xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 808643ea)
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 == -1ull &&
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, -1ull);
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 	free_vpid(vmx->nested.vpid02);
294 	vmx->nested.posted_intr_nv = -1;
295 	vmx->nested.current_vmptr = -1ull;
296 	if (enable_shadow_vmcs) {
297 		vmx_disable_shadow_vmcs(vmx);
298 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
299 		free_vmcs(vmx->vmcs01.shadow_vmcs);
300 		vmx->vmcs01.shadow_vmcs = NULL;
301 	}
302 	kfree(vmx->nested.cached_vmcs12);
303 	vmx->nested.cached_vmcs12 = NULL;
304 	kfree(vmx->nested.cached_shadow_vmcs12);
305 	vmx->nested.cached_shadow_vmcs12 = NULL;
306 	/* Unpin physical memory we referred to in the vmcs02 */
307 	if (vmx->nested.apic_access_page) {
308 		kvm_release_page_clean(vmx->nested.apic_access_page);
309 		vmx->nested.apic_access_page = NULL;
310 	}
311 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
312 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
313 	vmx->nested.pi_desc = NULL;
314 
315 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
316 
317 	nested_release_evmcs(vcpu);
318 
319 	free_loaded_vmcs(&vmx->nested.vmcs02);
320 }
321 
322 /*
323  * Ensure that the current vmcs of the logical processor is the
324  * vmcs01 of the vcpu before calling free_nested().
325  */
326 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
327 {
328 	vcpu_load(vcpu);
329 	vmx_leave_nested(vcpu);
330 	vcpu_put(vcpu);
331 }
332 
333 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
334 		struct x86_exception *fault)
335 {
336 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
337 	struct vcpu_vmx *vmx = to_vmx(vcpu);
338 	u32 vm_exit_reason;
339 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
340 
341 	if (vmx->nested.pml_full) {
342 		vm_exit_reason = EXIT_REASON_PML_FULL;
343 		vmx->nested.pml_full = false;
344 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
345 	} else if (fault->error_code & PFERR_RSVD_MASK)
346 		vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
347 	else
348 		vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
349 
350 	nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
351 	vmcs12->guest_physical_address = fault->address;
352 }
353 
354 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
355 {
356 	kvm_init_shadow_ept_mmu(vcpu,
357 				to_vmx(vcpu)->nested.msrs.ept_caps &
358 				VMX_EPT_EXECUTE_ONLY_BIT,
359 				nested_ept_ad_enabled(vcpu),
360 				nested_ept_get_eptp(vcpu));
361 }
362 
363 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
364 {
365 	WARN_ON(mmu_is_nested(vcpu));
366 
367 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
368 	nested_ept_new_eptp(vcpu);
369 	vcpu->arch.mmu->get_guest_pgd     = nested_ept_get_eptp;
370 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
371 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
372 
373 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
374 }
375 
376 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
377 {
378 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
379 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
380 }
381 
382 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
383 					    u16 error_code)
384 {
385 	bool inequality, bit;
386 
387 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
388 	inequality =
389 		(error_code & vmcs12->page_fault_error_code_mask) !=
390 		 vmcs12->page_fault_error_code_match;
391 	return inequality ^ bit;
392 }
393 
394 
395 /*
396  * KVM wants to inject page-faults which it got to the guest. This function
397  * checks whether in a nested guest, we need to inject them to L1 or L2.
398  */
399 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
400 {
401 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
402 	unsigned int nr = vcpu->arch.exception.nr;
403 	bool has_payload = vcpu->arch.exception.has_payload;
404 	unsigned long payload = vcpu->arch.exception.payload;
405 
406 	if (nr == PF_VECTOR) {
407 		if (vcpu->arch.exception.nested_apf) {
408 			*exit_qual = vcpu->arch.apf.nested_apf_token;
409 			return 1;
410 		}
411 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
412 						    vcpu->arch.exception.error_code)) {
413 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
414 			return 1;
415 		}
416 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
417 		if (nr == DB_VECTOR) {
418 			if (!has_payload) {
419 				payload = vcpu->arch.dr6;
420 				payload &= ~DR6_BT;
421 				payload ^= DR6_ACTIVE_LOW;
422 			}
423 			*exit_qual = payload;
424 		} else
425 			*exit_qual = 0;
426 		return 1;
427 	}
428 
429 	return 0;
430 }
431 
432 
433 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
434 		struct x86_exception *fault)
435 {
436 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
437 
438 	WARN_ON(!is_guest_mode(vcpu));
439 
440 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
441 		!to_vmx(vcpu)->nested.nested_run_pending) {
442 		vmcs12->vm_exit_intr_error_code = fault->error_code;
443 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
444 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
445 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
446 				  fault->address);
447 	} else {
448 		kvm_inject_page_fault(vcpu, fault);
449 	}
450 }
451 
452 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
453 					       struct vmcs12 *vmcs12)
454 {
455 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
456 		return 0;
457 
458 	if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
459 	    CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
460 		return -EINVAL;
461 
462 	return 0;
463 }
464 
465 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
466 						struct vmcs12 *vmcs12)
467 {
468 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
469 		return 0;
470 
471 	if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
472 		return -EINVAL;
473 
474 	return 0;
475 }
476 
477 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
478 						struct vmcs12 *vmcs12)
479 {
480 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
481 		return 0;
482 
483 	if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
484 		return -EINVAL;
485 
486 	return 0;
487 }
488 
489 /*
490  * Check if MSR is intercepted for L01 MSR bitmap.
491  */
492 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
493 {
494 	unsigned long *msr_bitmap;
495 	int f = sizeof(unsigned long);
496 
497 	if (!cpu_has_vmx_msr_bitmap())
498 		return true;
499 
500 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
501 
502 	if (msr <= 0x1fff) {
503 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
504 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
505 		msr &= 0x1fff;
506 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
507 	}
508 
509 	return true;
510 }
511 
512 /*
513  * If a msr is allowed by L0, we should check whether it is allowed by L1.
514  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
515  */
516 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
517 					       unsigned long *msr_bitmap_nested,
518 					       u32 msr, int type)
519 {
520 	int f = sizeof(unsigned long);
521 
522 	/*
523 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
524 	 * have the write-low and read-high bitmap offsets the wrong way round.
525 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
526 	 */
527 	if (msr <= 0x1fff) {
528 		if (type & MSR_TYPE_R &&
529 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
530 			/* read-low */
531 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
532 
533 		if (type & MSR_TYPE_W &&
534 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
535 			/* write-low */
536 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
537 
538 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
539 		msr &= 0x1fff;
540 		if (type & MSR_TYPE_R &&
541 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
542 			/* read-high */
543 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
544 
545 		if (type & MSR_TYPE_W &&
546 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
547 			/* write-high */
548 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
549 
550 	}
551 }
552 
553 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
554 {
555 	int msr;
556 
557 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
558 		unsigned word = msr / BITS_PER_LONG;
559 
560 		msr_bitmap[word] = ~0;
561 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
562 	}
563 }
564 
565 /*
566  * Merge L0's and L1's MSR bitmap, return false to indicate that
567  * we do not use the hardware.
568  */
569 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
570 						 struct vmcs12 *vmcs12)
571 {
572 	int msr;
573 	unsigned long *msr_bitmap_l1;
574 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
575 	struct kvm_host_map *map = &to_vmx(vcpu)->nested.msr_bitmap_map;
576 
577 	/* Nothing to do if the MSR bitmap is not in use.  */
578 	if (!cpu_has_vmx_msr_bitmap() ||
579 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
580 		return false;
581 
582 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
583 		return false;
584 
585 	msr_bitmap_l1 = (unsigned long *)map->hva;
586 
587 	/*
588 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
589 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
590 	 * the x2APIC MSR range and selectively disable them below.
591 	 */
592 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
593 
594 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
595 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
596 			/*
597 			 * L0 need not intercept reads for MSRs between 0x800
598 			 * and 0x8ff, it just lets the processor take the value
599 			 * from the virtual-APIC page; take those 256 bits
600 			 * directly from the L1 bitmap.
601 			 */
602 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
603 				unsigned word = msr / BITS_PER_LONG;
604 
605 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
606 			}
607 		}
608 
609 		nested_vmx_disable_intercept_for_msr(
610 			msr_bitmap_l1, msr_bitmap_l0,
611 			X2APIC_MSR(APIC_TASKPRI),
612 			MSR_TYPE_R | MSR_TYPE_W);
613 
614 		if (nested_cpu_has_vid(vmcs12)) {
615 			nested_vmx_disable_intercept_for_msr(
616 				msr_bitmap_l1, msr_bitmap_l0,
617 				X2APIC_MSR(APIC_EOI),
618 				MSR_TYPE_W);
619 			nested_vmx_disable_intercept_for_msr(
620 				msr_bitmap_l1, msr_bitmap_l0,
621 				X2APIC_MSR(APIC_SELF_IPI),
622 				MSR_TYPE_W);
623 		}
624 	}
625 
626 	/* KVM unconditionally exposes the FS/GS base MSRs to L1. */
627 #ifdef CONFIG_X86_64
628 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
629 					     MSR_FS_BASE, MSR_TYPE_RW);
630 
631 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
632 					     MSR_GS_BASE, MSR_TYPE_RW);
633 
634 	nested_vmx_disable_intercept_for_msr(msr_bitmap_l1, msr_bitmap_l0,
635 					     MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
636 #endif
637 
638 	/*
639 	 * Checking the L0->L1 bitmap is trying to verify two things:
640 	 *
641 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
642 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
643 	 *    from the L12 MSR bitmap that is too permissive.
644 	 * 2. That L1 or L2s have actually used the MSR. This avoids
645 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
646 	 *    works properly because we only update the L01 MSR bitmap lazily.
647 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
648 	 *    updated to reflect this when L1 (or its L2s) actually write to
649 	 *    the MSR.
650 	 */
651 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL))
652 		nested_vmx_disable_intercept_for_msr(
653 					msr_bitmap_l1, msr_bitmap_l0,
654 					MSR_IA32_SPEC_CTRL,
655 					MSR_TYPE_R | MSR_TYPE_W);
656 
657 	if (!msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD))
658 		nested_vmx_disable_intercept_for_msr(
659 					msr_bitmap_l1, msr_bitmap_l0,
660 					MSR_IA32_PRED_CMD,
661 					MSR_TYPE_W);
662 
663 	kvm_vcpu_unmap(vcpu, &to_vmx(vcpu)->nested.msr_bitmap_map, false);
664 
665 	return true;
666 }
667 
668 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
669 				       struct vmcs12 *vmcs12)
670 {
671 	struct kvm_host_map map;
672 	struct vmcs12 *shadow;
673 
674 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
675 	    vmcs12->vmcs_link_pointer == -1ull)
676 		return;
677 
678 	shadow = get_shadow_vmcs12(vcpu);
679 
680 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map))
681 		return;
682 
683 	memcpy(shadow, map.hva, VMCS12_SIZE);
684 	kvm_vcpu_unmap(vcpu, &map, false);
685 }
686 
687 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
688 					      struct vmcs12 *vmcs12)
689 {
690 	struct vcpu_vmx *vmx = to_vmx(vcpu);
691 
692 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
693 	    vmcs12->vmcs_link_pointer == -1ull)
694 		return;
695 
696 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
697 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
698 }
699 
700 /*
701  * In nested virtualization, check if L1 has set
702  * VM_EXIT_ACK_INTR_ON_EXIT
703  */
704 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
705 {
706 	return get_vmcs12(vcpu)->vm_exit_controls &
707 		VM_EXIT_ACK_INTR_ON_EXIT;
708 }
709 
710 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
711 					  struct vmcs12 *vmcs12)
712 {
713 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
714 	    CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
715 		return -EINVAL;
716 	else
717 		return 0;
718 }
719 
720 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
721 					   struct vmcs12 *vmcs12)
722 {
723 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
724 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
725 	    !nested_cpu_has_vid(vmcs12) &&
726 	    !nested_cpu_has_posted_intr(vmcs12))
727 		return 0;
728 
729 	/*
730 	 * If virtualize x2apic mode is enabled,
731 	 * virtualize apic access must be disabled.
732 	 */
733 	if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
734 	       nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
735 		return -EINVAL;
736 
737 	/*
738 	 * If virtual interrupt delivery is enabled,
739 	 * we must exit on external interrupts.
740 	 */
741 	if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
742 		return -EINVAL;
743 
744 	/*
745 	 * bits 15:8 should be zero in posted_intr_nv,
746 	 * the descriptor address has been already checked
747 	 * in nested_get_vmcs12_pages.
748 	 *
749 	 * bits 5:0 of posted_intr_desc_addr should be zero.
750 	 */
751 	if (nested_cpu_has_posted_intr(vmcs12) &&
752 	   (CC(!nested_cpu_has_vid(vmcs12)) ||
753 	    CC(!nested_exit_intr_ack_set(vcpu)) ||
754 	    CC((vmcs12->posted_intr_nv & 0xff00)) ||
755 	    CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
756 		return -EINVAL;
757 
758 	/* tpr shadow is needed by all apicv features. */
759 	if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
760 		return -EINVAL;
761 
762 	return 0;
763 }
764 
765 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
766 				       u32 count, u64 addr)
767 {
768 	if (count == 0)
769 		return 0;
770 
771 	if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
772 	    !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
773 		return -EINVAL;
774 
775 	return 0;
776 }
777 
778 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
779 						     struct vmcs12 *vmcs12)
780 {
781 	if (CC(nested_vmx_check_msr_switch(vcpu,
782 					   vmcs12->vm_exit_msr_load_count,
783 					   vmcs12->vm_exit_msr_load_addr)) ||
784 	    CC(nested_vmx_check_msr_switch(vcpu,
785 					   vmcs12->vm_exit_msr_store_count,
786 					   vmcs12->vm_exit_msr_store_addr)))
787 		return -EINVAL;
788 
789 	return 0;
790 }
791 
792 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
793                                                       struct vmcs12 *vmcs12)
794 {
795 	if (CC(nested_vmx_check_msr_switch(vcpu,
796 					   vmcs12->vm_entry_msr_load_count,
797 					   vmcs12->vm_entry_msr_load_addr)))
798                 return -EINVAL;
799 
800 	return 0;
801 }
802 
803 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
804 					 struct vmcs12 *vmcs12)
805 {
806 	if (!nested_cpu_has_pml(vmcs12))
807 		return 0;
808 
809 	if (CC(!nested_cpu_has_ept(vmcs12)) ||
810 	    CC(!page_address_valid(vcpu, vmcs12->pml_address)))
811 		return -EINVAL;
812 
813 	return 0;
814 }
815 
816 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
817 							struct vmcs12 *vmcs12)
818 {
819 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
820 	       !nested_cpu_has_ept(vmcs12)))
821 		return -EINVAL;
822 	return 0;
823 }
824 
825 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
826 							 struct vmcs12 *vmcs12)
827 {
828 	if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
829 	       !nested_cpu_has_ept(vmcs12)))
830 		return -EINVAL;
831 	return 0;
832 }
833 
834 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
835 						 struct vmcs12 *vmcs12)
836 {
837 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
838 		return 0;
839 
840 	if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
841 	    CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
842 		return -EINVAL;
843 
844 	return 0;
845 }
846 
847 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
848 				       struct vmx_msr_entry *e)
849 {
850 	/* x2APIC MSR accesses are not allowed */
851 	if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
852 		return -EINVAL;
853 	if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
854 	    CC(e->index == MSR_IA32_UCODE_REV))
855 		return -EINVAL;
856 	if (CC(e->reserved != 0))
857 		return -EINVAL;
858 	return 0;
859 }
860 
861 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
862 				     struct vmx_msr_entry *e)
863 {
864 	if (CC(e->index == MSR_FS_BASE) ||
865 	    CC(e->index == MSR_GS_BASE) ||
866 	    CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
867 	    nested_vmx_msr_check_common(vcpu, e))
868 		return -EINVAL;
869 	return 0;
870 }
871 
872 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
873 				      struct vmx_msr_entry *e)
874 {
875 	if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
876 	    nested_vmx_msr_check_common(vcpu, e))
877 		return -EINVAL;
878 	return 0;
879 }
880 
881 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
882 {
883 	struct vcpu_vmx *vmx = to_vmx(vcpu);
884 	u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
885 				       vmx->nested.msrs.misc_high);
886 
887 	return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
888 }
889 
890 /*
891  * Load guest's/host's msr at nested entry/exit.
892  * return 0 for success, entry index for failure.
893  *
894  * One of the failure modes for MSR load/store is when a list exceeds the
895  * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
896  * as possible, process all valid entries before failing rather than precheck
897  * for a capacity violation.
898  */
899 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
900 {
901 	u32 i;
902 	struct vmx_msr_entry e;
903 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
904 
905 	for (i = 0; i < count; i++) {
906 		if (unlikely(i >= max_msr_list_size))
907 			goto fail;
908 
909 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
910 					&e, sizeof(e))) {
911 			pr_debug_ratelimited(
912 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
913 				__func__, i, gpa + i * sizeof(e));
914 			goto fail;
915 		}
916 		if (nested_vmx_load_msr_check(vcpu, &e)) {
917 			pr_debug_ratelimited(
918 				"%s check failed (%u, 0x%x, 0x%x)\n",
919 				__func__, i, e.index, e.reserved);
920 			goto fail;
921 		}
922 		if (kvm_set_msr(vcpu, e.index, e.value)) {
923 			pr_debug_ratelimited(
924 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
925 				__func__, i, e.index, e.value);
926 			goto fail;
927 		}
928 	}
929 	return 0;
930 fail:
931 	/* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
932 	return i + 1;
933 }
934 
935 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
936 					    u32 msr_index,
937 					    u64 *data)
938 {
939 	struct vcpu_vmx *vmx = to_vmx(vcpu);
940 
941 	/*
942 	 * If the L0 hypervisor stored a more accurate value for the TSC that
943 	 * does not include the time taken for emulation of the L2->L1
944 	 * VM-exit in L0, use the more accurate value.
945 	 */
946 	if (msr_index == MSR_IA32_TSC) {
947 		int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
948 						    MSR_IA32_TSC);
949 
950 		if (i >= 0) {
951 			u64 val = vmx->msr_autostore.guest.val[i].value;
952 
953 			*data = kvm_read_l1_tsc(vcpu, val);
954 			return true;
955 		}
956 	}
957 
958 	if (kvm_get_msr(vcpu, msr_index, data)) {
959 		pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
960 			msr_index);
961 		return false;
962 	}
963 	return true;
964 }
965 
966 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
967 				     struct vmx_msr_entry *e)
968 {
969 	if (kvm_vcpu_read_guest(vcpu,
970 				gpa + i * sizeof(*e),
971 				e, 2 * sizeof(u32))) {
972 		pr_debug_ratelimited(
973 			"%s cannot read MSR entry (%u, 0x%08llx)\n",
974 			__func__, i, gpa + i * sizeof(*e));
975 		return false;
976 	}
977 	if (nested_vmx_store_msr_check(vcpu, e)) {
978 		pr_debug_ratelimited(
979 			"%s check failed (%u, 0x%x, 0x%x)\n",
980 			__func__, i, e->index, e->reserved);
981 		return false;
982 	}
983 	return true;
984 }
985 
986 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
987 {
988 	u64 data;
989 	u32 i;
990 	struct vmx_msr_entry e;
991 	u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
992 
993 	for (i = 0; i < count; i++) {
994 		if (unlikely(i >= max_msr_list_size))
995 			return -EINVAL;
996 
997 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
998 			return -EINVAL;
999 
1000 		if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1001 			return -EINVAL;
1002 
1003 		if (kvm_vcpu_write_guest(vcpu,
1004 					 gpa + i * sizeof(e) +
1005 					     offsetof(struct vmx_msr_entry, value),
1006 					 &data, sizeof(data))) {
1007 			pr_debug_ratelimited(
1008 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1009 				__func__, i, e.index, data);
1010 			return -EINVAL;
1011 		}
1012 	}
1013 	return 0;
1014 }
1015 
1016 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1017 {
1018 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1019 	u32 count = vmcs12->vm_exit_msr_store_count;
1020 	u64 gpa = vmcs12->vm_exit_msr_store_addr;
1021 	struct vmx_msr_entry e;
1022 	u32 i;
1023 
1024 	for (i = 0; i < count; i++) {
1025 		if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1026 			return false;
1027 
1028 		if (e.index == msr_index)
1029 			return true;
1030 	}
1031 	return false;
1032 }
1033 
1034 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1035 					   u32 msr_index)
1036 {
1037 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1038 	struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1039 	bool in_vmcs12_store_list;
1040 	int msr_autostore_slot;
1041 	bool in_autostore_list;
1042 	int last;
1043 
1044 	msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1045 	in_autostore_list = msr_autostore_slot >= 0;
1046 	in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1047 
1048 	if (in_vmcs12_store_list && !in_autostore_list) {
1049 		if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1050 			/*
1051 			 * Emulated VMEntry does not fail here.  Instead a less
1052 			 * accurate value will be returned by
1053 			 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1054 			 * instead of reading the value from the vmcs02 VMExit
1055 			 * MSR-store area.
1056 			 */
1057 			pr_warn_ratelimited(
1058 				"Not enough msr entries in msr_autostore.  Can't add msr %x\n",
1059 				msr_index);
1060 			return;
1061 		}
1062 		last = autostore->nr++;
1063 		autostore->val[last].index = msr_index;
1064 	} else if (!in_vmcs12_store_list && in_autostore_list) {
1065 		last = --autostore->nr;
1066 		autostore->val[msr_autostore_slot] = autostore->val[last];
1067 	}
1068 }
1069 
1070 /*
1071  * Load guest's/host's cr3 at nested entry/exit.  @nested_ept is true if we are
1072  * emulating VM-Entry into a guest with EPT enabled.  On failure, the expected
1073  * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1074  * @entry_failure_code.
1075  */
1076 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1077 			       bool nested_ept, bool reload_pdptrs,
1078 			       enum vm_entry_failure_code *entry_failure_code)
1079 {
1080 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3))) {
1081 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
1082 		return -EINVAL;
1083 	}
1084 
1085 	/*
1086 	 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1087 	 * must not be dereferenced.
1088 	 */
1089 	if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1090 	    CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))) {
1091 		*entry_failure_code = ENTRY_FAIL_PDPTE;
1092 		return -EINVAL;
1093 	}
1094 
1095 	if (!nested_ept)
1096 		kvm_mmu_new_pgd(vcpu, cr3);
1097 
1098 	vcpu->arch.cr3 = cr3;
1099 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1100 
1101 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1102 	kvm_init_mmu(vcpu);
1103 
1104 	return 0;
1105 }
1106 
1107 /*
1108  * Returns if KVM is able to config CPU to tag TLB entries
1109  * populated by L2 differently than TLB entries populated
1110  * by L1.
1111  *
1112  * If L0 uses EPT, L1 and L2 run with different EPTP because
1113  * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1114  * are tagged with different EPTP.
1115  *
1116  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1117  * with different VPID (L1 entries are tagged with vmx->vpid
1118  * while L2 entries are tagged with vmx->nested.vpid02).
1119  */
1120 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1121 {
1122 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1123 
1124 	return enable_ept ||
1125 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1126 }
1127 
1128 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1129 					    struct vmcs12 *vmcs12,
1130 					    bool is_vmenter)
1131 {
1132 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1133 
1134 	/*
1135 	 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1136 	 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1137 	 * full TLB flush from the guest's perspective.  This is required even
1138 	 * if VPID is disabled in the host as KVM may need to synchronize the
1139 	 * MMU in response to the guest TLB flush.
