
501 }
502 
503 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
504 						struct vmcs12 *vmcs12)
505 {
506 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
507 		return 0;
508 
509 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
510 		return -EINVAL;
511 
512 	return 0;
513 }
514 
515 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
516 						struct vmcs12 *vmcs12)
517 {
518 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
519 		return 0;
520 
521 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
522 		return -EINVAL;
523 
524 	return 0;
525 }
526 
527 /*
528  * For x2APIC MSRs, ignore the vmcs01 bitmap.  L1 can enable x2APIC without L1
529  * itself utilizing x2APIC.  All MSRs were previously set to be intercepted,
530  * only the "disable intercept" case needs to be handled.
531  */
532 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
533 							unsigned long *msr_bitmap_l0,
534 							u32 msr, int type)
535 {
536 	if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
537 		vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
538 
539 	if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
540 		vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
541 }
542 
543 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
544 {
545 	int msr;
546 
547 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
548 		unsigned word = msr / BITS_PER_LONG;
549 
550 		msr_bitmap[word] = ~0;
551 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
552 	}
553 }
554 
555 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw)					\
556 static inline									\
557 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx,			\
558 					 unsigned long *msr_bitmap_l1,		\
559 					 unsigned long *msr_bitmap_l0, u32 msr)	\
560 {										\
561 	if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) ||		\
562 	    vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr))			\
563 		vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
564 	else									\
565 		vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr);			\
566 }
567 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
568 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
569 
570 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
571 						    unsigned long *msr_bitmap_l1,
572 						    unsigned long *msr_bitmap_l0,
573 						    u32 msr, int types)
574 {
575 	if (types & MSR_TYPE_R)
576 		nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
577 						  msr_bitmap_l0, msr);
578 	if (types & MSR_TYPE_W)
579 		nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
580 						   msr_bitmap_l0, msr);
581 }
582 
583 /*
584  * Merge L0's and L1's MSR bitmap, return false to indicate that
585  * we do not use the hardware.
586  */
587 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
588 						 struct vmcs12 *vmcs12)
589 {
590 	struct vcpu_vmx *vmx = to_vmx(vcpu);
591 	int msr;
592 	unsigned long *msr_bitmap_l1;
593 	unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
594 	struct kvm_host_map *map = &vmx->nested.msr_bitmap_map;
595 
596 	/* Nothing to do if the MSR bitmap is not in use.  */
597 	if (!cpu_has_vmx_msr_bitmap() ||
598 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
599 		return false;
600 
601 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
602 		return false;
603 
604 	msr_bitmap_l1 = (unsigned long *)map->hva;
605 
606 	/*
607 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
608 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
609 	 * the x2APIC MSR range and selectively toggle those relevant to L2.
610 	 */
611 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
612 
613 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
614 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
615 			/*
616 			 * L0 need not intercept reads for MSRs between 0x800
617 			 * and 0x8ff, it just lets the processor take the value
618 			 * from the virtual-APIC page; take those 256 bits
619 			 * directly from the L1 bitmap.
620 			 */
621 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
622 				unsigned word = msr / BITS_PER_LONG;
623 
624 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
625 			}
626 		}
627 
628 		nested_vmx_disable_intercept_for_x2apic_msr(
629 			msr_bitmap_l1, msr_bitmap_l0,
630 			X2APIC_MSR(APIC_TASKPRI),
631 			MSR_TYPE_R | MSR_TYPE_W);
632 
633 		if (nested_cpu_has_vid(vmcs12)) {
634 			nested_vmx_disable_intercept_for_x2apic_msr(
635 				msr_bitmap_l1, msr_bitmap_l0,
636 				X2APIC_MSR(APIC_EOI),
637 				MSR_TYPE_W);
638 			nested_vmx_disable_intercept_for_x2apic_msr(
639 				msr_bitmap_l1, msr_bitmap_l0,
640 				X2APIC_MSR(APIC_SELF_IPI),
641 				MSR_TYPE_W);
642 		}
643 	}
644 
645 	/*
646 	 * Always check vmcs01's bitmap to honor userspace MSR filters and any
647 	 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
648 	 */
649 #ifdef CONFIG_X86_64
650 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
651 					 MSR_FS_BASE, MSR_TYPE_RW);
652 
653 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
654 					 MSR_GS_BASE, MSR_TYPE_RW);
655 
656 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
657 					 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
658 #endif
659 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
660 					 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
661 
662 	nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
663 					 MSR_IA32_PRED_CMD, MSR_TYPE_W);
664 
665 	kvm_vcpu_unmap(vcpu, &vmx->nested.msr_bitmap_map, false);
666 
667 	return true;
668 }
669 
670 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
671 				       struct vmcs12 *vmcs12)
672 {
673 	struct vcpu_vmx *vmx = to_vmx(vcpu);
674 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
675 
676 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
677 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
678 		return;
679 
680 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
681 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
682 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
683 		return;
684 
685 	kvm_read_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
686 			      VMCS12_SIZE);
687 }
688 
689 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
690 					      struct vmcs12 *vmcs12)
691 {
692 	struct vcpu_vmx *vmx = to_vmx(vcpu);
693 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
694 
695 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
696 	    vmcs12->vmcs_link_pointer == INVALID_GPA)
697 		return;
698 
699 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
700 	    kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
701 				      vmcs12->vmcs_link_pointer, VMCS12_SIZE))
702 		return;
703 
704 	kvm_write_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
705 			       VMCS12_SIZE);
706 }
707 
708 /*
709  * In nested virtualization, check if L1 has set
710  * VM_EXIT_ACK_INTR_ON_EXIT
711  */
712 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
713 {
714 	return get_vmcs12(vcpu)->vm_exit_controls &
715 		VM_EXIT_ACK_INTR_ON_EXIT;
716 }
717 
718 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
719 					  struct vmcs12 *vmcs12)
720 {
721 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
722 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
723 		return -EINVAL;
724 	else
725 		return 0;
726 }
727 
728 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
729 					   struct vmcs12 *vmcs12)
730 {
731 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
732 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
733 	    !nested_cpu_has_vid(vmcs12) &&
734 	    !nested_cpu_has_posted_intr(vmcs12))
735 		return 0;
736 
737 	/*
738 	 * If virtualize x2apic mode is enabled,
739 	 * virtualize apic access must be disabled.
740 	 */
741 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
742 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
743 		return -EINVAL;
744 
745 	/*
746 	 * If virtual interrupt delivery is enabled,
747 	 * we must exit on external interrupts.
748 	 */
749 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
750 		return -EINVAL;
751 
752 	/*
753 	 * bits 15:8 should be zero in posted_intr_nv,
754 	 * the descriptor address has been already checked
755 	 * in nested_get_vmcs12_pages.
756 	 *
757 	 * bits 5:0 of posted_intr_desc_addr should be zero.
758 	 */
759 	if (nested_cpu_has_posted_intr(vmcs12) &&
760 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
761 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
762 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
763 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
764 		return -EINVAL;
765 
766 	/* tpr shadow is needed by all apicv features. */
767 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
768 		return -EINVAL;
769 
770 	return 0;
771 }
772 
773 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
774 				       u32 count, u64 addr)
775 {
776 	if (count == 0)
777 		return 0;
778 
779 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
780 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
781 		return -EINVAL;
782 
783 	return 0;
784 }
785 
786 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
787 						     struct vmcs12 *vmcs12)
788 {
789 	if (CC(nested_vmx_check_msr_switch(vcpu,
790 					   vmcs12->vm_exit_msr_load_count,
791 					   vmcs12->vm_exit_msr_load_addr)) ||
792 	    CC(nested_vmx_check_msr_switch(vcpu,
793 					   vmcs12->vm_exit_msr_store_count,
794 					   vmcs12->vm_exit_msr_store_addr)))
795 		return -EINVAL;
796 
797 	return 0;
798 }
799 
800 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
801                                                       struct vmcs12 *vmcs12)
802 {
803 	if (CC(nested_vmx_check_msr_switch(vcpu,
804 					   vmcs12->vm_entry_msr_load_count,
805 					   vmcs12->vm_entry_msr_load_addr)))
806                 return -EINVAL;
807 
808 	return 0;
809 }
810 
811 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
812 					 struct vmcs12 *vmcs12)
813 {
814 	if (!nested_cpu_has_pml(vmcs12))
815 		return 0;
816 
817 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
818 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
819 		return -EINVAL;
820 
821 	return 0;
822 }
823 
824 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
825 							struct vmcs12 *vmcs12)
826 {
827 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
828 	       !nested_cpu_has_ept(vmcs12)))
829 		return -EINVAL;
830 	return 0;
831 }
832 
833 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
834 							 struct vmcs12 *vmcs12)
835 {
836 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
837 	       !nested_cpu_has_ept(vmcs12)))
838 		return -EINVAL;
839 	return 0;
840 }
841 
842 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
843 						 struct vmcs12 *vmcs12)
844 {
845 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
846 		return 0;
847 
848 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
849 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
850 		return -EINVAL;
851 
852 	return 0;
853 }
854 
855 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
856 				       struct vmx_msr_entry *e)
857 {
858 	/* x2APIC MSR accesses are not allowed */
859 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
860 		return -EINVAL;
861 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
862 	    CC(e->index == MSR_IA32_UCODE_REV))
863 		return -EINVAL;
864 	if (CC(e->reserved != 0))
865 		return -EINVAL;
866 	return 0;
867 }
868 
869 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
870 				     struct vmx_msr_entry *e)
871 {
872 	if (CC(e->index == MSR_FS_BASE) ||
873 	    CC(e->index == MSR_GS_BASE) ||
874 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
875 	    nested_vmx_msr_check_common(vcpu, e))
876 		return -EINVAL;
877 	return 0;
878 }
879 
880 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
881 				      struct vmx_msr_entry *e)
882 {
883 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
884 	    nested_vmx_msr_check_common(vcpu, e))
885 		return -EINVAL;
886 	return 0;
887 }
888 
889 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
890 {
891 	struct vcpu_vmx *vmx = to_vmx(vcpu);
892 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
893 				       vmx->nested.msrs.misc_high);
894 
895 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
896 }
897 
898 /*
899  * Load guest's/host's msr at nested entry/exit.
900  * return 0 for success, entry index for failure.
901  *
902  * One of the failure modes for MSR load/store is when a list exceeds the
903  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
904  * as possible, process all valid entries before failing rather than precheck
905  * for a capacity violation.
906  */
907 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
908 {
909 	u32 i;
910 	struct vmx_msr_entry e;
911 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
912 
913 	for (i = 0; i < count; i++) {
914 		if (unlikely(i >= max_msr_list_size))
915 			goto fail;
916 
917 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
918 					&e, sizeof(e))) {
919 			pr_debug_ratelimited(
920 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
921 				__func__, i, gpa + i * sizeof(e));
922 			goto fail;
923 		}
924 		if (nested_vmx_load_msr_check(vcpu, &e)) {
925 			pr_debug_ratelimited(
926 				"%s check failed (%u, 0x%x, 0x%x)\n",
927 				__func__, i, e.index, e.reserved);
928 			goto fail;
929 		}
930 		if (kvm_set_msr(vcpu, e.index, e.value)) {
931 			pr_debug_ratelimited(
932 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
933 				__func__, i, e.index, e.value);
934 			goto fail;
935 		}
936 	}
937 	return 0;
938 fail:
939 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
940 	return i + 1;
941 }
942 
943 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
944 					    u32 msr_index,
945 					    u64 *data)
946 {
947 	struct vcpu_vmx *vmx = to_vmx(vcpu);
948 
949 	/*
950 	 * If the L0 hypervisor stored a more accurate value for the TSC that
951 	 * does not include the time taken for emulation of the L2->L1
952 	 * VM-exit in L0, use the more accurate value.
953 	 */
954 	if (msr_index == MSR_IA32_TSC) {
955 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
956 						    MSR_IA32_TSC);
957 
958 		if (i >= 0) {
959 			u64 val = vmx->msr_autostore.guest.val[i].value;
960 
961 			*data = kvm_read_l1_tsc(vcpu, val);
962 			return true;
963 		}
964 	}
965 
966 	if (kvm_get_msr(vcpu, msr_index, data)) {
967 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
968 			msr_index);
969 		return false;
970 	}
971 	return true;
972 }
973 
974 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
975 				     struct vmx_msr_entry *e)
976 {
977 	if (kvm_vcpu_read_guest(vcpu,
978 				gpa + i * sizeof(*e),
979 				e, 2 * sizeof(u32))) {
980 		pr_debug_ratelimited(
981 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
982 			__func__, i, gpa + i * sizeof(*e));
983 		return false;
984 	}
985 	if (nested_vmx_store_msr_check(vcpu, e)) {
986 		pr_debug_ratelimited(
987 			"%s check failed (%u, 0x%x, 0x%x)\n",
988 			__func__, i, e->index, e->reserved);
989 		return false;
990 	}
991 	return true;
992 }
993 
994 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
995 {
996 	u64 data;
997 	u32 i;
998 	struct vmx_msr_entry e;
999 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1000 
1001 	for (i = 0; i < count; i++) {
1002 		if (unlikely(i >= max_msr_list_size))
1003 			return -EINVAL;
1004 
1005 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1006 			return -EINVAL;
1007 
1008 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1009 			return -EINVAL;
1010 
1011 		if (kvm_vcpu_write_guest(vcpu,
1012 					 gpa + i * sizeof(e) +
1013 					     offsetof(struct vmx_msr_entry, value),
1014 					 &data, sizeof(data))) {
1015 			pr_debug_ratelimited(
1016 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1017 				__func__, i, e.index, data);
1018 			return -EINVAL;
1019 		}
1020 	}
1021 	return 0;
1022 }
1023 
1024 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1025 {
1026 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1027 	u32 count = vmcs12->vm_exit_msr_store_count;
1028 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1029 	struct vmx_msr_entry e;
1030 	u32 i;
1031 
1032 	for (i = 0; i < count; i++) {
1033 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1034 			return false;
1035 
1036 		if (e.index == msr_index)
1037 			return true;
1038 	}
1039 	return false;
1040 }
1041 
1042 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1043 					   u32 msr_index)
1044 {
1045 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1046 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1047 	bool in_vmcs12_store_list;
1048 	int msr_autostore_slot;
1049 	bool in_autostore_list;
1050 	int last;
1051 
1052 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1053 	in_autostore_list = msr_autostore_slot >= 0;
1054 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1055 
1056 	if (in_vmcs12_store_list && !in_autostore_list) {
1057 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1058 			/*
1059 			 * Emulated VMEntry does not fail here.  Instead a less
1060 			 * accurate value will be returned by
1061 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1062 			 * instead of reading the value from the vmcs02 VMExit
1063 			 * MSR-store area.
1064 			 */
1065 			pr_warn_ratelimited(
1066 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1067 				msr_index);
1068 			return;
1069 		}
1070 		last = autostore->nr++;
1071 		autostore->val[last].index = msr_index;
1072 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1073 		last = --autostore->nr;
1074 		autostore->val[msr_autostore_slot] = autostore->val[last];
1075 	}
1076 }
1077 
1078 /*
1079  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1080  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1081  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1082  * @entry_failure_code.
1083  */
1084 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1085 			       bool nested_ept, bool reload_pdptrs,
1086 			       enum vm_entry_failure_code *entry_failure_code)
1087 {
1088 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1089 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1090 		return -EINVAL;
1091 	}
1092 
1093 	/*
1094 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1095 	 * must not be dereferenced.
1096 	 */
1097 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1098 	    CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1099 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1100 		return -EINVAL;
1101 	}
1102 
1103 	if (!nested_ept)
1104 		kvm_mmu_new_pgd(vcpu, cr3);
1105 
1106 	vcpu->arch.cr3 = cr3;
1107 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1108 
1109 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1110 	kvm_init_mmu(vcpu);
1111 
1112 	return 0;
1113 }
1114 
1115 /*
1116  * Returns if KVM is able to config CPU to tag TLB entries
1117  * populated by L2 differently than TLB entries populated
1118  * by L1.
1119  *
1120  * If L0 uses EPT, L1 and L2 run with different EPTP because
1121  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1122  * are tagged with different EPTP.
1123  *
1124  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1125  * with different VPID (L1 entries are tagged with vmx->vpid
1126  * while L2 entries are tagged with vmx->nested.vpid02).
1127  */
1128 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1129 {
1130 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1131 
1132 	return enable_ept ||
1133 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1134 }
1135 
1136 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1137 					    struct vmcs12 *vmcs12,
1138 					    bool is_vmenter)
1139 {
1140 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1141 
1142 	/*
1143 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1144 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1145 	 * full TLB flush from the guest's perspective.  This is required even
1146 	 * if VPID is disabled in the host as KVM may need to synchronize the
1147 	 * MMU in response to the guest TLB flush.
1148 	 *
1149 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1150 	 * EPT is a special snowflake, as guest-physical mappings aren't
1151 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1152 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1153 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1154 	 * those mappings.
1155 	 */
1156 	if (!nested_cpu_has_vpid(vmcs12)) {
1157 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1158 		return;
1159 	}
1160 
1161 	/* L2 should never have a VPID if VPID is disabled. */
1162 	WARN_ON(!enable_vpid);
1163 
1164 	/*
1165 	 * VPID is enabled and in use by vmcs12.  If vpid12 is changing, then
1166 	 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1167 	 * is the VPID incorporated into the MMU context.  I.e. KVM must assume
1168 	 * that the new vpid12 has never been used and thus represents a new
1169 	 * guest ASID that cannot have entries in the TLB.
1170 	 */
1171 	if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1172 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1173 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1174 		return;
1175 	}
1176 
1177 	/*
1178 	 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1179 	 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1180 	 * KVM was unable to allocate a VPID for L2, flush the current context
1181 	 * as the effective ASID is common to both L1 and L2.
1182 	 */
1183 	if (!nested_has_guest_tlb_tag(vcpu))
1184 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1185 }
1186 
1187 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1188 {
1189 	superset &= mask;
1190 	subset &= mask;
1191 
1192 	return (superset | subset) == superset;
1193 }
1194 
1195 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1196 {
1197 	const u64 feature_and_reserved =
1198 		/* feature (except bit 48; see below) */
1199 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1200 		/* reserved */
1201 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1202 	u64 vmx_basic = vmx->nested.msrs.basic;
1203 
1204 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1205 		return -EINVAL;
1206 
1207 	/*
1208 	 * KVM does not emulate a version of VMX that constrains physical
1209 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1210 	 */
1211 	if (data & BIT_ULL(48))
1212 		return -EINVAL;
1213 
1214 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1215 	    vmx_basic_vmcs_revision_id(data))
1216 		return -EINVAL;
1217 
1218 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1219 		return -EINVAL;
1220 
1221 	vmx->nested.msrs.basic = data;
1222 	return 0;
1223 }
1224 
1225 static int
1226 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1227 {
1228 	u64 supported;
1229 	u32 *lowp, *highp;
1230 
1231 	switch (msr_index) {
1232 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1233 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1234 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1235 		break;
1236 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1237 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1238 		highp = &vmx->nested.msrs.procbased_ctls_high;
1239 		break;
1240 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1241 		lowp = &vmx->nested.msrs.exit_ctls_low;
1242 		highp = &vmx->nested.msrs.exit_ctls_high;
1243 		break;
1244 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1245 		lowp = &vmx->nested.msrs.entry_ctls_low;
1246 		highp = &vmx->nested.msrs.entry_ctls_high;
1247 		break;
1248 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1249 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1250 		highp = &vmx->nested.msrs.secondary_ctls_high;
1251 		break;
1252 	default:
1253 		BUG();
1254 	}
1255 
1256 	supported = vmx_control_msr(*lowp, *highp);
1257 
1258 	/* Check must-be-1 bits are still 1. */
1259 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1260 		return -EINVAL;
1261 
1262 	/* Check must-be-0 bits are still 0. */
1263 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1264 		return -EINVAL;
1265 
1266 	*lowp = data;
1267 	*highp = data >> 32;
1268 	return 0;
1269 }
1270 
1271 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1272 {
1273 	const u64 feature_and_reserved_bits =
1274 		/* feature */
1275 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1276 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1277 		/* reserved */
1278 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1279 	u64 vmx_misc;
1280 
1281 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1282 				   vmx->nested.msrs.misc_high);
1283 
1284 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1285 		return -EINVAL;
1286 
1287 	if ((vmx->nested.msrs.pinbased_ctls_high &
1288 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1289 	    vmx_misc_preemption_timer_rate(data) !=
1290 	    vmx_misc_preemption_timer_rate(vmx_misc))
1291 		return -EINVAL;
1292 
1293 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1294 		return -EINVAL;
1295 
1296 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1297 		return -EINVAL;
1298 
1299 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1300 		return -EINVAL;
1301 
1302 	vmx->nested.msrs.misc_low = data;
1303 	vmx->nested.msrs.misc_high = data >> 32;
1304 
1305 	return 0;
1306 }
1307 
1308 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1309 {
1310 	u64 vmx_ept_vpid_cap;
1311 
1312 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1313 					   vmx->nested.msrs.vpid_caps);
1314 
1315 	/* Every bit is either reserved or a feature bit. */
1316 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1317 		return -EINVAL;
1318 
1319 	vmx->nested.msrs.ept_caps = data;
1320 	vmx->nested.msrs.vpid_caps = data >> 32;
1321 	return 0;
1322 }
1323 
1324 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1325 {
1326 	u64 *msr;
1327 
1328 	switch (msr_index) {
1329 	case MSR_IA32_VMX_CR0_FIXED0:
1330 		msr = &vmx->nested.msrs.cr0_fixed0;
1331 		break;
1332 	case MSR_IA32_VMX_CR4_FIXED0:
1333 		msr = &vmx->nested.msrs.cr4_fixed0;
1334 		break;
1335 	default:
1336 		BUG();
1337 	}
1338 
1339 	/*
1340 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1341 	 * must be 1 in the restored value.
