xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision dd5b2498)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/frame.h>
4 #include <linux/percpu.h>
5 
6 #include <asm/debugreg.h>
7 #include <asm/mmu_context.h>
8 
9 #include "cpuid.h"
10 #include "hyperv.h"
11 #include "mmu.h"
12 #include "nested.h"
13 #include "trace.h"
14 #include "x86.h"
15 
16 static bool __read_mostly enable_shadow_vmcs = 1;
17 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
18 
19 static bool __read_mostly nested_early_check = 0;
20 module_param(nested_early_check, bool, S_IRUGO);
21 
22 /*
23  * Hyper-V requires all of these, so mark them as supported even though
24  * they are just treated the same as all-context.
25  */
26 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
27 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
28 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
29 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
30 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
31 
32 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
33 
34 enum {
35 	VMX_VMREAD_BITMAP,
36 	VMX_VMWRITE_BITMAP,
37 	VMX_BITMAP_NR
38 };
39 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
40 
41 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
42 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
43 
44 static u16 shadow_read_only_fields[] = {
45 #define SHADOW_FIELD_RO(x) x,
46 #include "vmcs_shadow_fields.h"
47 };
48 static int max_shadow_read_only_fields =
49 	ARRAY_SIZE(shadow_read_only_fields);
50 
51 static u16 shadow_read_write_fields[] = {
52 #define SHADOW_FIELD_RW(x) x,
53 #include "vmcs_shadow_fields.h"
54 };
55 static int max_shadow_read_write_fields =
56 	ARRAY_SIZE(shadow_read_write_fields);
57 
58 static void init_vmcs_shadow_fields(void)
59 {
60 	int i, j;
61 
62 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
63 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
64 
65 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
66 		u16 field = shadow_read_only_fields[i];
67 
68 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
69 		    (i + 1 == max_shadow_read_only_fields ||
70 		     shadow_read_only_fields[i + 1] != field + 1))
71 			pr_err("Missing field from shadow_read_only_field %x\n",
72 			       field + 1);
73 
74 		clear_bit(field, vmx_vmread_bitmap);
75 #ifdef CONFIG_X86_64
76 		if (field & 1)
77 			continue;
78 #endif
79 		if (j < i)
80 			shadow_read_only_fields[j] = field;
81 		j++;
82 	}
83 	max_shadow_read_only_fields = j;
84 
85 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
86 		u16 field = shadow_read_write_fields[i];
87 
88 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
89 		    (i + 1 == max_shadow_read_write_fields ||
90 		     shadow_read_write_fields[i + 1] != field + 1))
91 			pr_err("Missing field from shadow_read_write_field %x\n",
92 			       field + 1);
93 
94 		/*
95 		 * PML and the preemption timer can be emulated, but the
96 		 * processor cannot vmwrite to fields that don't exist
97 		 * on bare metal.
98 		 */
99 		switch (field) {
100 		case GUEST_PML_INDEX:
101 			if (!cpu_has_vmx_pml())
102 				continue;
103 			break;
104 		case VMX_PREEMPTION_TIMER_VALUE:
105 			if (!cpu_has_vmx_preemption_timer())
106 				continue;
107 			break;
108 		case GUEST_INTR_STATUS:
109 			if (!cpu_has_vmx_apicv())
110 				continue;
111 			break;
112 		default:
113 			break;
114 		}
115 
116 		clear_bit(field, vmx_vmwrite_bitmap);
117 		clear_bit(field, vmx_vmread_bitmap);
118 #ifdef CONFIG_X86_64
119 		if (field & 1)
120 			continue;
121 #endif
122 		if (j < i)
123 			shadow_read_write_fields[j] = field;
124 		j++;
125 	}
126 	max_shadow_read_write_fields = j;
127 }
128 
129 /*
130  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
131  * set the success or error code of an emulated VMX instruction (as specified
132  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
133  * instruction.