1140 	 *
1141 	 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1142 	 * EPT is a special snowflake, as guest-physical mappings aren't
1143 	 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1144 	 * VPID disabled.  As a result, KVM _never_ needs to sync nEPT
1145 	 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1146 	 * those mappings.
1147 	 */
1148 	if (!nested_cpu_has_vpid(vmcs12)) {
1149 		kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1150 		return;
1151 	}
1152 
1153 	/* L2 should never have a VPID if VPID is disabled. */
1154 	WARN_ON(!enable_vpid);
1155 
1156 	/*
1157 	 * If VPID is enabled and used by vmc12, but L2 does not have a unique
1158 	 * TLB tag (ASID), i.e. EPT is disabled and KVM was unable to allocate
1159 	 * a VPID for L2, flush the current context as the effective ASID is
1160 	 * common to both L1 and L2.
1161 	 *
1162 	 * Defer the flush so that it runs after vmcs02.EPTP has been set by
1163 	 * KVM_REQ_LOAD_MMU_PGD (if nested EPT is enabled) and to avoid
1164 	 * redundant flushes further down the nested pipeline.
1165 	 *
1166 	 * If a TLB flush isn't required due to any of the above, and vpid12 is
1167 	 * changing then the new "virtual" VPID (vpid12) will reuse the same
1168 	 * "real" VPID (vpid02), and so needs to be flushed.  There's no direct
1169 	 * mapping between vpid02 and vpid12, vpid02 is per-vCPU and reused for
1170 	 * all nested vCPUs.  Remember, a flush on VM-Enter does not invalidate
1171 	 * guest-physical mappings, so there is no need to sync the nEPT MMU.
1172 	 */
1173 	if (!nested_has_guest_tlb_tag(vcpu)) {
1174 		kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1175 	} else if (is_vmenter &&
1176 		   vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1177 		vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1178 		vpid_sync_context(nested_get_vpid02(vcpu));
1179 	}
1180 }
1181 
1182 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1183 {
1184 	superset &= mask;
1185 	subset &= mask;
1186 
1187 	return (superset | subset) == superset;
1188 }
1189 
1190 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1191 {
1192 	const u64 feature_and_reserved =
1193 		/* feature (except bit 48; see below) */
1194 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1195 		/* reserved */
1196 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1197 	u64 vmx_basic = vmx->nested.msrs.basic;
1198 
1199 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1200 		return -EINVAL;
1201 
1202 	/*
1203 	 * KVM does not emulate a version of VMX that constrains physical
1204 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1205 	 */
1206 	if (data & BIT_ULL(48))
1207 		return -EINVAL;
1208 
1209 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1210 	    vmx_basic_vmcs_revision_id(data))
1211 		return -EINVAL;
1212 
1213 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1214 		return -EINVAL;
1215 
1216 	vmx->nested.msrs.basic = data;
1217 	return 0;
1218 }
1219 
1220 static int
1221 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1222 {
1223 	u64 supported;
1224 	u32 *lowp, *highp;
1225 
1226 	switch (msr_index) {
1227 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1228 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1229 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1230 		break;
1231 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1232 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1233 		highp = &vmx->nested.msrs.procbased_ctls_high;
1234 		break;
1235 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1236 		lowp = &vmx->nested.msrs.exit_ctls_low;
1237 		highp = &vmx->nested.msrs.exit_ctls_high;
1238 		break;
1239 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1240 		lowp = &vmx->nested.msrs.entry_ctls_low;
1241 		highp = &vmx->nested.msrs.entry_ctls_high;
1242 		break;
1243 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1244 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1245 		highp = &vmx->nested.msrs.secondary_ctls_high;
1246 		break;
1247 	default:
1248 		BUG();
1249 	}
1250 
1251 	supported = vmx_control_msr(*lowp, *highp);
1252 
1253 	/* Check must-be-1 bits are still 1. */
1254 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1255 		return -EINVAL;
1256 
1257 	/* Check must-be-0 bits are still 0. */
1258 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1259 		return -EINVAL;
1260 
1261 	*lowp = data;
1262 	*highp = data >> 32;
1263 	return 0;
1264 }
1265 
1266 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1267 {
1268 	const u64 feature_and_reserved_bits =
1269 		/* feature */
1270 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1271 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1272 		/* reserved */
1273 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1274 	u64 vmx_misc;
1275 
1276 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1277 				   vmx->nested.msrs.misc_high);
1278 
1279 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1280 		return -EINVAL;
1281 
1282 	if ((vmx->nested.msrs.pinbased_ctls_high &
1283 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1284 	    vmx_misc_preemption_timer_rate(data) !=
1285 	    vmx_misc_preemption_timer_rate(vmx_misc))
1286 		return -EINVAL;
1287 
1288 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1289 		return -EINVAL;
1290 
1291 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1292 		return -EINVAL;
1293 
1294 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1295 		return -EINVAL;
1296 
1297 	vmx->nested.msrs.misc_low = data;
1298 	vmx->nested.msrs.misc_high = data >> 32;
1299 
1300 	return 0;
1301 }
1302 
1303 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1304 {
1305 	u64 vmx_ept_vpid_cap;
1306 
1307 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1308 					   vmx->nested.msrs.vpid_caps);
1309 
1310 	/* Every bit is either reserved or a feature bit. */
1311 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1312 		return -EINVAL;
1313 
1314 	vmx->nested.msrs.ept_caps = data;
1315 	vmx->nested.msrs.vpid_caps = data >> 32;
1316 	return 0;
1317 }
1318 
1319 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1320 {
1321 	u64 *msr;
1322 
1323 	switch (msr_index) {
1324 	case MSR_IA32_VMX_CR0_FIXED0:
1325 		msr = &vmx->nested.msrs.cr0_fixed0;
1326 		break;
1327 	case MSR_IA32_VMX_CR4_FIXED0:
1328 		msr = &vmx->nested.msrs.cr4_fixed0;
1329 		break;
1330 	default:
1331 		BUG();
1332 	}
1333 
1334 	/*
1335 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1336 	 * must be 1 in the restored value.
1337 	 */
1338 	if (!is_bitwise_subset(data, *msr, -1ULL))
1339 		return -EINVAL;
1340 
1341 	*msr = data;
1342 	return 0;
1343 }
1344 
1345 /*
1346  * Called when userspace is restoring VMX MSRs.
1347  *
1348  * Returns 0 on success, non-0 otherwise.
1349  */
1350 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1351 {
1352 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1353 
1354 	/*
1355 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1356 	 * is in VMX operation.
1357 	 */
1358 	if (vmx->nested.vmxon)
1359 		return -EBUSY;
1360 
1361 	switch (msr_index) {
1362 	case MSR_IA32_VMX_BASIC:
1363 		return vmx_restore_vmx_basic(vmx, data);
1364 	case MSR_IA32_VMX_PINBASED_CTLS:
1365 	case MSR_IA32_VMX_PROCBASED_CTLS:
1366 	case MSR_IA32_VMX_EXIT_CTLS:
1367 	case MSR_IA32_VMX_ENTRY_CTLS:
1368 		/*
1369 		 * The "non-true" VMX capability MSRs are generated from the
1370 		 * "true" MSRs, so we do not support restoring them directly.
1371 		 *
1372 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1373 		 * should restore the "true" MSRs with the must-be-1 bits
1374 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1375 		 * DEFAULT SETTINGS".
1376 		 */
1377 		return -EINVAL;
1378 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1379 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1380 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1381 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1382 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1383 		return vmx_restore_control_msr(vmx, msr_index, data);
1384 	case MSR_IA32_VMX_MISC:
1385 		return vmx_restore_vmx_misc(vmx, data);
1386 	case MSR_IA32_VMX_CR0_FIXED0:
1387 	case MSR_IA32_VMX_CR4_FIXED0:
1388 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1389 	case MSR_IA32_VMX_CR0_FIXED1:
1390 	case MSR_IA32_VMX_CR4_FIXED1:
1391 		/*
1392 		 * These MSRs are generated based on the vCPU's CPUID, so we
1393 		 * do not support restoring them directly.
1394 		 */
1395 		return -EINVAL;
1396 	case MSR_IA32_VMX_EPT_VPID_CAP:
1397 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1398 	case MSR_IA32_VMX_VMCS_ENUM:
1399 		vmx->nested.msrs.vmcs_enum = data;
1400 		return 0;
1401 	case MSR_IA32_VMX_VMFUNC:
1402 		if (data & ~vmx->nested.msrs.vmfunc_controls)
1403 			return -EINVAL;
1404 		vmx->nested.msrs.vmfunc_controls = data;
1405 		return 0;
1406 	default:
1407 		/*
1408 		 * The rest of the VMX capability MSRs do not support restore.
1409 		 */
1410 		return -EINVAL;
1411 	}
1412 }
1413 
1414 /* Returns 0 on success, non-0 otherwise. */
1415 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1416 {
1417 	switch (msr_index) {
1418 	case MSR_IA32_VMX_BASIC:
1419 		*pdata = msrs->basic;
1420 		break;
1421 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1422 	case MSR_IA32_VMX_PINBASED_CTLS:
1423 		*pdata = vmx_control_msr(
1424 			msrs->pinbased_ctls_low,
1425 			msrs->pinbased_ctls_high);
1426 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1427 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1428 		break;
1429 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1430 	case MSR_IA32_VMX_PROCBASED_CTLS:
1431 		*pdata = vmx_control_msr(
1432 			msrs->procbased_ctls_low,
1433 			msrs->procbased_ctls_high);
1434 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1435 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1436 		break;
1437 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1438 	case MSR_IA32_VMX_EXIT_CTLS:
1439 		*pdata = vmx_control_msr(
1440 			msrs->exit_ctls_low,
1441 			msrs->exit_ctls_high);
1442 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1443 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1444 		break;
1445 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1446 	case MSR_IA32_VMX_ENTRY_CTLS:
1447 		*pdata = vmx_control_msr(
1448 			msrs->entry_ctls_low,
1449 			msrs->entry_ctls_high);
1450 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1451 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1452 		break;
1453 	case MSR_IA32_VMX_MISC:
1454 		*pdata = vmx_control_msr(
1455 			msrs->misc_low,
1456 			msrs->misc_high);
1457 		break;
1458 	case MSR_IA32_VMX_CR0_FIXED0:
1459 		*pdata = msrs->cr0_fixed0;
1460 		break;
1461 	case MSR_IA32_VMX_CR0_FIXED1:
1462 		*pdata = msrs->cr0_fixed1;
1463 		break;
1464 	case MSR_IA32_VMX_CR4_FIXED0:
1465 		*pdata = msrs->cr4_fixed0;
1466 		break;
1467 	case MSR_IA32_VMX_CR4_FIXED1:
1468 		*pdata = msrs->cr4_fixed1;
1469 		break;
1470 	case MSR_IA32_VMX_VMCS_ENUM:
1471 		*pdata = msrs->vmcs_enum;
1472 		break;
1473 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1474 		*pdata = vmx_control_msr(
1475 			msrs->secondary_ctls_low,
1476 			msrs->secondary_ctls_high);
1477 		break;
1478 	case MSR_IA32_VMX_EPT_VPID_CAP:
1479 		*pdata = msrs->ept_caps |
1480 			((u64)msrs->vpid_caps << 32);
1481 		break;
1482 	case MSR_IA32_VMX_VMFUNC:
1483 		*pdata = msrs->vmfunc_controls;
1484 		break;
1485 	default:
1486 		return 1;
1487 	}
1488 
1489 	return 0;
1490 }
1491 
1492 /*
1493  * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1494  * been modified by the L1 guest.  Note, "writable" in this context means
1495  * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1496  * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1497  * VM-exit information fields (which are actually writable if the vCPU is
1498  * configured to support "VMWRITE to any supported field in the VMCS").
1499  */
1500 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1501 {
1502 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1503 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1504 	struct shadow_vmcs_field field;
1505 	unsigned long val;
1506 	int i;
1507 
1508 	if (WARN_ON(!shadow_vmcs))
1509 		return;
1510 
1511 	preempt_disable();
1512 
1513 	vmcs_load(shadow_vmcs);
1514 
1515 	for (i = 0; i < max_shadow_read_write_fields; i++) {
1516 		field = shadow_read_write_fields[i];
1517 		val = __vmcs_readl(field.encoding);
1518 		vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1519 	}
1520 
1521 	vmcs_clear(shadow_vmcs);
1522 	vmcs_load(vmx->loaded_vmcs->vmcs);
1523 
1524 	preempt_enable();
1525 }
1526 
1527 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1528 {
1529 	const struct shadow_vmcs_field *fields[] = {
1530 		shadow_read_write_fields,
1531 		shadow_read_only_fields
1532 	};
1533 	const int max_fields[] = {
1534 		max_shadow_read_write_fields,
1535 		max_shadow_read_only_fields
1536 	};
1537 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1538 	struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1539 	struct shadow_vmcs_field field;
1540 	unsigned long val;
1541 	int i, q;
1542 
1543 	if (WARN_ON(!shadow_vmcs))
1544 		return;
1545 
1546 	vmcs_load(shadow_vmcs);
1547 
1548 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1549 		for (i = 0; i < max_fields[q]; i++) {
1550 			field = fields[q][i];
1551 			val = vmcs12_read_any(vmcs12, field.encoding,
1552 					      field.offset);
1553 			__vmcs_writel(field.encoding, val);
1554 		}
1555 	}
1556 
1557 	vmcs_clear(shadow_vmcs);
1558 	vmcs_load(vmx->loaded_vmcs->vmcs);
1559 }
1560 
1561 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1562 {
1563 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1564 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1565 
1566 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1567 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1568 	vmcs12->guest_rip = evmcs->guest_rip;
1569 
1570 	if (unlikely(!(hv_clean_fields &
1571 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1572 		vmcs12->guest_rsp = evmcs->guest_rsp;
1573 		vmcs12->guest_rflags = evmcs->guest_rflags;
1574 		vmcs12->guest_interruptibility_info =
1575 			evmcs->guest_interruptibility_info;
1576 	}
1577 
1578 	if (unlikely(!(hv_clean_fields &
1579 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1580 		vmcs12->cpu_based_vm_exec_control =
1581 			evmcs->cpu_based_vm_exec_control;
1582 	}
1583 
1584 	if (unlikely(!(hv_clean_fields &
1585 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1586 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1587 	}
1588 
1589 	if (unlikely(!(hv_clean_fields &
1590 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1591 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1592 	}
1593 
1594 	if (unlikely(!(hv_clean_fields &
1595 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1596 		vmcs12->vm_entry_intr_info_field =
1597 			evmcs->vm_entry_intr_info_field;
1598 		vmcs12->vm_entry_exception_error_code =
1599 			evmcs->vm_entry_exception_error_code;
1600 		vmcs12->vm_entry_instruction_len =
1601 			evmcs->vm_entry_instruction_len;
1602 	}
1603 
1604 	if (unlikely(!(hv_clean_fields &
1605 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1606 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1607 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1608 		vmcs12->host_cr0 = evmcs->host_cr0;
1609 		vmcs12->host_cr3 = evmcs->host_cr3;
1610 		vmcs12->host_cr4 = evmcs->host_cr4;
1611 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1612 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1613 		vmcs12->host_rip = evmcs->host_rip;
1614 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1615 		vmcs12->host_es_selector = evmcs->host_es_selector;
1616 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1617 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1618 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1619 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1620 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1621 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1622 	}
1623 
1624 	if (unlikely(!(hv_clean_fields &
1625 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1626 		vmcs12->pin_based_vm_exec_control =
1627 			evmcs->pin_based_vm_exec_control;
1628 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1629 		vmcs12->secondary_vm_exec_control =
1630 			evmcs->secondary_vm_exec_control;
1631 	}
1632 
1633 	if (unlikely(!(hv_clean_fields &
1634 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1635 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1636 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1637 	}
1638 
1639 	if (unlikely(!(hv_clean_fields &
1640 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1641 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1642 	}
1643 
1644 	if (unlikely(!(hv_clean_fields &
1645 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1646 		vmcs12->guest_es_base = evmcs->guest_es_base;
1647 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1648 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1649 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1650 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1651 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1652 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1653 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1654 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1655 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1656 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1657 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1658 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1659 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1660 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1661 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1662 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1663 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1664 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1665 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1666 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1667 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1668 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1669 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1670 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1671 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1672 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1673 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1674 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1675 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1676 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1677 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1678 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1679 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1680 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1681 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1682 	}
1683 
1684 	if (unlikely(!(hv_clean_fields &
1685 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1686 		vmcs12->tsc_offset = evmcs->tsc_offset;
1687 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1688 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1689 	}
1690 
1691 	if (unlikely(!(hv_clean_fields &
1692 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1693 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1694 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1695 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1696 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1697 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1698 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1699 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1700 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1701 	}
1702 
1703 	if (unlikely(!(hv_clean_fields &
1704 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1705 		vmcs12->host_fs_base = evmcs->host_fs_base;
1706 		vmcs12->host_gs_base = evmcs->host_gs_base;
1707 		vmcs12->host_tr_base = evmcs->host_tr_base;
1708 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1709 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1710 		vmcs12->host_rsp = evmcs->host_rsp;
1711 	}
1712 
1713 	if (unlikely(!(hv_clean_fields &
1714 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1715 		vmcs12->ept_pointer = evmcs->ept_pointer;
1716 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1717 	}
1718 
1719 	if (unlikely(!(hv_clean_fields &
1720 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1721 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1722 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1723 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1724 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1725 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1726 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1727 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1728 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1729 		vmcs12->guest_pending_dbg_exceptions =
1730 			evmcs->guest_pending_dbg_exceptions;
1731 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1732 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1733 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1734 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1735 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1736 	}
1737 
1738 	/*
1739 	 * Not used?
1740 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1741 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1742 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1743 	 * vmcs12->page_fault_error_code_mask =
1744 	 *		evmcs->page_fault_error_code_mask;
1745 	 * vmcs12->page_fault_error_code_match =
1746 	 *		evmcs->page_fault_error_code_match;
1747 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1748 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1749 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1750 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1751 	 */
1752 
1753 	/*
1754 	 * Read only fields:
1755 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1756 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1757 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1758 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1759 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1760 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1761 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1762 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1763 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1764 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1765 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1766 	 *
1767 	 * Not present in struct vmcs12:
1768 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1769 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1770 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1771 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1772 	 */
1773 
1774 	return;
1775 }
1776 
1777 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1778 {
1779 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1780 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1781 
1782 	/*
1783 	 * Should not be changed by KVM:
1784 	 *
1785 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1786 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1787 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1788 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1789 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1790 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1791 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1792 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1793 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1794 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1795 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1796 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1797 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1798 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1799 	 * evmcs->host_rip = vmcs12->host_rip;
1800 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1801 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1802 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1803 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1804 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1805 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1806 	 * evmcs->host_rsp = vmcs12->host_rsp;
1807 	 * sync_vmcs02_to_vmcs12() doesn't read these:
1808 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1809 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1810 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1811 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1812 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1813 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1814 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1815 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1816 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1817 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1818 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1819 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1820 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1821 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1822 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1823 	 * evmcs->page_fault_error_code_mask =
1824 	 *		vmcs12->page_fault_error_code_mask;
1825 	 * evmcs->page_fault_error_code_match =
1826 	 *		vmcs12->page_fault_error_code_match;
1827 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1828 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1829 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1830 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1831 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1832 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1833 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1834 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1835 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1836 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1837 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1838 	 *
1839 	 * Not present in struct vmcs12:
1840 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1841 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1842 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1843 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1844 	 */
1845 
1846 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1847 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1848 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1849 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1850 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1851 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1852 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1853 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1854 
1855 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1856 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1857 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1858 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1859 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1860 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1861 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1862 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1863 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1864 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1865 
1866 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1867 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1868 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1869 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1870 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1871 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1872 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1873 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1874 
1875 	evmcs->guest_es_base = vmcs12->guest_es_base;
1876 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1877 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1878 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1879 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1880 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1881 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1882 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1883 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1884 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1885 
1886 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1887 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1888 
1889 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1890 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1891 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1892 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1893 
1894 	evmcs->guest_pending_dbg_exceptions =
1895 		vmcs12->guest_pending_dbg_exceptions;
1896 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1897 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1898 
1899 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1900 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1901 
1902 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1903 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1904 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1905 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1906 
1907 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1908 
1909 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1910 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1911 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1912 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1913 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1914 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1915 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1916 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1917 
1918 	evmcs->exit_qualification = vmcs12->exit_qualification;
1919 
1920 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1921 	evmcs->guest_rsp = vmcs12->guest_rsp;
1922 	evmcs->guest_rflags = vmcs12->guest_rflags;
1923 
1924 	evmcs->guest_interruptibility_info =
1925 		vmcs12->guest_interruptibility_info;
1926 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1927 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1928 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1929 	evmcs->vm_entry_exception_error_code =
1930 		vmcs12->vm_entry_exception_error_code;
1931 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1932 
1933 	evmcs->guest_rip = vmcs12->guest_rip;
1934 
1935 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1936 
1937 	return;
1938 }
1939 
1940 /*
1941  * This is an equivalent of the nested hypervisor executing the vmptrld
1942  * instruction.
1943  */
1944 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
1945 	struct kvm_vcpu *vcpu, bool from_launch)
1946 {
1947 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1948 	bool evmcs_gpa_changed = false;
1949 	u64 evmcs_gpa;
1950 
1951 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1952 		return EVMPTRLD_DISABLED;
1953 
1954 	if (!nested_enlightened_vmentry(vcpu, &evmcs_gpa)) {
1955 		nested_release_evmcs(vcpu);
1956 		return EVMPTRLD_DISABLED;
1957 	}
1958 
1959 	if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
1960 		vmx->nested.current_vmptr = -1ull;
1961 
1962 		nested_release_evmcs(vcpu);
1963 
1964 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
1965 				 &vmx->nested.hv_evmcs_map))
1966 			return EVMPTRLD_ERROR;
1967 
1968 		vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
1969 
1970 		/*
1971 		 * Currently, KVM only supports eVMCS version 1
1972 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1973 		 * value to first u32 field of eVMCS which should specify eVMCS
1974 		 * VersionNumber.
1975 		 *
1976 		 * Guest should be aware of supported eVMCS versions by host by
1977 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1978 		 * expected to set this CPUID leaf according to the value
1979 		 * returned in vmcs_version from nested_enable_evmcs().
1980 		 *
1981 		 * However, it turns out that Microsoft Hyper-V fails to comply
1982 		 * to their own invented interface: When Hyper-V use eVMCS, it
1983 		 * just sets first u32 field of eVMCS to revision_id specified
1984 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1985 		 * which is one of the supported versions specified in
1986 		 * CPUID.0x4000000A.EAX[0:15].
1987 		 *
1988 		 * To overcome Hyper-V bug, we accept here either a supported
1989 		 * eVMCS version or VMCS12 revision_id as valid values for first
1990 		 * u32 field of eVMCS.
1991 		 */
1992 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1993 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1994 			nested_release_evmcs(vcpu);
1995 			return EVMPTRLD_VMFAIL;
1996 		}
1997 
1998 		vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
1999 
2000 		evmcs_gpa_changed = true;
2001 		/*
2002 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2003 		 * reloaded from guest's memory (read only fields, fields not
2004 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
2005 		 * are no leftovers.