1342 	 */
1343 	if (!is_bitwise_subset(data, *msr, -1ULL))
1344 		return -EINVAL;
1345 
1346 	*msr = data;
1347 	return 0;
1348 }
1349 
1350 /*
1351  * Called when userspace is restoring VMX MSRs.
1352  *
1353  * Returns 0 on success, non-0 otherwise.
1354  */
1355 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1356 {
1357 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1358 
1359 	/*
1360 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1361 	 * is in VMX operation.
1362 	 */
1363 	if (vmx->nested.vmxon)
1364 		return -EBUSY;
1365 
1366 	switch (msr_index) {
1367 	case MSR_IA32_VMX_BASIC:
1368 		return vmx_restore_vmx_basic(vmx, data);
1369 	case MSR_IA32_VMX_PINBASED_CTLS:
1370 	case MSR_IA32_VMX_PROCBASED_CTLS:
1371 	case MSR_IA32_VMX_EXIT_CTLS:
1372 	case MSR_IA32_VMX_ENTRY_CTLS:
1373 		/*
1374 		 * The "non-true" VMX capability MSRs are generated from the
1375 		 * "true" MSRs, so we do not support restoring them directly.
1376 		 *
1377 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1378 		 * should restore the "true" MSRs with the must-be-1 bits
1379 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1380 		 * DEFAULT SETTINGS".
1381 		 */
1382 		return -EINVAL;
1383 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1384 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1385 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1386 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1387 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1388 		return vmx_restore_control_msr(vmx, msr_index, data);
1389 	case MSR_IA32_VMX_MISC:
1390 		return vmx_restore_vmx_misc(vmx, data);
1391 	case MSR_IA32_VMX_CR0_FIXED0:
1392 	case MSR_IA32_VMX_CR4_FIXED0:
1393 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1394 	case MSR_IA32_VMX_CR0_FIXED1:
1395 	case MSR_IA32_VMX_CR4_FIXED1:
1396 		/*
1397 		 * These MSRs are generated based on the vCPU's CPUID, so we
1398 		 * do not support restoring them directly.
1399 		 */
1400 		return -EINVAL;
1401 	case MSR_IA32_VMX_EPT_VPID_CAP:
1402 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1403 	case MSR_IA32_VMX_VMCS_ENUM:
1404 		vmx->nested.msrs.vmcs_enum = data;
1405 		return 0;
1406 	case MSR_IA32_VMX_VMFUNC:
1407 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1408 			return -EINVAL;
1409 		vmx->nested.msrs.vmfunc_controls = data;
1410 		return 0;
1411 	default:
1412 		/*
1413 		 * The rest of the VMX capability MSRs do not support restore.
1414 		 */
1415 		return -EINVAL;
1416 	}
1417 }
1418 
1419 /* Returns 0 on success, non-0 otherwise. */
1420 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1421 {
1422 	switch (msr_index) {
1423 	case MSR_IA32_VMX_BASIC:
1424 		*pdata = msrs->basic;
1425 		break;
1426 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1427 	case MSR_IA32_VMX_PINBASED_CTLS:
1428 		*pdata = vmx_control_msr(
1429 			msrs->pinbased_ctls_low,
1430 			msrs->pinbased_ctls_high);
1431 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1432 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1433 		break;
1434 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1435 	case MSR_IA32_VMX_PROCBASED_CTLS:
1436 		*pdata = vmx_control_msr(
1437 			msrs->procbased_ctls_low,
1438 			msrs->procbased_ctls_high);
1439 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1440 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1441 		break;
1442 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1443 	case MSR_IA32_VMX_EXIT_CTLS:
1444 		*pdata = vmx_control_msr(
1445 			msrs->exit_ctls_low,
1446 			msrs->exit_ctls_high);
1447 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1448 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1449 		break;
1450 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1451 	case MSR_IA32_VMX_ENTRY_CTLS:
1452 		*pdata = vmx_control_msr(
1453 			msrs->entry_ctls_low,
1454 			msrs->entry_ctls_high);
1455 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1456 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1457 		break;
1458 	case MSR_IA32_VMX_MISC:
1459 		*pdata = vmx_control_msr(
1460 			msrs->misc_low,
1461 			msrs->misc_high);
1462 		break;
1463 	case MSR_IA32_VMX_CR0_FIXED0:
1464 		*pdata = msrs->cr0_fixed0;
1465 		break;
1466 	case MSR_IA32_VMX_CR0_FIXED1:
1467 		*pdata = msrs->cr0_fixed1;
1468 		break;
1469 	case MSR_IA32_VMX_CR4_FIXED0:
1470 		*pdata = msrs->cr4_fixed0;
1471 		break;
1472 	case MSR_IA32_VMX_CR4_FIXED1:
1473 		*pdata = msrs->cr4_fixed1;
1474 		break;
1475 	case MSR_IA32_VMX_VMCS_ENUM:
1476 		*pdata = msrs->vmcs_enum;
1477 		break;
1478 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1479 		*pdata = vmx_control_msr(
1480 			msrs->secondary_ctls_low,
1481 			msrs->secondary_ctls_high);
1482 		break;
1483 	case MSR_IA32_VMX_EPT_VPID_CAP:
1484 		*pdata = msrs->ept_caps |
1485 			((u64)msrs->vpid_caps << 32);
1486 		break;
1487 	case MSR_IA32_VMX_VMFUNC:
1488 		*pdata = msrs->vmfunc_controls;
1489 		break;
1490 	default:
1491 		return 1;
1492 	}
1493 
1494 	return 0;
1495 }
1496 
1497 /*
1498  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1499  * been modified by the L1 guest.  Note, "writable" in this context means
1500  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1501  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1502  * VM-exit information fields (which are actually writable if the vCPU is
1503  * configured to support "VMWRITE to any supported field in the VMCS").
1504  */
1505 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1506 {
1507 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1508 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1509 	struct shadow_vmcs_field field;
1510 	unsigned long val;
1511 	int i;
1512 
1513 	if (WARN_ON(!shadow_vmcs))
1514 		return;
1515 
1516 	preempt_disable();
1517 
1518 	vmcs_load(shadow_vmcs);
1519 
1520 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1521 		field = shadow_read_write_fields[i];
1522 		val = __vmcs_readl(field.encoding);
1523 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1524 	}
1525 
1526 	vmcs_clear(shadow_vmcs);
1527 	vmcs_load(vmx->loaded_vmcs->vmcs);
1528 
1529 	preempt_enable();
1530 }
1531 
1532 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1533 {
1534 	const struct shadow_vmcs_field *fields[] = {
1535 		shadow_read_write_fields,
1536 		shadow_read_only_fields
1537 	};
1538 	const int max_fields[] = {
1539 		max_shadow_read_write_fields,
1540 		max_shadow_read_only_fields
1541 	};
1542 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1543 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1544 	struct shadow_vmcs_field field;
1545 	unsigned long val;
1546 	int i, q;
1547 
1548 	if (WARN_ON(!shadow_vmcs))
1549 		return;
1550 
1551 	vmcs_load(shadow_vmcs);
1552 
1553 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1554 		for (i = 0; i < max_fields[q]; i++) {
1555 			field = fields[q][i];
1556 			val = vmcs12_read_any(vmcs12, field.encoding,
1557 					      field.offset);
1558 			__vmcs_writel(field.encoding, val);
1559 		}
1560 	}
1561 
1562 	vmcs_clear(shadow_vmcs);
1563 	vmcs_load(vmx->loaded_vmcs->vmcs);
1564 }
1565 
1566 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1567 {
1568 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1569 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1570 
1571 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1572 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1573 	vmcs12->guest_rip = evmcs->guest_rip;
1574 
1575 	if (unlikely(!(hv_clean_fields &
1576 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1577 		vmcs12->guest_rsp = evmcs->guest_rsp;
1578 		vmcs12->guest_rflags = evmcs->guest_rflags;
1579 		vmcs12->guest_interruptibility_info =
1580 			evmcs->guest_interruptibility_info;
1581 	}
1582 
1583 	if (unlikely(!(hv_clean_fields &
1584 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1585 		vmcs12->cpu_based_vm_exec_control =
1586 			evmcs->cpu_based_vm_exec_control;
1587 	}
1588 
1589 	if (unlikely(!(hv_clean_fields &
1590 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1591 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1592 	}
1593 
1594 	if (unlikely(!(hv_clean_fields &
1595 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1596 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1597 	}
1598 
1599 	if (unlikely(!(hv_clean_fields &
1600 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1601 		vmcs12->vm_entry_intr_info_field =
1602 			evmcs->vm_entry_intr_info_field;
1603 		vmcs12->vm_entry_exception_error_code =
1604 			evmcs->vm_entry_exception_error_code;
1605 		vmcs12->vm_entry_instruction_len =
1606 			evmcs->vm_entry_instruction_len;
1607 	}
1608 
1609 	if (unlikely(!(hv_clean_fields &
1610 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1611 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1612 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1613 		vmcs12->host_cr0 = evmcs->host_cr0;
1614 		vmcs12->host_cr3 = evmcs->host_cr3;
1615 		vmcs12->host_cr4 = evmcs->host_cr4;
1616 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1617 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1618 		vmcs12->host_rip = evmcs->host_rip;
1619 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1620 		vmcs12->host_es_selector = evmcs->host_es_selector;
1621 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1622 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1623 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1624 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1625 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1626 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1627 	}
1628 
1629 	if (unlikely(!(hv_clean_fields &
1630 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1631 		vmcs12->pin_based_vm_exec_control =
1632 			evmcs->pin_based_vm_exec_control;
1633 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1634 		vmcs12->secondary_vm_exec_control =
1635 			evmcs->secondary_vm_exec_control;
1636 	}
1637 
1638 	if (unlikely(!(hv_clean_fields &
1639 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1640 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1641 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1642 	}
1643 
1644 	if (unlikely(!(hv_clean_fields &
1645 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1646 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1647 	}
1648 
1649 	if (unlikely(!(hv_clean_fields &
1650 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1651 		vmcs12->guest_es_base = evmcs->guest_es_base;
1652 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1653 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1654 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1655 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1656 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1657 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1658 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1659 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1660 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1661 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1662 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1663 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1664 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1665 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1666 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1667 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1668 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1669 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1670 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1671 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1672 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1673 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1674 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1675 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1676 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1677 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1678 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1679 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1680 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1681 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1682 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1683 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1684 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1685 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1686 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1687 	}
1688 
1689 	if (unlikely(!(hv_clean_fields &
1690 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1691 		vmcs12->tsc_offset = evmcs->tsc_offset;
1692 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1693 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1694 	}
1695 
1696 	if (unlikely(!(hv_clean_fields &
1697 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1698 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1699 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1700 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1701 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1702 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1703 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1704 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1705 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1706 	}
1707 
1708 	if (unlikely(!(hv_clean_fields &
1709 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1710 		vmcs12->host_fs_base = evmcs->host_fs_base;
1711 		vmcs12->host_gs_base = evmcs->host_gs_base;
1712 		vmcs12->host_tr_base = evmcs->host_tr_base;
1713 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1714 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1715 		vmcs12->host_rsp = evmcs->host_rsp;
1716 	}
1717 
1718 	if (unlikely(!(hv_clean_fields &
1719 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1720 		vmcs12->ept_pointer = evmcs->ept_pointer;
1721 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1722 	}
1723 
1724 	if (unlikely(!(hv_clean_fields &
1725 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1726 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1727 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1728 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1729 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1730 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1731 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1732 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1733 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1734 		vmcs12->guest_pending_dbg_exceptions =
1735 			evmcs->guest_pending_dbg_exceptions;
1736 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1737 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1738 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1739 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1740 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1741 	}
1742 
1743 	/*
1744 	 * Not used?
1745 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1746 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1747 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1748 	 * vmcs12->page_fault_error_code_mask =
1749 	 *		evmcs->page_fault_error_code_mask;
1750 	 * vmcs12->page_fault_error_code_match =
1751 	 *		evmcs->page_fault_error_code_match;
1752 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1753 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1754 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1755 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1756 	 */
1757 
1758 	/*
1759 	 * Read only fields:
1760 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1761 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1762 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1763 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1764 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1765 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1766 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1767 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1768 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1769 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1770 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1771 	 *
1772 	 * Not present in struct vmcs12:
1773 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1774 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1775 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1776 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1777 	 */
1778 
1779 	return;
1780 }
1781 
1782 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1783 {
1784 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1785 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1786 
1787 	/*
1788 	 * Should not be changed by KVM:
1789 	 *
1790 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1791 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1792 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1793 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1794 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1795 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1796 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1797 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1798 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1799 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1800 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1801 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1802 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1803 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1804 	 * evmcs->host_rip = vmcs12->host_rip;
1805 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1806 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1807 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1808 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1809 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1810 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1811 	 * evmcs->host_rsp = vmcs12->host_rsp;
1812 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1813 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1814 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1815 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1816 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1817 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1818 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1819 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1820 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1821 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1822 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1823 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1824 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1825 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1826 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1827 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1828 	 * evmcs->page_fault_error_code_mask =
1829 	 *		vmcs12->page_fault_error_code_mask;
1830 	 * evmcs->page_fault_error_code_match =
1831 	 *		vmcs12->page_fault_error_code_match;
1832 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1833 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1834 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1835 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1836 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1837 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1838 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1839 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1840 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1841 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1842 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1843 	 *
1844 	 * Not present in struct vmcs12:
1845 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1846 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1847 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1848 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1849 	 */
1850 
1851 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1852 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1853 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1854 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1855 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1856 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1857 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1858 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1859 
1860 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1861 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1862 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1863 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1864 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1865 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1866 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1867 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1868 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1869 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1870 
1871 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1872 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1873 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1874 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1875 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1876 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1877 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1878 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1879 
1880 	evmcs->guest_es_base = vmcs12->guest_es_base;
1881 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1882 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1883 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1884 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1885 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1886 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1887 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1888 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1889 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1890 
1891 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1892 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1893 
1894 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1895 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1896 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1897 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1898 
1899 	evmcs->guest_pending_dbg_exceptions =
1900 		vmcs12->guest_pending_dbg_exceptions;
1901 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1902 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1903 
1904 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1905 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1906 
1907 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1908 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1909 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1910 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1911 
1912 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1913 
1914 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1915 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1916 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1917 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1918 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1919 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1920 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1921 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1922 
1923 	evmcs->exit_qualification = vmcs12->exit_qualification;
1924 
1925 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1926 	evmcs->guest_rsp = vmcs12->guest_rsp;
1927 	evmcs->guest_rflags = vmcs12->guest_rflags;
1928 
1929 	evmcs->guest_interruptibility_info =
1930 		vmcs12->guest_interruptibility_info;
1931 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1932 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1933 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1934 	evmcs->vm_entry_exception_error_code =
1935 		vmcs12->vm_entry_exception_error_code;
1936 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1937 
1938 	evmcs->guest_rip = vmcs12->guest_rip;
1939 
1940 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1941 
1942 	return;
1943 }
1944 
1945 /*
1946  * This is an equivalent of the nested hypervisor executing the vmptrld
1947  * instruction.
1948  */
1949 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1950 	struct kvm_vcpu *vcpu, bool from_launch)
1951 {
1952 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1953 	bool evmcs_gpa_changed = false;
1954 	u64 evmcs_gpa;
1955 
1956 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1957 		return EVMPTRLD_DISABLED;
1958 
1959 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa)) {
1960 		nested_release_evmcs(vcpu);
1961 		return EVMPTRLD_DISABLED;
1962 	}
1963 
1964 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1965 		vmx->nested.current_vmptr = INVALID_GPA;
1966 
1967 		nested_release_evmcs(vcpu);
1968 
1969 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1970 				 &vmx->nested.hv_evmcs_map))
1971 			return EVMPTRLD_ERROR;
1972 
1973 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1974 
1975 		/*
1976 		 * Currently, KVM only supports eVMCS version 1
1977 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1978 		 * value to first u32 field of eVMCS which should specify eVMCS
1979 		 * VersionNumber.
1980 		 *
1981 		 * Guest should be aware of supported eVMCS versions by host by
1982 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1983 		 * expected to set this CPUID leaf according to the value
1984 		 * returned in vmcs_version from nested_enable_evmcs().
1985 		 *
1986 		 * However, it turns out that Microsoft Hyper-V fails to comply
1987 		 * to their own invented interface: When Hyper-V use eVMCS, it
1988 		 * just sets first u32 field of eVMCS to revision_id specified
1989 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1990 		 * which is one of the supported versions specified in
1991 		 * CPUID.0x4000000A.EAX[0:15].
1992 		 *
1993 		 * To overcome Hyper-V bug, we accept here either a supported
1994 		 * eVMCS version or VMCS12 revision_id as valid values for first
1995 		 * u32 field of eVMCS.
1996 		 */
1997 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1998 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1999 			nested_release_evmcs(vcpu);
2000 			return EVMPTRLD_VMFAIL;
2001 		}
2002 
2003 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2004 
2005 		evmcs_gpa_changed = true;
2006 		/*
2007 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2008 		 * reloaded from guest's memory (read only fields, fields not
2009 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2010 		 * are no leftovers.
2011 		 */
2012 		if (from_launch) {
2013 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2014 			memset(vmcs12, 0, sizeof(*vmcs12));
2015 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2016 		}
2017 
2018 	}
2019 
2020 	/*
2021 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2022 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2023 	 */
2024 	if (from_launch || evmcs_gpa_changed)
2025 		vmx->nested.hv_evmcs->hv_clean_fields &=
2026 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2027 
2028 	return EVMPTRLD_SUCCEEDED;
2029 }
2030 
2031 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2032 {
2033 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2034 
2035 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2036 		copy_vmcs12_to_enlightened(vmx);
2037 	else
2038 		copy_vmcs12_to_shadow(vmx);
2039 
2040 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2041 }
2042 
2043 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2044 {
2045 	struct vcpu_vmx *vmx =
2046 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2047 
2048 	vmx->nested.preemption_timer_expired = true;
2049 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2050 	kvm_vcpu_kick(&vmx->vcpu);
2051 
2052 	return HRTIMER_NORESTART;
2053 }
2054 
2055 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2056 {
2057 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2058 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2059 
2060 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2061 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2062 
2063 	if (!vmx->nested.has_preemption_timer_deadline) {
2064 		vmx->nested.preemption_timer_deadline =
2065 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2066 		vmx->nested.has_preemption_timer_deadline = true;
2067 	}
2068 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2069 }
2070 
2071 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2072 					u64 preemption_timeout)
2073 {
2074 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2075 
2076 	/*
2077 	 * A timer value of zero is architecturally guaranteed to cause
2078 	 * a VMExit prior to executing any instructions in the guest.
2079 	 */
2080 	if (preemption_timeout == 0) {
2081 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2082 		return;
2083 	}
2084 
2085 	if (vcpu->arch.virtual_tsc_khz == 0)
2086 		return;
2087 
2088 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2089 	preemption_timeout *= 1000000;
2090 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2091 	hrtimer_start(&vmx->nested.preemption_timer,
2092 		      ktime_add_ns(ktime_get(), preemption_timeout),
2093 		      HRTIMER_MODE_ABS_PINNED);
2094 }
2095 
2096 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2097 {
2098 	if (vmx->nested.nested_run_pending &&
2099 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2100 		return vmcs12->guest_ia32_efer;
2101 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2102 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2103 	else
2104 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2105 }
2106 
2107 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2108 {
2109 	/*
2110 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2111 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2112 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2113 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2114 	 */
2115 	if (vmx->nested.vmcs02_initialized)
2116 		return;
2117 	vmx->nested.vmcs02_initialized = true;
2118 
2119 	/*
2120 	 * We don't care what the EPTP value is we just need to guarantee
2121 	 * it's valid so we don't get a false positive when doing early
2122 	 * consistency checks.