134  */
135 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
136 {
137 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
138 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
139 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
140 	return kvm_skip_emulated_instruction(vcpu);
141 }
142 
143 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
144 {
145 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
146 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
147 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
148 			| X86_EFLAGS_CF);
149 	return kvm_skip_emulated_instruction(vcpu);
150 }
151 
152 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
153 				u32 vm_instruction_error)
154 {
155 	struct vcpu_vmx *vmx = to_vmx(vcpu);
156 
157 	/*
158 	 * failValid writes the error number to the current VMCS, which
159 	 * can't be done if there isn't a current VMCS.
160 	 */
161 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
162 		return nested_vmx_failInvalid(vcpu);
163 
164 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
165 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
166 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
167 			| X86_EFLAGS_ZF);
168 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
169 	/*
170 	 * We don't need to force a shadow sync because
171 	 * VM_INSTRUCTION_ERROR is not shadowed
172 	 */
173 	return kvm_skip_emulated_instruction(vcpu);
174 }
175 
176 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
177 {
178 	/* TODO: not to reset guest simply here. */
179 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
180 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
181 }
182 
183 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
184 {
185 	vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
186 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
187 }
188 
189 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
190 {
191 	struct vcpu_vmx *vmx = to_vmx(vcpu);
192 
193 	if (!vmx->nested.hv_evmcs)
194 		return;
195 
196 	kunmap(vmx->nested.hv_evmcs_page);
197 	kvm_release_page_dirty(vmx->nested.hv_evmcs_page);
198 	vmx->nested.hv_evmcs_vmptr = -1ull;
199 	vmx->nested.hv_evmcs_page = NULL;
200 	vmx->nested.hv_evmcs = NULL;
201 }
202 
203 /*
204  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
205  * just stops using VMX.
206  */
207 static void free_nested(struct kvm_vcpu *vcpu)
208 {
209 	struct vcpu_vmx *vmx = to_vmx(vcpu);
210 
211 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
212 		return;
213 
214 	vmx->nested.vmxon = false;
215 	vmx->nested.smm.vmxon = false;
216 	free_vpid(vmx->nested.vpid02);
217 	vmx->nested.posted_intr_nv = -1;
218 	vmx->nested.current_vmptr = -1ull;
219 	if (enable_shadow_vmcs) {
220 		vmx_disable_shadow_vmcs(vmx);
221 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
222 		free_vmcs(vmx->vmcs01.shadow_vmcs);
223 		vmx->vmcs01.shadow_vmcs = NULL;
224 	}
225 	kfree(vmx->nested.cached_vmcs12);
226 	kfree(vmx->nested.cached_shadow_vmcs12);
227 	/* Unpin physical memory we referred to in the vmcs02 */
228 	if (vmx->nested.apic_access_page) {
229 		kvm_release_page_dirty(vmx->nested.apic_access_page);
230 		vmx->nested.apic_access_page = NULL;
231 	}
232 	if (vmx->nested.virtual_apic_page) {
233 		kvm_release_page_dirty(vmx->nested.virtual_apic_page);
234 		vmx->nested.virtual_apic_page = NULL;
235 	}
236 	if (vmx->nested.pi_desc_page) {
237 		kunmap(vmx->nested.pi_desc_page);
238 		kvm_release_page_dirty(vmx->nested.pi_desc_page);
239 		vmx->nested.pi_desc_page = NULL;
240 		vmx->nested.pi_desc = NULL;
241 	}
242 
243 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
244 
245 	nested_release_evmcs(vcpu);
246 
247 	free_loaded_vmcs(&vmx->nested.vmcs02);
248 }
249 
250 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
251 {
252 	struct vcpu_vmx *vmx = to_vmx(vcpu);
253 	int cpu;
254 
255 	if (vmx->loaded_vmcs == vmcs)
256 		return;
257 
258 	cpu = get_cpu();
259 	vmx_vcpu_put(vcpu);
260 	vmx->loaded_vmcs = vmcs;
261 	vmx_vcpu_load(vcpu, cpu);
262 	put_cpu();
263 
264 	vm_entry_controls_reset_shadow(vmx);
265 	vm_exit_controls_reset_shadow(vmx);
266 	vmx_segment_cache_clear(vmx);
267 }
268 
269 /*
270  * Ensure that the current vmcs of the logical processor is the
271  * vmcs01 of the vcpu before calling free_nested().