2006 		 */
2007 		if (from_launch) {
2008 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2009 			memset(vmcs12, 0, sizeof(*vmcs12));
2010 			vmcs12->hdr.revision_id = VMCS12_REVISION;
2011 		}
2012 
2013 	}
2014 
2015 	/*
2016 	 * Clean fields data can't be used on VMLAUNCH and when we switch
2017 	 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2018 	 */
2019 	if (from_launch || evmcs_gpa_changed)
2020 		vmx->nested.hv_evmcs->hv_clean_fields &=
2021 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2022 
2023 	return EVMPTRLD_SUCCEEDED;
2024 }
2025 
2026 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2027 {
2028 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2029 
2030 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2031 		copy_vmcs12_to_enlightened(vmx);
2032 	else
2033 		copy_vmcs12_to_shadow(vmx);
2034 
2035 	vmx->nested.need_vmcs12_to_shadow_sync = false;
2036 }
2037 
2038 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2039 {
2040 	struct vcpu_vmx *vmx =
2041 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2042 
2043 	vmx->nested.preemption_timer_expired = true;
2044 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2045 	kvm_vcpu_kick(&vmx->vcpu);
2046 
2047 	return HRTIMER_NORESTART;
2048 }
2049 
2050 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2051 {
2052 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2053 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2054 
2055 	u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2056 			    VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2057 
2058 	if (!vmx->nested.has_preemption_timer_deadline) {
2059 		vmx->nested.preemption_timer_deadline =
2060 			vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2061 		vmx->nested.has_preemption_timer_deadline = true;
2062 	}
2063 	return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2064 }
2065 
2066 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2067 					u64 preemption_timeout)
2068 {
2069 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2070 
2071 	/*
2072 	 * A timer value of zero is architecturally guaranteed to cause
2073 	 * a VMExit prior to executing any instructions in the guest.
2074 	 */
2075 	if (preemption_timeout == 0) {
2076 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2077 		return;
2078 	}
2079 
2080 	if (vcpu->arch.virtual_tsc_khz == 0)
2081 		return;
2082 
2083 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2084 	preemption_timeout *= 1000000;
2085 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2086 	hrtimer_start(&vmx->nested.preemption_timer,
2087 		      ktime_add_ns(ktime_get(), preemption_timeout),
2088 		      HRTIMER_MODE_ABS_PINNED);
2089 }
2090 
2091 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2092 {
2093 	if (vmx->nested.nested_run_pending &&
2094 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2095 		return vmcs12->guest_ia32_efer;
2096 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2097 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2098 	else
2099 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2100 }
2101 
2102 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2103 {
2104 	/*
2105 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2106 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
2107 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2108 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
2109 	 */
2110 	if (vmx->nested.vmcs02_initialized)
2111 		return;
2112 	vmx->nested.vmcs02_initialized = true;
2113 
2114 	/*
2115 	 * We don't care what the EPTP value is we just need to guarantee
2116 	 * it's valid so we don't get a false positive when doing early
2117 	 * consistency checks.
2118 	 */
2119 	if (enable_ept && nested_early_check)
2120 		vmcs_write64(EPT_POINTER,
2121 			     construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2122 
2123 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
2124 	if (cpu_has_vmx_vmfunc())
2125 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
2126 
2127 	if (cpu_has_vmx_posted_intr())
2128 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2129 
2130 	if (cpu_has_vmx_msr_bitmap())
2131 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2132 
2133 	/*
2134 	 * PML is emulated for L2, but never enabled in hardware as the MMU
2135 	 * handles A/D emulation.  Disabling PML for L2 also avoids having to
2136 	 * deal with filtering out L2 GPAs from the buffer.
2137 	 */
2138 	if (enable_pml) {
2139 		vmcs_write64(PML_ADDRESS, 0);
2140 		vmcs_write16(GUEST_PML_INDEX, -1);
2141 	}
2142 
2143 	if (cpu_has_vmx_encls_vmexit())
2144 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2145 
2146 	/*
2147 	 * Set the MSR load/store lists to match L0's settings.  Only the
2148 	 * addresses are constant (for vmcs02), the counts can change based
2149 	 * on L2's behavior, e.g. switching to/from long mode.
2150 	 */
2151 	vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2152 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2153 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2154 
2155 	vmx_set_constant_host_state(vmx);
2156 }
2157 
2158 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2159 				      struct vmcs12 *vmcs12)
2160 {
2161 	prepare_vmcs02_constant_state(vmx);
2162 
2163 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
2164 
2165 	if (enable_vpid) {
2166 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2167 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2168 		else
2169 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2170 	}
2171 }
2172 
2173 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2174 {
2175 	u32 exec_control;
2176 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2177 
2178 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2179 		prepare_vmcs02_early_rare(vmx, vmcs12);
2180 
2181 	/*
2182 	 * PIN CONTROLS
2183 	 */
2184 	exec_control = vmx_pin_based_exec_ctrl(vmx);
2185 	exec_control |= (vmcs12->pin_based_vm_exec_control &
2186 			 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2187 
2188 	/* Posted interrupts setting is only taken from vmcs12.  */
2189 	if (nested_cpu_has_posted_intr(vmcs12)) {
2190 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2191 		vmx->nested.pi_pending = false;
2192 	} else {
2193 		exec_control &= ~PIN_BASED_POSTED_INTR;
2194 	}
2195 	pin_controls_set(vmx, exec_control);
2196 
2197 	/*
2198 	 * EXEC CONTROLS
2199 	 */
2200 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2201 	exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2202 	exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2203 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2204 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2205 
2206 	vmx->nested.l1_tpr_threshold = -1;
2207 	if (exec_control & CPU_BASED_TPR_SHADOW)
2208 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2209 #ifdef CONFIG_X86_64
2210 	else
2211 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2212 				CPU_BASED_CR8_STORE_EXITING;
2213 #endif
2214 
2215 	/*
2216 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2217 	 * for I/O port accesses.
2218 	 */
2219 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2220 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2221 
2222 	/*
2223 	 * This bit will be computed in nested_get_vmcs12_pages, because
2224 	 * we do not have access to L1's MSR bitmap yet.  For now, keep
2225 	 * the same bit as before, hoping to avoid multiple VMWRITEs that
2226 	 * only set/clear this bit.
2227 	 */
2228 	exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2229 	exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2230 
2231 	exec_controls_set(vmx, exec_control);
2232 
2233 	/*
2234 	 * SECONDARY EXEC CONTROLS
2235 	 */
2236 	if (cpu_has_secondary_exec_ctrls()) {
2237 		exec_control = vmx->secondary_exec_control;
2238 
2239 		/* Take the following fields only from vmcs12 */
2240 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2241 				  SECONDARY_EXEC_ENABLE_INVPCID |
2242 				  SECONDARY_EXEC_ENABLE_RDTSCP |
2243 				  SECONDARY_EXEC_XSAVES |
2244 				  SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2245 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2246 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2247 				  SECONDARY_EXEC_ENABLE_VMFUNC |
2248 				  SECONDARY_EXEC_TSC_SCALING);
2249 		if (nested_cpu_has(vmcs12,
2250 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2251 			exec_control |= vmcs12->secondary_vm_exec_control;
2252 
2253 		/* PML is emulated and never enabled in hardware for L2. */
2254 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2255 
2256 		/* VMCS shadowing for L2 is emulated for now */
2257 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2258 
2259 		/*
2260 		 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2261 		 * will not have to rewrite the controls just for this bit.
2262 		 */
2263 		if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated() &&
2264 		    (vmcs12->guest_cr4 & X86_CR4_UMIP))
2265 			exec_control |= SECONDARY_EXEC_DESC;
2266 
2267 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2268 			vmcs_write16(GUEST_INTR_STATUS,
2269 				vmcs12->guest_intr_status);
2270 
2271 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2272 		    exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2273 
2274 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2275 			vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2276 
2277 		secondary_exec_controls_set(vmx, exec_control);
2278 	}
2279 
2280 	/*
2281 	 * ENTRY CONTROLS
2282 	 *
2283 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2284 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2285 	 * on the related bits (if supported by the CPU) in the hope that
2286 	 * we can avoid VMWrites during vmx_set_efer().
2287 	 */
2288 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2289 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2290 	if (cpu_has_load_ia32_efer()) {
2291 		if (guest_efer & EFER_LMA)
2292 			exec_control |= VM_ENTRY_IA32E_MODE;
2293 		if (guest_efer != host_efer)
2294 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2295 	}
2296 	vm_entry_controls_set(vmx, exec_control);
2297 
2298 	/*
2299 	 * EXIT CONTROLS
2300 	 *
2301 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2302 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2303 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2304 	 */
2305 	exec_control = vmx_vmexit_ctrl();
2306 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2307 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2308 	vm_exit_controls_set(vmx, exec_control);
2309 
2310 	/*
2311 	 * Interrupt/Exception Fields
2312 	 */
2313 	if (vmx->nested.nested_run_pending) {
2314 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2315 			     vmcs12->vm_entry_intr_info_field);
2316 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2317 			     vmcs12->vm_entry_exception_error_code);
2318 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2319 			     vmcs12->vm_entry_instruction_len);
2320 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2321 			     vmcs12->guest_interruptibility_info);
2322 		vmx->loaded_vmcs->nmi_known_unmasked =
2323 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2324 	} else {
2325 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2326 	}
2327 }
2328 
2329 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2330 {
2331 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2332 
2333 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2334 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2335 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2336 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2337 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2338 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2339 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2340 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2341 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2342 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2343 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2344 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2345 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2346 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2347 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2348 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2349 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2350 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2351 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2352 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2353 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2354 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2355 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2356 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2357 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2358 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2359 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2360 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2361 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2362 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2363 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2364 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2365 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2366 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2367 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2368 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2369 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2370 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2371 
2372 		vmx->segment_cache.bitmask = 0;
2373 	}
2374 
2375 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2376 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2377 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2378 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2379 			    vmcs12->guest_pending_dbg_exceptions);
2380 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2381 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2382 
2383 		/*
2384 		 * L1 may access the L2's PDPTR, so save them to construct
2385 		 * vmcs12
2386 		 */
2387 		if (enable_ept) {
2388 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2389 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2390 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2391 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2392 		}
2393 
2394 		if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2395 		    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2396 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2397 	}
2398 
2399 	if (nested_cpu_has_xsaves(vmcs12))
2400 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2401 
2402 	/*
2403 	 * Whether page-faults are trapped is determined by a combination of
2404 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.  If L0
2405 	 * doesn't care about page faults then we should set all of these to
2406 	 * L1's desires. However, if L0 does care about (some) page faults, it
2407 	 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2408 	 * simply ask to exit on each and every L2 page fault. This is done by
2409 	 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2410 	 * Note that below we don't need special code to set EB.PF beyond the
2411 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2412 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2413 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2414 	 */
2415 	if (vmx_need_pf_intercept(&vmx->vcpu)) {
2416 		/*
2417 		 * TODO: if both L0 and L1 need the same MASK and MATCH,
2418 		 * go ahead and use it?
2419 		 */
2420 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2421 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2422 	} else {
2423 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2424 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2425 	}
2426 
2427 	if (cpu_has_vmx_apicv()) {
2428 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2429 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2430 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2431 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2432 	}
2433 
2434 	/*
2435 	 * Make sure the msr_autostore list is up to date before we set the
2436 	 * count in the vmcs02.
2437 	 */
2438 	prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2439 
2440 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2441 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2442 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2443 
2444 	set_cr4_guest_host_mask(vmx);
2445 }
2446 
2447 /*
2448  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2449  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2450  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2451  * guest in a way that will both be appropriate to L1's requests, and our
2452  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2453  * function also has additional necessary side-effects, like setting various
2454  * vcpu->arch fields.
2455  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2456  * is assigned to entry_failure_code on failure.
2457  */
2458 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2459 			  bool from_vmentry,
2460 			  enum vm_entry_failure_code *entry_failure_code)
2461 {
2462 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2463 	bool load_guest_pdptrs_vmcs12 = false;
2464 
2465 	if (vmx->nested.dirty_vmcs12 || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
2466 		prepare_vmcs02_rare(vmx, vmcs12);
2467 		vmx->nested.dirty_vmcs12 = false;
2468 
2469 		load_guest_pdptrs_vmcs12 = !evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) ||
2470 			!(vmx->nested.hv_evmcs->hv_clean_fields &
2471 			  HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2472 	}
2473 
2474 	if (vmx->nested.nested_run_pending &&
2475 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2476 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2477 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2478 	} else {
2479 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2480 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2481 	}
2482 	if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2483 	    !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2484 		vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2485 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2486 
2487 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2488 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2489 	 * trap. Note that CR0.TS also needs updating - we do this later.
2490 	 */
2491 	vmx_update_exception_bitmap(vcpu);
2492 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2493 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2494 
2495 	if (vmx->nested.nested_run_pending &&
2496 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2497 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2498 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2499 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2500 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2501 	}
2502 
2503 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2504 			vcpu->arch.l1_tsc_offset,
2505 			vmx_get_l2_tsc_offset(vcpu),
2506 			vmx_get_l2_tsc_multiplier(vcpu));
2507 
2508 	vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2509 			vcpu->arch.l1_tsc_scaling_ratio,
2510 			vmx_get_l2_tsc_multiplier(vcpu));
2511 
2512 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2513 	if (kvm_has_tsc_control)
2514 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2515 
2516 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2517 
2518 	if (nested_cpu_has_ept(vmcs12))
2519 		nested_ept_init_mmu_context(vcpu);
2520 
2521 	/*
2522 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2523 	 * bits which we consider mandatory enabled.
2524 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2525 	 * the specifications by L1; It's not enough to take
2526 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2527 	 * have more bits than L1 expected.
2528 	 */
2529 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2530 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2531 
2532 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2533 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2534 
2535 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2536 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2537 	vmx_set_efer(vcpu, vcpu->arch.efer);
2538 
2539 	/*
2540 	 * Guest state is invalid and unrestricted guest is disabled,
2541 	 * which means L1 attempted VMEntry to L2 with invalid state.
2542 	 * Fail the VMEntry.
2543 	 */
2544 	if (CC(!vmx_guest_state_valid(vcpu))) {
2545 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2546 		return -EINVAL;
2547 	}
2548 
2549 	/* Shadow page tables on either EPT or shadow page tables. */
2550 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2551 				from_vmentry, entry_failure_code))
2552 		return -EINVAL;
2553 
2554 	/*
2555 	 * Immediately write vmcs02.GUEST_CR3.  It will be propagated to vmcs12
2556 	 * on nested VM-Exit, which can occur without actually running L2 and
2557 	 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2558 	 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2559 	 * transition to HLT instead of running L2.
2560 	 */
2561 	if (enable_ept)
2562 		vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2563 
2564 	/* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2565 	if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2566 	    is_pae_paging(vcpu)) {
2567 		vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2568 		vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2569 		vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2570 		vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2571 	}
2572 
2573 	if (!enable_ept)
2574 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2575 
2576 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2577 	    WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2578 				     vmcs12->guest_ia32_perf_global_ctrl)))
2579 		return -EINVAL;
2580 
2581 	kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2582 	kvm_rip_write(vcpu, vmcs12->guest_rip);
2583 
2584 	/*
2585 	 * It was observed that genuine Hyper-V running in L1 doesn't reset
2586 	 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2587 	 * bits when it changes a field in eVMCS. Mark all fields as clean
2588 	 * here.
2589 	 */
2590 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
2591 		vmx->nested.hv_evmcs->hv_clean_fields |=
2592 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2593 
2594 	return 0;
2595 }
2596 
2597 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2598 {
2599 	if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2600 	       nested_cpu_has_virtual_nmis(vmcs12)))
2601 		return -EINVAL;
2602 
2603 	if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2604 	       nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2605 		return -EINVAL;
2606 
2607 	return 0;
2608 }
2609 
2610 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2611 {
2612 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2613 
2614 	/* Check for memory type validity */
2615 	switch (new_eptp & VMX_EPTP_MT_MASK) {
2616 	case VMX_EPTP_MT_UC:
2617 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2618 			return false;
2619 		break;
2620 	case VMX_EPTP_MT_WB:
2621 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2622 			return false;
2623 		break;
2624 	default:
2625 		return false;
2626 	}
2627 
2628 	/* Page-walk levels validity. */
2629 	switch (new_eptp & VMX_EPTP_PWL_MASK) {
2630 	case VMX_EPTP_PWL_5:
2631 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2632 			return false;
2633 		break;
2634 	case VMX_EPTP_PWL_4:
2635 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2636 			return false;
2637 		break;
2638 	default:
2639 		return false;
2640 	}
2641 
2642 	/* Reserved bits should not be set */
2643 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2644 		return false;
2645 
2646 	/* AD, if set, should be supported */
2647 	if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2648 		if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2649 			return false;
2650 	}
2651 
2652 	return true;
2653 }
2654 
2655 /*
2656  * Checks related to VM-Execution Control Fields
2657  */
2658 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2659                                               struct vmcs12 *vmcs12)
2660 {
2661 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2662 
2663 	if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2664 				   vmx->nested.msrs.pinbased_ctls_low,
2665 				   vmx->nested.msrs.pinbased_ctls_high)) ||
2666 	    CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2667 				   vmx->nested.msrs.procbased_ctls_low,
2668 				   vmx->nested.msrs.procbased_ctls_high)))
2669 		return -EINVAL;
2670 
2671 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2672 	    CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2673 				   vmx->nested.msrs.secondary_ctls_low,
2674 				   vmx->nested.msrs.secondary_ctls_high)))
2675 		return -EINVAL;
2676 
2677 	if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2678 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2679 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2680 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2681 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2682 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2683 	    nested_vmx_check_nmi_controls(vmcs12) ||
2684 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2685 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2686 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2687 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2688 	    CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2689 		return -EINVAL;
2690 
2691 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2692 	    nested_cpu_has_save_preemption_timer(vmcs12))
2693 		return -EINVAL;
2694 
2695 	if (nested_cpu_has_ept(vmcs12) &&
2696 	    CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2697 		return -EINVAL;
2698 
2699 	if (nested_cpu_has_vmfunc(vmcs12)) {
2700 		if (CC(vmcs12->vm_function_control &
2701 		       ~vmx->nested.msrs.vmfunc_controls))
2702 			return -EINVAL;
2703 
2704 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2705 			if (CC(!nested_cpu_has_ept(vmcs12)) ||
2706 			    CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2707 				return -EINVAL;
2708 		}
2709 	}
2710 
2711 	return 0;
2712 }
2713 
2714 /*
2715  * Checks related to VM-Exit Control Fields
2716  */
2717 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2718                                          struct vmcs12 *vmcs12)
2719 {
2720 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2721 
2722 	if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2723 				    vmx->nested.msrs.exit_ctls_low,
2724 				    vmx->nested.msrs.exit_ctls_high)) ||
2725 	    CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2726 		return -EINVAL;
2727 
2728 	return 0;
2729 }
2730 
2731 /*
2732  * Checks related to VM-Entry Control Fields
2733  */
2734 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2735 					  struct vmcs12 *vmcs12)
2736 {
2737 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2738 
2739 	if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2740 				    vmx->nested.msrs.entry_ctls_low,
2741 				    vmx->nested.msrs.entry_ctls_high)))
2742 		return -EINVAL;
2743 
2744 	/*
2745 	 * From the Intel SDM, volume 3:
2746 	 * Fields relevant to VM-entry event injection must be set properly.
2747 	 * These fields are the VM-entry interruption-information field, the
2748 	 * VM-entry exception error code, and the VM-entry instruction length.
2749 	 */
2750 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2751 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2752 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2753 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2754 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2755 		bool should_have_error_code;
2756 		bool urg = nested_cpu_has2(vmcs12,
2757 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2758 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2759 
2760 		/* VM-entry interruption-info field: interruption type */
2761 		if (CC(intr_type == INTR_TYPE_RESERVED) ||
2762 		    CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2763 		       !nested_cpu_supports_monitor_trap_flag(vcpu)))
2764 			return -EINVAL;
2765 
2766 		/* VM-entry interruption-info field: vector */
2767 		if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2768 		    CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2769 		    CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2770 			return -EINVAL;
2771 
2772 		/* VM-entry interruption-info field: deliver error code */
2773 		should_have_error_code =
2774 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2775 			x86_exception_has_error_code(vector);
2776 		if (CC(has_error_code != should_have_error_code))
2777 			return -EINVAL;
2778 
2779 		/* VM-entry exception error code */
2780 		if (CC(has_error_code &&
2781 		       vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2782 			return -EINVAL;
2783 
2784 		/* VM-entry interruption-info field: reserved bits */
2785 		if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2786 			return -EINVAL;
2787 
2788 		/* VM-entry instruction length */
2789 		switch (intr_type) {
2790 		case INTR_TYPE_SOFT_EXCEPTION:
2791 		case INTR_TYPE_SOFT_INTR:
2792 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2793 			if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2794 			    CC(vmcs12->vm_entry_instruction_len == 0 &&
2795 			    CC(!nested_cpu_has_zero_length_injection(vcpu))))
2796 				return -EINVAL;
2797 		}
2798 	}
2799 
2800 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2801 		return -EINVAL;
2802 
2803 	return 0;
2804 }
2805 
2806 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2807 				     struct vmcs12 *vmcs12)
2808 {
2809 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2810 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2811 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2812 		return -EINVAL;
2813 
2814 	if (to_vmx(vcpu)->nested.enlightened_vmcs_enabled)
2815 		return nested_evmcs_check_controls(vmcs12);
2816 
2817 	return 0;
2818 }
2819 
2820 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2821 				       struct vmcs12 *vmcs12)
2822 {
2823 	bool ia32e;
2824 
2825 	if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2826 	    CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2827 	    CC(kvm_vcpu_is_illegal_gpa(vcpu, vmcs12->host_cr3)))
2828 		return -EINVAL;
2829 
2830 	if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2831 	    CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2832 		return -EINVAL;
2833 
2834 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2835 	    CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2836 		return -EINVAL;
2837 
2838 	if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2839 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2840 					   vmcs12->host_ia32_perf_global_ctrl)))
2841 		return -EINVAL;
2842 
2843 #ifdef CONFIG_X86_64
2844 	ia32e = !!(vcpu->arch.efer & EFER_LMA);
2845 #else
2846 	ia32e = false;
2847 #endif
2848 
2849 	if (ia32e) {
2850 		if (CC(!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)) ||
2851 		    CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2852 			return -EINVAL;
2853 	} else {
2854 		if (CC(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) ||
2855 		    CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
2856 		    CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
2857 		    CC((vmcs12->host_rip) >> 32))
2858 			return -EINVAL;
2859 	}
2860 
2861 	if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2862 	    CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2863 	    CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2864 	    CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2865 	    CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2866 	    CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2867 	    CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
2868 	    CC(vmcs12->host_cs_selector == 0) ||
2869 	    CC(vmcs12->host_tr_selector == 0) ||
2870 	    CC(vmcs12->host_ss_selector == 0 && !ia32e))
2871 		return -EINVAL;
2872 
2873 	if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
2874 	    CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
2875 	    CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
2876 	    CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
2877 	    CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
2878 	    CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
2879 		return -EINVAL;
2880 
2881 	/*
2882 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2883 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2884 	 * the values of the LMA and LME bits in the field must each be that of
2885 	 * the host address-space size VM-exit control.