2123 	 */
2124 	if (enable_ept && nested_early_check)
2125 		vmcs_write64(EPT_POINTER,
2126 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2127 
2128 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2129 	if (cpu_has_vmx_vmfunc())
2130 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2131 
2132 	if (cpu_has_vmx_posted_intr())
2133 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2134 
2135 	if (cpu_has_vmx_msr_bitmap())
2136 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2137 
2138 	/*
2139 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2140 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2141 	 * deal with filtering out L2 GPAs from the buffer.
2142 	 */
2143 	if (enable_pml) {
2144 		vmcs_write64(PML_ADDRESS, 0);
2145 		vmcs_write16(GUEST_PML_INDEX, -1);
2146 	}
2147 
2148 	if (cpu_has_vmx_encls_vmexit())
2149 		vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2150 
2151 	/*
2152 	 * Set the MSR load/store lists to match L0's settings.  Only the
2153 	 * addresses are constant (for vmcs02), the counts can change based
2154 	 * on L2's behavior, e.g. switching to/from long mode.
2155 	 */
2156 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2157 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2158 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2159 
2160 	vmx_set_constant_host_state(vmx);
2161 }
2162 
2163 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2164 				      struct vmcs12 *vmcs12)
2165 {
2166 	prepare_vmcs02_constant_state(vmx);
2167 
2168 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2169 
2170 	if (enable_vpid) {
2171 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2172 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2173 		else
2174 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2175 	}
2176 }
2177 
2178 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2179 				 struct vmcs12 *vmcs12)
2180 {
2181 	u32 exec_control;
2182 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2183 
2184 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2185 		prepare_vmcs02_early_rare(vmx, vmcs12);
2186 
2187 	/*
2188 	 * PIN CONTROLS
2189 	 */
2190 	exec_control = __pin_controls_get(vmcs01);
2191 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2192 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2193 
2194 	/* Posted interrupts setting is only taken from vmcs12.  */
2195 	vmx->nested.pi_pending = false;
2196 	if (nested_cpu_has_posted_intr(vmcs12))
2197 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2198 	else
2199 		exec_control &= ~PIN_BASED_POSTED_INTR;
2200 	pin_controls_set(vmx, exec_control);
2201 
2202 	/*
2203 	 * EXEC CONTROLS
2204 	 */
2205 	exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2206 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2207 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2208 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2209 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2210 
2211 	vmx->nested.l1_tpr_threshold = -1;
2212 	if (exec_control & CPU_BASED_TPR_SHADOW)
2213 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2214 #ifdef CONFIG_X86_64
2215 	else
2216 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2217 				CPU_BASED_CR8_STORE_EXITING;
2218 #endif
2219 
2220 	/*
2221 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2222 	 * for I/O port accesses.
2223 	 */
2224 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2225 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2226 
2227 	/*
2228 	 * This bit will be computed in nested_get_vmcs12_pages, because
2229 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2230 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2231 	 * only set/clear this bit.
2232 	 */
2233 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2234 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2235 
2236 	exec_controls_set(vmx, exec_control);
2237 
2238 	/*
2239 	 * SECONDARY EXEC CONTROLS
2240 	 */
2241 	if (cpu_has_secondary_exec_ctrls()) {
2242 		exec_control = __secondary_exec_controls_get(vmcs01);
2243 
2244 		/* Take the following fields only from vmcs12 */
2245 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2246 				  SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2247 				  SECONDARY_EXEC_ENABLE_INVPCID |
2248 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2249 				  SECONDARY_EXEC_XSAVES |
2250 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2251 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2252 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2253 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2254 				  SECONDARY_EXEC_TSC_SCALING |
2255 				  SECONDARY_EXEC_DESC);
2256 
2257 		if (nested_cpu_has(vmcs12,
2258 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2259 			exec_control |= vmcs12->secondary_vm_exec_control;
2260 
2261 		/* PML is emulated and never enabled in hardware for L2. */
2262 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2263 
2264 		/* VMCS shadowing for L2 is emulated for now */
2265 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2266 
2267 		/*
2268 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2269 		 * will not have to rewrite the controls just for this bit.
2270 		 */
2271 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2272 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2273 			exec_control |= SECONDARY_EXEC_DESC;
2274 
2275 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2276 			vmcs_write16(GUEST_INTR_STATUS,
2277 				vmcs12->guest_intr_status);
2278 
2279 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2280 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2281 
2282 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2283 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2284 
2285 		secondary_exec_controls_set(vmx, exec_control);
2286 	}
2287 
2288 	/*
2289 	 * ENTRY CONTROLS
2290 	 *
2291 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2292 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2293 	 * on the related bits (if supported by the CPU) in the hope that
2294 	 * we can avoid VMWrites during vmx_set_efer().
2295 	 */
2296 	exec_control = __vm_entry_controls_get(vmcs01);
2297 	exec_control |= vmcs12->vm_entry_controls;
2298 	exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2299 	if (cpu_has_load_ia32_efer()) {
2300 		if (guest_efer & EFER_LMA)
2301 			exec_control |= VM_ENTRY_IA32E_MODE;
2302 		if (guest_efer != host_efer)
2303 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2304 	}
2305 	vm_entry_controls_set(vmx, exec_control);
2306 
2307 	/*
2308 	 * EXIT CONTROLS
2309 	 *
2310 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2311 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2312 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2313 	 */
2314 	exec_control = __vm_exit_controls_get(vmcs01);
2315 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2316 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2317 	else
2318 		exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2319 	vm_exit_controls_set(vmx, exec_control);
2320 
2321 	/*
2322 	 * Interrupt/Exception Fields
2323 	 */
2324 	if (vmx->nested.nested_run_pending) {
2325 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2326 			     vmcs12->vm_entry_intr_info_field);
2327 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2328 			     vmcs12->vm_entry_exception_error_code);
2329 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2330 			     vmcs12->vm_entry_instruction_len);
2331 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2332 			     vmcs12->guest_interruptibility_info);
2333 		vmx->loaded_vmcs->nmi_known_unmasked =
2334 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2335 	} else {
2336 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2337 	}
2338 }
2339 
2340 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2341 {
2342 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2343 
2344 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2345 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2346 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2347 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2348 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2349 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2350 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2351 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2352 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2353 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2354 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2355 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2356 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2357 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2358 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2359 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2360 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2361 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2362 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2363 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2364 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2365 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2366 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2367 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2368 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2369 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2370 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2371 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2372 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2373 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2374 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2375 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2376 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2377 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2378 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2379 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2380 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2381 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2382 
2383 		vmx->segment_cache.bitmask = 0;
2384 	}
2385 
2386 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2387 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2388 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2389 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2390 			    vmcs12->guest_pending_dbg_exceptions);
2391 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2392 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2393 
2394 		/*
2395 		 * L1 may access the L2's PDPTR, so save them to construct
2396 		 * vmcs12
2397 		 */
2398 		if (enable_ept) {
2399 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2400 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2401 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2402 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2403 		}
2404 
2405 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2406 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2407 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2408 	}
2409 
2410 	if (nested_cpu_has_xsaves(vmcs12))
2411 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2412 
2413 	/*
2414 	 * Whether page-faults are trapped is determined by a combination of
2415 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2416 	 * doesn't care about page faults then we should set all of these to
2417 	 * L1's desires. However, if L0 does care about (some) page faults, it
2418 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2419 	 * simply ask to exit on each and every L2 page fault. This is done by
2420 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2421 	 * Note that below we don't need special code to set EB.PF beyond the
2422 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2423 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2424 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2425 	 */
2426 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2427 		/*
2428 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2429 		 * go ahead and use it?
2430 		 */
2431 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2432 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2433 	} else {
2434 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2435 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2436 	}
2437 
2438 	if (cpu_has_vmx_apicv()) {
2439 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2440 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2441 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2442 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2443 	}
2444 
2445 	/*
2446 	 * Make sure the msr_autostore list is up to date before we set the
2447 	 * count in the vmcs02.
2448 	 */
2449 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2450 
2451 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2452 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2453 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2454 
2455 	set_cr4_guest_host_mask(vmx);
2456 }
2457 
2458 /*
2459  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2460  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2461  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2462  * guest in a way that will both be appropriate to L1's requests, and our
2463  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2464  * function also has additional necessary side-effects, like setting various
2465  * vcpu->arch fields.
2466  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2467  * is assigned to entry_failure_code on failure.
2468  */
2469 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2470 			  bool from_vmentry,
2471 			  enum vm_entry_failure_code *entry_failure_code)
2472 {
2473 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2474 	bool load_guest_pdptrs_vmcs12 = false;
2475 
2476 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2477 		prepare_vmcs02_rare(vmx, vmcs12);
2478 		vmx->nested.dirty_vmcs12 = false;
2479 
2480 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2481 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2482 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2483 	}
2484 
2485 	if (vmx->nested.nested_run_pending &&
2486 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2487 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2488 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2489 	} else {
2490 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2491 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2492 	}
2493 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2494 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2495 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2496 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2497 
2498 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2499 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2500 	 * trap. Note that CR0.TS also needs updating - we do this later.
2501 	 */
2502 	vmx_update_exception_bitmap(vcpu);
2503 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2504 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2505 
2506 	if (vmx->nested.nested_run_pending &&
2507 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2508 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2509 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2510 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2511 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2512 	}
2513 
2514 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2515 			vcpu->arch.l1_tsc_offset,
2516 			vmx_get_l2_tsc_offset(vcpu),
2517 			vmx_get_l2_tsc_multiplier(vcpu));
2518 
2519 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2520 			vcpu->arch.l1_tsc_scaling_ratio,
2521 			vmx_get_l2_tsc_multiplier(vcpu));
2522 
2523 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2524 	if (kvm_has_tsc_control)
2525 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2526 
2527 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2528 
2529 	if (nested_cpu_has_ept(vmcs12))
2530 		nested_ept_init_mmu_context(vcpu);
2531 
2532 	/*
2533 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2534 	 * bits which we consider mandatory enabled.
2535 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2536 	 * the specifications by L1; It's not enough to take
2537 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2538 	 * have more bits than L1 expected.
2539 	 */
2540 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2541 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2542 
2543 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2544 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2545 
2546 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2547 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2548 	vmx_set_efer(vcpu, vcpu->arch.efer);
2549 
2550 	/*
2551 	 * Guest state is invalid and unrestricted guest is disabled,
2552 	 * which means L1 attempted VMEntry to L2 with invalid state.
2553 	 * Fail the VMEntry.
2554 	 *
2555 	 * However when force loading the guest state (SMM exit or
2556 	 * loading nested state after migration, it is possible to
2557 	 * have invalid guest state now, which will be later fixed by
2558 	 * restoring L2 register state
2559 	 */
2560 	if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2561 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2562 		return -EINVAL;
2563 	}
2564 
2565 	/* Shadow page tables on either EPT or shadow page tables. */
2566 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2567 				from_vmentry, entry_failure_code))
2568 		return -EINVAL;
2569 
2570 	/*
2571 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2572 	 * on nested VM-Exit, which can occur without actually running L2 and
2573 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2574 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2575 	 * transition to HLT instead of running L2.
2576 	 */
2577 	if (enable_ept)
2578 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2579 
2580 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2581 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2582 	    is_pae_paging(vcpu)) {
2583 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2584 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2585 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2586 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2587 	}
2588 
2589 	if (!enable_ept)
2590 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2591 
2592 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2593 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2594 				     vmcs12->guest_ia32_perf_global_ctrl)))
2595 		return -EINVAL;
2596 
2597 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2598 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2599 
2600 	/*
2601 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2602 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2603 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2604 	 * here.
2605 	 */
2606 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2607 		vmx->nested.hv_evmcs->hv_clean_fields |=
2608 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2609 
2610 	return 0;
2611 }
2612 
2613 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2614 {
2615 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2616 	       nested_cpu_has_virtual_nmis(vmcs12)))
2617 		return -EINVAL;
2618 
2619 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2620 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2621 		return -EINVAL;
2622 
2623 	return 0;
2624 }
2625 
2626 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2627 {
2628 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2629 
2630 	/* Check for memory type validity */
2631 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2632 	case VMX_EPTP_MT_UC:
2633 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2634 			return false;
2635 		break;
2636 	case VMX_EPTP_MT_WB:
2637 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2638 			return false;
2639 		break;
2640 	default:
2641 		return false;
2642 	}
2643 
2644 	/* Page-walk levels validity. */
2645 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2646 	case VMX_EPTP_PWL_5:
2647 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2648 			return false;
2649 		break;
2650 	case VMX_EPTP_PWL_4:
2651 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2652 			return false;
2653 		break;
2654 	default:
2655 		return false;
2656 	}
2657 
2658 	/* Reserved bits should not be set */
2659 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2660 		return false;
2661 
2662 	/* AD, if set, should be supported */
2663 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2664 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2665 			return false;
2666 	}
2667 
2668 	return true;
2669 }
2670 
2671 /*
2672  * Checks related to VM-Execution Control Fields
2673  */
2674 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2675                                               struct vmcs12 *vmcs12)
2676 {
2677 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2678 
2679 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2680 				   vmx->nested.msrs.pinbased_ctls_low,
2681 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2682 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2683 				   vmx->nested.msrs.procbased_ctls_low,
2684 				   vmx->nested.msrs.procbased_ctls_high)))
2685 		return -EINVAL;
2686 
2687 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2688 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2689 				   vmx->nested.msrs.secondary_ctls_low,
2690 				   vmx->nested.msrs.secondary_ctls_high)))
2691 		return -EINVAL;
2692 
2693 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2694 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2695 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2696 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2697 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2698 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2699 	    nested_vmx_check_nmi_controls(vmcs12) ||
2700 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2701 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2702 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2703 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2704 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2705 		return -EINVAL;
2706 
2707 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2708 	    nested_cpu_has_save_preemption_timer(vmcs12))
2709 		return -EINVAL;
2710 
2711 	if (nested_cpu_has_ept(vmcs12) &&
2712 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2713 		return -EINVAL;
2714 
2715 	if (nested_cpu_has_vmfunc(vmcs12)) {
2716 		if (CC(vmcs12->vm_function_control &
2717 		       ~vmx->nested.msrs.vmfunc_controls))
2718 			return -EINVAL;
2719 
2720 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2721 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2722 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2723 				return -EINVAL;
2724 		}
2725 	}
2726 
2727 	return 0;
2728 }
2729 
2730 /*
2731  * Checks related to VM-Exit Control Fields
2732  */
2733 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2734                                          struct vmcs12 *vmcs12)
2735 {
2736 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2737 
2738 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2739 				    vmx->nested.msrs.exit_ctls_low,
2740 				    vmx->nested.msrs.exit_ctls_high)) ||
2741 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2742 		return -EINVAL;
2743 
2744 	return 0;
2745 }
2746 
2747 /*
2748  * Checks related to VM-Entry Control Fields
2749  */
2750 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2751 					  struct vmcs12 *vmcs12)
2752 {
2753 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2754 
2755 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2756 				    vmx->nested.msrs.entry_ctls_low,
2757 				    vmx->nested.msrs.entry_ctls_high)))
2758 		return -EINVAL;
2759 
2760 	/*
2761 	 * From the Intel SDM, volume 3:
2762 	 * Fields relevant to VM-entry event injection must be set properly.
2763 	 * These fields are the VM-entry interruption-information field, the
2764 	 * VM-entry exception error code, and the VM-entry instruction length.
2765 	 */
2766 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2767 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2768 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2769 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2770 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2771 		bool should_have_error_code;
2772 		bool urg = nested_cpu_has2(vmcs12,
2773 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2774 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2775 
2776 		/* VM-entry interruption-info field: interruption type */
2777 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2778 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2779 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2780 			return -EINVAL;
2781 
2782 		/* VM-entry interruption-info field: vector */
2783 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2784 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2785 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2786 			return -EINVAL;
2787 
2788 		/* VM-entry interruption-info field: deliver error code */
2789 		should_have_error_code =
2790 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2791 			x86_exception_has_error_code(vector);
2792 		if (CC(has_error_code != should_have_error_code))
2793 			return -EINVAL;
2794 
2795 		/* VM-entry exception error code */
2796 		if (CC(has_error_code &&
2797 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2798 			return -EINVAL;
2799 
2800 		/* VM-entry interruption-info field: reserved bits */
2801 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2802 			return -EINVAL;
2803 
2804 		/* VM-entry instruction length */
2805 		switch (intr_type) {
2806 		case INTR_TYPE_SOFT_EXCEPTION:
2807 		case INTR_TYPE_SOFT_INTR:
2808 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2809 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2810 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2811 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2812 				return -EINVAL;
2813 		}
2814 	}
2815 
2816 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2817 		return -EINVAL;
2818 
2819 	return 0;
2820 }
2821 
2822 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2823 				     struct vmcs12 *vmcs12)
2824 {
2825 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2826 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2827 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2828 		return -EINVAL;
2829 
2830 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2831 		return nested_evmcs_check_controls(vmcs12);
2832 
2833 	return 0;
2834 }
2835 
2836 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
2837 				       struct vmcs12 *vmcs12)
2838 {
2839 #ifdef CONFIG_X86_64
2840 	if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
2841 		!!(vcpu->arch.efer & EFER_LMA)))
2842 		return -EINVAL;
2843 #endif
2844 	return 0;
2845 }
2846 
2847 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2848 				       struct vmcs12 *vmcs12)
2849 {
2850 	bool ia32e;
2851 
2852 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2853 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2854 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2855 		return -EINVAL;
2856 
2857 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2858 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2859 		return -EINVAL;
2860 
2861 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2862 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2863 		return -EINVAL;
2864 
2865 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2866 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2867 					   vmcs12->host_ia32_perf_global_ctrl)))
2868 		return -EINVAL;
2869 
2870 #ifdef CONFIG_X86_64
2871 	ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
2872 #else
2873 	ia32e = false;
2874 #endif
2875 
2876 	if (ia32e) {
2877 		if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2878 			return -EINVAL;
2879 	} else {
2880 		if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2881 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2882 		    CC((vmcs12->host_rip) >> 32))
2883 			return -EINVAL;
2884 	}
2885 
2886 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2887 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2888 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2889 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2890 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2891 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2892 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2893 	    CC(vmcs12->host_cs_selector == 0) ||
2894 	    CC(vmcs12->host_tr_selector == 0) ||
2895 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2896 		return -EINVAL;
2897 
2898 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2899 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2900 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2901 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2902 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2903 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2904 		return -EINVAL;
2905 
2906 	/*
2907 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2908 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2909 	 * the values of the LMA and LME bits in the field must each be that of
2910 	 * the host address-space size VM-exit control.
2911 	 */
2912 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2913 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2914 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2915 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2916 			return -EINVAL;
2917 	}
2918 
2919 	return 0;
2920 }
2921 
2922 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2923 					  struct vmcs12 *vmcs12)
2924 {
2925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2926 	struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
2927 	struct vmcs_hdr hdr;
2928 
2929 	if (vmcs12->vmcs_link_pointer == INVALID_GPA)
2930 		return 0;
2931 
2932 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2933 		return -EINVAL;
2934 
2935 	if (ghc->gpa != vmcs12->vmcs_link_pointer &&
2936 	    CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
2937 					 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
2938                 return -EINVAL;
2939 
2940 	if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
2941 					    offsetof(struct vmcs12, hdr),
2942 					    sizeof(hdr))))
2943 		return -EINVAL;
2944 
2945 	if (CC(hdr.revision_id != VMCS12_REVISION) ||
2946 	    CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2947 		return -EINVAL;
2948 
2949 	return 0;
2950 }
2951 
2952 /*
2953  * Checks related to Guest Non-register State
2954  */
2955 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2956 {
2957 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2958 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2959 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2960 		return -EINVAL;
2961 
2962 	return 0;
2963 }
2964 
2965 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2966 					struct vmcs12 *vmcs12,
2967 					enum vm_entry_failure_code *entry_failure_code)
2968 {
2969 	bool ia32e;
2970 
2971 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
2972 
2973 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2974 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2975 		return -EINVAL;
2976 
2977 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2978 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2979 		return -EINVAL;
2980 
2981 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2982 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2983 		return -EINVAL;
2984 
2985 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2986 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
2987 		return -EINVAL;
2988 	}
2989 
2990 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2991 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2992 					   vmcs12->guest_ia32_perf_global_ctrl)))
2993 		return -EINVAL;
2994 
2995 	/*
2996 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2997 	 * are performed on the field for the IA32_EFER MSR:
2998 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2999 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3000 	 *   the IA-32e mode guest VM-exit control. It must also be identical
3001 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3002 	 *   CR0.PG) is 1.