272  */
273 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
274 {
275 	vcpu_load(vcpu);
276 	vmx_leave_nested(vcpu);
277 	vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
278 	free_nested(vcpu);
279 	vcpu_put(vcpu);
280 }
281 
282 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
283 		struct x86_exception *fault)
284 {
285 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
286 	struct vcpu_vmx *vmx = to_vmx(vcpu);
287 	u32 exit_reason;
288 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
289 
290 	if (vmx->nested.pml_full) {
291 		exit_reason = EXIT_REASON_PML_FULL;
292 		vmx->nested.pml_full = false;
293 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
294 	} else if (fault->error_code & PFERR_RSVD_MASK)
295 		exit_reason = EXIT_REASON_EPT_MISCONFIG;
296 	else
297 		exit_reason = EXIT_REASON_EPT_VIOLATION;
298 
299 	nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
300 	vmcs12->guest_physical_address = fault->address;
301 }
302 
303 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
304 {
305 	WARN_ON(mmu_is_nested(vcpu));
306 
307 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
308 	kvm_init_shadow_ept_mmu(vcpu,
309 			to_vmx(vcpu)->nested.msrs.ept_caps &
310 			VMX_EPT_EXECUTE_ONLY_BIT,
311 			nested_ept_ad_enabled(vcpu),
312 			nested_ept_get_cr3(vcpu));
313 	vcpu->arch.mmu->set_cr3           = vmx_set_cr3;
314 	vcpu->arch.mmu->get_cr3           = nested_ept_get_cr3;
315 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
316 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
317 
318 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
319 }
320 
321 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
322 {
323 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
324 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
325 }
326 
327 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
328 					    u16 error_code)
329 {
330 	bool inequality, bit;
331 
332 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
333 	inequality =
334 		(error_code & vmcs12->page_fault_error_code_mask) !=
335 		 vmcs12->page_fault_error_code_match;
336 	return inequality ^ bit;
337 }
338 
339 
340 /*
341  * KVM wants to inject page-faults which it got to the guest. This function
342  * checks whether in a nested guest, we need to inject them to L1 or L2.
343  */
344 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
345 {
346 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
347 	unsigned int nr = vcpu->arch.exception.nr;
348 	bool has_payload = vcpu->arch.exception.has_payload;
349 	unsigned long payload = vcpu->arch.exception.payload;
350 
351 	if (nr == PF_VECTOR) {
352 		if (vcpu->arch.exception.nested_apf) {
353 			*exit_qual = vcpu->arch.apf.nested_apf_token;
354 			return 1;
355 		}
356 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
357 						    vcpu->arch.exception.error_code)) {
358 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
359 			return 1;
360 		}
361 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
362 		if (nr == DB_VECTOR) {
363 			if (!has_payload) {
364 				payload = vcpu->arch.dr6;
365 				payload &= ~(DR6_FIXED_1 | DR6_BT);
366 				payload ^= DR6_RTM;
367 			}
368 			*exit_qual = payload;
369 		} else
370 			*exit_qual = 0;
371 		return 1;
372 	}
373 
374 	return 0;
375 }
376 
377 
378 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
379 		struct x86_exception *fault)
380 {
381 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
382 
383 	WARN_ON(!is_guest_mode(vcpu));
384 
385 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
386 		!to_vmx(vcpu)->nested.nested_run_pending) {
387 		vmcs12->vm_exit_intr_error_code = fault->error_code;
388 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
389 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
390 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
391 				  fault->address);
392 	} else {
393 		kvm_inject_page_fault(vcpu, fault);
394 	}
395 }
396 
397 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
398 {
399 	return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
400 }
401 
402 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
403 					       struct vmcs12 *vmcs12)
404 {
405 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
406 		return 0;
407 
408 	if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
409 	    !page_address_valid(vcpu, vmcs12->io_bitmap_b))
410 		return -EINVAL;
411 
412 	return 0;
413 }
414 
415 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
416 						struct vmcs12 *vmcs12)
417 {
418 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
419 		return 0;
420 
421 	if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
422 		return -EINVAL;
423 
424 	return 0;
425 }
426 
427 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
428 						struct vmcs12 *vmcs12)
429 {
430 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
431 		return 0;
432 
433 	if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
434 		return -EINVAL;
435 
436 	return 0;
437 }
438 
439 /*
440  * Check if MSR is intercepted for L01 MSR bitmap.