2886 	 */
2887 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2888 		if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
2889 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
2890 		    CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
2891 			return -EINVAL;
2892 	}
2893 
2894 	return 0;
2895 }
2896 
2897 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2898 					  struct vmcs12 *vmcs12)
2899 {
2900 	int r = 0;
2901 	struct vmcs12 *shadow;
2902 	struct kvm_host_map map;
2903 
2904 	if (vmcs12->vmcs_link_pointer == -1ull)
2905 		return 0;
2906 
2907 	if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
2908 		return -EINVAL;
2909 
2910 	if (CC(kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->vmcs_link_pointer), &map)))
2911 		return -EINVAL;
2912 
2913 	shadow = map.hva;
2914 
2915 	if (CC(shadow->hdr.revision_id != VMCS12_REVISION) ||
2916 	    CC(shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
2917 		r = -EINVAL;
2918 
2919 	kvm_vcpu_unmap(vcpu, &map, false);
2920 	return r;
2921 }
2922 
2923 /*
2924  * Checks related to Guest Non-register State
2925  */
2926 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2927 {
2928 	if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2929 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
2930 	       vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
2931 		return -EINVAL;
2932 
2933 	return 0;
2934 }
2935 
2936 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
2937 					struct vmcs12 *vmcs12,
2938 					enum vm_entry_failure_code *entry_failure_code)
2939 {
2940 	bool ia32e;
2941 
2942 	*entry_failure_code = ENTRY_FAIL_DEFAULT;
2943 
2944 	if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
2945 	    CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
2946 		return -EINVAL;
2947 
2948 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
2949 	    CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
2950 		return -EINVAL;
2951 
2952 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
2953 	    CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
2954 		return -EINVAL;
2955 
2956 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2957 		*entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
2958 		return -EINVAL;
2959 	}
2960 
2961 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2962 	    CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2963 					   vmcs12->guest_ia32_perf_global_ctrl)))
2964 		return -EINVAL;
2965 
2966 	/*
2967 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2968 	 * are performed on the field for the IA32_EFER MSR:
2969 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2970 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2971 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2972 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2973 	 *   CR0.PG) is 1.
2974 	 */
2975 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2976 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2977 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2978 		if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
2979 		    CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
2980 		    CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
2981 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
2982 			return -EINVAL;
2983 	}
2984 
2985 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2986 	    (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
2987 	     CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
2988 		return -EINVAL;
2989 
2990 	if (nested_check_guest_non_reg_state(vmcs12))
2991 		return -EINVAL;
2992 
2993 	return 0;
2994 }
2995 
2996 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2997 {
2998 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2999 	unsigned long cr3, cr4;
3000 	bool vm_fail;
3001 
3002 	if (!nested_early_check)
3003 		return 0;
3004 
3005 	if (vmx->msr_autoload.host.nr)
3006 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3007 	if (vmx->msr_autoload.guest.nr)
3008 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3009 
3010 	preempt_disable();
3011 
3012 	vmx_prepare_switch_to_guest(vcpu);
3013 
3014 	/*
3015 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3016 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
3017 	 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3018 	 * there is no need to preserve other bits or save/restore the field.
3019 	 */
3020 	vmcs_writel(GUEST_RFLAGS, 0);
3021 
3022 	cr3 = __get_current_cr3_fast();
3023 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3024 		vmcs_writel(HOST_CR3, cr3);
3025 		vmx->loaded_vmcs->host_state.cr3 = cr3;
3026 	}
3027 
3028 	cr4 = cr4_read_shadow();
3029 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3030 		vmcs_writel(HOST_CR4, cr4);
3031 		vmx->loaded_vmcs->host_state.cr4 = cr4;
3032 	}
3033 
3034 	vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3035 				 vmx->loaded_vmcs->launched);
3036 
3037 	if (vmx->msr_autoload.host.nr)
3038 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3039 	if (vmx->msr_autoload.guest.nr)
3040 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3041 
3042 	if (vm_fail) {
3043 		u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3044 
3045 		preempt_enable();
3046 
3047 		trace_kvm_nested_vmenter_failed(
3048 			"early hardware check VM-instruction error: ", error);
3049 		WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3050 		return 1;
3051 	}
3052 
3053 	/*
3054 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3055 	 */
3056 	if (hw_breakpoint_active())
3057 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
3058 	local_irq_enable();
3059 	preempt_enable();
3060 
3061 	/*
3062 	 * A non-failing VMEntry means we somehow entered guest mode with
3063 	 * an illegal RIP, and that's just the tip of the iceberg.  There
3064 	 * is no telling what memory has been modified or what state has
3065 	 * been exposed to unknown code.  Hitting this all but guarantees
3066 	 * a (very critical) hardware issue.
3067 	 */
3068 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3069 		VMX_EXIT_REASONS_FAILED_VMENTRY));
3070 
3071 	return 0;
3072 }
3073 
3074 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3075 {
3076 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3077 
3078 	/*
3079 	 * hv_evmcs may end up being not mapped after migration (when
3080 	 * L2 was running), map it here to make sure vmcs12 changes are
3081 	 * properly reflected.
3082 	 */
3083 	if (vmx->nested.enlightened_vmcs_enabled &&
3084 	    vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3085 		enum nested_evmptrld_status evmptrld_status =
3086 			nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3087 
3088 		if (evmptrld_status == EVMPTRLD_VMFAIL ||
3089 		    evmptrld_status == EVMPTRLD_ERROR)
3090 			return false;
3091 
3092 		/*
3093 		 * Post migration VMCS12 always provides the most actual
3094 		 * information, copy it to eVMCS upon entry.
3095 		 */
3096 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3097 	}
3098 
3099 	return true;
3100 }
3101 
3102 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3103 {
3104 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3105 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3106 	struct kvm_host_map *map;
3107 	struct page *page;
3108 	u64 hpa;
3109 
3110 	if (!vcpu->arch.pdptrs_from_userspace &&
3111 	    !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3112 		/*
3113 		 * Reload the guest's PDPTRs since after a migration
3114 		 * the guest CR3 might be restored prior to setting the nested
3115 		 * state which can lead to a load of wrong PDPTRs.
3116 		 */
3117 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3)))
3118 			return false;
3119 	}
3120 
3121 
3122 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3123 		/*
3124 		 * Translate L1 physical address to host physical
3125 		 * address for vmcs02. Keep the page pinned, so this
3126 		 * physical address remains valid. We keep a reference
3127 		 * to it so we can release it later.
3128 		 */
3129 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
3130 			kvm_release_page_clean(vmx->nested.apic_access_page);
3131 			vmx->nested.apic_access_page = NULL;
3132 		}
3133 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
3134 		if (!is_error_page(page)) {
3135 			vmx->nested.apic_access_page = page;
3136 			hpa = page_to_phys(vmx->nested.apic_access_page);
3137 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
3138 		} else {
3139 			pr_debug_ratelimited("%s: no backing 'struct page' for APIC-access address in vmcs12\n",
3140 					     __func__);
3141 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3142 			vcpu->run->internal.suberror =
3143 				KVM_INTERNAL_ERROR_EMULATION;
3144 			vcpu->run->internal.ndata = 0;
3145 			return false;
3146 		}
3147 	}
3148 
3149 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3150 		map = &vmx->nested.virtual_apic_map;
3151 
3152 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3153 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3154 		} else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3155 		           nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3156 			   !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3157 			/*
3158 			 * The processor will never use the TPR shadow, simply
3159 			 * clear the bit from the execution control.  Such a
3160 			 * configuration is useless, but it happens in tests.
3161 			 * For any other configuration, failing the vm entry is
3162 			 * _not_ what the processor does but it's basically the
3163 			 * only possibility we have.
3164 			 */
3165 			exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3166 		} else {
3167 			/*
3168 			 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3169 			 * force VM-Entry to fail.
3170 			 */
3171 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
3172 		}
3173 	}
3174 
3175 	if (nested_cpu_has_posted_intr(vmcs12)) {
3176 		map = &vmx->nested.pi_desc_map;
3177 
3178 		if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3179 			vmx->nested.pi_desc =
3180 				(struct pi_desc *)(((void *)map->hva) +
3181 				offset_in_page(vmcs12->posted_intr_desc_addr));
3182 			vmcs_write64(POSTED_INTR_DESC_ADDR,
3183 				     pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3184 		} else {
3185 			/*
3186 			 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3187 			 * access the contents of the VMCS12 posted interrupt
3188 			 * descriptor. (Note that KVM may do this when it
3189 			 * should not, per the architectural specification.)
3190 			 */
3191 			vmx->nested.pi_desc = NULL;
3192 			pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3193 		}
3194 	}
3195 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3196 		exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3197 	else
3198 		exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3199 
3200 	return true;
3201 }
3202 
3203 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3204 {
3205 	if (!nested_get_evmcs_page(vcpu)) {
3206 		pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3207 				     __func__);
3208 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3209 		vcpu->run->internal.suberror =
3210 			KVM_INTERNAL_ERROR_EMULATION;
3211 		vcpu->run->internal.ndata = 0;
3212 
3213 		return false;
3214 	}
3215 
3216 	if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3217 		return false;
3218 
3219 	return true;
3220 }
3221 
3222 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3223 {
3224 	struct vmcs12 *vmcs12;
3225 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3226 	gpa_t dst;
3227 
3228 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3229 		return 0;
3230 
3231 	if (WARN_ON_ONCE(vmx->nested.pml_full))
3232 		return 1;
3233 
3234 	/*
3235 	 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3236 	 * set is already checked as part of A/D emulation.
3237 	 */
3238 	vmcs12 = get_vmcs12(vcpu);
3239 	if (!nested_cpu_has_pml(vmcs12))
3240 		return 0;
3241 
3242 	if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3243 		vmx->nested.pml_full = true;
3244 		return 1;
3245 	}
3246 
3247 	gpa &= ~0xFFFull;
3248 	dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3249 
3250 	if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3251 				 offset_in_page(dst), sizeof(gpa)))
3252 		return 0;
3253 
3254 	vmcs12->guest_pml_index--;
3255 
3256 	return 0;
3257 }
3258 
3259 /*
3260  * Intel's VMX Instruction Reference specifies a common set of prerequisites
3261  * for running VMX instructions (except VMXON, whose prerequisites are
3262  * slightly different). It also specifies what exception to inject otherwise.
3263  * Note that many of these exceptions have priority over VM exits, so they
3264  * don't have to be checked again here.
3265  */
3266 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3267 {
3268 	if (!to_vmx(vcpu)->nested.vmxon) {
3269 		kvm_queue_exception(vcpu, UD_VECTOR);
3270 		return 0;
3271 	}
3272 
3273 	if (vmx_get_cpl(vcpu)) {
3274 		kvm_inject_gp(vcpu, 0);
3275 		return 0;
3276 	}
3277 
3278 	return 1;
3279 }
3280 
3281 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3282 {
3283 	u8 rvi = vmx_get_rvi();
3284 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3285 
3286 	return ((rvi & 0xf0) > (vppr & 0xf0));
3287 }
3288 
3289 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3290 				   struct vmcs12 *vmcs12);
3291 
3292 /*
3293  * If from_vmentry is false, this is being called from state restore (either RSM
3294  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
3295  *
3296  * Returns:
3297  *	NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3298  *	NVMX_VMENTRY_VMFAIL:  Consistency check VMFail
3299  *	NVMX_VMENTRY_VMEXIT:  Consistency check VMExit
3300  *	NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3301  */
3302 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3303 							bool from_vmentry)
3304 {
3305 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3306 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3307 	enum vm_entry_failure_code entry_failure_code;
3308 	bool evaluate_pending_interrupts;
3309 	union vmx_exit_reason exit_reason = {
3310 		.basic = EXIT_REASON_INVALID_STATE,
3311 		.failed_vmentry = 1,
3312 	};
3313 	u32 failed_index;
3314 
3315 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3316 		kvm_vcpu_flush_tlb_current(vcpu);
3317 
3318 	evaluate_pending_interrupts = exec_controls_get(vmx) &
3319 		(CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3320 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3321 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3322 
3323 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3324 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3325 	if (kvm_mpx_supported() &&
3326 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
3327 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3328 
3329 	/*
3330 	 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3331 	 * nested early checks are disabled.  In the event of a "late" VM-Fail,
3332 	 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3333 	 * software model to the pre-VMEntry host state.  When EPT is disabled,
3334 	 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3335 	 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3.  Stuffing
3336 	 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3337 	 * the correct value.  Smashing vmcs01.GUEST_CR3 is safe because nested
3338 	 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3339 	 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3340 	 * L1.  Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3341 	 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3342 	 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3343 	 * path would need to manually save/restore vmcs01.GUEST_CR3.
3344 	 */
3345 	if (!enable_ept && !nested_early_check)
3346 		vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3347 
3348 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3349 
3350 	prepare_vmcs02_early(vmx, vmcs12);
3351 
3352 	if (from_vmentry) {
3353 		if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3354 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3355 			return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3356 		}
3357 
3358 		if (nested_vmx_check_vmentry_hw(vcpu)) {
3359 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3360 			return NVMX_VMENTRY_VMFAIL;
3361 		}
3362 
3363 		if (nested_vmx_check_guest_state(vcpu, vmcs12,
3364 						 &entry_failure_code)) {
3365 			exit_reason.basic = EXIT_REASON_INVALID_STATE;
3366 			vmcs12->exit_qualification = entry_failure_code;
3367 			goto vmentry_fail_vmexit;
3368 		}
3369 	}
3370 
3371 	enter_guest_mode(vcpu);
3372 
3373 	if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3374 		exit_reason.basic = EXIT_REASON_INVALID_STATE;
3375 		vmcs12->exit_qualification = entry_failure_code;
3376 		goto vmentry_fail_vmexit_guest_mode;
3377 	}
3378 
3379 	if (from_vmentry) {
3380 		failed_index = nested_vmx_load_msr(vcpu,
3381 						   vmcs12->vm_entry_msr_load_addr,
3382 						   vmcs12->vm_entry_msr_load_count);
3383 		if (failed_index) {
3384 			exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3385 			vmcs12->exit_qualification = failed_index;
3386 			goto vmentry_fail_vmexit_guest_mode;
3387 		}
3388 	} else {
3389 		/*
3390 		 * The MMU is not initialized to point at the right entities yet and
3391 		 * "get pages" would need to read data from the guest (i.e. we will
3392 		 * need to perform gpa to hpa translation). Request a call
3393 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3394 		 * have already been set at vmentry time and should not be reset.
3395 		 */
3396 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3397 	}
3398 
3399 	/*
3400 	 * If L1 had a pending IRQ/NMI until it executed
3401 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3402 	 * disallowed (e.g. interrupts disabled), L0 needs to
3403 	 * evaluate if this pending event should cause an exit from L2
3404 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3405 	 * intercept EXTERNAL_INTERRUPT).
3406 	 *
3407 	 * Usually this would be handled by the processor noticing an
3408 	 * IRQ/NMI window request, or checking RVI during evaluation of
3409 	 * pending virtual interrupts.  However, this setting was done
3410 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3411 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3412 	 */
3413 	if (unlikely(evaluate_pending_interrupts))
3414 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3415 
3416 	/*
3417 	 * Do not start the preemption timer hrtimer until after we know
3418 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3419 	 * the timer.
3420 	 */
3421 	vmx->nested.preemption_timer_expired = false;
3422 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3423 		u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3424 		vmx_start_preemption_timer(vcpu, timer_value);
3425 	}
3426 
3427 	/*
3428 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3429 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3430 	 * returned as far as L1 is concerned. It will only return (and set
3431 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3432 	 */
3433 	return NVMX_VMENTRY_SUCCESS;
3434 
3435 	/*
3436 	 * A failed consistency check that leads to a VMExit during L1's
3437 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3438 	 * 26.7 "VM-entry failures during or after loading guest state".
3439 	 */
3440 vmentry_fail_vmexit_guest_mode:
3441 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3442 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3443 	leave_guest_mode(vcpu);
3444 
3445 vmentry_fail_vmexit:
3446 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3447 
3448 	if (!from_vmentry)
3449 		return NVMX_VMENTRY_VMEXIT;
3450 
3451 	load_vmcs12_host_state(vcpu, vmcs12);
3452 	vmcs12->vm_exit_reason = exit_reason.full;
3453 	if (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
3454 		vmx->nested.need_vmcs12_to_shadow_sync = true;
3455 	return NVMX_VMENTRY_VMEXIT;
3456 }
3457 
3458 /*
3459  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3460  * for running an L2 nested guest.
3461  */
3462 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3463 {
3464 	struct vmcs12 *vmcs12;
3465 	enum nvmx_vmentry_status status;
3466 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3467 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3468 	enum nested_evmptrld_status evmptrld_status;
3469 
3470 	if (!nested_vmx_check_permission(vcpu))
3471 		return 1;
3472 
3473 	evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3474 	if (evmptrld_status == EVMPTRLD_ERROR) {
3475 		kvm_queue_exception(vcpu, UD_VECTOR);
3476 		return 1;
3477 	} else if (CC(evmptrld_status == EVMPTRLD_VMFAIL)) {
3478 		return nested_vmx_failInvalid(vcpu);
3479 	}
3480 
3481 	if (CC(!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr) &&
3482 	       vmx->nested.current_vmptr == -1ull))
3483 		return nested_vmx_failInvalid(vcpu);
3484 
3485 	vmcs12 = get_vmcs12(vcpu);
3486 
3487 	/*
3488 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3489 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3490 	 * rather than RFLAGS.ZF, and no error number is stored to the
3491 	 * VM-instruction error field.
3492 	 */
3493 	if (CC(vmcs12->hdr.shadow_vmcs))
3494 		return nested_vmx_failInvalid(vcpu);
3495 
3496 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)) {
3497 		copy_enlightened_to_vmcs12(vmx, vmx->nested.hv_evmcs->hv_clean_fields);
3498 		/* Enlightened VMCS doesn't have launch state */
3499 		vmcs12->launch_state = !launch;
3500 	} else if (enable_shadow_vmcs) {
3501 		copy_shadow_to_vmcs12(vmx);
3502 	}
3503 
3504 	/*
3505 	 * The nested entry process starts with enforcing various prerequisites
3506 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3507 	 * they fail: As the SDM explains, some conditions should cause the
3508 	 * instruction to fail, while others will cause the instruction to seem
3509 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3510 	 * To speed up the normal (success) code path, we should avoid checking
3511 	 * for misconfigurations which will anyway be caught by the processor
3512 	 * when using the merged vmcs02.
3513 	 */
3514 	if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3515 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3516 
3517 	if (CC(vmcs12->launch_state == launch))
3518 		return nested_vmx_fail(vcpu,
3519 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3520 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3521 
3522 	if (nested_vmx_check_controls(vcpu, vmcs12))
3523 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3524 
3525 	if (nested_vmx_check_host_state(vcpu, vmcs12))
3526 		return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3527 
3528 	/*
3529 	 * We're finally done with prerequisite checking, and can start with
3530 	 * the nested entry.
3531 	 */
3532 	vmx->nested.nested_run_pending = 1;
3533 	vmx->nested.has_preemption_timer_deadline = false;
3534 	status = nested_vmx_enter_non_root_mode(vcpu, true);
3535 	if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3536 		goto vmentry_failed;
3537 
3538 	/* Emulate processing of posted interrupts on VM-Enter. */
3539 	if (nested_cpu_has_posted_intr(vmcs12) &&
3540 	    kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3541 		vmx->nested.pi_pending = true;
3542 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3543 		kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3544 	}
3545 
3546 	/* Hide L1D cache contents from the nested guest.  */
3547 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3548 
3549 	/*
3550 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3551 	 * also be used as part of restoring nVMX state for
3552 	 * snapshot restore (migration).
3553 	 *
3554 	 * In this flow, it is assumed that vmcs12 cache was
3555 	 * transferred as part of captured nVMX state and should
3556 	 * therefore not be read from guest memory (which may not
3557 	 * exist on destination host yet).
3558 	 */
3559 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3560 
3561 	switch (vmcs12->guest_activity_state) {
3562 	case GUEST_ACTIVITY_HLT:
3563 		/*
3564 		 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3565 		 * awakened by event injection or by an NMI-window VM-exit or
3566 		 * by an interrupt-window VM-exit, halt the vcpu.
3567 		 */
3568 		if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3569 		    !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3570 		    !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3571 		      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3572 			vmx->nested.nested_run_pending = 0;
3573 			return kvm_vcpu_halt(vcpu);
3574 		}
3575 		break;
3576 	case GUEST_ACTIVITY_WAIT_SIPI:
3577 		vmx->nested.nested_run_pending = 0;
3578 		vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3579 		break;
3580 	default:
3581 		break;
3582 	}
3583 
3584 	return 1;
3585 
3586 vmentry_failed:
3587 	vmx->nested.nested_run_pending = 0;
3588 	if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3589 		return 0;
3590 	if (status == NVMX_VMENTRY_VMEXIT)
3591 		return 1;
3592 	WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3593 	return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3594 }
3595 
3596 /*
3597  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3598  * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3599  * This function returns the new value we should put in vmcs12.guest_cr0.
3600  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3601  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3602  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3603  *     didn't trap the bit, because if L1 did, so would L0).
3604  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3605  *     been modified by L2, and L1 knows it. So just leave the old value of
3606  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3607  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3608  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3609  *     changed these bits, and therefore they need to be updated, but L0
3610  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3611  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3612  */
3613 static inline unsigned long
3614 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3615 {
3616 	return
3617 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3618 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3619 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3620 			vcpu->arch.cr0_guest_owned_bits));
3621 }
3622 
3623 static inline unsigned long
3624 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3625 {
3626 	return
3627 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3628 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3629 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3630 			vcpu->arch.cr4_guest_owned_bits));
3631 }
3632 
3633 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3634 				      struct vmcs12 *vmcs12)
3635 {
3636 	u32 idt_vectoring;
3637 	unsigned int nr;
3638 
3639 	if (vcpu->arch.exception.injected) {
3640 		nr = vcpu->arch.exception.nr;
3641 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3642 
3643 		if (kvm_exception_is_soft(nr)) {
3644 			vmcs12->vm_exit_instruction_len =
3645 				vcpu->arch.event_exit_inst_len;
3646 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3647 		} else
3648 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3649 
3650 		if (vcpu->arch.exception.has_error_code) {
3651 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3652 			vmcs12->idt_vectoring_error_code =
3653 				vcpu->arch.exception.error_code;
3654 		}
3655 
3656 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3657 	} else if (vcpu->arch.nmi_injected) {
3658 		vmcs12->idt_vectoring_info_field =
3659 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3660 	} else if (vcpu->arch.interrupt.injected) {
3661 		nr = vcpu->arch.interrupt.nr;
3662 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3663 
3664 		if (vcpu->arch.interrupt.soft) {
3665 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3666 			vmcs12->vm_entry_instruction_len =
3667 				vcpu->arch.event_exit_inst_len;
3668 		} else
3669 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3670 
3671 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3672 	}
3673 }
3674 
3675 
3676 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3677 {
3678 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3679 	gfn_t gfn;
3680 
3681 	/*
3682 	 * Don't need to mark the APIC access page dirty; it is never
3683 	 * written to by the CPU during APIC virtualization.