3003 	 */
3004 	if (to_vmx(vcpu)->nested.nested_run_pending &&
3005 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3006 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
3007 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3008 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3009 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3010 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3011 			return -EINVAL;
3012 	}
3013 
3014 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3015 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3016 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3017 		return -EINVAL;
3018 
3019 	if (nested_check_guest_non_reg_state(vmcs12))
3020 		return -EINVAL;
3021 
3022 	return 0;
3023 }
3024 
3025 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3026 {
3027 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3028 	unsigned long cr3, cr4;
3029 	bool vm_fail;
3030 
3031 	if (!nested_early_check)
3032 		return 0;
3033 
3034 	if (vmx->msr_autoload.host.nr)
3035 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3036 	if (vmx->msr_autoload.guest.nr)
3037 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3038 
3039 	preempt_disable();
3040 
3041 	vmx_prepare_switch_to_guest(vcpu);
3042 
3043 	/*
3044 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3045 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3046 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3047 	 * there is no need to preserve other bits or save/restore the field.
3048 	 */
3049 	vmcs_writel(GUEST_RFLAGS, 0);
3050 
3051 	cr3 = __get_current_cr3_fast();
3052 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3053 		vmcs_writel(HOST_CR3, cr3);
3054 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3055 	}
3056 
3057 	cr4 = cr4_read_shadow();
3058 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3059 		vmcs_writel(HOST_CR4, cr4);
3060 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3061 	}
3062 
3063 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3064 				 vmx->loaded_vmcs->launched);
3065 
3066 	if (vmx->msr_autoload.host.nr)
3067 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3068 	if (vmx->msr_autoload.guest.nr)
3069 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3070 
3071 	if (vm_fail) {
3072 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3073 
3074 		preempt_enable();
3075 
3076 		trace_kvm_nested_vmenter_failed(
3077 			"early hardware check VM-instruction error: ", error);
3078 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3079 		return 1;
3080 	}
3081 
3082 	/*
3083 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3084 	 */
3085 	if (hw_breakpoint_active())
3086 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3087 	local_irq_enable();
3088 	preempt_enable();
3089 
3090 	/*
3091 	 * A non-failing VMEntry means we somehow entered guest mode with
3092 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3093 	 * is no telling what memory has been modified or what state has
3094 	 * been exposed to unknown code.  Hitting this all but guarantees
3095 	 * a (very critical) hardware issue.
3096 	 */
3097 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3098 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3099 
3100 	return 0;
3101 }
3102 
3103 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3104 {
3105 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3106 
3107 	/*
3108 	 * hv_evmcs may end up being not mapped after migration (when
3109 	 * L2 was running), map it here to make sure vmcs12 changes are
3110 	 * properly reflected.
3111 	 */
3112 	if (vmx->nested.enlightened_vmcs_enabled &&
3113 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3114 		enum nested_evmptrld_status evmptrld_status =
3115 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3116 
3117 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3118 		    evmptrld_status == EVMPTRLD_ERROR)
3119 			return false;
3120 
3121 		/*
3122 		 * Post migration VMCS12 always provides the most actual
3123 		 * information, copy it to eVMCS upon entry.
3124 		 */
3125 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3126 	}
3127 
3128 	return true;
3129 }
3130 
3131 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3132 {
3133 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3134 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3135 	struct kvm_host_map *map;
3136 	struct page *page;
3137 	u64 hpa;
3138 
3139 	if (!vcpu->arch.pdptrs_from_userspace &&
3140 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3141 		/*
3142 		 * Reload the guest's PDPTRs since after a migration
3143 		 * the guest CR3 might be restored prior to setting the nested
3144 		 * state which can lead to a load of wrong PDPTRs.
3145 		 */
3146 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3)))
3147 			return false;
3148 	}
3149 
3150 
3151 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3152 		/*
3153 		 * Translate L1 physical address to host physical
3154 		 * address for vmcs02. Keep the page pinned, so this
3155 		 * physical address remains valid. We keep a reference
3156 		 * to it so we can release it later.
3157 		 */
3158 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3159 			kvm_release_page_clean(vmx->nested.apic_access_page);
3160 			vmx->nested.apic_access_page = NULL;
3161 		}
3162 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3163 		if (!is_error_page(page)) {
3164 			vmx->nested.apic_access_page = page;
3165 			hpa = page_to_phys(vmx->nested.apic_access_page);
3166 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3167 		} else {
3168 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3169 					     __func__);
3170 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3171 			vcpu->run->internal.suberror =
3172 				KVM_INTERNAL_ERROR_EMULATION;
3173 			vcpu->run->internal.ndata = 0;
3174 			return false;
3175 		}
3176 	}
3177 
3178 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3179 		map = &vmx->nested.virtual_apic_map;
3180 
3181 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3182 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3183 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3184 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3185 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3186 			/*
3187 			 * The processor will never use the TPR shadow, simply
3188 			 * clear the bit from the execution control.  Such a
3189 			 * configuration is useless, but it happens in tests.
3190 			 * For any other configuration, failing the vm entry is
3191 			 * _not_ what the processor does but it's basically the
3192 			 * only possibility we have.
3193 			 */
3194 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3195 		} else {
3196 			/*
3197 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3198 			 * force VM-Entry to fail.
3199 			 */
3200 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3201 		}
3202 	}
3203 
3204 	if (nested_cpu_has_posted_intr(vmcs12)) {
3205 		map = &vmx->nested.pi_desc_map;
3206 
3207 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3208 			vmx->nested.pi_desc =
3209 				(struct pi_desc *)(((void *)map->hva) +
3210 				offset_in_page(vmcs12->posted_intr_desc_addr));
3211 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3212 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3213 		} else {
3214 			/*
3215 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3216 			 * access the contents of the VMCS12 posted interrupt
3217 			 * descriptor. (Note that KVM may do this when it
3218 			 * should not, per the architectural specification.)
3219 			 */
3220 			vmx->nested.pi_desc = NULL;
3221 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3222 		}
3223 	}
3224 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3225 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3226 	else
3227 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3228 
3229 	return true;
3230 }
3231 
3232 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3233 {
3234 	if (!nested_get_evmcs_page(vcpu)) {
3235 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3236 				     __func__);
3237 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3238 		vcpu->run->internal.suberror =
3239 			KVM_INTERNAL_ERROR_EMULATION;
3240 		vcpu->run->internal.ndata = 0;
3241 
3242 		return false;
3243 	}
3244 
3245 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3246 		return false;
3247 
3248 	return true;
3249 }
3250 
3251 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3252 {
3253 	struct vmcs12 *vmcs12;
3254 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3255 	gpa_t dst;
3256 
3257 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3258 		return 0;
3259 
3260 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3261 		return 1;
3262 
3263 	/*
3264 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3265 	 * set is already checked as part of A/D emulation.
3266 	 */
3267 	vmcs12 = get_vmcs12(vcpu);
3268 	if (!nested_cpu_has_pml(vmcs12))
3269 		return 0;
3270 
3271 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3272 		vmx->nested.pml_full = true;
3273 		return 1;
3274 	}
3275 
3276 	gpa &= ~0xFFFull;
3277 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3278 
3279 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3280 				 offset_in_page(dst), sizeof(gpa)))
3281 		return 0;
3282 
3283 	vmcs12->guest_pml_index--;
3284 
3285 	return 0;
3286 }
3287 
3288 /*
3289  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3290  * for running VMX instructions (except VMXON, whose prerequisites are
3291  * slightly different). It also specifies what exception to inject otherwise.
3292  * Note that many of these exceptions have priority over VM exits, so they
3293  * don't have to be checked again here.
3294  */
3295 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3296 {
3297 	if (!to_vmx(vcpu)->nested.vmxon) {
3298 		kvm_queue_exception(vcpu, UD_VECTOR);
3299 		return 0;
3300 	}
3301 
3302 	if (vmx_get_cpl(vcpu)) {
3303 		kvm_inject_gp(vcpu, 0);
3304 		return 0;
3305 	}
3306 
3307 	return 1;
3308 }
3309 
3310 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3311 {
3312 	u8 rvi = vmx_get_rvi();
3313 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3314 
3315 	return ((rvi & 0xf0) > (vppr & 0xf0));
3316 }
3317 
3318 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3319 				   struct vmcs12 *vmcs12);
3320 
3321 /*
3322  * If from_vmentry is false, this is being called from state restore (either RSM
3323  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3324  *
3325  * Returns:
3326  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3327  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3328  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3329  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3330  */
3331 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3332 							bool from_vmentry)
3333 {
3334 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3335 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3336 	enum vm_entry_failure_code entry_failure_code;
3337 	bool evaluate_pending_interrupts;
3338 	union vmx_exit_reason exit_reason = {
3339 		.basic = EXIT_REASON_INVALID_STATE,
3340 		.failed_vmentry = 1,
3341 	};
3342 	u32 failed_index;
3343 
3344 	kvm_service_local_tlb_flush_requests(vcpu);
3345 
3346 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3347 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3348 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3349 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3350 
3351 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3352 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3353 	if (kvm_mpx_supported() &&
3354 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3355 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3356 
3357 	/*
3358 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3359 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3360 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3361 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3362 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3363 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3364 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3365 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3366 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3367 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3368 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3369 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3370 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3371 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3372 	 */
3373 	if (!enable_ept && !nested_early_check)
3374 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3375 
3376 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3377 
3378 	prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3379 
3380 	if (from_vmentry) {
3381 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3382 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3383 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3384 		}
3385 
3386 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3387 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3388 			return NVMX_VMENTRY_VMFAIL;
3389 		}
3390 
3391 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3392 						 &entry_failure_code)) {
3393 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3394 			vmcs12->exit_qualification = entry_failure_code;
3395 			goto vmentry_fail_vmexit;
3396 		}
3397 	}
3398 
3399 	enter_guest_mode(vcpu);
3400 
3401 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3402 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3403 		vmcs12->exit_qualification = entry_failure_code;
3404 		goto vmentry_fail_vmexit_guest_mode;
3405 	}
3406 
3407 	if (from_vmentry) {
3408 		failed_index = nested_vmx_load_msr(vcpu,
3409 						   vmcs12->vm_entry_msr_load_addr,
3410 						   vmcs12->vm_entry_msr_load_count);
3411 		if (failed_index) {
3412 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3413 			vmcs12->exit_qualification = failed_index;
3414 			goto vmentry_fail_vmexit_guest_mode;
3415 		}
3416 	} else {
3417 		/*
3418 		 * The MMU is not initialized to point at the right entities yet and
3419 		 * "get pages" would need to read data from the guest (i.e. we will
3420 		 * need to perform gpa to hpa translation). Request a call
3421 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3422 		 * have already been set at vmentry time and should not be reset.
3423 		 */
3424 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3425 	}
3426 
3427 	/*
3428 	 * If L1 had a pending IRQ/NMI until it executed
3429 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3430 	 * disallowed (e.g. interrupts disabled), L0 needs to
3431 	 * evaluate if this pending event should cause an exit from L2
3432 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3433 	 * intercept EXTERNAL_INTERRUPT).
3434 	 *
3435 	 * Usually this would be handled by the processor noticing an
3436 	 * IRQ/NMI window request, or checking RVI during evaluation of
3437 	 * pending virtual interrupts.  However, this setting was done
3438 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3439 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3440 	 */
3441 	if (unlikely(evaluate_pending_interrupts))
3442 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3443 
3444 	/*
3445 	 * Do not start the preemption timer hrtimer until after we know
3446 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3447 	 * the timer.
3448 	 */
3449 	vmx->nested.preemption_timer_expired = false;
3450 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3451 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3452 		vmx_start_preemption_timer(vcpu, timer_value);
3453 	}
3454 
3455 	/*
3456 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3457 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3458 	 * returned as far as L1 is concerned. It will only return (and set
3459 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3460 	 */
3461 	return NVMX_VMENTRY_SUCCESS;
3462 
3463 	/*
3464 	 * A failed consistency check that leads to a VMExit during L1's
3465 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3466 	 * 26.7 "VM-entry failures during or after loading guest state".
3467 	 */
3468 vmentry_fail_vmexit_guest_mode:
3469 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3470 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3471 	leave_guest_mode(vcpu);
3472 
3473 vmentry_fail_vmexit:
3474 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3475 
3476 	if (!from_vmentry)
3477 		return NVMX_VMENTRY_VMEXIT;
3478 
3479 	load_vmcs12_host_state(vcpu, vmcs12);
3480 	vmcs12->vm_exit_reason = exit_reason.full;
3481 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3482 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3483 	return NVMX_VMENTRY_VMEXIT;
3484 }
3485 
3486 /*
3487  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3488  * for running an L2 nested guest.
3489  */
3490 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3491 {
3492 	struct vmcs12 *vmcs12;
3493 	enum nvmx_vmentry_status status;
3494 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3495 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3496 	enum nested_evmptrld_status evmptrld_status;
3497 
3498 	if (!nested_vmx_check_permission(vcpu))
3499 		return 1;
3500 
3501 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3502 	if (evmptrld_status == EVMPTRLD_ERROR) {
3503 		kvm_queue_exception(vcpu, UD_VECTOR);
3504 		return 1;
3505 	} else if (CC(evmptrld_status == EVMPTRLD_VMFAIL)) {
3506 		return nested_vmx_failInvalid(vcpu);
3507 	}
3508 
3509 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3510 	       vmx->nested.current_vmptr == INVALID_GPA))
3511 		return nested_vmx_failInvalid(vcpu);
3512 
3513 	vmcs12 = get_vmcs12(vcpu);
3514 
3515 	/*
3516 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3517 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3518 	 * rather than RFLAGS.ZF, and no error number is stored to the
3519 	 * VM-instruction error field.
3520 	 */
3521 	if (CC(vmcs12->hdr.shadow_vmcs))
3522 		return nested_vmx_failInvalid(vcpu);
3523 
3524 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3525 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3526 		/* Enlightened VMCS doesn't have launch state */
3527 		vmcs12->launch_state = !launch;
3528 	} else if (enable_shadow_vmcs) {
3529 		copy_shadow_to_vmcs12(vmx);
3530 	}
3531 
3532 	/*
3533 	 * The nested entry process starts with enforcing various prerequisites
3534 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3535 	 * they fail: As the SDM explains, some conditions should cause the
3536 	 * instruction to fail, while others will cause the instruction to seem
3537 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3538 	 * To speed up the normal (success) code path, we should avoid checking
3539 	 * for misconfigurations which will anyway be caught by the processor
3540 	 * when using the merged vmcs02.
3541 	 */
3542 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3543 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3544 
3545 	if (CC(vmcs12->launch_state == launch))
3546 		return nested_vmx_fail(vcpu,
3547 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3548 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3549 
3550 	if (nested_vmx_check_controls(vcpu, vmcs12))
3551 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3552 
3553 	if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3554 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3555 
3556 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3557 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3558 
3559 	/*
3560 	 * We're finally done with prerequisite checking, and can start with
3561 	 * the nested entry.
3562 	 */
3563 	vmx->nested.nested_run_pending = 1;
3564 	vmx->nested.has_preemption_timer_deadline = false;
3565 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3566 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3567 		goto vmentry_failed;
3568 
3569 	/* Emulate processing of posted interrupts on VM-Enter. */
3570 	if (nested_cpu_has_posted_intr(vmcs12) &&
3571 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3572 		vmx->nested.pi_pending = true;
3573 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3574 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3575 	}
3576 
3577 	/* Hide L1D cache contents from the nested guest.  */
3578 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3579 
3580 	/*
3581 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3582 	 * also be used as part of restoring nVMX state for
3583 	 * snapshot restore (migration).
3584 	 *
3585 	 * In this flow, it is assumed that vmcs12 cache was
3586 	 * transferred as part of captured nVMX state and should
3587 	 * therefore not be read from guest memory (which may not
3588 	 * exist on destination host yet).
3589 	 */
3590 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3591 
3592 	switch (vmcs12->guest_activity_state) {
3593 	case GUEST_ACTIVITY_HLT:
3594 		/*
3595 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3596 		 * awakened by event injection or by an NMI-window VM-exit or
3597 		 * by an interrupt-window VM-exit, halt the vcpu.
3598 		 */
3599 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3600 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3601 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3602 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3603 			vmx->nested.nested_run_pending = 0;
3604 			return kvm_vcpu_halt(vcpu);
3605 		}
3606 		break;
3607 	case GUEST_ACTIVITY_WAIT_SIPI:
3608 		vmx->nested.nested_run_pending = 0;
3609 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3610 		break;
3611 	default:
3612 		break;
3613 	}
3614 
3615 	return 1;
3616 
3617 vmentry_failed:
3618 	vmx->nested.nested_run_pending = 0;
3619 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3620 		return 0;
3621 	if (status == NVMX_VMENTRY_VMEXIT)
3622 		return 1;
3623 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3624 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3625 }
3626 
3627 /*
3628  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3629  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3630  * This function returns the new value we should put in vmcs12.guest_cr0.
3631  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3632  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3633  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3634  *     didn't trap the bit, because if L1 did, so would L0).
3635  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3636  *     been modified by L2, and L1 knows it. So just leave the old value of
3637  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3638  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3639  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3640  *     changed these bits, and therefore they need to be updated, but L0
3641  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3642  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3643  */
3644 static inline unsigned long
3645 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3646 {
3647 	return
3648 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3649 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3650 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3651 			vcpu->arch.cr0_guest_owned_bits));
3652 }
3653 
3654 static inline unsigned long
3655 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3656 {
3657 	return
3658 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3659 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3660 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3661 			vcpu->arch.cr4_guest_owned_bits));
3662 }
3663 
3664 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3665 				      struct vmcs12 *vmcs12)
3666 {
3667 	u32 idt_vectoring;
3668 	unsigned int nr;
3669 
3670 	if (vcpu->arch.exception.injected) {
3671 		nr = vcpu->arch.exception.nr;
3672 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3673 
3674 		if (kvm_exception_is_soft(nr)) {
3675 			vmcs12->vm_exit_instruction_len =
3676 				vcpu->arch.event_exit_inst_len;
3677 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3678 		} else
3679 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3680 
3681 		if (vcpu->arch.exception.has_error_code) {
3682 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3683 			vmcs12->idt_vectoring_error_code =
3684 				vcpu->arch.exception.error_code;
3685 		}
3686 
3687 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3688 	} else if (vcpu->arch.nmi_injected) {
3689 		vmcs12->idt_vectoring_info_field =
3690 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3691 	} else if (vcpu->arch.interrupt.injected) {
3692 		nr = vcpu->arch.interrupt.nr;
3693 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3694 
3695 		if (vcpu->arch.interrupt.soft) {
3696 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3697 			vmcs12->vm_entry_instruction_len =
3698 				vcpu->arch.event_exit_inst_len;
3699 		} else
3700 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3701 
3702 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3703 	}
3704 }
3705 
3706 
3707 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3708 {
3709 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3710 	gfn_t gfn;
3711 
3712 	/*
3713 	 * Don't need to mark the APIC access page dirty; it is never
3714 	 * written to by the CPU during APIC virtualization.
3715 	 */
3716 
3717 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3718 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3719 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3720 	}
3721 
3722 	if (nested_cpu_has_posted_intr(vmcs12)) {
3723 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3724 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3725 	}
3726 }
3727 
3728 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3729 {
3730 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3731 	int max_irr;
3732 	void *vapic_page;
3733 	u16 status;
3734 
3735 	if (!vmx->nested.pi_pending)
3736 		return 0;
3737 
3738 	if (!vmx->nested.pi_desc)
3739 		goto mmio_needed;
3740 
3741 	vmx->nested.pi_pending = false;
3742 
3743 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3744 		return 0;
3745 
3746 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3747 	if (max_irr != 256) {
3748 		vapic_page = vmx->nested.virtual_apic_map.hva;
3749 		if (!vapic_page)
3750 			goto mmio_needed;
3751 
3752 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3753 			vapic_page, &max_irr);
3754 		status = vmcs_read16(GUEST_INTR_STATUS);
3755 		if ((u8)max_irr > ((u8)status & 0xff)) {
3756 			status &= ~0xff;
3757 			status |= (u8)max_irr;
3758 			vmcs_write16(GUEST_INTR_STATUS, status);
3759 		}
3760 	}
3761 
3762 	nested_mark_vmcs12_pages_dirty(vcpu);
3763 	return 0;
3764 
3765 mmio_needed:
3766 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3767 	return -ENXIO;
3768 }
3769 
3770 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3771 					       unsigned long exit_qual)
3772 {
3773 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3774 	unsigned int nr = vcpu->arch.exception.nr;
3775 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3776 
3777 	if (vcpu->arch.exception.has_error_code) {
3778 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3779 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3780 	}
3781 
3782 	if (kvm_exception_is_soft(nr))
3783 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3784 	else
3785 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3786 
3787 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3788 	    vmx_get_nmi_mask(vcpu))
3789 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3790 
3791 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3792 }
3793 
3794 /*
3795  * Returns true if a debug trap is pending delivery.
3796  *
3797  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3798  * exception may be inferred from the presence of an exception payload.