441  */
442 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
443 {
444 	unsigned long *msr_bitmap;
445 	int f = sizeof(unsigned long);
446 
447 	if (!cpu_has_vmx_msr_bitmap())
448 		return true;
449 
450 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
451 
452 	if (msr <= 0x1fff) {
453 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
454 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
455 		msr &= 0x1fff;
456 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
457 	}
458 
459 	return true;
460 }
461 
462 /*
463  * If a msr is allowed by L0, we should check whether it is allowed by L1.
464  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
465  */
466 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
467 					       unsigned long *msr_bitmap_nested,
468 					       u32 msr, int type)
469 {
470 	int f = sizeof(unsigned long);
471 
472 	/*
473 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
474 	 * have the write-low and read-high bitmap offsets the wrong way round.
475 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
476 	 */
477 	if (msr <= 0x1fff) {
478 		if (type & MSR_TYPE_R &&
479 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
480 			/* read-low */
481 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
482 
483 		if (type & MSR_TYPE_W &&
484 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
485 			/* write-low */
486 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
487 
488 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
489 		msr &= 0x1fff;
490 		if (type & MSR_TYPE_R &&
491 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
492 			/* read-high */
493 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
494 
495 		if (type & MSR_TYPE_W &&
496 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
497 			/* write-high */
498 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
499 
500 	}
501 }
502 
503 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap) {
504 	int msr;
505 
506 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
507 		unsigned word = msr / BITS_PER_LONG;
508 
509 		msr_bitmap[word] = ~0;
510 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
511 	}
512 }
513 
514 /*
515  * Merge L0's and L1's MSR bitmap, return false to indicate that
516  * we do not use the hardware.
517  */
518 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
519 						 struct vmcs12 *vmcs12)
520 {
521 	int msr;
522 	struct page *page;
523 	unsigned long *msr_bitmap_l1;
524 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
525 	/*
526 	 * pred_cmd & spec_ctrl are trying to verify two things:
527 	 *
528 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
529 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
530 	 *    from the L12 MSR bitmap that is too permissive.
531 	 * 2. That L1 or L2s have actually used the MSR. This avoids
532 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
533 	 *    works properly because we only update the L01 MSR bitmap lazily.
534 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
535 	 *    updated to reflect this when L1 (or its L2s) actually write to
536 	 *    the MSR.
537 	 */
538 	bool pred_cmd = !msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD);
539 	bool spec_ctrl = !msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL);
540 
541 	/* Nothing to do if the MSR bitmap is not in use.  */
542 	if (!cpu_has_vmx_msr_bitmap() ||
543 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
544 		return false;
545 
546 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
547 	    !pred_cmd && !spec_ctrl)
548 		return false;
549 
550 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
551 	if (is_error_page(page))
552 		return false;
553 
554 	msr_bitmap_l1 = (unsigned long *)kmap(page);
555 
556 	/*
557 	 * To keep the control flow simple, pay eight 8-byte writes (sixteen
558 	 * 4-byte writes on 32-bit systems) up front to enable intercepts for
559 	 * the x2APIC MSR range and selectively disable them below.
560 	 */
561 	enable_x2apic_msr_intercepts(msr_bitmap_l0);
562 
563 	if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
564 		if (nested_cpu_has_apic_reg_virt(vmcs12)) {
565 			/*
566 			 * L0 need not intercept reads for MSRs between 0x800
567 			 * and 0x8ff, it just lets the processor take the value
568 			 * from the virtual-APIC page; take those 256 bits
569 			 * directly from the L1 bitmap.