3684 	 */
3685 
3686 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3687 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3688 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3689 	}
3690 
3691 	if (nested_cpu_has_posted_intr(vmcs12)) {
3692 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3693 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3694 	}
3695 }
3696 
3697 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3698 {
3699 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3700 	int max_irr;
3701 	void *vapic_page;
3702 	u16 status;
3703 
3704 	if (!vmx->nested.pi_pending)
3705 		return 0;
3706 
3707 	if (!vmx->nested.pi_desc)
3708 		goto mmio_needed;
3709 
3710 	vmx->nested.pi_pending = false;
3711 
3712 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3713 		return 0;
3714 
3715 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3716 	if (max_irr != 256) {
3717 		vapic_page = vmx->nested.virtual_apic_map.hva;
3718 		if (!vapic_page)
3719 			goto mmio_needed;
3720 
3721 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3722 			vapic_page, &max_irr);
3723 		status = vmcs_read16(GUEST_INTR_STATUS);
3724 		if ((u8)max_irr > ((u8)status & 0xff)) {
3725 			status &= ~0xff;
3726 			status |= (u8)max_irr;
3727 			vmcs_write16(GUEST_INTR_STATUS, status);
3728 		}
3729 	}
3730 
3731 	nested_mark_vmcs12_pages_dirty(vcpu);
3732 	return 0;
3733 
3734 mmio_needed:
3735 	kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3736 	return -ENXIO;
3737 }
3738 
3739 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3740 					       unsigned long exit_qual)
3741 {
3742 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3743 	unsigned int nr = vcpu->arch.exception.nr;
3744 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3745 
3746 	if (vcpu->arch.exception.has_error_code) {
3747 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3748 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3749 	}
3750 
3751 	if (kvm_exception_is_soft(nr))
3752 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3753 	else
3754 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3755 
3756 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3757 	    vmx_get_nmi_mask(vcpu))
3758 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3759 
3760 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3761 }
3762 
3763 /*
3764  * Returns true if a debug trap is pending delivery.
3765  *
3766  * In KVM, debug traps bear an exception payload. As such, the class of a #DB
3767  * exception may be inferred from the presence of an exception payload.
3768  */
3769 static inline bool vmx_pending_dbg_trap(struct kvm_vcpu *vcpu)
3770 {
3771 	return vcpu->arch.exception.pending &&
3772 			vcpu->arch.exception.nr == DB_VECTOR &&
3773 			vcpu->arch.exception.payload;
3774 }
3775 
3776 /*
3777  * Certain VM-exits set the 'pending debug exceptions' field to indicate a
3778  * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
3779  * represents these debug traps with a payload that is said to be compatible
3780  * with the 'pending debug exceptions' field, write the payload to the VMCS
3781  * field if a VM-exit is delivered before the debug trap.
3782  */
3783 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
3784 {
3785 	if (vmx_pending_dbg_trap(vcpu))
3786 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
3787 			    vcpu->arch.exception.payload);
3788 }
3789 
3790 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
3791 {
3792 	return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3793 	       to_vmx(vcpu)->nested.preemption_timer_expired;
3794 }
3795 
3796 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
3797 {
3798 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3799 	unsigned long exit_qual;
3800 	bool block_nested_events =
3801 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3802 	bool mtf_pending = vmx->nested.mtf_pending;
3803 	struct kvm_lapic *apic = vcpu->arch.apic;
3804 
3805 	/*
3806 	 * Clear the MTF state. If a higher priority VM-exit is delivered first,
3807 	 * this state is discarded.
3808 	 */
3809 	if (!block_nested_events)
3810 		vmx->nested.mtf_pending = false;
3811 
3812 	if (lapic_in_kernel(vcpu) &&
3813 		test_bit(KVM_APIC_INIT, &apic->pending_events)) {
3814 		if (block_nested_events)
3815 			return -EBUSY;
3816 		nested_vmx_update_pending_dbg(vcpu);
3817 		clear_bit(KVM_APIC_INIT, &apic->pending_events);
3818 		if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
3819 			nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
3820 		return 0;
3821 	}
3822 
3823 	if (lapic_in_kernel(vcpu) &&
3824 	    test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
3825 		if (block_nested_events)
3826 			return -EBUSY;
3827 
3828 		clear_bit(KVM_APIC_SIPI, &apic->pending_events);
3829 		if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3830 			nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
3831 						apic->sipi_vector & 0xFFUL);
3832 		return 0;
3833 	}
3834 
3835 	/*
3836 	 * Process any exceptions that are not debug traps before MTF.
3837 	 *
3838 	 * Note that only a pending nested run can block a pending exception.
3839 	 * Otherwise an injected NMI/interrupt should either be
3840 	 * lost or delivered to the nested hypervisor in the IDT_VECTORING_INFO,
3841 	 * while delivering the pending exception.
3842 	 */
3843 
3844 	if (vcpu->arch.exception.pending && !vmx_pending_dbg_trap(vcpu)) {
3845 		if (vmx->nested.nested_run_pending)
3846 			return -EBUSY;
3847 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3848 			goto no_vmexit;
3849 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3850 		return 0;
3851 	}
3852 
3853 	if (mtf_pending) {
3854 		if (block_nested_events)
3855 			return -EBUSY;
3856 		nested_vmx_update_pending_dbg(vcpu);
3857 		nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
3858 		return 0;
3859 	}
3860 
3861 	if (vcpu->arch.exception.pending) {
3862 		if (vmx->nested.nested_run_pending)
3863 			return -EBUSY;
3864 		if (!nested_vmx_check_exception(vcpu, &exit_qual))
3865 			goto no_vmexit;
3866 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3867 		return 0;
3868 	}
3869 
3870 	if (nested_vmx_preemption_timer_pending(vcpu)) {
3871 		if (block_nested_events)
3872 			return -EBUSY;
3873 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3874 		return 0;
3875 	}
3876 
3877 	if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
3878 		if (block_nested_events)
3879 			return -EBUSY;
3880 		goto no_vmexit;
3881 	}
3882 
3883 	if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
3884 		if (block_nested_events)
3885 			return -EBUSY;
3886 		if (!nested_exit_on_nmi(vcpu))
3887 			goto no_vmexit;
3888 
3889 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3890 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3891 				  INTR_INFO_VALID_MASK, 0);
3892 		/*
3893 		 * The NMI-triggered VM exit counts as injection:
3894 		 * clear this one and block further NMIs.
3895 		 */
3896 		vcpu->arch.nmi_pending = 0;
3897 		vmx_set_nmi_mask(vcpu, true);
3898 		return 0;
3899 	}
3900 
3901 	if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
3902 		if (block_nested_events)
3903 			return -EBUSY;
3904 		if (!nested_exit_on_intr(vcpu))
3905 			goto no_vmexit;
3906 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3907 		return 0;
3908 	}
3909 
3910 no_vmexit:
3911 	return vmx_complete_nested_posted_interrupt(vcpu);
3912 }
3913 
3914 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3915 {
3916 	ktime_t remaining =
3917 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3918 	u64 value;
3919 
3920 	if (ktime_to_ns(remaining) <= 0)
3921 		return 0;
3922 
3923 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3924 	do_div(value, 1000000);
3925 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3926 }
3927 
3928 static bool is_vmcs12_ext_field(unsigned long field)
3929 {
3930 	switch (field) {
3931 	case GUEST_ES_SELECTOR:
3932 	case GUEST_CS_SELECTOR:
3933 	case GUEST_SS_SELECTOR:
3934 	case GUEST_DS_SELECTOR:
3935 	case GUEST_FS_SELECTOR:
3936 	case GUEST_GS_SELECTOR:
3937 	case GUEST_LDTR_SELECTOR:
3938 	case GUEST_TR_SELECTOR:
3939 	case GUEST_ES_LIMIT:
3940 	case GUEST_CS_LIMIT:
3941 	case GUEST_SS_LIMIT:
3942 	case GUEST_DS_LIMIT:
3943 	case GUEST_FS_LIMIT:
3944 	case GUEST_GS_LIMIT:
3945 	case GUEST_LDTR_LIMIT:
3946 	case GUEST_TR_LIMIT:
3947 	case GUEST_GDTR_LIMIT:
3948 	case GUEST_IDTR_LIMIT:
3949 	case GUEST_ES_AR_BYTES:
3950 	case GUEST_DS_AR_BYTES:
3951 	case GUEST_FS_AR_BYTES:
3952 	case GUEST_GS_AR_BYTES:
3953 	case GUEST_LDTR_AR_BYTES:
3954 	case GUEST_TR_AR_BYTES:
3955 	case GUEST_ES_BASE:
3956 	case GUEST_CS_BASE:
3957 	case GUEST_SS_BASE:
3958 	case GUEST_DS_BASE:
3959 	case GUEST_FS_BASE:
3960 	case GUEST_GS_BASE:
3961 	case GUEST_LDTR_BASE:
3962 	case GUEST_TR_BASE:
3963 	case GUEST_GDTR_BASE:
3964 	case GUEST_IDTR_BASE:
3965 	case GUEST_PENDING_DBG_EXCEPTIONS:
3966 	case GUEST_BNDCFGS:
3967 		return true;
3968 	default:
3969 		break;
3970 	}
3971 
3972 	return false;
3973 }
3974 
3975 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
3976 				       struct vmcs12 *vmcs12)
3977 {
3978 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3979 
3980 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3981 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3982 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3983 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3984 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3985 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3986 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3987 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3988 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3989 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3990 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3991 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3992 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3993 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3994 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3995 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3996 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3997 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3998 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3999 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4000 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4001 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4002 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4003 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4004 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4005 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4006 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4007 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4008 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4009 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4010 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4011 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4012 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4013 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4014 	vmcs12->guest_pending_dbg_exceptions =
4015 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4016 	if (kvm_mpx_supported())
4017 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
4018 
4019 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4020 }
4021 
4022 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4023 				       struct vmcs12 *vmcs12)
4024 {
4025 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4026 	int cpu;
4027 
4028 	if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4029 		return;
4030 
4031 
4032 	WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4033 
4034 	cpu = get_cpu();
4035 	vmx->loaded_vmcs = &vmx->nested.vmcs02;
4036 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4037 
4038 	sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4039 
4040 	vmx->loaded_vmcs = &vmx->vmcs01;
4041 	vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4042 	put_cpu();
4043 }
4044 
4045 /*
4046  * Update the guest state fields of vmcs12 to reflect changes that
4047  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4048  * VM-entry controls is also updated, since this is really a guest
4049  * state bit.)
4050  */
4051 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4052 {
4053 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4054 
4055 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
4056 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4057 
4058 	vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4059 		!evmptr_is_valid(vmx->nested.hv_evmcs_vmptr);
4060 
4061 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4062 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4063 
4064 	vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4065 	vmcs12->guest_rip = kvm_rip_read(vcpu);
4066 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4067 
4068 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4069 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4070 
4071 	vmcs12->guest_interruptibility_info =
4072 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4073 
4074 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4075 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4076 	else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4077 		vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4078 	else
4079 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4080 
4081 	if (nested_cpu_has_preemption_timer(vmcs12) &&
4082 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4083 	    !vmx->nested.nested_run_pending)
4084 		vmcs12->vmx_preemption_timer_value =
4085 			vmx_get_preemption_timer_value(vcpu);
4086 
4087 	/*
4088 	 * In some cases (usually, nested EPT), L2 is allowed to change its
4089 	 * own CR3 without exiting. If it has changed it, we must keep it.
4090 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4091 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4092 	 *
4093 	 * Additionally, restore L2's PDPTR to vmcs12.
4094 	 */
4095 	if (enable_ept) {
4096 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4097 		if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4098 			vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4099 			vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4100 			vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4101 			vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4102 		}
4103 	}
4104 
4105 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4106 
4107 	if (nested_cpu_has_vid(vmcs12))
4108 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4109 
4110 	vmcs12->vm_entry_controls =
4111 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4112 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4113 
4114 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4115 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
4116 
4117 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4118 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
4119 }
4120 
4121 /*
4122  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4123  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4124  * and this function updates it to reflect the changes to the guest state while
4125  * L2 was running (and perhaps made some exits which were handled directly by L0
4126  * without going back to L1), and to reflect the exit reason.
4127  * Note that we do not have to copy here all VMCS fields, just those that
4128  * could have changed by the L2 guest or the exit - i.e., the guest-state and
4129  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4130  * which already writes to vmcs12 directly.
4131  */
4132 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4133 			   u32 vm_exit_reason, u32 exit_intr_info,
4134 			   unsigned long exit_qualification)
4135 {
4136 	/* update exit information fields: */
4137 	vmcs12->vm_exit_reason = vm_exit_reason;
4138 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
4139 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4140 	vmcs12->exit_qualification = exit_qualification;
4141 	vmcs12->vm_exit_intr_info = exit_intr_info;
4142 
4143 	vmcs12->idt_vectoring_info_field = 0;
4144 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4145 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4146 
4147 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4148 		vmcs12->launch_state = 1;
4149 
4150 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
4151 		 * instead of reading the real value. */
4152 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4153 
4154 		/*
4155 		 * Transfer the event that L0 or L1 may wanted to inject into
4156 		 * L2 to IDT_VECTORING_INFO_FIELD.
4157 		 */
4158 		vmcs12_save_pending_event(vcpu, vmcs12);
4159 
4160 		/*
4161 		 * According to spec, there's no need to store the guest's
4162 		 * MSRs if the exit is due to a VM-entry failure that occurs
4163 		 * during or after loading the guest state. Since this exit
4164 		 * does not fall in that category, we need to save the MSRs.
4165 		 */
4166 		if (nested_vmx_store_msr(vcpu,
4167 					 vmcs12->vm_exit_msr_store_addr,
4168 					 vmcs12->vm_exit_msr_store_count))
4169 			nested_vmx_abort(vcpu,
4170 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4171 	}
4172 
4173 	/*
4174 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
4175 	 * preserved above and would only end up incorrectly in L1.
4176 	 */
4177 	vcpu->arch.nmi_injected = false;
4178 	kvm_clear_exception_queue(vcpu);
4179 	kvm_clear_interrupt_queue(vcpu);
4180 }
4181 
4182 /*
4183  * A part of what we need to when the nested L2 guest exits and we want to
4184  * run its L1 parent, is to reset L1's guest state to the host state specified
4185  * in vmcs12.
4186  * This function is to be called not only on normal nested exit, but also on
4187  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4188  * Failures During or After Loading Guest State").
4189  * This function should be called when the active VMCS is L1's (vmcs01).
4190  */
4191 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4192 				   struct vmcs12 *vmcs12)
4193 {
4194 	enum vm_entry_failure_code ignored;
4195 	struct kvm_segment seg;
4196 
4197 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4198 		vcpu->arch.efer = vmcs12->host_ia32_efer;
4199 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4200 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4201 	else
4202 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4203 	vmx_set_efer(vcpu, vcpu->arch.efer);
4204 
4205 	kvm_rsp_write(vcpu, vmcs12->host_rsp);
4206 	kvm_rip_write(vcpu, vmcs12->host_rip);
4207 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4208 	vmx_set_interrupt_shadow(vcpu, 0);
4209 
4210 	/*
4211 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4212 	 * actually changed, because vmx_set_cr0 refers to efer set above.
4213 	 *
4214 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4215 	 * (KVM doesn't change it);
4216 	 */
4217 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4218 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
4219 
4220 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
4221 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4222 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
4223 
4224 	nested_ept_uninit_mmu_context(vcpu);
4225 
4226 	/*
4227 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4228 	 * couldn't have changed.
4229 	 */
4230 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4231 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4232 
4233 	nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4234 
4235 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4236 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4237 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4238 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4239 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4240 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4241 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4242 
4243 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
4244 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4245 		vmcs_write64(GUEST_BNDCFGS, 0);
4246 
4247 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4248 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4249 		vcpu->arch.pat = vmcs12->host_ia32_pat;
4250 	}
4251 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
4252 		WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4253 					 vmcs12->host_ia32_perf_global_ctrl));
4254 
4255 	/* Set L1 segment info according to Intel SDM
4256 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
4257 	seg = (struct kvm_segment) {
4258 		.base = 0,
4259 		.limit = 0xFFFFFFFF,
4260 		.selector = vmcs12->host_cs_selector,
4261 		.type = 11,
4262 		.present = 1,
4263 		.s = 1,
4264 		.g = 1
4265 	};
4266 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4267 		seg.l = 1;
4268 	else
4269 		seg.db = 1;
4270 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4271 	seg = (struct kvm_segment) {
4272 		.base = 0,
4273 		.limit = 0xFFFFFFFF,
4274 		.type = 3,
4275 		.present = 1,
4276 		.s = 1,
4277 		.db = 1,
4278 		.g = 1
4279 	};
4280 	seg.selector = vmcs12->host_ds_selector;
4281 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4282 	seg.selector = vmcs12->host_es_selector;
4283 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4284 	seg.selector = vmcs12->host_ss_selector;
4285 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4286 	seg.selector = vmcs12->host_fs_selector;
4287 	seg.base = vmcs12->host_fs_base;
4288 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4289 	seg.selector = vmcs12->host_gs_selector;
4290 	seg.base = vmcs12->host_gs_base;
4291 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4292 	seg = (struct kvm_segment) {
4293 		.base = vmcs12->host_tr_base,
4294 		.limit = 0x67,
4295 		.selector = vmcs12->host_tr_selector,
4296 		.type = 11,
4297 		.present = 1
4298 	};
4299 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4300 
4301 	kvm_set_dr(vcpu, 7, 0x400);
4302 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4303 
4304 	if (cpu_has_vmx_msr_bitmap())
4305 		vmx_update_msr_bitmap(vcpu);
4306 
4307 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4308 				vmcs12->vm_exit_msr_load_count))
4309 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4310 }
4311 
4312 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4313 {
4314 	struct vmx_uret_msr *efer_msr;
4315 	unsigned int i;
4316 
4317 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4318 		return vmcs_read64(GUEST_IA32_EFER);
4319 
4320 	if (cpu_has_load_ia32_efer())
4321 		return host_efer;
4322 
4323 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4324 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4325 			return vmx->msr_autoload.guest.val[i].value;
4326 	}
4327 
4328 	efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4329 	if (efer_msr)
4330 		return efer_msr->data;
4331 
4332 	return host_efer;
4333 }
4334 
4335 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4336 {
4337 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4338 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4339 	struct vmx_msr_entry g, h;
4340 	gpa_t gpa;
4341 	u32 i, j;
4342 
4343 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4344 
4345 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4346 		/*
4347 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4348 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
4349 		 * and vcpu->arch.dr7 is not squirreled away before the
4350 		 * nested VMENTER (not worth adding a variable in nested_vmx).
4351 		 */
4352 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4353 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4354 		else
4355 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4356 	}
4357 
4358 	/*
4359 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4360 	 * handle a variety of side effects to KVM's software model.
4361 	 */
4362 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4363 
4364 	vcpu->arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4365 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4366 
4367 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4368 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4369 
4370 	nested_ept_uninit_mmu_context(vcpu);
4371 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4372 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4373 
4374 	/*
4375 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4376 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
4377 	 * VMFail, like everything else we just need to ensure our
4378 	 * software model is up-to-date.
4379 	 */
4380 	if (enable_ept && is_pae_paging(vcpu))
4381 		ept_save_pdptrs(vcpu);
4382 
4383 	kvm_mmu_reset_context(vcpu);
4384 
4385 	if (cpu_has_vmx_msr_bitmap())
4386 		vmx_update_msr_bitmap(vcpu);
4387 
4388 	/*
4389 	 * This nasty bit of open coding is a compromise between blindly
4390 	 * loading L1's MSRs using the exit load lists (incorrect emulation
4391 	 * of VMFail), leaving the nested VM's MSRs in the software model
4392 	 * (incorrect behavior) and snapshotting the modified MSRs (too
4393 	 * expensive since the lists are unbound by hardware).  For each
4394 	 * MSR that was (prematurely) loaded from the nested VMEntry load
4395 	 * list, reload it from the exit load list if it exists and differs
4396 	 * from the guest value.  The intent is to stuff host state as
4397 	 * silently as possible, not to fully process the exit load list.
4398 	 */
4399 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4400 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4401 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4402 			pr_debug_ratelimited(
4403 				"%s read MSR index failed (%u, 0x%08llx)\n",
4404 				__func__, i, gpa);
4405 			goto vmabort;
4406 		}
4407 
4408 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4409 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4410 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4411 				pr_debug_ratelimited(
4412 					"%s read MSR failed (%u, 0x%08llx)\n",
4413 					__func__, j, gpa);
4414 				goto vmabort;
4415 			}
4416 			if (h.index != g.index)
4417 				continue;
4418 			if (h.value == g.value)
4419 				break;
4420 
4421 			if (nested_vmx_load_msr_check(vcpu, &h)) {
4422 				pr_debug_ratelimited(
4423 					"%s check failed (%u, 0x%x, 0x%x)\n",
4424 					__func__, j, h.index, h.reserved);
4425 				goto vmabort;
4426 			}
4427 
4428 			if (kvm_set_msr(vcpu, h.index, h.value)) {
4429 				pr_debug_ratelimited(
4430 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4431 					__func__, j, h.index, h.value);
4432 				goto vmabort;
4433 			}
4434 		}
4435 	}
4436 
4437 	return;
4438 
4439 vmabort:
4440 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4441 }
4442 
4443 /*
4444  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4445  * and modify vmcs12 to make it see what it would expect to see there if
4446  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4447  */
4448 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4449 		       u32 exit_intr_info, unsigned long exit_qualification)
4450 {
4451 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4452 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4453 
4454 	/* trying to cancel vmlaunch/vmresume is a bug */
4455 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
4456 
4457 	/* Similarly, triple faults in L2 should never escape. */
4458 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
4459 
4460 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4461 		/*
4462 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4463 		 * Enlightened VMCS after migration and we still need to
4464 		 * do that when something is forcing L2->L1 exit prior to
4465 		 * the first L2 run.
4466 		 */
4467 		(void)nested_get_evmcs_page(vcpu);
4468 	}
4469 
4470 	/* Service the TLB flush request for L2 before switching to L1. */
4471 	if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
4472 		kvm_vcpu_flush_tlb_current(vcpu);
4473 
4474 	/*
4475 	 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4476 	 * now and the new vmentry.  Ensure that the VMCS02 PDPTR fields are
4477 	 * up-to-date before switching to L1.
4478 	 */
4479 	if (enable_ept && is_pae_paging(vcpu))
4480 		vmx_ept_load_pdptrs(vcpu);
4481 
4482 	leave_guest_mode(vcpu);
4483 
4484 	if (nested_cpu_has_preemption_timer(vmcs12))
4485 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4486 
4487 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4488 		vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4489 		if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4490 			vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4491 	}
4492 
4493 	if (likely(!vmx->fail)) {
4494 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4495 
4496 		if (vm_exit_reason != -1)
4497 			prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4498 				       exit_intr_info, exit_qualification);
4499 
4500 		/*
4501 		 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4502 		 * also be used to capture vmcs12 cache as part of
4503 		 * capturing nVMX state for snapshot (migration).
4504 		 *
4505 		 * Otherwise, this flush will dirty guest memory at a
4506 		 * point it is already assumed by user-space to be
4507 		 * immutable.
4508 		 */
4509 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4510 	} else {
4511 		/*
4512 		 * The only expected VM-instruction error is "VM entry with
4513 		 * invalid control field(s)." Anything else indicates a
4514 		 * problem with L0.  And we should never get here with a
4515 		 * VMFail of any type if early consistency checks are enabled.