3799  */
3800 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3801 {
3802 	return vcpu->arch.exception.pending &&
3803 			vcpu->arch.exception.nr == DB_VECTOR &&
3804 			vcpu->arch.exception.payload;
3805 }
3806 
3807 /*
3808  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3809  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3810  * represents these debug traps with a payload that is said to be compatible
3811  * with the 'pending debug exceptions' field, write the payload to the VMCS
3812  * field if a VM-exit is delivered before the debug trap.
3813  */
3814 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3815 {
3816 	if (vmx_pending_dbg_trap(vcpu))
3817 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3818 			    vcpu->arch.exception.payload);
3819 }
3820 
3821 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3822 {
3823 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3824 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3825 }
3826 
3827 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3828 {
3829 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3830 	unsigned long exit_qual;
3831 	bool block_nested_events =
3832 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3833 	bool mtf_pending = vmx->nested.mtf_pending;
3834 	struct kvm_lapic *apic = vcpu->arch.apic;
3835 
3836 	/*
3837 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3838 	 * this state is discarded.
3839 	 */
3840 	if (!block_nested_events)
3841 		vmx->nested.mtf_pending = false;
3842 
3843 	if (lapic_in_kernel(vcpu) &&
3844 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3845 		if (block_nested_events)
3846 			return -EBUSY;
3847 		nested_vmx_update_pending_dbg(vcpu);
3848 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3849 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3850 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3851 		return 0;
3852 	}
3853 
3854 	if (lapic_in_kernel(vcpu) &&
3855 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3856 		if (block_nested_events)
3857 			return -EBUSY;
3858 
3859 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3860 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3861 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3862 						apic->sipi_vector & 0xFFUL);
3863 		return 0;
3864 	}
3865 
3866 	/*
3867 	 * Process any exceptions that are not debug traps before MTF.
3868 	 *
3869 	 * Note that only a pending nested run can block a pending exception.
3870 	 * Otherwise an injected NMI/interrupt should either be
3871 	 * lost or delivered to the nested hypervisor in the IDT_VECTORING_INFO,
3872 	 * while delivering the pending exception.
3873 	 */
3874 
3875 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3876 		if (vmx->nested.nested_run_pending)
3877 			return -EBUSY;
3878 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3879 			goto no_vmexit;
3880 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3881 		return 0;
3882 	}
3883 
3884 	if (mtf_pending) {
3885 		if (block_nested_events)
3886 			return -EBUSY;
3887 		nested_vmx_update_pending_dbg(vcpu);
3888 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3889 		return 0;
3890 	}
3891 
3892 	if (vcpu->arch.exception.pending) {
3893 		if (vmx->nested.nested_run_pending)
3894 			return -EBUSY;
3895 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3896 			goto no_vmexit;
3897 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3898 		return 0;
3899 	}
3900 
3901 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3902 		if (block_nested_events)
3903 			return -EBUSY;
3904 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3905 		return 0;
3906 	}
3907 
3908 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3909 		if (block_nested_events)
3910 			return -EBUSY;
3911 		goto no_vmexit;
3912 	}
3913 
3914 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3915 		if (block_nested_events)
3916 			return -EBUSY;
3917 		if (!nested_exit_on_nmi(vcpu))
3918 			goto no_vmexit;
3919 
3920 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3921 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3922 				  INTR_INFO_VALID_MASK, 0);
3923 		/*
3924 		 * The NMI-triggered VM exit counts as injection:
3925 		 * clear this one and block further NMIs.
3926 		 */
3927 		vcpu->arch.nmi_pending = 0;
3928 		vmx_set_nmi_mask(vcpu, true);
3929 		return 0;
3930 	}
3931 
3932 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3933 		if (block_nested_events)
3934 			return -EBUSY;
3935 		if (!nested_exit_on_intr(vcpu))
3936 			goto no_vmexit;
3937 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3938 		return 0;
3939 	}
3940 
3941 no_vmexit:
3942 	return vmx_complete_nested_posted_interrupt(vcpu);
3943 }
3944 
3945 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3946 {
3947 	ktime_t remaining =
3948 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3949 	u64 value;
3950 
3951 	if (ktime_to_ns(remaining) <= 0)
3952 		return 0;
3953 
3954 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3955 	do_div(value, 1000000);
3956 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3957 }
3958 
3959 static bool is_vmcs12_ext_field(unsigned long field)
3960 {
3961 	switch (field) {
3962 	case GUEST_ES_SELECTOR:
3963 	case GUEST_CS_SELECTOR:
3964 	case GUEST_SS_SELECTOR:
3965 	case GUEST_DS_SELECTOR:
3966 	case GUEST_FS_SELECTOR:
3967 	case GUEST_GS_SELECTOR:
3968 	case GUEST_LDTR_SELECTOR:
3969 	case GUEST_TR_SELECTOR:
3970 	case GUEST_ES_LIMIT:
3971 	case GUEST_CS_LIMIT:
3972 	case GUEST_SS_LIMIT:
3973 	case GUEST_DS_LIMIT:
3974 	case GUEST_FS_LIMIT:
3975 	case GUEST_GS_LIMIT:
3976 	case GUEST_LDTR_LIMIT:
3977 	case GUEST_TR_LIMIT:
3978 	case GUEST_GDTR_LIMIT:
3979 	case GUEST_IDTR_LIMIT:
3980 	case GUEST_ES_AR_BYTES:
3981 	case GUEST_DS_AR_BYTES:
3982 	case GUEST_FS_AR_BYTES:
3983 	case GUEST_GS_AR_BYTES:
3984 	case GUEST_LDTR_AR_BYTES:
3985 	case GUEST_TR_AR_BYTES:
3986 	case GUEST_ES_BASE:
3987 	case GUEST_CS_BASE:
3988 	case GUEST_SS_BASE:
3989 	case GUEST_DS_BASE:
3990 	case GUEST_FS_BASE:
3991 	case GUEST_GS_BASE:
3992 	case GUEST_LDTR_BASE:
3993 	case GUEST_TR_BASE:
3994 	case GUEST_GDTR_BASE:
3995 	case GUEST_IDTR_BASE:
3996 	case GUEST_PENDING_DBG_EXCEPTIONS:
3997 	case GUEST_BNDCFGS:
3998 		return true;
3999 	default:
4000 		break;
4001 	}
4002 
4003 	return false;
4004 }
4005 
4006 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4007 				       struct vmcs12 *vmcs12)
4008 {
4009 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4010 
4011 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4012 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4013 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4014 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4015 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4016 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4017 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4018 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4019 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4020 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4021 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4022 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4023 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4024 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4025 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4026 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4027 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4028 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4029 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4030 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4031 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4032 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4033 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4034 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4035 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4036 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4037 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4038 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4039 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4040 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4041 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4042 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4043 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4044 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4045 	vmcs12->guest_pending_dbg_exceptions =
4046 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4047 	if (kvm_mpx_supported())
4048 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
4049 
4050 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4051 }
4052 
4053 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4054 				       struct vmcs12 *vmcs12)
4055 {
4056 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4057 	int cpu;
4058 
4059 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4060 		return;
4061 
4062 
4063 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4064 
4065 	cpu = get_cpu();
4066 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4067 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4068 
4069 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4070 
4071 	vmx->loaded_vmcs = &vmx->vmcs01;
4072 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4073 	put_cpu();
4074 }
4075 
4076 /*
4077  * Update the guest state fields of vmcs12 to reflect changes that
4078  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4079  * VM-entry controls is also updated, since this is really a guest
4080  * state bit.)
4081  */
4082 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4083 {
4084 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4085 
4086 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4087 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4088 
4089 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4090 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4091 
4092 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4093 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4094 
4095 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4096 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4097 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4098 
4099 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4100 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4101 
4102 	vmcs12->guest_interruptibility_info =
4103 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4104 
4105 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4106 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4107 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4108 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4109 	else
4110 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4111 
4112 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4113 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4114 	    !vmx->nested.nested_run_pending)
4115 		vmcs12->vmx_preemption_timer_value =
4116 			vmx_get_preemption_timer_value(vcpu);
4117 
4118 	/*
4119 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4120 	 * own CR3 without exiting. If it has changed it, we must keep it.
4121 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4122 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4123 	 *
4124 	 * Additionally, restore L2's PDPTR to vmcs12.
4125 	 */
4126 	if (enable_ept) {
4127 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4128 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4129 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4130 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4131 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4132 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4133 		}
4134 	}
4135 
4136 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4137 
4138 	if (nested_cpu_has_vid(vmcs12))
4139 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4140 
4141 	vmcs12->vm_entry_controls =
4142 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4143 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4144 
4145 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4146 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4147 
4148 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4149 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4150 }
4151 
4152 /*
4153  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4154  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4155  * and this function updates it to reflect the changes to the guest state while
4156  * L2 was running (and perhaps made some exits which were handled directly by L0
4157  * without going back to L1), and to reflect the exit reason.
4158  * Note that we do not have to copy here all VMCS fields, just those that
4159  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4160  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4161  * which already writes to vmcs12 directly.
4162  */
4163 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4164 			   u32 vm_exit_reason, u32 exit_intr_info,
4165 			   unsigned long exit_qualification)
4166 {
4167 	/* update exit information fields: */
4168 	vmcs12->vm_exit_reason = vm_exit_reason;
4169 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4170 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4171 	vmcs12->exit_qualification = exit_qualification;
4172 	vmcs12->vm_exit_intr_info = exit_intr_info;
4173 
4174 	vmcs12->idt_vectoring_info_field = 0;
4175 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4176 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4177 
4178 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4179 		vmcs12->launch_state = 1;
4180 
4181 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4182 		 * instead of reading the real value. */
4183 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4184 
4185 		/*
4186 		 * Transfer the event that L0 or L1 may wanted to inject into
4187 		 * L2 to IDT_VECTORING_INFO_FIELD.
4188 		 */
4189 		vmcs12_save_pending_event(vcpu, vmcs12);
4190 
4191 		/*
4192 		 * According to spec, there's no need to store the guest's
4193 		 * MSRs if the exit is due to a VM-entry failure that occurs
4194 		 * during or after loading the guest state. Since this exit
4195 		 * does not fall in that category, we need to save the MSRs.
4196 		 */
4197 		if (nested_vmx_store_msr(vcpu,
4198 					 vmcs12->vm_exit_msr_store_addr,
4199 					 vmcs12->vm_exit_msr_store_count))
4200 			nested_vmx_abort(vcpu,
4201 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4202 	}
4203 
4204 	/*
4205 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4206 	 * preserved above and would only end up incorrectly in L1.
4207 	 */
4208 	vcpu->arch.nmi_injected = false;
4209 	kvm_clear_exception_queue(vcpu);
4210 	kvm_clear_interrupt_queue(vcpu);
4211 }
4212 
4213 /*
4214  * A part of what we need to when the nested L2 guest exits and we want to
4215  * run its L1 parent, is to reset L1's guest state to the host state specified
4216  * in vmcs12.
4217  * This function is to be called not only on normal nested exit, but also on
4218  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4219  * Failures During or After Loading Guest State").
4220  * This function should be called when the active VMCS is L1's (vmcs01).
4221  */
4222 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4223 				   struct vmcs12 *vmcs12)
4224 {
4225 	enum vm_entry_failure_code ignored;
4226 	struct kvm_segment seg;
4227 
4228 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4229 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4230 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4231 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4232 	else
4233 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4234 	vmx_set_efer(vcpu, vcpu->arch.efer);
4235 
4236 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4237 	kvm_rip_write(vcpu, vmcs12->host_rip);
4238 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4239 	vmx_set_interrupt_shadow(vcpu, 0);
4240 
4241 	/*
4242 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4243 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4244 	 *
4245 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4246 	 * (KVM doesn't change it);
4247 	 */
4248 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4249 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4250 
4251 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4252 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4253 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4254 
4255 	nested_ept_uninit_mmu_context(vcpu);
4256 
4257 	/*
4258 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4259 	 * couldn't have changed.
4260 	 */
4261 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4262 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4263 
4264 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4265 
4266 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4267 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4268 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4269 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4270 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4271 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4272 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4273 
4274 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4275 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4276 		vmcs_write64(GUEST_BNDCFGS, 0);
4277 
4278 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4279 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4280 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4281 	}
4282 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4283 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4284 					 vmcs12->host_ia32_perf_global_ctrl));
4285 
4286 	/* Set L1 segment info according to Intel SDM
4287 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4288 	seg = (struct kvm_segment) {
4289 		.base = 0,
4290 		.limit = 0xFFFFFFFF,
4291 		.selector = vmcs12->host_cs_selector,
4292 		.type = 11,
4293 		.present = 1,
4294 		.s = 1,
4295 		.g = 1
4296 	};
4297 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4298 		seg.l = 1;
4299 	else
4300 		seg.db = 1;
4301 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4302 	seg = (struct kvm_segment) {
4303 		.base = 0,
4304 		.limit = 0xFFFFFFFF,
4305 		.type = 3,
4306 		.present = 1,
4307 		.s = 1,
4308 		.db = 1,
4309 		.g = 1
4310 	};
4311 	seg.selector = vmcs12->host_ds_selector;
4312 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4313 	seg.selector = vmcs12->host_es_selector;
4314 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4315 	seg.selector = vmcs12->host_ss_selector;
4316 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4317 	seg.selector = vmcs12->host_fs_selector;
4318 	seg.base = vmcs12->host_fs_base;
4319 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4320 	seg.selector = vmcs12->host_gs_selector;
4321 	seg.base = vmcs12->host_gs_base;
4322 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4323 	seg = (struct kvm_segment) {
4324 		.base = vmcs12->host_tr_base,
4325 		.limit = 0x67,
4326 		.selector = vmcs12->host_tr_selector,
4327 		.type = 11,
4328 		.present = 1
4329 	};
4330 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4331 
4332 	memset(&seg, 0, sizeof(seg));
4333 	seg.unusable = 1;
4334 	__vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4335 
4336 	kvm_set_dr(vcpu, 7, 0x400);
4337 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4338 
4339 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4340 				vmcs12->vm_exit_msr_load_count))
4341 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4342 
4343 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4344 }
4345 
4346 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4347 {
4348 	struct vmx_uret_msr *efer_msr;
4349 	unsigned int i;
4350 
4351 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4352 		return vmcs_read64(GUEST_IA32_EFER);
4353 
4354 	if (cpu_has_load_ia32_efer())
4355 		return host_efer;
4356 
4357 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4358 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4359 			return vmx->msr_autoload.guest.val[i].value;
4360 	}
4361 
4362 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4363 	if (efer_msr)
4364 		return efer_msr->data;
4365 
4366 	return host_efer;
4367 }
4368 
4369 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4370 {
4371 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4372 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4373 	struct vmx_msr_entry g, h;
4374 	gpa_t gpa;
4375 	u32 i, j;
4376 
4377 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4378 
4379 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4380 		/*
4381 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4382 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4383 		 * and vcpu->arch.dr7 is not squirreled away before the
4384 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4385 		 */
4386 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4387 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4388 		else
4389 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4390 	}
4391 
4392 	/*
4393 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4394 	 * handle a variety of side effects to KVM's software model.
4395 	 */
4396 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4397 
4398 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4399 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4400 
4401 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4402 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4403 
4404 	nested_ept_uninit_mmu_context(vcpu);
4405 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4406 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4407 
4408 	/*
4409 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4410 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4411 	 * VMFail, like everything else we just need to ensure our
4412 	 * software model is up-to-date.
4413 	 */
4414 	if (enable_ept && is_pae_paging(vcpu))
4415 		ept_save_pdptrs(vcpu);
4416 
4417 	kvm_mmu_reset_context(vcpu);
4418 
4419 	/*
4420 	 * This nasty bit of open coding is a compromise between blindly
4421 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4422 	 * of VMFail), leaving the nested VM's MSRs in the software model
4423 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4424 	 * expensive since the lists are unbound by hardware).  For each
4425 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4426 	 * list, reload it from the exit load list if it exists and differs
4427 	 * from the guest value.  The intent is to stuff host state as
4428 	 * silently as possible, not to fully process the exit load list.
4429 	 */
4430 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4431 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4432 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4433 			pr_debug_ratelimited(
4434 				"%s read MSR index failed (%u, 0x%08llx)\n",
4435 				__func__, i, gpa);
4436 			goto vmabort;
4437 		}
4438 
4439 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4440 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4441 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4442 				pr_debug_ratelimited(
4443 					"%s read MSR failed (%u, 0x%08llx)\n",
4444 					__func__, j, gpa);
4445 				goto vmabort;
4446 			}
4447 			if (h.index != g.index)
4448 				continue;
4449 			if (h.value == g.value)
4450 				break;
4451 
4452 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4453 				pr_debug_ratelimited(
4454 					"%s check failed (%u, 0x%x, 0x%x)\n",
4455 					__func__, j, h.index, h.reserved);
4456 				goto vmabort;
4457 			}
4458 
4459 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4460 				pr_debug_ratelimited(
4461 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4462 					__func__, j, h.index, h.value);
4463 				goto vmabort;
4464 			}
4465 		}
4466 	}
4467 
4468 	return;
4469 
4470 vmabort:
4471 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4472 }
4473 
4474 /*
4475  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4476  * and modify vmcs12 to make it see what it would expect to see there if
4477  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4478  */
4479 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4480 		       u32 exit_intr_info, unsigned long exit_qualification)
4481 {
4482 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4483 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4484 
4485 	/* trying to cancel vmlaunch/vmresume is a bug */
4486 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4487 
4488 	/* Similarly, triple faults in L2 should never escape. */
4489 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
4490 
4491 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4492 		/*
4493 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4494 		 * Enlightened VMCS after migration and we still need to
4495 		 * do that when something is forcing L2->L1 exit prior to
4496 		 * the first L2 run.
4497 		 */
4498 		(void)nested_get_evmcs_page(vcpu);
4499 	}
4500 
4501 	/* Service pending TLB flush requests for L2 before switching to L1. */
4502 	kvm_service_local_tlb_flush_requests(vcpu);
4503 
4504 	/*
4505 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4506 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4507 	 * up-to-date before switching to L1.
4508 	 */
4509 	if (enable_ept && is_pae_paging(vcpu))
4510 		vmx_ept_load_pdptrs(vcpu);
4511 
4512 	leave_guest_mode(vcpu);
4513 
4514 	if (nested_cpu_has_preemption_timer(vmcs12))
4515 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4516 
4517 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4518 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4519 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4520 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4521 	}
4522 
4523 	if (likely(!vmx->fail)) {
4524 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4525 
4526 		if (vm_exit_reason != -1)
4527 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4528 				       exit_intr_info, exit_qualification);
4529 
4530 		/*
4531 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4532 		 * also be used to capture vmcs12 cache as part of
4533 		 * capturing nVMX state for snapshot (migration).
4534 		 *
4535 		 * Otherwise, this flush will dirty guest memory at a
4536 		 * point it is already assumed by user-space to be
4537 		 * immutable.
4538 		 */
4539 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4540 	} else {
4541 		/*
4542 		 * The only expected VM-instruction error is "VM entry with
4543 		 * invalid control field(s)." Anything else indicates a
4544 		 * problem with L0.  And we should never get here with a
4545 		 * VMFail of any type if early consistency checks are enabled.
4546 		 */
4547 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4548 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4549 		WARN_ON_ONCE(nested_early_check);
4550 	}
4551 
4552 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4553 
4554 	/* Update any VMCS fields that might have changed while L2 ran */
4555 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4556 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4557 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4558 	if (kvm_has_tsc_control)
4559 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4560 
4561 	if (vmx->nested.l1_tpr_threshold != -1)
4562 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4563 
4564 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4565 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4566 		vmx_set_virtual_apic_mode(vcpu);
4567 	}
4568 
4569 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4570 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4571 		vmx_update_cpu_dirty_logging(vcpu);
4572 	}
4573 
4574 	/* Unpin physical memory we referred to in vmcs02 */
4575 	if (vmx->nested.apic_access_page) {
4576 		kvm_release_page_clean(vmx->nested.apic_access_page);
4577 		vmx->nested.apic_access_page = NULL;
4578 	}
4579 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4580 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4581 	vmx->nested.pi_desc = NULL;
4582 
4583 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4584 		vmx->nested.reload_vmcs01_apic_access_page = false;
4585 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4586 	}
4587 
4588 	if ((vm_exit_reason != -1) &&
4589 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4590 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4591 
4592 	/* in case we halted in L2 */
4593 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4594 
4595 	if (likely(!vmx->fail)) {
4596 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4597 		    nested_exit_intr_ack_set(vcpu)) {
4598 			int irq = kvm_cpu_get_interrupt(vcpu);
4599 			WARN_ON(irq < 0);
4600 			vmcs12->vm_exit_intr_info = irq |
4601 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4602 		}
4603 
4604 		if (vm_exit_reason != -1)
4605 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4606 						       vmcs12->exit_qualification,
4607 						       vmcs12->idt_vectoring_info_field,
4608 						       vmcs12->vm_exit_intr_info,
4609 						       vmcs12->vm_exit_intr_error_code,
4610 						       KVM_ISA_VMX);
4611 
4612 		load_vmcs12_host_state(vcpu, vmcs12);
4613 
4614 		return;
4615 	}
4616 
4617 	/*
4618 	 * After an early L2 VM-entry failure, we're now back
4619 	 * in L1 which thinks it just finished a VMLAUNCH or
4620 	 * VMRESUME instruction, so we need to set the failure
4621 	 * flag and the VM-instruction error field of the VMCS
4622 	 * accordingly, and skip the emulated instruction.