570 			 */
571 			for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
572 				unsigned word = msr / BITS_PER_LONG;
573 
574 				msr_bitmap_l0[word] = msr_bitmap_l1[word];
575 			}
576 		}
577 
578 		nested_vmx_disable_intercept_for_msr(
579 			msr_bitmap_l1, msr_bitmap_l0,
580 			X2APIC_MSR(APIC_TASKPRI),
581 			MSR_TYPE_R | MSR_TYPE_W);
582 
583 		if (nested_cpu_has_vid(vmcs12)) {
584 			nested_vmx_disable_intercept_for_msr(
585 				msr_bitmap_l1, msr_bitmap_l0,
586 				X2APIC_MSR(APIC_EOI),
587 				MSR_TYPE_W);
588 			nested_vmx_disable_intercept_for_msr(
589 				msr_bitmap_l1, msr_bitmap_l0,
590 				X2APIC_MSR(APIC_SELF_IPI),
591 				MSR_TYPE_W);
592 		}
593 	}
594 
595 	if (spec_ctrl)
596 		nested_vmx_disable_intercept_for_msr(
597 					msr_bitmap_l1, msr_bitmap_l0,
598 					MSR_IA32_SPEC_CTRL,
599 					MSR_TYPE_R | MSR_TYPE_W);
600 
601 	if (pred_cmd)
602 		nested_vmx_disable_intercept_for_msr(
603 					msr_bitmap_l1, msr_bitmap_l0,
604 					MSR_IA32_PRED_CMD,
605 					MSR_TYPE_W);
606 
607 	kunmap(page);
608 	kvm_release_page_clean(page);
609 
610 	return true;
611 }
612 
613 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
614 				       struct vmcs12 *vmcs12)
615 {
616 	struct vmcs12 *shadow;
617 	struct page *page;
618 
619 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
620 	    vmcs12->vmcs_link_pointer == -1ull)
621 		return;
622 
623 	shadow = get_shadow_vmcs12(vcpu);
624 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->vmcs_link_pointer);
625 
626 	memcpy(shadow, kmap(page), VMCS12_SIZE);
627 
628 	kunmap(page);
629 	kvm_release_page_clean(page);
630 }
631 
632 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
633 					      struct vmcs12 *vmcs12)
634 {
635 	struct vcpu_vmx *vmx = to_vmx(vcpu);
636 
637 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
638 	    vmcs12->vmcs_link_pointer == -1ull)
639 		return;
640 
641 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
642 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
643 }
644 
645 /*
646  * In nested virtualization, check if L1 has set
647  * VM_EXIT_ACK_INTR_ON_EXIT
648  */
649 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
650 {
651 	return get_vmcs12(vcpu)->vm_exit_controls &
652 		VM_EXIT_ACK_INTR_ON_EXIT;
653 }
654 
655 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
656 {
657 	return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
658 }
659 
660 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
661 					  struct vmcs12 *vmcs12)
662 {
663 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
664 	    !page_address_valid(vcpu, vmcs12->apic_access_addr))
665 		return -EINVAL;
666 	else
667 		return 0;
668 }
669 
670 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
671 					   struct vmcs12 *vmcs12)
672 {
673 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
674 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
675 	    !nested_cpu_has_vid(vmcs12) &&
676 	    !nested_cpu_has_posted_intr(vmcs12))
677 		return 0;
678 
679 	/*
680 	 * If virtualize x2apic mode is enabled,
681 	 * virtualize apic access must be disabled.
682 	 */
683 	if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
684 	    nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
685 		return -EINVAL;
686 
687 	/*
688 	 * If virtual interrupt delivery is enabled,
689 	 * we must exit on external interrupts.