4516 		 */
4517 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4518 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4519 		WARN_ON_ONCE(nested_early_check);
4520 	}
4521 
4522 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4523 
4524 	/* Update any VMCS fields that might have changed while L2 ran */
4525 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4526 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4527 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4528 	if (kvm_has_tsc_control)
4529 		vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4530 
4531 	if (vmx->nested.l1_tpr_threshold != -1)
4532 		vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4533 
4534 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4535 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
4536 		vmx_set_virtual_apic_mode(vcpu);
4537 	}
4538 
4539 	if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4540 		vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4541 		vmx_update_cpu_dirty_logging(vcpu);
4542 	}
4543 
4544 	/* Unpin physical memory we referred to in vmcs02 */
4545 	if (vmx->nested.apic_access_page) {
4546 		kvm_release_page_clean(vmx->nested.apic_access_page);
4547 		vmx->nested.apic_access_page = NULL;
4548 	}
4549 	kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4550 	kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4551 	vmx->nested.pi_desc = NULL;
4552 
4553 	if (vmx->nested.reload_vmcs01_apic_access_page) {
4554 		vmx->nested.reload_vmcs01_apic_access_page = false;
4555 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4556 	}
4557 
4558 	if ((vm_exit_reason != -1) &&
4559 	    (enable_shadow_vmcs || evmptr_is_valid(vmx->nested.hv_evmcs_vmptr)))
4560 		vmx->nested.need_vmcs12_to_shadow_sync = true;
4561 
4562 	/* in case we halted in L2 */
4563 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4564 
4565 	if (likely(!vmx->fail)) {
4566 		if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4567 		    nested_exit_intr_ack_set(vcpu)) {
4568 			int irq = kvm_cpu_get_interrupt(vcpu);
4569 			WARN_ON(irq < 0);
4570 			vmcs12->vm_exit_intr_info = irq |
4571 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4572 		}
4573 
4574 		if (vm_exit_reason != -1)
4575 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4576 						       vmcs12->exit_qualification,
4577 						       vmcs12->idt_vectoring_info_field,
4578 						       vmcs12->vm_exit_intr_info,
4579 						       vmcs12->vm_exit_intr_error_code,
4580 						       KVM_ISA_VMX);
4581 
4582 		load_vmcs12_host_state(vcpu, vmcs12);
4583 
4584 		return;
4585 	}
4586 
4587 	/*
4588 	 * After an early L2 VM-entry failure, we're now back
4589 	 * in L1 which thinks it just finished a VMLAUNCH or
4590 	 * VMRESUME instruction, so we need to set the failure
4591 	 * flag and the VM-instruction error field of the VMCS
4592 	 * accordingly, and skip the emulated instruction.
4593 	 */
4594 	(void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4595 
4596 	/*
4597 	 * Restore L1's host state to KVM's software model.  We're here
4598 	 * because a consistency check was caught by hardware, which
4599 	 * means some amount of guest state has been propagated to KVM's
4600 	 * model and needs to be unwound to the host's state.
4601 	 */
4602 	nested_vmx_restore_host_state(vcpu);
4603 
4604 	vmx->fail = 0;
4605 }
4606 
4607 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
4608 {
4609 	nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
4610 }
4611 
4612 /*
4613  * Decode the memory-address operand of a vmx instruction, as recorded on an
4614  * exit caused by such an instruction (run by a guest hypervisor).
4615  * On success, returns 0. When the operand is invalid, returns 1 and throws
4616  * #UD, #GP, or #SS.
4617  */
4618 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4619 			u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
4620 {
4621 	gva_t off;
4622 	bool exn;
4623 	struct kvm_segment s;
4624 
4625 	/*
4626 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4627 	 * Execution", on an exit, vmx_instruction_info holds most of the
4628 	 * addressing components of the operand. Only the displacement part
4629 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4630 	 * For how an actual address is calculated from all these components,
4631 	 * refer to Vol. 1, "Operand Addressing".
4632 	 */
4633 	int  scaling = vmx_instruction_info & 3;
4634 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4635 	bool is_reg = vmx_instruction_info & (1u << 10);
4636 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4637 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4638 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4639 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4640 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4641 
4642 	if (is_reg) {
4643 		kvm_queue_exception(vcpu, UD_VECTOR);
4644 		return 1;
4645 	}
4646 
4647 	/* Addr = segment_base + offset */
4648 	/* offset = base + [index * scale] + displacement */
4649 	off = exit_qualification; /* holds the displacement */
4650 	if (addr_size == 1)
4651 		off = (gva_t)sign_extend64(off, 31);
4652 	else if (addr_size == 0)
4653 		off = (gva_t)sign_extend64(off, 15);
4654 	if (base_is_valid)
4655 		off += kvm_register_read(vcpu, base_reg);
4656 	if (index_is_valid)
4657 		off += kvm_register_read(vcpu, index_reg) << scaling;
4658 	vmx_get_segment(vcpu, &s, seg_reg);
4659 
4660 	/*
4661 	 * The effective address, i.e. @off, of a memory operand is truncated
4662 	 * based on the address size of the instruction.  Note that this is
4663 	 * the *effective address*, i.e. the address prior to accounting for
4664 	 * the segment's base.
4665 	 */
4666 	if (addr_size == 1) /* 32 bit */
4667 		off &= 0xffffffff;
4668 	else if (addr_size == 0) /* 16 bit */
4669 		off &= 0xffff;
4670 
4671 	/* Checks for #GP/#SS exceptions. */
4672 	exn = false;
4673 	if (is_long_mode(vcpu)) {
4674 		/*
4675 		 * The virtual/linear address is never truncated in 64-bit
4676 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4677 		 * address when using FS/GS with a non-zero base.
4678 		 */
4679 		if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
4680 			*ret = s.base + off;
4681 		else
4682 			*ret = off;
4683 
4684 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4685 		 * non-canonical form. This is the only check on the memory
4686 		 * destination for long mode!
4687 		 */
4688 		exn = is_noncanonical_address(*ret, vcpu);
4689 	} else {
4690 		/*
4691 		 * When not in long mode, the virtual/linear address is
4692 		 * unconditionally truncated to 32 bits regardless of the
4693 		 * address size.
4694 		 */
4695 		*ret = (s.base + off) & 0xffffffff;
4696 
4697 		/* Protected mode: apply checks for segment validity in the
4698 		 * following order:
4699 		 * - segment type check (#GP(0) may be thrown)
4700 		 * - usability check (#GP(0)/#SS(0))
4701 		 * - limit check (#GP(0)/#SS(0))
4702 		 */
4703 		if (wr)
4704 			/* #GP(0) if the destination operand is located in a
4705 			 * read-only data segment or any code segment.
4706 			 */
4707 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4708 		else
4709 			/* #GP(0) if the source operand is located in an
4710 			 * execute-only code segment
4711 			 */
4712 			exn = ((s.type & 0xa) == 8);
4713 		if (exn) {
4714 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4715 			return 1;
4716 		}
4717 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4718 		 */
4719 		exn = (s.unusable != 0);
4720 
4721 		/*
4722 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4723 		 * outside the segment limit.  All CPUs that support VMX ignore
4724 		 * limit checks for flat segments, i.e. segments with base==0,
4725 		 * limit==0xffffffff and of type expand-up data or code.
4726 		 */
4727 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4728 		     ((s.type & 8) || !(s.type & 4))))
4729 			exn = exn || ((u64)off + len - 1 > s.limit);
4730 	}
4731 	if (exn) {
4732 		kvm_queue_exception_e(vcpu,
4733 				      seg_reg == VCPU_SREG_SS ?
4734 						SS_VECTOR : GP_VECTOR,
4735 				      0);
4736 		return 1;
4737 	}
4738 
4739 	return 0;
4740 }
4741 
4742 void nested_vmx_pmu_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
4743 {
4744 	struct vcpu_vmx *vmx;
4745 
4746 	if (!nested_vmx_allowed(vcpu))
4747 		return;
4748 
4749 	vmx = to_vmx(vcpu);
4750 	if (kvm_x86_ops.pmu_ops->is_valid_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL)) {
4751 		vmx->nested.msrs.entry_ctls_high |=
4752 				VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4753 		vmx->nested.msrs.exit_ctls_high |=
4754 				VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4755 	} else {
4756 		vmx->nested.msrs.entry_ctls_high &=
4757 				~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4758 		vmx->nested.msrs.exit_ctls_high &=
4759 				~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4760 	}
4761 }
4762 
4763 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
4764 				int *ret)
4765 {
4766 	gva_t gva;
4767 	struct x86_exception e;
4768 	int r;
4769 
4770 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
4771 				vmcs_read32(VMX_INSTRUCTION_INFO), false,
4772 				sizeof(*vmpointer), &gva)) {
4773 		*ret = 1;
4774 		return -EINVAL;
4775 	}
4776 
4777 	r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
4778 	if (r != X86EMUL_CONTINUE) {
4779 		*ret = kvm_handle_memory_failure(vcpu, r, &e);
4780 		return -EINVAL;
4781 	}
4782 
4783 	return 0;
4784 }
4785 
4786 /*
4787  * Allocate a shadow VMCS and associate it with the currently loaded
4788  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4789  * VMCS is also VMCLEARed, so that it is ready for use.
4790  */
4791 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4792 {
4793 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4794 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4795 
4796 	/*
4797 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4798 	 * executes VMXON and free it when L1 executes VMXOFF.
4799 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4800 	 * here when vmcs01 already have an allocated shadow vmcs.
4801 	 */
4802 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4803 
4804 	if (!loaded_vmcs->shadow_vmcs) {
4805 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4806 		if (loaded_vmcs->shadow_vmcs)
4807 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4808 	}
4809 	return loaded_vmcs->shadow_vmcs;
4810 }
4811 
4812 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4813 {
4814 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4815 	int r;
4816 
4817 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4818 	if (r < 0)
4819 		goto out_vmcs02;
4820 
4821 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4822 	if (!vmx->nested.cached_vmcs12)
4823 		goto out_cached_vmcs12;
4824 
4825 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4826 	if (!vmx->nested.cached_shadow_vmcs12)
4827 		goto out_cached_shadow_vmcs12;
4828 
4829 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4830 		goto out_shadow_vmcs;
4831 
4832 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4833 		     HRTIMER_MODE_ABS_PINNED);
4834 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4835 
4836 	vmx->nested.vpid02 = allocate_vpid();
4837 
4838 	vmx->nested.vmcs02_initialized = false;
4839 	vmx->nested.vmxon = true;
4840 
4841 	if (vmx_pt_mode_is_host_guest()) {
4842 		vmx->pt_desc.guest.ctl = 0;
4843 		pt_update_intercept_for_msr(vcpu);
4844 	}
4845 
4846 	return 0;
4847 
4848 out_shadow_vmcs:
4849 	kfree(vmx->nested.cached_shadow_vmcs12);
4850 
4851 out_cached_shadow_vmcs12:
4852 	kfree(vmx->nested.cached_vmcs12);
4853 
4854 out_cached_vmcs12:
4855 	free_loaded_vmcs(&vmx->nested.vmcs02);
4856 
4857 out_vmcs02:
4858 	return -ENOMEM;
4859 }
4860 
4861 /*
4862  * Emulate the VMXON instruction.
4863  * Currently, we just remember that VMX is active, and do not save or even
4864  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4865  * do not currently need to store anything in that guest-allocated memory
4866  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4867  * argument is different from the VMXON pointer (which the spec says they do).
4868  */
4869 static int handle_vmon(struct kvm_vcpu *vcpu)
4870 {
4871 	int ret;
4872 	gpa_t vmptr;
4873 	uint32_t revision;
4874 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4875 	const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
4876 		| FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
4877 
4878 	/*
4879 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4880 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4881 	 * 1 (see vmx_is_valid_cr4() for when we allow the guest to set this).
4882 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4883 	 * have already been checked by hardware, prior to the VM-exit for
4884 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4885 	 * that bit set to 1 in non-root mode.
4886 	 */
4887 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4888 		kvm_queue_exception(vcpu, UD_VECTOR);
4889 		return 1;
4890 	}
4891 
4892 	/* CPL=0 must be checked manually. */
4893 	if (vmx_get_cpl(vcpu)) {
4894 		kvm_inject_gp(vcpu, 0);
4895 		return 1;
4896 	}
4897 
4898 	if (vmx->nested.vmxon)
4899 		return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4900 
4901 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4902 			!= VMXON_NEEDED_FEATURES) {
4903 		kvm_inject_gp(vcpu, 0);
4904 		return 1;
4905 	}
4906 
4907 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
4908 		return ret;
4909 
4910 	/*
4911 	 * SDM 3: 24.11.5
4912 	 * The first 4 bytes of VMXON region contain the supported
4913 	 * VMCS revision identifier
4914 	 *
4915 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4916 	 * which replaces physical address width with 32
4917 	 */
4918 	if (!page_address_valid(vcpu, vmptr))
4919 		return nested_vmx_failInvalid(vcpu);
4920 
4921 	if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
4922 	    revision != VMCS12_REVISION)
4923 		return nested_vmx_failInvalid(vcpu);
4924 
4925 	vmx->nested.vmxon_ptr = vmptr;
4926 	ret = enter_vmx_operation(vcpu);
4927 	if (ret)
4928 		return ret;
4929 
4930 	return nested_vmx_succeed(vcpu);
4931 }
4932 
4933 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4934 {
4935 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4936 
4937 	if (vmx->nested.current_vmptr == -1ull)
4938 		return;
4939 
4940 	copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
4941 
4942 	if (enable_shadow_vmcs) {
4943 		/* copy to memory all shadowed fields in case
4944 		   they were modified */
4945 		copy_shadow_to_vmcs12(vmx);
4946 		vmx_disable_shadow_vmcs(vmx);
4947 	}
4948 	vmx->nested.posted_intr_nv = -1;
4949 
4950 	/* Flush VMCS12 to guest memory */
4951 	kvm_vcpu_write_guest_page(vcpu,
4952 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4953 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4954 
4955 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4956 
4957 	vmx->nested.current_vmptr = -1ull;
4958 }
4959 
4960 /* Emulate the VMXOFF instruction */
4961 static int handle_vmoff(struct kvm_vcpu *vcpu)
4962 {
4963 	if (!nested_vmx_check_permission(vcpu))
4964 		return 1;
4965 
4966 	free_nested(vcpu);
4967 
4968 	/* Process a latched INIT during time CPU was in VMX operation */
4969 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4970 
4971 	return nested_vmx_succeed(vcpu);
4972 }
4973 
4974 /* Emulate the VMCLEAR instruction */
4975 static int handle_vmclear(struct kvm_vcpu *vcpu)
4976 {
4977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4978 	u32 zero = 0;
4979 	gpa_t vmptr;
4980 	u64 evmcs_gpa;
4981 	int r;
4982 
4983 	if (!nested_vmx_check_permission(vcpu))
4984 		return 1;
4985 
4986 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
4987 		return r;
4988 
4989 	if (!page_address_valid(vcpu, vmptr))
4990 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
4991 
4992 	if (vmptr == vmx->nested.vmxon_ptr)
4993 		return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
4994 
4995 	/*
4996 	 * When Enlightened VMEntry is enabled on the calling CPU we treat
4997 	 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
4998 	 * way to distinguish it from VMCS12) and we must not corrupt it by
4999 	 * writing to the non-existent 'launch_state' field. The area doesn't
5000 	 * have to be the currently active EVMCS on the calling CPU and there's
5001 	 * nothing KVM has to do to transition it from 'active' to 'non-active'
5002 	 * state. It is possible that the area will stay mapped as
5003 	 * vmx->nested.hv_evmcs but this shouldn't be a problem.
5004 	 */
5005 	if (likely(!vmx->nested.enlightened_vmcs_enabled ||
5006 		   !nested_enlightened_vmentry(vcpu, &evmcs_gpa))) {
5007 		if (vmptr == vmx->nested.current_vmptr)
5008 			nested_release_vmcs12(vcpu);
5009 
5010 		kvm_vcpu_write_guest(vcpu,
5011 				     vmptr + offsetof(struct vmcs12,
5012 						      launch_state),
5013 				     &zero, sizeof(zero));
5014 	} else if (vmx->nested.hv_evmcs && vmptr == vmx->nested.hv_evmcs_vmptr) {
5015 		nested_release_evmcs(vcpu);
5016 	}
5017 
5018 	return nested_vmx_succeed(vcpu);
5019 }
5020 
5021 /* Emulate the VMLAUNCH instruction */
5022 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5023 {
5024 	return nested_vmx_run(vcpu, true);
5025 }
5026 
5027 /* Emulate the VMRESUME instruction */
5028 static int handle_vmresume(struct kvm_vcpu *vcpu)
5029 {
5030 
5031 	return nested_vmx_run(vcpu, false);
5032 }
5033 
5034 static int handle_vmread(struct kvm_vcpu *vcpu)
5035 {
5036 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5037 						    : get_vmcs12(vcpu);
5038 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5039 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5040 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5041 	struct x86_exception e;
5042 	unsigned long field;
5043 	u64 value;
5044 	gva_t gva = 0;
5045 	short offset;
5046 	int len, r;
5047 
5048 	if (!nested_vmx_check_permission(vcpu))
5049 		return 1;
5050 
5051 	/*
5052 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
5053 	 * any VMREAD sets the ALU flags for VMfailInvalid.
5054 	 */
5055 	if (vmx->nested.current_vmptr == -1ull ||
5056 	    (is_guest_mode(vcpu) &&
5057 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
5058 		return nested_vmx_failInvalid(vcpu);
5059 
5060 	/* Decode instruction info and find the field to read */
5061 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5062 
5063 	offset = vmcs_field_to_offset(field);
5064 	if (offset < 0)
5065 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5066 
5067 	if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5068 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5069 
5070 	/* Read the field, zero-extended to a u64 value */
5071 	value = vmcs12_read_any(vmcs12, field, offset);
5072 
5073 	/*
5074 	 * Now copy part of this value to register or memory, as requested.
5075 	 * Note that the number of bits actually copied is 32 or 64 depending
5076 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
5077 	 */
5078 	if (instr_info & BIT(10)) {
5079 		kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5080 	} else {
5081 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5082 		if (get_vmx_mem_address(vcpu, exit_qualification,
5083 					instr_info, true, len, &gva))
5084 			return 1;
5085 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
5086 		r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5087 		if (r != X86EMUL_CONTINUE)
5088 			return kvm_handle_memory_failure(vcpu, r, &e);
5089 	}
5090 
5091 	return nested_vmx_succeed(vcpu);
5092 }
5093 
5094 static bool is_shadow_field_rw(unsigned long field)
5095 {
5096 	switch (field) {
5097 #define SHADOW_FIELD_RW(x, y) case x:
5098 #include "vmcs_shadow_fields.h"
5099 		return true;
5100 	default:
5101 		break;
5102 	}
5103 	return false;
5104 }
5105 
5106 static bool is_shadow_field_ro(unsigned long field)
5107 {
5108 	switch (field) {
5109 #define SHADOW_FIELD_RO(x, y) case x:
5110 #include "vmcs_shadow_fields.h"
5111 		return true;
5112 	default:
5113 		break;
5114 	}
5115 	return false;
5116 }
5117 
5118 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5119 {
5120 	struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5121 						    : get_vmcs12(vcpu);
5122 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5123 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5124 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5125 	struct x86_exception e;
5126 	unsigned long field;
5127 	short offset;
5128 	gva_t gva;
5129 	int len, r;
5130 
5131 	/*
5132 	 * The value to write might be 32 or 64 bits, depending on L1's long
5133 	 * mode, and eventually we need to write that into a field of several
5134 	 * possible lengths. The code below first zero-extends the value to 64
5135 	 * bit (value), and then copies only the appropriate number of
5136 	 * bits into the vmcs12 field.
5137 	 */
5138 	u64 value = 0;
5139 
5140 	if (!nested_vmx_check_permission(vcpu))
5141 		return 1;
5142 
5143 	/*
5144 	 * In VMX non-root operation, when the VMCS-link pointer is -1ull,
5145 	 * any VMWRITE sets the ALU flags for VMfailInvalid.
5146 	 */
5147 	if (vmx->nested.current_vmptr == -1ull ||
5148 	    (is_guest_mode(vcpu) &&
5149 	     get_vmcs12(vcpu)->vmcs_link_pointer == -1ull))
5150 		return nested_vmx_failInvalid(vcpu);
5151 
5152 	if (instr_info & BIT(10))
5153 		value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5154 	else {
5155 		len = is_64_bit_mode(vcpu) ? 8 : 4;
5156 		if (get_vmx_mem_address(vcpu, exit_qualification,
5157 					instr_info, false, len, &gva))
5158 			return 1;
5159 		r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5160 		if (r != X86EMUL_CONTINUE)
5161 			return kvm_handle_memory_failure(vcpu, r, &e);
5162 	}
5163 
5164 	field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5165 
5166 	offset = vmcs_field_to_offset(field);
5167 	if (offset < 0)
5168 		return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5169 
5170 	/*
5171 	 * If the vCPU supports "VMWRITE to any supported field in the
5172 	 * VMCS," then the "read-only" fields are actually read/write.
5173 	 */
5174 	if (vmcs_field_readonly(field) &&
5175 	    !nested_cpu_has_vmwrite_any_field(vcpu))
5176 		return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5177 
5178 	/*
5179 	 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5180 	 * vmcs12, else we may crush a field or consume a stale value.
5181 	 */
5182 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5183 		copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5184 
5185 	/*
5186 	 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5187 	 * fields on VMWRITE.  Emulate this behavior to ensure consistent KVM
5188 	 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5189 	 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5190 	 * from L1 will return a different value than VMREAD from L2 (L1 sees
5191 	 * the stripped down value, L2 sees the full value as stored by KVM).
5192 	 */
5193 	if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5194 		value &= 0x1f0ff;
5195 
5196 	vmcs12_write_any(vmcs12, field, offset, value);
5197 
5198 	/*
5199 	 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5200 	 * dirty shadow vmcs12 instead of vmcs12.  Fields that can be updated
5201 	 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5202 	 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5203 	 */
5204 	if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5205 		/*
5206 		 * L1 can read these fields without exiting, ensure the
5207 		 * shadow VMCS is up-to-date.
5208 		 */
5209 		if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5210 			preempt_disable();
5211 			vmcs_load(vmx->vmcs01.shadow_vmcs);
5212 
5213 			__vmcs_writel(field, value);
5214 
5215 			vmcs_clear(vmx->vmcs01.shadow_vmcs);
5216 			vmcs_load(vmx->loaded_vmcs->vmcs);
5217 			preempt_enable();
5218 		}
5219 		vmx->nested.dirty_vmcs12 = true;
5220 	}
5221 
5222 	return nested_vmx_succeed(vcpu);
5223 }
5224 
5225 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5226 {
5227 	vmx->nested.current_vmptr = vmptr;
5228 	if (enable_shadow_vmcs) {
5229 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5230 		vmcs_write64(VMCS_LINK_POINTER,
5231 			     __pa(vmx->vmcs01.shadow_vmcs));
5232 		vmx->nested.need_vmcs12_to_shadow_sync = true;
5233 	}
5234 	vmx->nested.dirty_vmcs12 = true;
5235 }
5236 
5237 /* Emulate the VMPTRLD instruction */
5238 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5239 {
5240 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5241 	gpa_t vmptr;
5242 	int r;
5243 
5244 	if (!nested_vmx_check_permission(vcpu))
5245 		return 1;
5246 
5247 	if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5248 		return r;
5249 
5250 	if (!page_address_valid(vcpu, vmptr))
5251 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5252 
5253 	if (vmptr == vmx->nested.vmxon_ptr)
5254 		return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5255 
5256 	/* Forbid normal VMPTRLD if Enlightened version was used */
5257 	if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
5258 		return 1;
5259 
5260 	if (vmx->nested.current_vmptr != vmptr) {
5261 		struct kvm_host_map map;
5262 		struct vmcs12 *new_vmcs12;
5263 
5264 		if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmptr), &map)) {
5265 			/*
5266 			 * Reads from an unbacked page return all 1s,
5267 			 * which means that the 32 bits located at the
5268 			 * given physical address won't match the required
5269 			 * VMCS12_REVISION identifier.