4623 	 */
4624 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4625 
4626 	/*
4627 	 * Restore L1's host state to KVM's software model.  We're here
4628 	 * because a consistency check was caught by hardware, which
4629 	 * means some amount of guest state has been propagated to KVM's
4630 	 * model and needs to be unwound to the host's state.
4631 	 */
4632 	nested_vmx_restore_host_state(vcpu);
4633 
4634 	vmx->fail = 0;
4635 }
4636 
4637 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4638 {
4639 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4640 }
4641 
4642 /*
4643  * Decode the memory-address operand of a vmx instruction, as recorded on an
4644  * exit caused by such an instruction (run by a guest hypervisor).
4645  * On success, returns 0. When the operand is invalid, returns 1 and throws
4646  * #UD, #GP, or #SS.
4647  */
4648 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4649 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4650 {
4651 	gva_t off;
4652 	bool exn;
4653 	struct kvm_segment s;
4654 
4655 	/*
4656 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4657 	 * Execution", on an exit, vmx_instruction_info holds most of the
4658 	 * addressing components of the operand. Only the displacement part
4659 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4660 	 * For how an actual address is calculated from all these components,
4661 	 * refer to Vol. 1, "Operand Addressing".
4662 	 */
4663 	int  scaling = vmx_instruction_info & 3;
4664 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4665 	bool is_reg = vmx_instruction_info & (1u << 10);
4666 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4667 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4668 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4669 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4670 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4671 
4672 	if (is_reg) {
4673 		kvm_queue_exception(vcpu, UD_VECTOR);
4674 		return 1;
4675 	}
4676 
4677 	/* Addr = segment_base + offset */
4678 	/* offset = base + [index * scale] + displacement */
4679 	off = exit_qualification; /* holds the displacement */
4680 	if (addr_size == 1)
4681 		off = (gva_t)sign_extend64(off, 31);
4682 	else if (addr_size == 0)
4683 		off = (gva_t)sign_extend64(off, 15);
4684 	if (base_is_valid)
4685 		off += kvm_register_read(vcpu, base_reg);
4686 	if (index_is_valid)
4687 		off += kvm_register_read(vcpu, index_reg) << scaling;
4688 	vmx_get_segment(vcpu, &s, seg_reg);
4689 
4690 	/*
4691 	 * The effective address, i.e. @off, of a memory operand is truncated
4692 	 * based on the address size of the instruction.  Note that this is
4693 	 * the *effective address*, i.e. the address prior to accounting for
4694 	 * the segment's base.
4695 	 */
4696 	if (addr_size == 1) /* 32 bit */
4697 		off &= 0xffffffff;
4698 	else if (addr_size == 0) /* 16 bit */
4699 		off &= 0xffff;
4700 
4701 	/* Checks for #GP/#SS exceptions. */
4702 	exn = false;
4703 	if (is_long_mode(vcpu)) {
4704 		/*
4705 		 * The virtual/linear address is never truncated in 64-bit
4706 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4707 		 * address when using FS/GS with a non-zero base.
4708 		 */
4709 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4710 			*ret = s.base + off;
4711 		else
4712 			*ret = off;
4713 
4714 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4715 		 * non-canonical form. This is the only check on the memory
4716 		 * destination for long mode!
4717 		 */
4718 		exn = is_noncanonical_address(*ret, vcpu);
4719 	} else {
4720 		/*
4721 		 * When not in long mode, the virtual/linear address is
4722 		 * unconditionally truncated to 32 bits regardless of the
4723 		 * address size.
4724 		 */
4725 		*ret = (s.base + off) & 0xffffffff;
4726 
4727 		/* Protected mode: apply checks for segment validity in the
4728 		 * following order:
4729 		 * - segment type check (#GP(0) may be thrown)
4730 		 * - usability check (#GP(0)/#SS(0))
4731 		 * - limit check (#GP(0)/#SS(0))
4732 		 */
4733 		if (wr)
4734 			/* #GP(0) if the destination operand is located in a
4735 			 * read-only data segment or any code segment.
4736 			 */
4737 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4738 		else
4739 			/* #GP(0) if the source operand is located in an
4740 			 * execute-only code segment
4741 			 */
4742 			exn = ((s.type & 0xa) == 8);
4743 		if (exn) {
4744 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4745 			return 1;
4746 		}
4747 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4748 		 */
4749 		exn = (s.unusable != 0);
4750 
4751 		/*
4752 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4753 		 * outside the segment limit.  All CPUs that support VMX ignore
4754 		 * limit checks for flat segments, i.e. segments with base==0,
4755 		 * limit==0xffffffff and of type expand-up data or code.
4756 		 */
4757 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4758 		     ((s.type & 8) || !(s.type & 4))))
4759 			exn = exn || ((u64)off + len - 1 > s.limit);
4760 	}
4761 	if (exn) {
4762 		kvm_queue_exception_e(vcpu,
4763 				      seg_reg == VCPU_SREG_SS ?
4764 						SS_VECTOR : GP_VECTOR,
4765 				      0);
4766 		return 1;
4767 	}
4768 
4769 	return 0;
4770 }
4771 
4772 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4773 {
4774 	struct vcpu_vmx *vmx;
4775 
4776 	if (!nested_vmx_allowed(vcpu))
4777 		return;
4778 
4779 	vmx = to_vmx(vcpu);
4780 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4781 		vmx->nested.msrs.entry_ctls_high |=
4782 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4783 		vmx->nested.msrs.exit_ctls_high |=
4784 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4785 	} else {
4786 		vmx->nested.msrs.entry_ctls_high &=
4787 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4788 		vmx->nested.msrs.exit_ctls_high &=
4789 				~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4790 	}
4791 }
4792 
4793 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4794 				int *ret)
4795 {
4796 	gva_t gva;
4797 	struct x86_exception e;
4798 	int r;
4799 
4800 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4801 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4802 				sizeof(*vmpointer), &gva)) {
4803 		*ret = 1;
4804 		return -EINVAL;
4805 	}
4806 
4807 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4808 	if (r != X86EMUL_CONTINUE) {
4809 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4810 		return -EINVAL;
4811 	}
4812 
4813 	return 0;
4814 }
4815 
4816 /*
4817  * Allocate a shadow VMCS and associate it with the currently loaded
4818  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4819  * VMCS is also VMCLEARed, so that it is ready for use.
4820  */
4821 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4822 {
4823 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4824 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4825 
4826 	/*
4827 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4828 	 * executes VMXON and free it when L1 executes VMXOFF.
4829 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4830 	 * here when vmcs01 already have an allocated shadow vmcs.
4831 	 */
4832 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4833 
4834 	if (!loaded_vmcs->shadow_vmcs) {
4835 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4836 		if (loaded_vmcs->shadow_vmcs)
4837 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4838 	}
4839 	return loaded_vmcs->shadow_vmcs;
4840 }
4841 
4842 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4843 {
4844 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4845 	int r;
4846 
4847 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4848 	if (r < 0)
4849 		goto out_vmcs02;
4850 
4851 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4852 	if (!vmx->nested.cached_vmcs12)
4853 		goto out_cached_vmcs12;
4854 
4855 	vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
4856 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4857 	if (!vmx->nested.cached_shadow_vmcs12)
4858 		goto out_cached_shadow_vmcs12;
4859 
4860 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4861 		goto out_shadow_vmcs;
4862 
4863 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4864 		     HRTIMER_MODE_ABS_PINNED);
4865 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4866 
4867 	vmx->nested.vpid02 = allocate_vpid();
4868 
4869 	vmx->nested.vmcs02_initialized = false;
4870 	vmx->nested.vmxon = true;
4871 
4872 	if (vmx_pt_mode_is_host_guest()) {
4873 		vmx->pt_desc.guest.ctl = 0;
4874 		pt_update_intercept_for_msr(vcpu);
4875 	}
4876 
4877 	return 0;
4878 
4879 out_shadow_vmcs:
4880 	kfree(vmx->nested.cached_shadow_vmcs12);
4881 
4882 out_cached_shadow_vmcs12:
4883 	kfree(vmx->nested.cached_vmcs12);
4884 
4885 out_cached_vmcs12:
4886 	free_loaded_vmcs(&vmx->nested.vmcs02);
4887 
4888 out_vmcs02:
4889 	return -ENOMEM;
4890 }
4891 
4892 /* Emulate the VMXON instruction. */
4893 static int handle_vmon(struct kvm_vcpu *vcpu)
4894 {
4895 	int ret;
4896 	gpa_t vmptr;
4897 	uint32_t revision;
4898 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4899 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4900 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4901 
4902 	/*
4903 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4904 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4905 	 * 1 (see vmx_is_valid_cr4() for when we allow the guest to set this).
4906 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4907 	 * have already been checked by hardware, prior to the VM-exit for
4908 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4909 	 * that bit set to 1 in non-root mode.
4910 	 */
4911 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4912 		kvm_queue_exception(vcpu, UD_VECTOR);
4913 		return 1;
4914 	}
4915 
4916 	/* CPL=0 must be checked manually. */
4917 	if (vmx_get_cpl(vcpu)) {
4918 		kvm_inject_gp(vcpu, 0);
4919 		return 1;
4920 	}
4921 
4922 	if (vmx->nested.vmxon)
4923 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4924 
4925 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4926 			!= VMXON_NEEDED_FEATURES) {
4927 		kvm_inject_gp(vcpu, 0);
4928 		return 1;
4929 	}
4930 
4931 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4932 		return ret;
4933 
4934 	/*
4935 	 * SDM 3: 24.11.5
4936 	 * The first 4 bytes of VMXON region contain the supported
4937 	 * VMCS revision identifier
4938 	 *
4939 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4940 	 * which replaces physical address width with 32
4941 	 */
4942 	if (!page_address_valid(vcpu, vmptr))
4943 		return nested_vmx_failInvalid(vcpu);
4944 
4945 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4946 	    revision != VMCS12_REVISION)
4947 		return nested_vmx_failInvalid(vcpu);
4948 
4949 	vmx->nested.vmxon_ptr = vmptr;
4950 	ret = enter_vmx_operation(vcpu);
4951 	if (ret)
4952 		return ret;
4953 
4954 	return nested_vmx_succeed(vcpu);
4955 }
4956 
4957 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4958 {
4959 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4960 
4961 	if (vmx->nested.current_vmptr == INVALID_GPA)
4962 		return;
4963 
4964 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4965 
4966 	if (enable_shadow_vmcs) {
4967 		/* copy to memory all shadowed fields in case
4968 		   they were modified */
4969 		copy_shadow_to_vmcs12(vmx);
4970 		vmx_disable_shadow_vmcs(vmx);
4971 	}
4972 	vmx->nested.posted_intr_nv = -1;
4973 
4974 	/* Flush VMCS12 to guest memory */
4975 	kvm_vcpu_write_guest_page(vcpu,
4976 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4977 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4978 
4979 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4980 
4981 	vmx->nested.current_vmptr = INVALID_GPA;
4982 }
4983 
4984 /* Emulate the VMXOFF instruction */
4985 static int handle_vmoff(struct kvm_vcpu *vcpu)
4986 {
4987 	if (!nested_vmx_check_permission(vcpu))
4988 		return 1;
4989 
4990 	free_nested(vcpu);
4991 
4992 	/* Process a latched INIT during time CPU was in VMX operation */
4993 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4994 
4995 	return nested_vmx_succeed(vcpu);
4996 }
4997 
4998 /* Emulate the VMCLEAR instruction */
4999 static int handle_vmclear(struct kvm_vcpu *vcpu)
5000 {
5001 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5002 	u32 zero = 0;
5003 	gpa_t vmptr;
5004 	u64 evmcs_gpa;
5005 	int r;
5006 
5007 	if (!nested_vmx_check_permission(vcpu))
5008 		return 1;
5009 
5010 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5011 		return r;
5012 
5013 	if (!page_address_valid(vcpu, vmptr))
5014 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5015 
5016 	if (vmptr == vmx->nested.vmxon_ptr)
5017 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5018 
5019 	/*
5020 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
5021 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
5022 	 * way to distinguish it from VMCS12) and we must not corrupt it by
5023 	 * writing to the non-existent 'launch_state' field. The area doesn't
5024 	 * have to be the currently active EVMCS on the calling CPU and there's
5025 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5026 	 * state. It is possible that the area will stay mapped as
5027 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5028 	 */
5029 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
5030 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
5031 		if (vmptr == vmx->nested.current_vmptr)
5032 			nested_release_vmcs12(vcpu);
5033 
5034 		kvm_vcpu_write_guest(vcpu,
5035 				     vmptr + offsetof(struct vmcs12,
5036 						      launch_state),
5037 				     &zero, sizeof(zero));
5038 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5039 		nested_release_evmcs(vcpu);
5040 	}
5041 
5042 	return nested_vmx_succeed(vcpu);
5043 }
5044 
5045 /* Emulate the VMLAUNCH instruction */
5046 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5047 {
5048 	return nested_vmx_run(vcpu, true);
5049 }
5050 
5051 /* Emulate the VMRESUME instruction */
5052 static int handle_vmresume(struct kvm_vcpu *vcpu)
5053 {
5054 
5055 	return nested_vmx_run(vcpu, false);
5056 }
5057 
5058 static int handle_vmread(struct kvm_vcpu *vcpu)
5059 {
5060 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5061 						    : get_vmcs12(vcpu);
5062 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5063 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5064 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5065 	struct x86_exception e;
5066 	unsigned long field;
5067 	u64 value;
5068 	gva_t gva = 0;
5069 	short offset;
5070 	int len, r;
5071 
5072 	if (!nested_vmx_check_permission(vcpu))
5073 		return 1;
5074 
5075 	/*
5076 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5077 	 * any VMREAD sets the ALU flags for VMfailInvalid.
5078 	 */
5079 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5080 	    (is_guest_mode(vcpu) &&
5081 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5082 		return nested_vmx_failInvalid(vcpu);
5083 
5084 	/* Decode instruction info and find the field to read */
5085 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5086 
5087 	offset = vmcs_field_to_offset(field);
5088 	if (offset < 0)
5089 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5090 
5091 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5092 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5093 
5094 	/* Read the field, zero-extended to a u64 value */
5095 	value = vmcs12_read_any(vmcs12, field, offset);
5096 
5097 	/*
5098 	 * Now copy part of this value to register or memory, as requested.
5099 	 * Note that the number of bits actually copied is 32 or 64 depending
5100 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5101 	 */
5102 	if (instr_info & BIT(10)) {
5103 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5104 	} else {
5105 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5106 		if (get_vmx_mem_address(vcpu, exit_qualification,
5107 					instr_info, true, len, &gva))
5108 			return 1;
5109 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5110 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5111 		if (r != X86EMUL_CONTINUE)
5112 			return kvm_handle_memory_failure(vcpu, r, &e);
5113 	}
5114 
5115 	return nested_vmx_succeed(vcpu);
5116 }
5117 
5118 static bool is_shadow_field_rw(unsigned long field)
5119 {
5120 	switch (field) {
5121 #define SHADOW_FIELD_RW(x, y) case x:
5122 #include "vmcs_shadow_fields.h"
5123 		return true;
5124 	default:
5125 		break;
5126 	}
5127 	return false;
5128 }
5129 
5130 static bool is_shadow_field_ro(unsigned long field)
5131 {
5132 	switch (field) {
5133 #define SHADOW_FIELD_RO(x, y) case x:
5134 #include "vmcs_shadow_fields.h"
5135 		return true;
5136 	default:
5137 		break;
5138 	}
5139 	return false;
5140 }
5141 
5142 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5143 {
5144 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5145 						    : get_vmcs12(vcpu);
5146 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5147 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5148 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5149 	struct x86_exception e;
5150 	unsigned long field;
5151 	short offset;
5152 	gva_t gva;
5153 	int len, r;
5154 
5155 	/*
5156 	 * The value to write might be 32 or 64 bits, depending on L1's long
5157 	 * mode, and eventually we need to write that into a field of several
5158 	 * possible lengths. The code below first zero-extends the value to 64
5159 	 * bit (value), and then copies only the appropriate number of
5160 	 * bits into the vmcs12 field.
5161 	 */
5162 	u64 value = 0;
5163 
5164 	if (!nested_vmx_check_permission(vcpu))
5165 		return 1;
5166 
5167 	/*
5168 	 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5169 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5170 	 */
5171 	if (vmx->nested.current_vmptr == INVALID_GPA ||
5172 	    (is_guest_mode(vcpu) &&
5173 	     get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5174 		return nested_vmx_failInvalid(vcpu);
5175 
5176 	if (instr_info & BIT(10))
5177 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5178 	else {
5179 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5180 		if (get_vmx_mem_address(vcpu, exit_qualification,
5181 					instr_info, false, len, &gva))
5182 			return 1;
5183 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5184 		if (r != X86EMUL_CONTINUE)
5185 			return kvm_handle_memory_failure(vcpu, r, &e);
5186 	}
5187 
5188 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5189 
5190 	offset = vmcs_field_to_offset(field);
5191 	if (offset < 0)
5192 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5193 
5194 	/*
5195 	 * If the vCPU supports "VMWRITE to any supported field in the
5196 	 * VMCS," then the "read-only" fields are actually read/write.
5197 	 */
5198 	if (vmcs_field_readonly(field) &&
5199 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5200 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5201 
5202 	/*
5203 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5204 	 * vmcs12, else we may crush a field or consume a stale value.
5205 	 */
5206 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5207 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5208 
5209 	/*
5210 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5211 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5212 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5213 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5214 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5215 	 * the stripped down value, L2 sees the full value as stored by KVM).
5216 	 */
5217 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5218 		value &= 0x1f0ff;
5219 
5220 	vmcs12_write_any(vmcs12, field, offset, value);
5221 
5222 	/*
5223 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5224 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5225 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5226 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5227 	 */
5228 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5229 		/*
5230 		 * L1 can read these fields without exiting, ensure the
5231 		 * shadow VMCS is up-to-date.
5232 		 */
5233 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5234 			preempt_disable();
5235 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5236 
5237 			__vmcs_writel(field, value);
5238 
5239 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5240 			vmcs_load(vmx->loaded_vmcs->vmcs);
5241 			preempt_enable();
5242 		}
5243 		vmx->nested.dirty_vmcs12 = true;
5244 	}
5245 
5246 	return nested_vmx_succeed(vcpu);
5247 }
5248 
5249 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5250 {
5251 	vmx->nested.current_vmptr = vmptr;
5252 	if (enable_shadow_vmcs) {
5253 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5254 		vmcs_write64(VMCS_LINK_POINTER,
5255 			     __pa(vmx->vmcs01.shadow_vmcs));
5256 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5257 	}
5258 	vmx->nested.dirty_vmcs12 = true;
5259 }
5260 
5261 /* Emulate the VMPTRLD instruction */
5262 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5263 {
5264 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5265 	gpa_t vmptr;
5266 	int r;
5267 
5268 	if (!nested_vmx_check_permission(vcpu))
5269 		return 1;
5270 
5271 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5272 		return r;
5273 
5274 	if (!page_address_valid(vcpu, vmptr))
5275 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5276 
5277 	if (vmptr == vmx->nested.vmxon_ptr)
5278 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5279 
5280 	/* Forbid normal VMPTRLD if Enlightened version was used */
5281 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5282 		return 1;
5283 
5284 	if (vmx->nested.current_vmptr != vmptr) {
5285 		struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5286 		struct vmcs_hdr hdr;
5287 
5288 		if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5289 			/*
5290 			 * Reads from an unbacked page return all 1s,
5291 			 * which means that the 32 bits located at the
5292 			 * given physical address won't match the required
5293 			 * VMCS12_REVISION identifier.
5294 			 */
5295 			return nested_vmx_fail(vcpu,
5296 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5297 		}
5298 
5299 		if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5300 						 offsetof(struct vmcs12, hdr),
5301 						 sizeof(hdr))) {
5302 			return nested_vmx_fail(vcpu,
5303 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5304 		}
5305 
5306 		if (hdr.revision_id != VMCS12_REVISION ||
5307 		    (hdr.shadow_vmcs &&
5308 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5309 			return nested_vmx_fail(vcpu,
5310 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5311 		}
5312 
5313 		nested_release_vmcs12(vcpu);
5314 
5315 		/*
5316 		 * Load VMCS12 from guest memory since it is not already
5317 		 * cached.