5270 			 */
5271 			return nested_vmx_fail(vcpu,
5272 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5273 		}
5274 
5275 		new_vmcs12 = map.hva;
5276 
5277 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5278 		    (new_vmcs12->hdr.shadow_vmcs &&
5279 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5280 			kvm_vcpu_unmap(vcpu, &map, false);
5281 			return nested_vmx_fail(vcpu,
5282 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5283 		}
5284 
5285 		nested_release_vmcs12(vcpu);
5286 
5287 		/*
5288 		 * Load VMCS12 from guest memory since it is not already
5289 		 * cached.
5290 		 */
5291 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
5292 		kvm_vcpu_unmap(vcpu, &map, false);
5293 
5294 		set_current_vmptr(vmx, vmptr);
5295 	}
5296 
5297 	return nested_vmx_succeed(vcpu);
5298 }
5299 
5300 /* Emulate the VMPTRST instruction */
5301 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5302 {
5303 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5304 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5305 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5306 	struct x86_exception e;
5307 	gva_t gva;
5308 	int r;
5309 
5310 	if (!nested_vmx_check_permission(vcpu))
5311 		return 1;
5312 
5313 	if (unlikely(evmptr_is_valid(to_vmx(vcpu)->nested.hv_evmcs_vmptr)))
5314 		return 1;
5315 
5316 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5317 				true, sizeof(gpa_t), &gva))
5318 		return 1;
5319 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5320 	r = kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
5321 					sizeof(gpa_t), &e);
5322 	if (r != X86EMUL_CONTINUE)
5323 		return kvm_handle_memory_failure(vcpu, r, &e);
5324 
5325 	return nested_vmx_succeed(vcpu);
5326 }
5327 
5328 #define EPTP_PA_MASK   GENMASK_ULL(51, 12)
5329 
5330 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
5331 {
5332 	return VALID_PAGE(root_hpa) &&
5333 		((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
5334 }
5335 
5336 /* Emulate the INVEPT instruction */
5337 static int handle_invept(struct kvm_vcpu *vcpu)
5338 {
5339 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5340 	u32 vmx_instruction_info, types;
5341 	unsigned long type, roots_to_free;
5342 	struct kvm_mmu *mmu;
5343 	gva_t gva;
5344 	struct x86_exception e;
5345 	struct {
5346 		u64 eptp, gpa;
5347 	} operand;
5348 	int i, r;
5349 
5350 	if (!(vmx->nested.msrs.secondary_ctls_high &
5351 	      SECONDARY_EXEC_ENABLE_EPT) ||
5352 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5353 		kvm_queue_exception(vcpu, UD_VECTOR);
5354 		return 1;
5355 	}
5356 
5357 	if (!nested_vmx_check_permission(vcpu))
5358 		return 1;
5359 
5360 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5361 	type = kvm_register_read(vcpu, (vmx_instruction_info >> 28) & 0xf);
5362 
5363 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5364 
5365 	if (type >= 32 || !(types & (1 << type)))
5366 		return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5367 
5368 	/* According to the Intel VMX instruction reference, the memory
5369 	 * operand is read even if it isn't needed (e.g., for type==global)
5370 	 */
5371 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5372 			vmx_instruction_info, false, sizeof(operand), &gva))
5373 		return 1;
5374 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5375 	if (r != X86EMUL_CONTINUE)
5376 		return kvm_handle_memory_failure(vcpu, r, &e);
5377 
5378 	/*
5379 	 * Nested EPT roots are always held through guest_mmu,
5380 	 * not root_mmu.
5381 	 */
5382 	mmu = &vcpu->arch.guest_mmu;
5383 
5384 	switch (type) {
5385 	case VMX_EPT_EXTENT_CONTEXT:
5386 		if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5387 			return nested_vmx_fail(vcpu,
5388 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5389 
5390 		roots_to_free = 0;
5391 		if (nested_ept_root_matches(mmu->root_hpa, mmu->root_pgd,
5392 					    operand.eptp))
5393 			roots_to_free |= KVM_MMU_ROOT_CURRENT;
5394 
5395 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5396 			if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5397 						    mmu->prev_roots[i].pgd,
5398 						    operand.eptp))
5399 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5400 		}
5401 		break;
5402 	case VMX_EPT_EXTENT_GLOBAL:
5403 		roots_to_free = KVM_MMU_ROOTS_ALL;
5404 		break;
5405 	default:
5406 		BUG();
5407 		break;
5408 	}
5409 
5410 	if (roots_to_free)
5411 		kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
5412 
5413 	return nested_vmx_succeed(vcpu);
5414 }
5415 
5416 static int handle_invvpid(struct kvm_vcpu *vcpu)
5417 {
5418 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5419 	u32 vmx_instruction_info;
5420 	unsigned long type, types;
5421 	gva_t gva;
5422 	struct x86_exception e;
5423 	struct {
5424 		u64 vpid;
5425 		u64 gla;
5426 	} operand;
5427 	u16 vpid02;
5428 	int r;
5429 
5430 	if (!(vmx->nested.msrs.secondary_ctls_high &
5431 	      SECONDARY_EXEC_ENABLE_VPID) ||
5432 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5433 		kvm_queue_exception(vcpu, UD_VECTOR);
5434 		return 1;
5435 	}
5436 
5437 	if (!nested_vmx_check_permission(vcpu))
5438 		return 1;
5439 
5440 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5441 	type = kvm_register_read(vcpu, (vmx_instruction_info >> 28) & 0xf);
5442 
5443 	types = (vmx->nested.msrs.vpid_caps &
5444 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5445 
5446 	if (type >= 32 || !(types & (1 << type)))
5447 		return nested_vmx_fail(vcpu,
5448 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5449 
5450 	/* according to the intel vmx instruction reference, the memory
5451 	 * operand is read even if it isn't needed (e.g., for type==global)
5452 	 */
5453 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5454 			vmx_instruction_info, false, sizeof(operand), &gva))
5455 		return 1;
5456 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5457 	if (r != X86EMUL_CONTINUE)
5458 		return kvm_handle_memory_failure(vcpu, r, &e);
5459 
5460 	if (operand.vpid >> 16)
5461 		return nested_vmx_fail(vcpu,
5462 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5463 
5464 	vpid02 = nested_get_vpid02(vcpu);
5465 	switch (type) {
5466 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5467 		if (!operand.vpid ||
5468 		    is_noncanonical_address(operand.gla, vcpu))
5469 			return nested_vmx_fail(vcpu,
5470 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5471 		vpid_sync_vcpu_addr(vpid02, operand.gla);
5472 		break;
5473 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5474 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5475 		if (!operand.vpid)
5476 			return nested_vmx_fail(vcpu,
5477 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5478 		vpid_sync_context(vpid02);
5479 		break;
5480 	case VMX_VPID_EXTENT_ALL_CONTEXT:
5481 		vpid_sync_context(vpid02);
5482 		break;
5483 	default:
5484 		WARN_ON_ONCE(1);
5485 		return kvm_skip_emulated_instruction(vcpu);
5486 	}
5487 
5488 	/*
5489 	 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5490 	 * linear mappings for L2 (tagged with L2's VPID).  Free all guest
5491 	 * roots as VPIDs are not tracked in the MMU role.
5492 	 *
5493 	 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5494 	 * an MMU when EPT is disabled.
5495 	 *
5496 	 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5497 	 */
5498 	if (!enable_ept)
5499 		kvm_mmu_free_guest_mode_roots(vcpu, &vcpu->arch.root_mmu);
5500 
5501 	return nested_vmx_succeed(vcpu);
5502 }
5503 
5504 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5505 				     struct vmcs12 *vmcs12)
5506 {
5507 	u32 index = kvm_rcx_read(vcpu);
5508 	u64 new_eptp;
5509 
5510 	if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5511 		return 1;
5512 	if (index >= VMFUNC_EPTP_ENTRIES)
5513 		return 1;
5514 
5515 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5516 				     &new_eptp, index * 8, 8))
5517 		return 1;
5518 
5519 	/*
5520 	 * If the (L2) guest does a vmfunc to the currently
5521 	 * active ept pointer, we don't have to do anything else
5522 	 */
5523 	if (vmcs12->ept_pointer != new_eptp) {
5524 		if (!nested_vmx_check_eptp(vcpu, new_eptp))
5525 			return 1;
5526 
5527 		vmcs12->ept_pointer = new_eptp;
5528 		nested_ept_new_eptp(vcpu);
5529 
5530 		if (!nested_cpu_has_vpid(vmcs12))
5531 			kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5532 	}
5533 
5534 	return 0;
5535 }
5536 
5537 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5538 {
5539 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5540 	struct vmcs12 *vmcs12;
5541 	u32 function = kvm_rax_read(vcpu);
5542 
5543 	/*
5544 	 * VMFUNC is only supported for nested guests, but we always enable the
5545 	 * secondary control for simplicity; for non-nested mode, fake that we
5546 	 * didn't by injecting #UD.
5547 	 */
5548 	if (!is_guest_mode(vcpu)) {
5549 		kvm_queue_exception(vcpu, UD_VECTOR);
5550 		return 1;
5551 	}
5552 
5553 	vmcs12 = get_vmcs12(vcpu);
5554 
5555 	/*
5556 	 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5557 	 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5558 	 */
5559 	if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5560 		kvm_queue_exception(vcpu, UD_VECTOR);
5561 		return 1;
5562 	}
5563 
5564 	if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5565 		goto fail;
5566 
5567 	switch (function) {
5568 	case 0:
5569 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
5570 			goto fail;
5571 		break;
5572 	default:
5573 		goto fail;
5574 	}
5575 	return kvm_skip_emulated_instruction(vcpu);
5576 
5577 fail:
5578 	/*
5579 	 * This is effectively a reflected VM-Exit, as opposed to a synthesized
5580 	 * nested VM-Exit.  Pass the original exit reason, i.e. don't hardcode
5581 	 * EXIT_REASON_VMFUNC as the exit reason.
5582 	 */
5583 	nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
5584 			  vmx_get_intr_info(vcpu),
5585 			  vmx_get_exit_qual(vcpu));
5586 	return 1;
5587 }
5588 
5589 /*
5590  * Return true if an IO instruction with the specified port and size should cause
5591  * a VM-exit into L1.
5592  */
5593 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
5594 				 int size)
5595 {
5596 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5597 	gpa_t bitmap, last_bitmap;
5598 	u8 b;
5599 
5600 	last_bitmap = (gpa_t)-1;
5601 	b = -1;
5602 
5603 	while (size > 0) {
5604 		if (port < 0x8000)
5605 			bitmap = vmcs12->io_bitmap_a;
5606 		else if (port < 0x10000)
5607 			bitmap = vmcs12->io_bitmap_b;
5608 		else
5609 			return true;
5610 		bitmap += (port & 0x7fff) / 8;
5611 
5612 		if (last_bitmap != bitmap)
5613 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
5614 				return true;
5615 		if (b & (1 << (port & 7)))
5616 			return true;
5617 
5618 		port++;
5619 		size--;
5620 		last_bitmap = bitmap;
5621 	}
5622 
5623 	return false;
5624 }
5625 
5626 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
5627 				       struct vmcs12 *vmcs12)
5628 {
5629 	unsigned long exit_qualification;
5630 	unsigned short port;
5631 	int size;
5632 
5633 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
5634 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
5635 
5636 	exit_qualification = vmx_get_exit_qual(vcpu);
5637 
5638 	port = exit_qualification >> 16;
5639 	size = (exit_qualification & 7) + 1;
5640 
5641 	return nested_vmx_check_io_bitmaps(vcpu, port, size);
5642 }
5643 
5644 /*
5645  * Return 1 if we should exit from L2 to L1 to handle an MSR access,
5646  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5647  * disinterest in the current event (read or write a specific MSR) by using an
5648  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5649  */
5650 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5651 					struct vmcs12 *vmcs12,
5652 					union vmx_exit_reason exit_reason)
5653 {
5654 	u32 msr_index = kvm_rcx_read(vcpu);
5655 	gpa_t bitmap;
5656 
5657 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
5658 		return true;
5659 
5660 	/*
5661 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5662 	 * for the four combinations of read/write and low/high MSR numbers.
5663 	 * First we need to figure out which of the four to use:
5664 	 */
5665 	bitmap = vmcs12->msr_bitmap;
5666 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
5667 		bitmap += 2048;
5668 	if (msr_index >= 0xc0000000) {
5669 		msr_index -= 0xc0000000;
5670 		bitmap += 1024;
5671 	}
5672 
5673 	/* Then read the msr_index'th bit from this bitmap: */
5674 	if (msr_index < 1024*8) {
5675 		unsigned char b;
5676 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
5677 			return true;
5678 		return 1 & (b >> (msr_index & 7));
5679 	} else
5680 		return true; /* let L1 handle the wrong parameter */
5681 }
5682 
5683 /*
5684  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5685  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5686  * intercept (via guest_host_mask etc.) the current event.
5687  */
5688 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5689 	struct vmcs12 *vmcs12)
5690 {
5691 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5692 	int cr = exit_qualification & 15;
5693 	int reg;
5694 	unsigned long val;
5695 
5696 	switch ((exit_qualification >> 4) & 3) {
5697 	case 0: /* mov to cr */
5698 		reg = (exit_qualification >> 8) & 15;
5699 		val = kvm_register_read(vcpu, reg);
5700 		switch (cr) {
5701 		case 0:
5702 			if (vmcs12->cr0_guest_host_mask &
5703 			    (val ^ vmcs12->cr0_read_shadow))
5704 				return true;
5705 			break;
5706 		case 3:
5707 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5708 				return true;
5709 			break;
5710 		case 4:
5711 			if (vmcs12->cr4_guest_host_mask &
5712 			    (vmcs12->cr4_read_shadow ^ val))
5713 				return true;
5714 			break;
5715 		case 8:
5716 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5717 				return true;
5718 			break;
5719 		}
5720 		break;
5721 	case 2: /* clts */
5722 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5723 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5724 			return true;
5725 		break;
5726 	case 1: /* mov from cr */
5727 		switch (cr) {
5728 		case 3:
5729 			if (vmcs12->cpu_based_vm_exec_control &
5730 			    CPU_BASED_CR3_STORE_EXITING)
5731 				return true;
5732 			break;
5733 		case 8:
5734 			if (vmcs12->cpu_based_vm_exec_control &
5735 			    CPU_BASED_CR8_STORE_EXITING)
5736 				return true;
5737 			break;
5738 		}
5739 		break;
5740 	case 3: /* lmsw */
5741 		/*
5742 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5743 		 * cr0. Other attempted changes are ignored, with no exit.
5744 		 */
5745 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5746 		if (vmcs12->cr0_guest_host_mask & 0xe &
5747 		    (val ^ vmcs12->cr0_read_shadow))
5748 			return true;
5749 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5750 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5751 		    (val & 0x1))
5752 			return true;
5753 		break;
5754 	}
5755 	return false;
5756 }
5757 
5758 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
5759 					  struct vmcs12 *vmcs12)
5760 {
5761 	u32 encls_leaf;
5762 
5763 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
5764 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
5765 		return false;
5766 
5767 	encls_leaf = kvm_rax_read(vcpu);
5768 	if (encls_leaf > 62)
5769 		encls_leaf = 63;
5770 	return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
5771 }
5772 
5773 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5774 	struct vmcs12 *vmcs12, gpa_t bitmap)
5775 {
5776 	u32 vmx_instruction_info;
5777 	unsigned long field;
5778 	u8 b;
5779 
5780 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5781 		return true;
5782 
5783 	/* Decode instruction info and find the field to access */
5784 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5785 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5786 
5787 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5788 	if (field >> 15)
5789 		return true;
5790 
5791 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5792 		return true;
5793 
5794 	return 1 & (b >> (field & 7));
5795 }
5796 
5797 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
5798 {
5799 	u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
5800 
5801 	if (nested_cpu_has_mtf(vmcs12))
5802 		return true;
5803 
5804 	/*
5805 	 * An MTF VM-exit may be injected into the guest by setting the
5806 	 * interruption-type to 7 (other event) and the vector field to 0. Such
5807 	 * is the case regardless of the 'monitor trap flag' VM-execution
5808 	 * control.
5809 	 */
5810 	return entry_intr_info == (INTR_INFO_VALID_MASK
5811 				   | INTR_TYPE_OTHER_EVENT);
5812 }
5813 
5814 /*
5815  * Return true if L0 wants to handle an exit from L2 regardless of whether or not
5816  * L1 wants the exit.  Only call this when in is_guest_mode (L2).
5817  */
5818 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
5819 				     union vmx_exit_reason exit_reason)
5820 {
5821 	u32 intr_info;
5822 
5823 	switch ((u16)exit_reason.basic) {
5824 	case EXIT_REASON_EXCEPTION_NMI:
5825 		intr_info = vmx_get_intr_info(vcpu);
5826 		if (is_nmi(intr_info))
5827 			return true;
5828 		else if (is_page_fault(intr_info))
5829 			return vcpu->arch.apf.host_apf_flags || !enable_ept;
5830 		else if (is_debug(intr_info) &&
5831 			 vcpu->guest_debug &
5832 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5833 			return true;
5834 		else if (is_breakpoint(intr_info) &&
5835 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5836 			return true;
5837 		else if (is_alignment_check(intr_info) &&
5838 			 !vmx_guest_inject_ac(vcpu))
5839 			return true;
5840 		return false;
5841 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5842 		return true;
5843 	case EXIT_REASON_MCE_DURING_VMENTRY:
5844 		return true;
5845 	case EXIT_REASON_EPT_VIOLATION:
5846 		/*
5847 		 * L0 always deals with the EPT violation. If nested EPT is
5848 		 * used, and the nested mmu code discovers that the address is
5849 		 * missing in the guest EPT table (EPT12), the EPT violation
5850 		 * will be injected with nested_ept_inject_page_fault()
5851 		 */
5852 		return true;
5853 	case EXIT_REASON_EPT_MISCONFIG:
5854 		/*
5855 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5856 		 * table (shadow on EPT) or a merged EPT table that L0 built
5857 		 * (EPT on EPT). So any problems with the structure of the
5858 		 * table is L0's fault.
5859 		 */
5860 		return true;
5861 	case EXIT_REASON_PREEMPTION_TIMER:
5862 		return true;
5863 	case EXIT_REASON_PML_FULL:
5864 		/*
5865 		 * PML is emulated for an L1 VMM and should never be enabled in
5866 		 * vmcs02, always "handle" PML_FULL by exiting to userspace.
5867 		 */
5868 		return true;
5869 	case EXIT_REASON_VMFUNC:
5870 		/* VM functions are emulated through L2->L0 vmexits. */
5871 		return true;
5872 	default:
5873 		break;
5874 	}
5875 	return false;
5876 }
5877 
5878 /*
5879  * Return 1 if L1 wants to intercept an exit from L2.  Only call this when in
5880  * is_guest_mode (L2).
5881  */
5882 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
5883 				     union vmx_exit_reason exit_reason)
5884 {
5885 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5886 	u32 intr_info;
5887 
5888 	switch ((u16)exit_reason.basic) {
5889 	case EXIT_REASON_EXCEPTION_NMI:
5890 		intr_info = vmx_get_intr_info(vcpu);
5891 		if (is_nmi(intr_info))
5892 			return true;
5893 		else if (is_page_fault(intr_info))
5894 			return true;
5895 		return vmcs12->exception_bitmap &
5896 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5897 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5898 		return nested_exit_on_intr(vcpu);
5899 	case EXIT_REASON_TRIPLE_FAULT:
5900 		return true;
5901 	case EXIT_REASON_INTERRUPT_WINDOW:
5902 		return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
5903 	case EXIT_REASON_NMI_WINDOW:
5904 		return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
5905 	case EXIT_REASON_TASK_SWITCH:
5906 		return true;
5907 	case EXIT_REASON_CPUID:
5908 		return true;
5909 	case EXIT_REASON_HLT:
5910 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5911 	case EXIT_REASON_INVD:
5912 		return true;
5913 	case EXIT_REASON_INVLPG:
5914 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5915 	case EXIT_REASON_RDPMC:
5916 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5917 	case EXIT_REASON_RDRAND:
5918 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5919 	case EXIT_REASON_RDSEED:
5920 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5921 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5922 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5923 	case EXIT_REASON_VMREAD:
5924 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5925 			vmcs12->vmread_bitmap);
5926 	case EXIT_REASON_VMWRITE:
5927 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5928 			vmcs12->vmwrite_bitmap);
5929 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5930 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5931 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5932 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5933 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5934 		/*
5935 		 * VMX instructions trap unconditionally. This allows L1 to
5936 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5937 		 */
5938 		return true;
5939 	case EXIT_REASON_CR_ACCESS:
5940 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5941 	case EXIT_REASON_DR_ACCESS:
5942 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5943 	case EXIT_REASON_IO_INSTRUCTION:
5944 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5945 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5946 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5947 	case EXIT_REASON_MSR_READ:
5948 	case EXIT_REASON_MSR_WRITE:
5949 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5950 	case EXIT_REASON_INVALID_STATE:
5951 		return true;
5952 	case EXIT_REASON_MWAIT_INSTRUCTION:
5953 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5954 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5955 		return nested_vmx_exit_handled_mtf(vmcs12);
5956 	case EXIT_REASON_MONITOR_INSTRUCTION:
5957 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5958 	case EXIT_REASON_PAUSE_INSTRUCTION:
5959 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5960 			nested_cpu_has2(vmcs12,
5961 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5962 	case EXIT_REASON_MCE_DURING_VMENTRY:
5963 		return true;
5964 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5965 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5966 	case EXIT_REASON_APIC_ACCESS:
5967 	case EXIT_REASON_APIC_WRITE:
5968 	case EXIT_REASON_EOI_INDUCED:
5969 		/*
5970 		 * The controls for "virtualize APIC accesses," "APIC-
5971 		 * register virtualization," and "virtual-interrupt
5972 		 * delivery" only come from vmcs12.
5973 		 */
5974 		return true;
5975 	case EXIT_REASON_INVPCID:
5976 		return
5977 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5978 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5979 	case EXIT_REASON_WBINVD:
5980 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5981 	case EXIT_REASON_XSETBV:
5982 		return true;
5983 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5984 		/*
5985 		 * This should never happen, since it is not possible to
5986 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5987 		 * If if it were, XSS would have to be checked against
5988 		 * the XSS exit bitmap in vmcs12.