5318 		 */
5319 		if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5320 					  VMCS12_SIZE)) {
5321 			return nested_vmx_fail(vcpu,
5322 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5323 		}
5324 
5325 		set_current_vmptr(vmx, vmptr);
5326 	}
5327 
5328 	return nested_vmx_succeed(vcpu);
5329 }
5330 
5331 /* Emulate the VMPTRST instruction */
5332 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5333 {
5334 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5335 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5336 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5337 	struct x86_exception e;
5338 	gva_t gva;
5339 	int r;
5340 
5341 	if (!nested_vmx_check_permission(vcpu))
5342 		return 1;
5343 
5344 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5345 		return 1;
5346 
5347 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5348 				true, sizeof(gpa_t), &gva))
5349 		return 1;
5350 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5351 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5352 					sizeof(gpa_t), &e);
5353 	if (r != X86EMUL_CONTINUE)
5354 		return kvm_handle_memory_failure(vcpu, r, &e);
5355 
5356 	return nested_vmx_succeed(vcpu);
5357 }
5358 
5359 /* Emulate the INVEPT instruction */
5360 static int handle_invept(struct kvm_vcpu *vcpu)
5361 {
5362 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5363 	u32 vmx_instruction_info, types;
5364 	unsigned long type, roots_to_free;
5365 	struct kvm_mmu *mmu;
5366 	gva_t gva;
5367 	struct x86_exception e;
5368 	struct {
5369 		u64 eptp, gpa;
5370 	} operand;
5371 	int i, r, gpr_index;
5372 
5373 	if (!(vmx->nested.msrs.secondary_ctls_high &
5374 	      SECONDARY_EXEC_ENABLE_EPT) ||
5375 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5376 		kvm_queue_exception(vcpu, UD_VECTOR);
5377 		return 1;
5378 	}
5379 
5380 	if (!nested_vmx_check_permission(vcpu))
5381 		return 1;
5382 
5383 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5384 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5385 	type = kvm_register_read(vcpu, gpr_index);
5386 
5387 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5388 
5389 	if (type >= 32 || !(types & (1 << type)))
5390 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5391 
5392 	/* According to the Intel VMX instruction reference, the memory
5393 	 * operand is read even if it isn't needed (e.g., for type==global)
5394 	 */
5395 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5396 			vmx_instruction_info, false, sizeof(operand), &gva))
5397 		return 1;
5398 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5399 	if (r != X86EMUL_CONTINUE)
5400 		return kvm_handle_memory_failure(vcpu, r, &e);
5401 
5402 	/*
5403 	 * Nested EPT roots are always held through guest_mmu,
5404 	 * not root_mmu.
5405 	 */
5406 	mmu = &vcpu->arch.guest_mmu;
5407 
5408 	switch (type) {
5409 	case VMX_EPT_EXTENT_CONTEXT:
5410 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5411 			return nested_vmx_fail(vcpu,
5412 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5413 
5414 		roots_to_free = 0;
5415 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
5416 					    operand.eptp))
5417 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5418 
5419 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5420 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5421 						    mmu->prev_roots[i].pgd,
5422 						    operand.eptp))
5423 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5424 		}
5425 		break;
5426 	case VMX_EPT_EXTENT_GLOBAL:
5427 		roots_to_free = KVM_MMU_ROOTS_ALL;
5428 		break;
5429 	default:
5430 		BUG();
5431 		break;
5432 	}
5433 
5434 	if (roots_to_free)
5435 		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
5436 
5437 	return nested_vmx_succeed(vcpu);
5438 }
5439 
5440 static int handle_invvpid(struct kvm_vcpu *vcpu)
5441 {
5442 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5443 	u32 vmx_instruction_info;
5444 	unsigned long type, types;
5445 	gva_t gva;
5446 	struct x86_exception e;
5447 	struct {
5448 		u64 vpid;
5449 		u64 gla;
5450 	} operand;
5451 	u16 vpid02;
5452 	int r, gpr_index;
5453 
5454 	if (!(vmx->nested.msrs.secondary_ctls_high &
5455 	      SECONDARY_EXEC_ENABLE_VPID) ||
5456 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5457 		kvm_queue_exception(vcpu, UD_VECTOR);
5458 		return 1;
5459 	}
5460 
5461 	if (!nested_vmx_check_permission(vcpu))
5462 		return 1;
5463 
5464 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5465 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5466 	type = kvm_register_read(vcpu, gpr_index);
5467 
5468 	types = (vmx->nested.msrs.vpid_caps &
5469 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5470 
5471 	if (type >= 32 || !(types & (1 << type)))
5472 		return nested_vmx_fail(vcpu,
5473 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5474 
5475 	/* according to the intel vmx instruction reference, the memory
5476 	 * operand is read even if it isn't needed (e.g., for type==global)
5477 	 */
5478 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5479 			vmx_instruction_info, false, sizeof(operand), &gva))
5480 		return 1;
5481 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5482 	if (r != X86EMUL_CONTINUE)
5483 		return kvm_handle_memory_failure(vcpu, r, &e);
5484 
5485 	if (operand.vpid >> 16)
5486 		return nested_vmx_fail(vcpu,
5487 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5488 
5489 	vpid02 = nested_get_vpid02(vcpu);
5490 	switch (type) {
5491 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5492 		if (!operand.vpid ||
5493 		    is_noncanonical_address(operand.gla, vcpu))
5494 			return nested_vmx_fail(vcpu,
5495 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5496 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5497 		break;
5498 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5499 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5500 		if (!operand.vpid)
5501 			return nested_vmx_fail(vcpu,
5502 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5503 		vpid_sync_context(vpid02);
5504 		break;
5505 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5506 		vpid_sync_context(vpid02);
5507 		break;
5508 	default:
5509 		WARN_ON_ONCE(1);
5510 		return kvm_skip_emulated_instruction(vcpu);
5511 	}
5512 
5513 	/*
5514 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5515 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5516 	 * roots as VPIDs are not tracked in the MMU role.
5517 	 *
5518 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5519 	 * an MMU when EPT is disabled.
5520 	 *
5521 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5522 	 */
5523 	if (!enable_ept)
5524 		kvm_mmu_free_guest_mode_roots(vcpu, &vcpu->arch.root_mmu);
5525 
5526 	return nested_vmx_succeed(vcpu);
5527 }
5528 
5529 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5530 				     struct vmcs12 *vmcs12)
5531 {
5532 	u32 index = kvm_rcx_read(vcpu);
5533 	u64 new_eptp;
5534 
5535 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5536 		return 1;
5537 	if (index >= VMFUNC_EPTP_ENTRIES)
5538 		return 1;
5539 
5540 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5541 				     &new_eptp, index * 8, 8))
5542 		return 1;
5543 
5544 	/*
5545 	 * If the (L2) guest does a vmfunc to the currently
5546 	 * active ept pointer, we don't have to do anything else
5547 	 */
5548 	if (vmcs12->ept_pointer != new_eptp) {
5549 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5550 			return 1;
5551 
5552 		vmcs12->ept_pointer = new_eptp;
5553 		nested_ept_new_eptp(vcpu);
5554 
5555 		if (!nested_cpu_has_vpid(vmcs12))
5556 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5557 	}
5558 
5559 	return 0;
5560 }
5561 
5562 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5563 {
5564 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5565 	struct vmcs12 *vmcs12;
5566 	u32 function = kvm_rax_read(vcpu);
5567 
5568 	/*
5569 	 * VMFUNC is only supported for nested guests, but we always enable the
5570 	 * secondary control for simplicity; for non-nested mode, fake that we
5571 	 * didn't by injecting #UD.
5572 	 */
5573 	if (!is_guest_mode(vcpu)) {
5574 		kvm_queue_exception(vcpu, UD_VECTOR);
5575 		return 1;
5576 	}
5577 
5578 	vmcs12 = get_vmcs12(vcpu);
5579 
5580 	/*
5581 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5582 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5583 	 */
5584 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5585 		kvm_queue_exception(vcpu, UD_VECTOR);
5586 		return 1;
5587 	}
5588 
5589 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5590 		goto fail;
5591 
5592 	switch (function) {
5593 	case 0:
5594 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5595 			goto fail;
5596 		break;
5597 	default:
5598 		goto fail;
5599 	}
5600 	return kvm_skip_emulated_instruction(vcpu);
5601 
5602 fail:
5603 	/*
5604 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5605 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5606 	 * EXIT_REASON_VMFUNC as the exit reason.
5607 	 */
5608 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5609 			  vmx_get_intr_info(vcpu),
5610 			  vmx_get_exit_qual(vcpu));
5611 	return 1;
5612 }
5613 
5614 /*
5615  * Return true if an IO instruction with the specified port and size should cause
5616  * a VM-exit into L1.
5617  */
5618 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5619 				 int size)
5620 {
5621 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5622 	gpa_t bitmap, last_bitmap;
5623 	u8 b;
5624 
5625 	last_bitmap = INVALID_GPA;
5626 	b = -1;
5627 
5628 	while (size > 0) {
5629 		if (port < 0x8000)
5630 			bitmap = vmcs12->io_bitmap_a;
5631 		else if (port < 0x10000)
5632 			bitmap = vmcs12->io_bitmap_b;
5633 		else
5634 			return true;
5635 		bitmap += (port & 0x7fff) / 8;
5636 
5637 		if (last_bitmap != bitmap)
5638 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5639 				return true;
5640 		if (b & (1 << (port & 7)))
5641 			return true;
5642 
5643 		port++;
5644 		size--;
5645 		last_bitmap = bitmap;
5646 	}
5647 
5648 	return false;
5649 }
5650 
5651 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5652 				       struct vmcs12 *vmcs12)
5653 {
5654 	unsigned long exit_qualification;
5655 	unsigned short port;
5656 	int size;
5657 
5658 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5659 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5660 
5661 	exit_qualification = vmx_get_exit_qual(vcpu);
5662 
5663 	port = exit_qualification >> 16;
5664 	size = (exit_qualification & 7) + 1;
5665 
5666 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5667 }
5668 
5669 /*
5670  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5671  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5672  * disinterest in the current event (read or write a specific MSR) by using an
5673  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5674  */
5675 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5676 					struct vmcs12 *vmcs12,
5677 					union vmx_exit_reason exit_reason)
5678 {
5679 	u32 msr_index = kvm_rcx_read(vcpu);
5680 	gpa_t bitmap;
5681 
5682 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5683 		return true;
5684 
5685 	/*
5686 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5687 	 * for the four combinations of read/write and low/high MSR numbers.
5688 	 * First we need to figure out which of the four to use:
5689 	 */
5690 	bitmap = vmcs12->msr_bitmap;
5691 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5692 		bitmap += 2048;
5693 	if (msr_index >= 0xc0000000) {
5694 		msr_index -= 0xc0000000;
5695 		bitmap += 1024;
5696 	}
5697 
5698 	/* Then read the msr_index'th bit from this bitmap: */
5699 	if (msr_index < 1024*8) {
5700 		unsigned char b;
5701 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5702 			return true;
5703 		return 1 & (b >> (msr_index & 7));
5704 	} else
5705 		return true; /* let L1 handle the wrong parameter */
5706 }
5707 
5708 /*
5709  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5710  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5711  * intercept (via guest_host_mask etc.) the current event.
5712  */
5713 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5714 	struct vmcs12 *vmcs12)
5715 {
5716 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5717 	int cr = exit_qualification & 15;
5718 	int reg;
5719 	unsigned long val;
5720 
5721 	switch ((exit_qualification >> 4) & 3) {
5722 	case 0: /* mov to cr */
5723 		reg = (exit_qualification >> 8) & 15;
5724 		val = kvm_register_read(vcpu, reg);
5725 		switch (cr) {
5726 		case 0:
5727 			if (vmcs12->cr0_guest_host_mask &
5728 			    (val ^ vmcs12->cr0_read_shadow))
5729 				return true;
5730 			break;
5731 		case 3:
5732 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5733 				return true;
5734 			break;
5735 		case 4:
5736 			if (vmcs12->cr4_guest_host_mask &
5737 			    (vmcs12->cr4_read_shadow ^ val))
5738 				return true;
5739 			break;
5740 		case 8:
5741 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5742 				return true;
5743 			break;
5744 		}
5745 		break;
5746 	case 2: /* clts */
5747 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5748 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5749 			return true;
5750 		break;
5751 	case 1: /* mov from cr */
5752 		switch (cr) {
5753 		case 3:
5754 			if (vmcs12->cpu_based_vm_exec_control &
5755 			    CPU_BASED_CR3_STORE_EXITING)
5756 				return true;
5757 			break;
5758 		case 8:
5759 			if (vmcs12->cpu_based_vm_exec_control &
5760 			    CPU_BASED_CR8_STORE_EXITING)
5761 				return true;
5762 			break;
5763 		}
5764 		break;
5765 	case 3: /* lmsw */
5766 		/*
5767 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5768 		 * cr0. Other attempted changes are ignored, with no exit.
5769 		 */
5770 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5771 		if (vmcs12->cr0_guest_host_mask & 0xe &
5772 		    (val ^ vmcs12->cr0_read_shadow))
5773 			return true;
5774 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5775 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5776 		    (val & 0x1))
5777 			return true;
5778 		break;
5779 	}
5780 	return false;
5781 }
5782 
5783 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
5784 					  struct vmcs12 *vmcs12)
5785 {
5786 	u32 encls_leaf;
5787 
5788 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
5789 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
5790 		return false;
5791 
5792 	encls_leaf = kvm_rax_read(vcpu);
5793 	if (encls_leaf > 62)
5794 		encls_leaf = 63;
5795 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
5796 }
5797 
5798 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5799 	struct vmcs12 *vmcs12, gpa_t bitmap)
5800 {
5801 	u32 vmx_instruction_info;
5802 	unsigned long field;
5803 	u8 b;
5804 
5805 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5806 		return true;
5807 
5808 	/* Decode instruction info and find the field to access */
5809 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5810 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5811 
5812 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5813 	if (field >> 15)
5814 		return true;
5815 
5816 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5817 		return true;
5818 
5819 	return 1 & (b >> (field & 7));
5820 }
5821 
5822 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5823 {
5824 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5825 
5826 	if (nested_cpu_has_mtf(vmcs12))
5827 		return true;
5828 
5829 	/*
5830 	 * An MTF VM-exit may be injected into the guest by setting the
5831 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5832 	 * is the case regardless of the 'monitor trap flag' VM-execution
5833 	 * control.
5834 	 */
5835 	return entry_intr_info == (INTR_INFO_VALID_MASK
5836 				   | INTR_TYPE_OTHER_EVENT);
5837 }
5838 
5839 /*
5840  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5841  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5842  */
5843 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
5844 				     union vmx_exit_reason exit_reason)
5845 {
5846 	u32 intr_info;
5847 
5848 	switch ((u16)exit_reason.basic) {
5849 	case EXIT_REASON_EXCEPTION_NMI:
5850 		intr_info = vmx_get_intr_info(vcpu);
5851 		if (is_nmi(intr_info))
5852 			return true;
5853 		else if (is_page_fault(intr_info))
5854 			return vcpu->arch.apf.host_apf_flags ||
5855 			       vmx_need_pf_intercept(vcpu);
5856 		else if (is_debug(intr_info) &&
5857 			 vcpu->guest_debug &
5858 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5859 			return true;
5860 		else if (is_breakpoint(intr_info) &&
5861 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5862 			return true;
5863 		else if (is_alignment_check(intr_info) &&
5864 			 !vmx_guest_inject_ac(vcpu))
5865 			return true;
5866 		return false;
5867 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5868 		return true;
5869 	case EXIT_REASON_MCE_DURING_VMENTRY:
5870 		return true;
5871 	case EXIT_REASON_EPT_VIOLATION:
5872 		/*
5873 		 * L0 always deals with the EPT violation. If nested EPT is
5874 		 * used, and the nested mmu code discovers that the address is
5875 		 * missing in the guest EPT table (EPT12), the EPT violation
5876 		 * will be injected with nested_ept_inject_page_fault()
5877 		 */
5878 		return true;
5879 	case EXIT_REASON_EPT_MISCONFIG:
5880 		/*
5881 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5882 		 * table (shadow on EPT) or a merged EPT table that L0 built
5883 		 * (EPT on EPT). So any problems with the structure of the
5884 		 * table is L0's fault.
5885 		 */
5886 		return true;
5887 	case EXIT_REASON_PREEMPTION_TIMER:
5888 		return true;
5889 	case EXIT_REASON_PML_FULL:
5890 		/*
5891 		 * PML is emulated for an L1 VMM and should never be enabled in
5892 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
5893 		 */
5894 		return true;
5895 	case EXIT_REASON_VMFUNC:
5896 		/* VM functions are emulated through L2->L0 vmexits. */
5897 		return true;
5898 	case EXIT_REASON_BUS_LOCK:
5899 		/*
5900 		 * At present, bus lock VM exit is never exposed to L1.
5901 		 * Handle L2's bus locks in L0 directly.
5902 		 */
5903 		return true;
5904 	default:
5905 		break;
5906 	}
5907 	return false;
5908 }
5909 
5910 /*
5911  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5912  * is_guest_mode (L2).
5913  */
5914 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
5915 				     union vmx_exit_reason exit_reason)
5916 {
5917 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5918 	u32 intr_info;
5919 
5920 	switch ((u16)exit_reason.basic) {
5921 	case EXIT_REASON_EXCEPTION_NMI:
5922 		intr_info = vmx_get_intr_info(vcpu);
5923 		if (is_nmi(intr_info))
5924 			return true;
5925 		else if (is_page_fault(intr_info))
5926 			return true;
5927 		return vmcs12->exception_bitmap &
5928 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5929 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5930 		return nested_exit_on_intr(vcpu);
5931 	case EXIT_REASON_TRIPLE_FAULT:
5932 		return true;
5933 	case EXIT_REASON_INTERRUPT_WINDOW:
5934 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5935 	case EXIT_REASON_NMI_WINDOW:
5936 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5937 	case EXIT_REASON_TASK_SWITCH:
5938 		return true;
5939 	case EXIT_REASON_CPUID:
5940 		return true;
5941 	case EXIT_REASON_HLT:
5942 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5943 	case EXIT_REASON_INVD:
5944 		return true;
5945 	case EXIT_REASON_INVLPG:
5946 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5947 	case EXIT_REASON_RDPMC:
5948 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5949 	case EXIT_REASON_RDRAND:
5950 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5951 	case EXIT_REASON_RDSEED:
5952 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5953 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5954 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5955 	case EXIT_REASON_VMREAD:
5956 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5957 			vmcs12->vmread_bitmap);
5958 	case EXIT_REASON_VMWRITE:
5959 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5960 			vmcs12->vmwrite_bitmap);
5961 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5962 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5963 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5964 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5965 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5966 		/*
5967 		 * VMX instructions trap unconditionally. This allows L1 to
5968 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5969 		 */
5970 		return true;
5971 	case EXIT_REASON_CR_ACCESS:
5972 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5973 	case EXIT_REASON_DR_ACCESS:
5974 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5975 	case EXIT_REASON_IO_INSTRUCTION:
5976 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5977 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5978 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5979 	case EXIT_REASON_MSR_READ:
5980 	case EXIT_REASON_MSR_WRITE:
5981 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5982 	case EXIT_REASON_INVALID_STATE:
5983 		return true;
5984 	case EXIT_REASON_MWAIT_INSTRUCTION:
5985 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5986 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5987 		return nested_vmx_exit_handled_mtf(vmcs12);
5988 	case EXIT_REASON_MONITOR_INSTRUCTION:
5989 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5990 	case EXIT_REASON_PAUSE_INSTRUCTION:
5991 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5992 			nested_cpu_has2(vmcs12,
5993 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5994 	case EXIT_REASON_MCE_DURING_VMENTRY:
5995 		return true;
5996 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5997 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5998 	case EXIT_REASON_APIC_ACCESS:
5999 	case EXIT_REASON_APIC_WRITE:
6000 	case EXIT_REASON_EOI_INDUCED:
6001 		/*
6002 		 * The controls for "virtualize APIC accesses," "APIC-
6003 		 * register virtualization," and "virtual-interrupt
6004 		 * delivery" only come from vmcs12.
6005 		 */
6006 		return true;
6007 	case EXIT_REASON_INVPCID:
6008 		return
6009 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6010 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6011 	case EXIT_REASON_WBINVD:
6012 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6013 	case EXIT_REASON_XSETBV:
6014 		return true;
6015 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6016 		/*
6017 		 * This should never happen, since it is not possible to
6018 		 * set XSS to a non-zero value---neither in L1 nor in L2.
6019 		 * If if it were, XSS would have to be checked against
6020 		 * the XSS exit bitmap in vmcs12.