5989 		 */
5990 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5991 	case EXIT_REASON_UMWAIT:
5992 	case EXIT_REASON_TPAUSE:
5993 		return nested_cpu_has2(vmcs12,
5994 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
5995 	case EXIT_REASON_ENCLS:
5996 		return nested_vmx_exit_handled_encls(vcpu, vmcs12);
5997 	default:
5998 		return true;
5999 	}
6000 }
6001 
6002 /*
6003  * Conditionally reflect a VM-Exit into L1.  Returns %true if the VM-Exit was
6004  * reflected into L1.
6005  */
6006 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6007 {
6008 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6009 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6010 	unsigned long exit_qual;
6011 	u32 exit_intr_info;
6012 
6013 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
6014 
6015 	/*
6016 	 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6017 	 * has already loaded L2's state.
6018 	 */
6019 	if (unlikely(vmx->fail)) {
6020 		trace_kvm_nested_vmenter_failed(
6021 			"hardware VM-instruction error: ",
6022 			vmcs_read32(VM_INSTRUCTION_ERROR));
6023 		exit_intr_info = 0;
6024 		exit_qual = 0;
6025 		goto reflect_vmexit;
6026 	}
6027 
6028 	trace_kvm_nested_vmexit(exit_reason.full, vcpu, KVM_ISA_VMX);
6029 
6030 	/* If L0 (KVM) wants the exit, it trumps L1's desires. */
6031 	if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6032 		return false;
6033 
6034 	/* If L1 doesn't want the exit, handle it in L0. */
6035 	if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6036 		return false;
6037 
6038 	/*
6039 	 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits.  For
6040 	 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6041 	 * need to be synthesized by querying the in-kernel LAPIC, but external
6042 	 * interrupts are never reflected to L1 so it's a non-issue.
6043 	 */
6044 	exit_intr_info = vmx_get_intr_info(vcpu);
6045 	if (is_exception_with_error_code(exit_intr_info)) {
6046 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6047 
6048 		vmcs12->vm_exit_intr_error_code =
6049 			vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6050 	}
6051 	exit_qual = vmx_get_exit_qual(vcpu);
6052 
6053 reflect_vmexit:
6054 	nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6055 	return true;
6056 }
6057 
6058 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6059 				struct kvm_nested_state __user *user_kvm_nested_state,
6060 				u32 user_data_size)
6061 {
6062 	struct vcpu_vmx *vmx;
6063 	struct vmcs12 *vmcs12;
6064 	struct kvm_nested_state kvm_state = {
6065 		.flags = 0,
6066 		.format = KVM_STATE_NESTED_FORMAT_VMX,
6067 		.size = sizeof(kvm_state),
6068 		.hdr.vmx.flags = 0,
6069 		.hdr.vmx.vmxon_pa = -1ull,
6070 		.hdr.vmx.vmcs12_pa = -1ull,
6071 		.hdr.vmx.preemption_timer_deadline = 0,
6072 	};
6073 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6074 		&user_kvm_nested_state->data.vmx[0];
6075 
6076 	if (!vcpu)
6077 		return kvm_state.size + sizeof(*user_vmx_nested_state);
6078 
6079 	vmx = to_vmx(vcpu);
6080 	vmcs12 = get_vmcs12(vcpu);
6081 
6082 	if (nested_vmx_allowed(vcpu) &&
6083 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6084 		kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6085 		kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6086 
6087 		if (vmx_has_valid_vmcs12(vcpu)) {
6088 			kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6089 
6090 			/* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6091 			if (vmx->nested.hv_evmcs_vmptr != EVMPTR_INVALID)
6092 				kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6093 
6094 			if (is_guest_mode(vcpu) &&
6095 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
6096 			    vmcs12->vmcs_link_pointer != -1ull)
6097 				kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6098 		}
6099 
6100 		if (vmx->nested.smm.vmxon)
6101 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6102 
6103 		if (vmx->nested.smm.guest_mode)
6104 			kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6105 
6106 		if (is_guest_mode(vcpu)) {
6107 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6108 
6109 			if (vmx->nested.nested_run_pending)
6110 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6111 
6112 			if (vmx->nested.mtf_pending)
6113 				kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6114 
6115 			if (nested_cpu_has_preemption_timer(vmcs12) &&
6116 			    vmx->nested.has_preemption_timer_deadline) {
6117 				kvm_state.hdr.vmx.flags |=
6118 					KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6119 				kvm_state.hdr.vmx.preemption_timer_deadline =
6120 					vmx->nested.preemption_timer_deadline;
6121 			}
6122 		}
6123 	}
6124 
6125 	if (user_data_size < kvm_state.size)
6126 		goto out;
6127 
6128 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6129 		return -EFAULT;
6130 
6131 	if (!vmx_has_valid_vmcs12(vcpu))
6132 		goto out;
6133 
6134 	/*
6135 	 * When running L2, the authoritative vmcs12 state is in the
6136 	 * vmcs02. When running L1, the authoritative vmcs12 state is
6137 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
6138 	 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6139 	 * vmcs12 state is in the vmcs12 already.
6140 	 */
6141 	if (is_guest_mode(vcpu)) {
6142 		sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6143 		sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6144 	} else  {
6145 		copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6146 		if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6147 			if (evmptr_is_valid(vmx->nested.hv_evmcs_vmptr))
6148 				/*
6149 				 * L1 hypervisor is not obliged to keep eVMCS
6150 				 * clean fields data always up-to-date while
6151 				 * not in guest mode, 'hv_clean_fields' is only
6152 				 * supposed to be actual upon vmentry so we need
6153 				 * to ignore it here and do full copy.
6154 				 */
6155 				copy_enlightened_to_vmcs12(vmx, 0);
6156 			else if (enable_shadow_vmcs)
6157 				copy_shadow_to_vmcs12(vmx);
6158 		}
6159 	}
6160 
6161 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6162 	BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6163 
6164 	/*
6165 	 * Copy over the full allocated size of vmcs12 rather than just the size
6166 	 * of the struct.
6167 	 */
6168 	if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6169 		return -EFAULT;
6170 
6171 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6172 	    vmcs12->vmcs_link_pointer != -1ull) {
6173 		if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6174 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6175 			return -EFAULT;
6176 	}
6177 out:
6178 	return kvm_state.size;
6179 }
6180 
6181 /*
6182  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
6183  */
6184 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6185 {
6186 	if (is_guest_mode(vcpu)) {
6187 		to_vmx(vcpu)->nested.nested_run_pending = 0;
6188 		nested_vmx_vmexit(vcpu, -1, 0, 0);
6189 	}
6190 	free_nested(vcpu);
6191 }
6192 
6193 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6194 				struct kvm_nested_state __user *user_kvm_nested_state,
6195 				struct kvm_nested_state *kvm_state)
6196 {
6197 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6198 	struct vmcs12 *vmcs12;
6199 	enum vm_entry_failure_code ignored;
6200 	struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6201 		&user_kvm_nested_state->data.vmx[0];
6202 	int ret;
6203 
6204 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6205 		return -EINVAL;
6206 
6207 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull) {
6208 		if (kvm_state->hdr.vmx.smm.flags)
6209 			return -EINVAL;
6210 
6211 		if (kvm_state->hdr.vmx.vmcs12_pa != -1ull)
6212 			return -EINVAL;
6213 
6214 		/*
6215 		 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6216 		 * enable eVMCS capability on vCPU. However, since then
6217 		 * code was changed such that flag signals vmcs12 should
6218 		 * be copied into eVMCS in guest memory.
6219 		 *
6220 		 * To preserve backwards compatability, allow user
6221 		 * to set this flag even when there is no VMXON region.
6222 		 */
6223 		if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6224 			return -EINVAL;
6225 	} else {
6226 		if (!nested_vmx_allowed(vcpu))
6227 			return -EINVAL;
6228 
6229 		if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6230 			return -EINVAL;
6231 	}
6232 
6233 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6234 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6235 		return -EINVAL;
6236 
6237 	if (kvm_state->hdr.vmx.smm.flags &
6238 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6239 		return -EINVAL;
6240 
6241 	if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6242 		return -EINVAL;
6243 
6244 	/*
6245 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
6246 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
6247 	 * must be zero.
6248 	 */
6249 	if (is_smm(vcpu) ?
6250 		(kvm_state->flags &
6251 		 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6252 		: kvm_state->hdr.vmx.smm.flags)
6253 		return -EINVAL;
6254 
6255 	if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6256 	    !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6257 		return -EINVAL;
6258 
6259 	if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6260 		(!nested_vmx_allowed(vcpu) || !vmx->nested.enlightened_vmcs_enabled))
6261 			return -EINVAL;
6262 
6263 	vmx_leave_nested(vcpu);
6264 
6265 	if (kvm_state->hdr.vmx.vmxon_pa == -1ull)
6266 		return 0;
6267 
6268 	vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6269 	ret = enter_vmx_operation(vcpu);
6270 	if (ret)
6271 		return ret;
6272 
6273 	/* Empty 'VMXON' state is permitted if no VMCS loaded */
6274 	if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6275 		/* See vmx_has_valid_vmcs12.  */
6276 		if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6277 		    (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6278 		    (kvm_state->hdr.vmx.vmcs12_pa != -1ull))
6279 			return -EINVAL;
6280 		else
6281 			return 0;
6282 	}
6283 
6284 	if (kvm_state->hdr.vmx.vmcs12_pa != -1ull) {
6285 		if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6286 		    !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6287 			return -EINVAL;
6288 
6289 		set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6290 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6291 		/*
6292 		 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6293 		 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6294 		 * restored yet. EVMCS will be mapped from
6295 		 * nested_get_vmcs12_pages().
6296 		 */
6297 		vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6298 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6299 	} else {
6300 		return -EINVAL;
6301 	}
6302 
6303 	if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6304 		vmx->nested.smm.vmxon = true;
6305 		vmx->nested.vmxon = false;
6306 
6307 		if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6308 			vmx->nested.smm.guest_mode = true;
6309 	}
6310 
6311 	vmcs12 = get_vmcs12(vcpu);
6312 	if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6313 		return -EFAULT;
6314 
6315 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6316 		return -EINVAL;
6317 
6318 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6319 		return 0;
6320 
6321 	vmx->nested.nested_run_pending =
6322 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6323 
6324 	vmx->nested.mtf_pending =
6325 		!!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6326 
6327 	ret = -EINVAL;
6328 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6329 	    vmcs12->vmcs_link_pointer != -1ull) {
6330 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6331 
6332 		if (kvm_state->size <
6333 		    sizeof(*kvm_state) +
6334 		    sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6335 			goto error_guest_mode;
6336 
6337 		if (copy_from_user(shadow_vmcs12,
6338 				   user_vmx_nested_state->shadow_vmcs12,
6339 				   sizeof(*shadow_vmcs12))) {
6340 			ret = -EFAULT;
6341 			goto error_guest_mode;
6342 		}
6343 
6344 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6345 		    !shadow_vmcs12->hdr.shadow_vmcs)
6346 			goto error_guest_mode;
6347 	}
6348 
6349 	vmx->nested.has_preemption_timer_deadline = false;
6350 	if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6351 		vmx->nested.has_preemption_timer_deadline = true;
6352 		vmx->nested.preemption_timer_deadline =
6353 			kvm_state->hdr.vmx.preemption_timer_deadline;
6354 	}
6355 
6356 	if (nested_vmx_check_controls(vcpu, vmcs12) ||
6357 	    nested_vmx_check_host_state(vcpu, vmcs12) ||
6358 	    nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6359 		goto error_guest_mode;
6360 
6361 	vmx->nested.dirty_vmcs12 = true;
6362 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
6363 	if (ret)
6364 		goto error_guest_mode;
6365 
6366 	return 0;
6367 
6368 error_guest_mode:
6369 	vmx->nested.nested_run_pending = 0;
6370 	return ret;
6371 }
6372 
6373 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6374 {
6375 	if (enable_shadow_vmcs) {
6376 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6377 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6378 	}
6379 }
6380 
6381 /*
6382  * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6.  Undo
6383  * that madness to get the encoding for comparison.
6384  */
6385 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6386 
6387 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6388 {
6389 	/*
6390 	 * Note these are the so called "index" of the VMCS field encoding, not
6391 	 * the index into vmcs12.
6392 	 */
6393 	unsigned int max_idx, idx;
6394 	int i;
6395 
6396 	/*
6397 	 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6398 	 * vmcs12, regardless of whether or not the associated feature is
6399 	 * exposed to L1.  Simply find the field with the highest index.
6400 	 */
6401 	max_idx = 0;
6402 	for (i = 0; i < nr_vmcs12_fields; i++) {
6403 		/* The vmcs12 table is very, very sparsely populated. */
6404 		if (!vmcs_field_to_offset_table[i])
6405 			continue;
6406 
6407 		idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6408 		if (idx > max_idx)
6409 			max_idx = idx;
6410 	}
6411 
6412 	return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6413 }
6414 
6415 /*
6416  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
6417  * returned for the various VMX controls MSRs when nested VMX is enabled.
6418  * The same values should also be used to verify that vmcs12 control fields are
6419  * valid during nested entry from L1 to L2.
6420  * Each of these control msrs has a low and high 32-bit half: A low bit is on
6421  * if the corresponding bit in the (32-bit) control field *must* be on, and a
6422  * bit in the high half is on if the corresponding bit in the control field
6423  * may be on. See also vmx_control_verify().
6424  */
6425 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps)
6426 {
6427 	/*
6428 	 * Note that as a general rule, the high half of the MSRs (bits in
6429 	 * the control fields which may be 1) should be initialized by the
6430 	 * intersection of the underlying hardware's MSR (i.e., features which
6431 	 * can be supported) and the list of features we want to expose -
6432 	 * because they are known to be properly supported in our code.
6433 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
6434 	 * be set to 0, meaning that L1 may turn off any of these bits. The
6435 	 * reason is that if one of these bits is necessary, it will appear
6436 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
6437 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
6438 	 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
6439 	 * These rules have exceptions below.
6440 	 */
6441 
6442 	/* pin-based controls */
6443 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
6444 		msrs->pinbased_ctls_low,
6445 		msrs->pinbased_ctls_high);
6446 	msrs->pinbased_ctls_low |=
6447 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6448 	msrs->pinbased_ctls_high &=
6449 		PIN_BASED_EXT_INTR_MASK |
6450 		PIN_BASED_NMI_EXITING |
6451 		PIN_BASED_VIRTUAL_NMIS |
6452 		(enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6453 	msrs->pinbased_ctls_high |=
6454 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6455 		PIN_BASED_VMX_PREEMPTION_TIMER;
6456 
6457 	/* exit controls */
6458 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
6459 		msrs->exit_ctls_low,
6460 		msrs->exit_ctls_high);
6461 	msrs->exit_ctls_low =
6462 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6463 
6464 	msrs->exit_ctls_high &=
6465 #ifdef CONFIG_X86_64
6466 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
6467 #endif
6468 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6469 		VM_EXIT_CLEAR_BNDCFGS | VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6470 	msrs->exit_ctls_high |=
6471 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6472 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6473 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
6474 
6475 	/* We support free control of debug control saving. */
6476 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6477 
6478 	/* entry controls */
6479 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
6480 		msrs->entry_ctls_low,
6481 		msrs->entry_ctls_high);
6482 	msrs->entry_ctls_low =
6483 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6484 	msrs->entry_ctls_high &=
6485 #ifdef CONFIG_X86_64
6486 		VM_ENTRY_IA32E_MODE |
6487 #endif
6488 		VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS |
6489 		VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
6490 	msrs->entry_ctls_high |=
6491 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
6492 
6493 	/* We support free control of debug control loading. */
6494 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6495 
6496 	/* cpu-based controls */
6497 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
6498 		msrs->procbased_ctls_low,
6499 		msrs->procbased_ctls_high);
6500 	msrs->procbased_ctls_low =
6501 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6502 	msrs->procbased_ctls_high &=
6503 		CPU_BASED_INTR_WINDOW_EXITING |
6504 		CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6505 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6506 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6507 		CPU_BASED_CR3_STORE_EXITING |
6508 #ifdef CONFIG_X86_64
6509 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6510 #endif
6511 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6512 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6513 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6514 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6515 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6516 	/*
6517 	 * We can allow some features even when not supported by the
6518 	 * hardware. For example, L1 can specify an MSR bitmap - and we
6519 	 * can use it to avoid exits to L1 - even when L0 runs L2
6520 	 * without MSR bitmaps.
6521 	 */
6522 	msrs->procbased_ctls_high |=
6523 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6524 		CPU_BASED_USE_MSR_BITMAPS;
6525 
6526 	/* We support free control of CR3 access interception. */
6527 	msrs->procbased_ctls_low &=
6528 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6529 
6530 	/*
6531 	 * secondary cpu-based controls.  Do not include those that
6532 	 * depend on CPUID bits, they are added later by
6533 	 * vmx_vcpu_after_set_cpuid.
6534 	 */
6535 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
6536 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
6537 		      msrs->secondary_ctls_low,
6538 		      msrs->secondary_ctls_high);
6539 
6540 	msrs->secondary_ctls_low = 0;
6541 	msrs->secondary_ctls_high &=
6542 		SECONDARY_EXEC_DESC |
6543 		SECONDARY_EXEC_ENABLE_RDTSCP |
6544 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6545 		SECONDARY_EXEC_WBINVD_EXITING |
6546 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
6547 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6548 		SECONDARY_EXEC_RDRAND_EXITING |
6549 		SECONDARY_EXEC_ENABLE_INVPCID |
6550 		SECONDARY_EXEC_RDSEED_EXITING |
6551 		SECONDARY_EXEC_XSAVES |
6552 		SECONDARY_EXEC_TSC_SCALING;
6553 
6554 	/*
6555 	 * We can emulate "VMCS shadowing," even if the hardware
6556 	 * doesn't support it.
6557 	 */
6558 	msrs->secondary_ctls_high |=
6559 		SECONDARY_EXEC_SHADOW_VMCS;
6560 
6561 	if (enable_ept) {
6562 		/* nested EPT: emulate EPT also to L1 */
6563 		msrs->secondary_ctls_high |=
6564 			SECONDARY_EXEC_ENABLE_EPT;
6565 		msrs->ept_caps =
6566 			VMX_EPT_PAGE_WALK_4_BIT |
6567 			VMX_EPT_PAGE_WALK_5_BIT |
6568 			VMX_EPTP_WB_BIT |
6569 			VMX_EPT_INVEPT_BIT |
6570 			VMX_EPT_EXECUTE_ONLY_BIT;
6571 
6572 		msrs->ept_caps &= ept_caps;
6573 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
6574 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
6575 			VMX_EPT_1GB_PAGE_BIT;
6576 		if (enable_ept_ad_bits) {
6577 			msrs->secondary_ctls_high |=
6578 				SECONDARY_EXEC_ENABLE_PML;
6579 			msrs->ept_caps |= VMX_EPT_AD_BIT;
6580 		}
6581 	}
6582 
6583 	if (cpu_has_vmx_vmfunc()) {
6584 		msrs->secondary_ctls_high |=
6585 			SECONDARY_EXEC_ENABLE_VMFUNC;
6586 		/*
6587 		 * Advertise EPTP switching unconditionally
6588 		 * since we emulate it
6589 		 */
6590 		if (enable_ept)
6591 			msrs->vmfunc_controls =
6592 				VMX_VMFUNC_EPTP_SWITCHING;
6593 	}
6594 
6595 	/*
6596 	 * Old versions of KVM use the single-context version without
6597 	 * checking for support, so declare that it is supported even
6598 	 * though it is treated as global context.  The alternative is
6599 	 * not failing the single-context invvpid, and it is worse.
6600 	 */
6601 	if (enable_vpid) {
6602 		msrs->secondary_ctls_high |=
6603 			SECONDARY_EXEC_ENABLE_VPID;
6604 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
6605 			VMX_VPID_EXTENT_SUPPORTED_MASK;
6606 	}
6607 
6608 	if (enable_unrestricted_guest)
6609 		msrs->secondary_ctls_high |=
6610 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
6611 
6612 	if (flexpriority_enabled)
6613 		msrs->secondary_ctls_high |=
6614 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6615 
6616 	if (enable_sgx)
6617 		msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
6618 
6619 	/* miscellaneous data */
6620 	rdmsr(MSR_IA32_VMX_MISC,
6621 		msrs->misc_low,
6622 		msrs->misc_high);
6623 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
6624 	msrs->misc_low |=
6625 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
6626 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
6627 		VMX_MISC_ACTIVITY_HLT |
6628 		VMX_MISC_ACTIVITY_WAIT_SIPI;
6629 	msrs->misc_high = 0;
6630 
6631 	/*
6632 	 * This MSR reports some information about VMX support. We
6633 	 * should return information about the VMX we emulate for the
6634 	 * guest, and the VMCS structure we give it - not about the
6635 	 * VMX support of the underlying hardware.
6636 	 */
6637 	msrs->basic =
6638 		VMCS12_REVISION |
6639 		VMX_BASIC_TRUE_CTLS |
6640 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
6641 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
6642 
6643 	if (cpu_has_vmx_basic_inout())
6644 		msrs->basic |= VMX_BASIC_INOUT;
6645 
6646 	/*
6647 	 * These MSRs specify bits which the guest must keep fixed on
6648 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
6649 	 * We picked the standard core2 setting.
6650 	 */
6651 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
6652 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
6653 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
6654 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
6655 
6656 	/* These MSRs specify bits which the guest must keep fixed off. */
6657 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
6658 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
6659 
6660 	msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
6661 }
6662 
6663 void nested_vmx_hardware_unsetup(void)
6664 {
6665 	int i;
6666 
6667 	if (enable_shadow_vmcs) {
6668 		for (i = 0; i < VMX_BITMAP_NR; i++)
6669 			free_page((unsigned long)vmx_bitmap[i]);
6670 	}
6671 }
6672 
6673 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
6674 {
6675 	int i;
6676 
6677 	if (!cpu_has_vmx_shadow_vmcs())
6678 		enable_shadow_vmcs = 0;
6679 	if (enable_shadow_vmcs) {
6680 		for (i = 0; i < VMX_BITMAP_NR; i++) {
6681 			/*
6682 			 * The vmx_bitmap is not tied to a VM and so should
6683 			 * not be charged to a memcg.
6684 			 */
6685 			vmx_bitmap[i] = (unsigned long *)
6686 				__get_free_page(GFP_KERNEL);
6687 			if (!vmx_bitmap[i]) {
6688 				nested_vmx_hardware_unsetup();
6689 				return -ENOMEM;
6690 			}
6691 		}
6692 
6693 		init_vmcs_shadow_fields();
6694 	}
6695 
6696 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear;
6697 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch;
6698 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld;
6699 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst;
6700 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread;
6701 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume;
6702 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite;
6703 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff;
6704 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon;
6705 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept;
6706 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid;
6707 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc;
6708 
6709 	return 0;
6710 }
6711 
6712 struct kvm_x86_nested_ops vmx_nested_ops = {
6713 	.check_events = vmx_check_nested_events,
6714 	.hv_timer_pending = nested_vmx_preemption_timer_pending,
6715 	.triple_fault = nested_vmx_triple_fault,
6716 	.get_state = vmx_get_nested_state,
6717 	.set_state = vmx_set_nested_state,
6718 	.get_nested_state_pages = vmx_get_nested_state_pages,
6719 	.write_log_dirty = nested_vmx_write_pml_buffer,
6720 	.enable_evmcs = nested_enable_evmcs,
6721 	.get_evmcs_version = nested_get_evmcs_version,
6722 };
6723