6021 		 */
6022 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
6023 	case EXIT_REASON_UMWAIT:
6024 	case EXIT_REASON_TPAUSE:
6025 		return nested_cpu_has2(vmcs12,
6026 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6027 	case EXIT_REASON_ENCLS:
6028 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6029 	default:
6030 		return true;
6031 	}
6032 }
6033 
6034 /*
6035  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6036  * reflected into L1.
6037  */
6038 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6039 {
6040 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6041 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6042 	unsigned long exit_qual;
6043 	u32 exit_intr_info;
6044 
6045 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6046 
6047 	/*
6048 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6049 	 * has already loaded L2's state.
6050 	 */
6051 	if (unlikely(vmx->fail)) {
6052 		trace_kvm_nested_vmenter_failed(
6053 			"hardware VM-instruction error: ",
6054 			vmcs_read32(VM_INSTRUCTION_ERROR));
6055 		exit_intr_info = 0;
6056 		exit_qual = 0;
6057 		goto reflect_vmexit;
6058 	}
6059 
6060 	trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6061 
6062 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6063 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6064 		return false;
6065 
6066 	/* If L1 doesn't want the exit, handle it in L0. */
6067 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6068 		return false;
6069 
6070 	/*
6071 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6072 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6073 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6074 	 * interrupts are never reflected to L1 so it's a non-issue.
6075 	 */
6076 	exit_intr_info = vmx_get_intr_info(vcpu);
6077 	if (is_exception_with_error_code(exit_intr_info)) {
6078 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6079 
6080 		vmcs12->vm_exit_intr_error_code =
6081 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6082 	}
6083 	exit_qual = vmx_get_exit_qual(vcpu);
6084 
6085 reflect_vmexit:
6086 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6087 	return true;
6088 }
6089 
6090 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6091 				struct kvm_nested_state __user *user_kvm_nested_state,
6092 				u32 user_data_size)
6093 {
6094 	struct vcpu_vmx *vmx;
6095 	struct vmcs12 *vmcs12;
6096 	struct kvm_nested_state kvm_state = {
6097 		.flags = 0,
6098 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6099 		.size = sizeof(kvm_state),
6100 		.hdr.vmx.flags = 0,
6101 		.hdr.vmx.vmxon_pa = INVALID_GPA,
6102 		.hdr.vmx.vmcs12_pa = INVALID_GPA,
6103 		.hdr.vmx.preemption_timer_deadline = 0,
6104 	};
6105 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6106 		&user_kvm_nested_state->data.vmx[0];
6107 
6108 	if (!vcpu)
6109 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6110 
6111 	vmx = to_vmx(vcpu);
6112 	vmcs12 = get_vmcs12(vcpu);
6113 
6114 	if (nested_vmx_allowed(vcpu) &&
6115 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6116 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6117 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6118 
6119 		if (vmx_has_valid_vmcs12(vcpu)) {
6120 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6121 
6122 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6123 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6124 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6125 
6126 			if (is_guest_mode(vcpu) &&
6127 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6128 			    vmcs12->vmcs_link_pointer != INVALID_GPA)
6129 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6130 		}
6131 
6132 		if (vmx->nested.smm.vmxon)
6133 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6134 
6135 		if (vmx->nested.smm.guest_mode)
6136 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6137 
6138 		if (is_guest_mode(vcpu)) {
6139 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6140 
6141 			if (vmx->nested.nested_run_pending)
6142 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6143 
6144 			if (vmx->nested.mtf_pending)
6145 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6146 
6147 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6148 			    vmx->nested.has_preemption_timer_deadline) {
6149 				kvm_state.hdr.vmx.flags |=
6150 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6151 				kvm_state.hdr.vmx.preemption_timer_deadline =
6152 					vmx->nested.preemption_timer_deadline;
6153 			}
6154 		}
6155 	}
6156 
6157 	if (user_data_size < kvm_state.size)
6158 		goto out;
6159 
6160 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6161 		return -EFAULT;
6162 
6163 	if (!vmx_has_valid_vmcs12(vcpu))
6164 		goto out;
6165 
6166 	/*
6167 	 * When running L2, the authoritative vmcs12 state is in the
6168 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6169 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6170 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6171 	 * vmcs12 state is in the vmcs12 already.
6172 	 */
6173 	if (is_guest_mode(vcpu)) {
6174 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6175 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6176 	} else  {
6177 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6178 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6179 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6180 				/*
6181 				 * L1 hypervisor is not obliged to keep eVMCS
6182 				 * clean fields data always up-to-date while
6183 				 * not in guest mode, 'hv_clean_fields' is only
6184 				 * supposed to be actual upon vmentry so we need
6185 				 * to ignore it here and do full copy.
6186 				 */
6187 				copy_enlightened_to_vmcs12(vmx, 0);
6188 			else if (enable_shadow_vmcs)
6189 				copy_shadow_to_vmcs12(vmx);
6190 		}
6191 	}
6192 
6193 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6194 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6195 
6196 	/*
6197 	 * Copy over the full allocated size of vmcs12 rather than just the size
6198 	 * of the struct.
6199 	 */
6200 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6201 		return -EFAULT;
6202 
6203 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6204 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6205 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6206 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6207 			return -EFAULT;
6208 	}
6209 out:
6210 	return kvm_state.size;
6211 }
6212 
6213 /*
6214  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6215  */
6216 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6217 {
6218 	if (is_guest_mode(vcpu)) {
6219 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6220 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6221 	}
6222 	free_nested(vcpu);
6223 }
6224 
6225 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6226 				struct kvm_nested_state __user *user_kvm_nested_state,
6227 				struct kvm_nested_state *kvm_state)
6228 {
6229 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6230 	struct vmcs12 *vmcs12;
6231 	enum vm_entry_failure_code ignored;
6232 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6233 		&user_kvm_nested_state->data.vmx[0];
6234 	int ret;
6235 
6236 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6237 		return -EINVAL;
6238 
6239 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6240 		if (kvm_state->hdr.vmx.smm.flags)
6241 			return -EINVAL;
6242 
6243 		if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6244 			return -EINVAL;
6245 
6246 		/*
6247 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6248 		 * enable eVMCS capability on vCPU. However, since then
6249 		 * code was changed such that flag signals vmcs12 should
6250 		 * be copied into eVMCS in guest memory.
6251 		 *
6252 		 * To preserve backwards compatability, allow user
6253 		 * to set this flag even when there is no VMXON region.
6254 		 */
6255 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6256 			return -EINVAL;
6257 	} else {
6258 		if (!nested_vmx_allowed(vcpu))
6259 			return -EINVAL;
6260 
6261 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6262 			return -EINVAL;
6263 	}
6264 
6265 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6266 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6267 		return -EINVAL;
6268 
6269 	if (kvm_state->hdr.vmx.smm.flags &
6270 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6271 		return -EINVAL;
6272 
6273 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6274 		return -EINVAL;
6275 
6276 	/*
6277 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6278 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6279 	 * must be zero.
6280 	 */
6281 	if (is_smm(vcpu) ?
6282 		(kvm_state->flags &
6283 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6284 		: kvm_state->hdr.vmx.smm.flags)
6285 		return -EINVAL;
6286 
6287 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6288 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6289 		return -EINVAL;
6290 
6291 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6292 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6293 			return -EINVAL;
6294 
6295 	vmx_leave_nested(vcpu);
6296 
6297 	if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6298 		return 0;
6299 
6300 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6301 	ret = enter_vmx_operation(vcpu);
6302 	if (ret)
6303 		return ret;
6304 
6305 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6306 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6307 		/* See vmx_has_valid_vmcs12.  */
6308 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6309 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6310 		    (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6311 			return -EINVAL;
6312 		else
6313 			return 0;
6314 	}
6315 
6316 	if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6317 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6318 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6319 			return -EINVAL;
6320 
6321 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6322 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6323 		/*
6324 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6325 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6326 		 * restored yet. EVMCS will be mapped from
6327 		 * nested_get_vmcs12_pages().
6328 		 */
6329 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6330 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6331 	} else {
6332 		return -EINVAL;
6333 	}
6334 
6335 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6336 		vmx->nested.smm.vmxon = true;
6337 		vmx->nested.vmxon = false;
6338 
6339 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6340 			vmx->nested.smm.guest_mode = true;
6341 	}
6342 
6343 	vmcs12 = get_vmcs12(vcpu);
6344 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6345 		return -EFAULT;
6346 
6347 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6348 		return -EINVAL;
6349 
6350 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6351 		return 0;
6352 
6353 	vmx->nested.nested_run_pending =
6354 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6355 
6356 	vmx->nested.mtf_pending =
6357 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6358 
6359 	ret = -EINVAL;
6360 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6361 	    vmcs12->vmcs_link_pointer != INVALID_GPA) {
6362 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6363 
6364 		if (kvm_state->size <
6365 		    sizeof(*kvm_state) +
6366 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6367 			goto error_guest_mode;
6368 
6369 		if (copy_from_user(shadow_vmcs12,
6370 				   user_vmx_nested_state->shadow_vmcs12,
6371 				   sizeof(*shadow_vmcs12))) {
6372 			ret = -EFAULT;
6373 			goto error_guest_mode;
6374 		}
6375 
6376 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6377 		    !shadow_vmcs12->hdr.shadow_vmcs)
6378 			goto error_guest_mode;
6379 	}
6380 
6381 	vmx->nested.has_preemption_timer_deadline = false;
6382 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6383 		vmx->nested.has_preemption_timer_deadline = true;
6384 		vmx->nested.preemption_timer_deadline =
6385 			kvm_state->hdr.vmx.preemption_timer_deadline;
6386 	}
6387 
6388 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6389 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6390 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6391 		goto error_guest_mode;
6392 
6393 	vmx->nested.dirty_vmcs12 = true;
6394 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6395 	if (ret)
6396 		goto error_guest_mode;
6397 
6398 	return 0;
6399 
6400 error_guest_mode:
6401 	vmx->nested.nested_run_pending = 0;
6402 	return ret;
6403 }
6404 
6405 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6406 {
6407 	if (enable_shadow_vmcs) {
6408 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6409 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6410 	}
6411 }
6412 
6413 /*
6414  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6415  * that madness to get the encoding for comparison.
6416  */
6417 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6418 
6419 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6420 {
6421 	/*
6422 	 * Note these are the so called "index" of the VMCS field encoding, not
6423 	 * the index into vmcs12.
6424 	 */
6425 	unsigned int max_idx, idx;
6426 	int i;
6427 
6428 	/*
6429 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6430 	 * vmcs12, regardless of whether or not the associated feature is
6431 	 * exposed to L1.  Simply find the field with the highest index.
6432 	 */
6433 	max_idx = 0;
6434 	for (i = 0; i < nr_vmcs12_fields; i++) {
6435 		/* The vmcs12 table is very, very sparsely populated. */
6436 		if (!vmcs_field_to_offset_table[i])
6437 			continue;
6438 
6439 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6440 		if (idx > max_idx)
6441 			max_idx = idx;
6442 	}
6443 
6444 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6445 }
6446 
6447 /*
6448  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6449  * returned for the various VMX controls MSRs when nested VMX is enabled.
6450  * The same values should also be used to verify that vmcs12 control fields are
6451  * valid during nested entry from L1 to L2.
6452  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6453  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6454  * bit in the high half is on if the corresponding bit in the control field
6455  * may be on. See also vmx_control_verify().
6456  */
6457 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6458 {
6459 	/*
6460 	 * Note that as a general rule, the high half of the MSRs (bits in
6461 	 * the control fields which may be 1) should be initialized by the
6462 	 * intersection of the underlying hardware's MSR (i.e., features which
6463 	 * can be supported) and the list of features we want to expose -
6464 	 * because they are known to be properly supported in our code.
6465 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6466 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6467 	 * reason is that if one of these bits is necessary, it will appear
6468 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6469 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6470 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6471 	 * These rules have exceptions below.
6472 	 */
6473 
6474 	/* pin-based controls */
6475 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6476 		msrs->pinbased_ctls_low,
6477 		msrs->pinbased_ctls_high);
6478 	msrs->pinbased_ctls_low |=
6479 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6480 	msrs->pinbased_ctls_high &=
6481 		PIN_BASED_EXT_INTR_MASK |
6482 		PIN_BASED_NMI_EXITING |
6483 		PIN_BASED_VIRTUAL_NMIS |
6484 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6485 	msrs->pinbased_ctls_high |=
6486 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6487 		PIN_BASED_VMX_PREEMPTION_TIMER;
6488 
6489 	/* exit controls */
6490 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6491 		msrs->exit_ctls_low,
6492 		msrs->exit_ctls_high);
6493 	msrs->exit_ctls_low =
6494 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6495 
6496 	msrs->exit_ctls_high &=
6497 #ifdef CONFIG_X86_64
6498 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6499 #endif
6500 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6501 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6502 	msrs->exit_ctls_high |=
6503 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6504 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6505 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6506 
6507 	/* We support free control of debug control saving. */
6508 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6509 
6510 	/* entry controls */
6511 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6512 		msrs->entry_ctls_low,
6513 		msrs->entry_ctls_high);
6514 	msrs->entry_ctls_low =
6515 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6516 	msrs->entry_ctls_high &=
6517 #ifdef CONFIG_X86_64
6518 		VM_ENTRY_IA32E_MODE |
6519 #endif
6520 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6521 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6522 	msrs->entry_ctls_high |=
6523 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6524 
6525 	/* We support free control of debug control loading. */
6526 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6527 
6528 	/* cpu-based controls */
6529 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6530 		msrs->procbased_ctls_low,
6531 		msrs->procbased_ctls_high);
6532 	msrs->procbased_ctls_low =
6533 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6534 	msrs->procbased_ctls_high &=
6535 		CPU_BASED_INTR_WINDOW_EXITING |
6536 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6537 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6538 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6539 		CPU_BASED_CR3_STORE_EXITING |
6540 #ifdef CONFIG_X86_64
6541 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6542 #endif
6543 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6544 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6545 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6546 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6547 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6548 	/*
6549 	 * We can allow some features even when not supported by the
6550 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6551 	 * can use it to avoid exits to L1 - even when L0 runs L2
6552 	 * without MSR bitmaps.
6553 	 */
6554 	msrs->procbased_ctls_high |=
6555 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6556 		CPU_BASED_USE_MSR_BITMAPS;
6557 
6558 	/* We support free control of CR3 access interception. */
6559 	msrs->procbased_ctls_low &=
6560 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6561 
6562 	/*
6563 	 * secondary cpu-based controls.  Do not include those that
6564 	 * depend on CPUID bits, they are added later by
6565 	 * vmx_vcpu_after_set_cpuid.
6566 	 */
6567 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6568 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6569 		      msrs->secondary_ctls_low,
6570 		      msrs->secondary_ctls_high);
6571 
6572 	msrs->secondary_ctls_low = 0;
6573 	msrs->secondary_ctls_high &=
6574 		SECONDARY_EXEC_DESC |
6575 		SECONDARY_EXEC_ENABLE_RDTSCP |
6576 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6577 		SECONDARY_EXEC_WBINVD_EXITING |
6578 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6579 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6580 		SECONDARY_EXEC_RDRAND_EXITING |
6581 		SECONDARY_EXEC_ENABLE_INVPCID |
6582 		SECONDARY_EXEC_RDSEED_EXITING |
6583 		SECONDARY_EXEC_XSAVES |
6584 		SECONDARY_EXEC_TSC_SCALING;
6585 
6586 	/*
6587 	 * We can emulate "VMCS shadowing," even if the hardware
6588 	 * doesn't support it.
6589 	 */
6590 	msrs->secondary_ctls_high |=
6591 		SECONDARY_EXEC_SHADOW_VMCS;
6592 
6593 	if (enable_ept) {
6594 		/* nested EPT: emulate EPT also to L1 */
6595 		msrs->secondary_ctls_high |=
6596 			SECONDARY_EXEC_ENABLE_EPT;
6597 		msrs->ept_caps =
6598 			VMX_EPT_PAGE_WALK_4_BIT |
6599 			VMX_EPT_PAGE_WALK_5_BIT |
6600 			VMX_EPTP_WB_BIT |
6601 			VMX_EPT_INVEPT_BIT |
6602 			VMX_EPT_EXECUTE_ONLY_BIT;
6603 
6604 		msrs->ept_caps &= ept_caps;
6605 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6606 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6607 			VMX_EPT_1GB_PAGE_BIT;
6608 		if (enable_ept_ad_bits) {
6609 			msrs->secondary_ctls_high |=
6610 				SECONDARY_EXEC_ENABLE_PML;
6611 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6612 		}
6613 	}
6614 
6615 	if (cpu_has_vmx_vmfunc()) {
6616 		msrs->secondary_ctls_high |=
6617 			SECONDARY_EXEC_ENABLE_VMFUNC;
6618 		/*
6619 		 * Advertise EPTP switching unconditionally
6620 		 * since we emulate it
6621 		 */
6622 		if (enable_ept)
6623 			msrs->vmfunc_controls =
6624 				VMX_VMFUNC_EPTP_SWITCHING;
6625 	}
6626 
6627 	/*
6628 	 * Old versions of KVM use the single-context version without
6629 	 * checking for support, so declare that it is supported even
6630 	 * though it is treated as global context.  The alternative is
6631 	 * not failing the single-context invvpid, and it is worse.
6632 	 */
6633 	if (enable_vpid) {
6634 		msrs->secondary_ctls_high |=
6635 			SECONDARY_EXEC_ENABLE_VPID;
6636 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6637 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6638 	}
6639 
6640 	if (enable_unrestricted_guest)
6641 		msrs->secondary_ctls_high |=
6642 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6643 
6644 	if (flexpriority_enabled)
6645 		msrs->secondary_ctls_high |=
6646 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6647 
6648 	if (enable_sgx)
6649 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6650 
6651 	/* miscellaneous data */
6652 	rdmsr(MSR_IA32_VMX_MISC,
6653 		msrs->misc_low,
6654 		msrs->misc_high);
6655 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6656 	msrs->misc_low |=
6657 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6658 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6659 		VMX_MISC_ACTIVITY_HLT |
6660 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6661 	msrs->misc_high = 0;
6662 
6663 	/*
6664 	 * This MSR reports some information about VMX support. We
6665 	 * should return information about the VMX we emulate for the
6666 	 * guest, and the VMCS structure we give it - not about the
6667 	 * VMX support of the underlying hardware.
6668 	 */
6669 	msrs->basic =
6670 		VMCS12_REVISION |
6671 		VMX_BASIC_TRUE_CTLS |
6672 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6673 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6674 
6675 	if (cpu_has_vmx_basic_inout())
6676 		msrs->basic |= VMX_BASIC_INOUT;
6677 
6678 	/*
6679 	 * These MSRs specify bits which the guest must keep fixed on
6680 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6681 	 * We picked the standard core2 setting.
6682 	 */
6683 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6684 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6685 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6686 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6687 
6688 	/* These MSRs specify bits which the guest must keep fixed off. */
6689 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6690 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6691 
6692 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
6693 }
6694 
6695 void nested_vmx_hardware_unsetup(void)
6696 {
6697 	int i;
6698 
6699 	if (enable_shadow_vmcs) {
6700 		for (i = 0; i < VMX_BITMAP_NR; i++)
6701 			free_page((unsigned long)vmx_bitmap[i]);
6702 	}
6703 }
6704 
6705 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6706 {
6707 	int i;
6708 
6709 	if (!cpu_has_vmx_shadow_vmcs())
6710 		enable_shadow_vmcs = 0;
6711 	if (enable_shadow_vmcs) {
6712 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6713 			/*
6714 			 * The vmx_bitmap is not tied to a VM and so should
6715 			 * not be charged to a memcg.
6716 			 */
6717 			vmx_bitmap[i] = (unsigned long *)
6718 				__get_free_page(GFP_KERNEL);
6719 			if (!vmx_bitmap[i]) {
6720 				nested_vmx_hardware_unsetup();
6721 				return -ENOMEM;
6722 			}
6723 		}
6724 
6725 		init_vmcs_shadow_fields();
6726 	}
6727 
6728 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6729 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6730 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6731 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6732 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6733 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6734 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6735 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6736 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6737 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6738 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6739 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6740 
6741 	return 0;
6742 }
6743 
6744 struct kvm_x86_nested_ops vmx_nested_ops = {
6745 	.check_events = vmx_check_nested_events,
6746 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6747 	.triple_fault = nested_vmx_triple_fault,
6748 	.get_state = vmx_get_nested_state,
6749 	.set_state = vmx_set_nested_state,
6750 	.get_nested_state_pages = vmx_get_nested_state_pages,
6751 	.write_log_dirty = nested_vmx_write_pml_buffer,
6752 	.enable_evmcs = nested_enable_evmcs,
6753 	.get_evmcs_version = nested_get_evmcs_version,
6754 };
6755