690 	 */
691 	if (nested_cpu_has_vid(vmcs12) &&
692 	   !nested_exit_on_intr(vcpu))
693 		return -EINVAL;
694 
695 	/*
696 	 * bits 15:8 should be zero in posted_intr_nv,
697 	 * the descriptor address has been already checked
698 	 * in nested_get_vmcs12_pages.
699 	 *
700 	 * bits 5:0 of posted_intr_desc_addr should be zero.
701 	 */
702 	if (nested_cpu_has_posted_intr(vmcs12) &&
703 	   (!nested_cpu_has_vid(vmcs12) ||
704 	    !nested_exit_intr_ack_set(vcpu) ||
705 	    (vmcs12->posted_intr_nv & 0xff00) ||
706 	    (vmcs12->posted_intr_desc_addr & 0x3f) ||
707 	    (vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu))))
708 		return -EINVAL;
709 
710 	/* tpr shadow is needed by all apicv features. */
711 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
712 		return -EINVAL;
713 
714 	return 0;
715 }
716 
717 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
718 				       u32 count, u64 addr)
719 {
720 	int maxphyaddr;
721 
722 	if (count == 0)
723 		return 0;
724 	maxphyaddr = cpuid_maxphyaddr(vcpu);
725 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
726 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
727 		return -EINVAL;
728 
729 	return 0;
730 }
731 
732 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
733 						     struct vmcs12 *vmcs12)
734 {
735 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_load_count,
736 					vmcs12->vm_exit_msr_load_addr) ||
737 	    nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_store_count,
738 					vmcs12->vm_exit_msr_store_addr))
739 		return -EINVAL;
740 
741 	return 0;
742 }
743 
744 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
745                                                       struct vmcs12 *vmcs12)
746 {
747 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_entry_msr_load_count,
748                                         vmcs12->vm_entry_msr_load_addr))
749                 return -EINVAL;
750 
751 	return 0;
752 }
753 
754 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
755 					 struct vmcs12 *vmcs12)
756 {
757 	if (!nested_cpu_has_pml(vmcs12))
758 		return 0;
759 
760 	if (!nested_cpu_has_ept(vmcs12) ||
761 	    !page_address_valid(vcpu, vmcs12->pml_address))
762 		return -EINVAL;
763 
764 	return 0;
765 }
766 
767 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
768 							struct vmcs12 *vmcs12)
769 {
770 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
771 	    !nested_cpu_has_ept(vmcs12))
772 		return -EINVAL;
773 	return 0;
774 }
775 
776 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
777 							 struct vmcs12 *vmcs12)
778 {
779 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
780 	    !nested_cpu_has_ept(vmcs12))
781 		return -EINVAL;
782 	return 0;
783 }
784 
785 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
786 						 struct vmcs12 *vmcs12)
787 {
788 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
789 		return 0;
790 
791 	if (!page_address_valid(vcpu, vmcs12->vmread_bitmap) ||
792 	    !page_address_valid(vcpu, vmcs12->vmwrite_bitmap))
793 		return -EINVAL;
794 
795 	return 0;
796 }
797 
798 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
799 				       struct vmx_msr_entry *e)
800 {
801 	/* x2APIC MSR accesses are not allowed */
802 	if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
803 		return -EINVAL;
804 	if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
805 	    e->index == MSR_IA32_UCODE_REV)
806 		return -EINVAL;
807 	if (e->reserved != 0)
808 		return -EINVAL;
809 	return 0;
810 }
811 
812 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
813 				     struct vmx_msr_entry *e)
814 {
815 	if (e->index == MSR_FS_BASE ||
816 	    e->index == MSR_GS_BASE ||
817 	    e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
818 	    nested_vmx_msr_check_common(vcpu, e))
819 		return -EINVAL;
820 	return 0;
821 }
822 
823 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
824 				      struct vmx_msr_entry *e)
825 {
826 	if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
827 	    nested_vmx_msr_check_common(vcpu, e))
828 		return -EINVAL;
829 	return 0;
830 }
831 
832 /*
833  * Load guest's/host's msr at nested entry/exit.
834  * return 0 for success, entry index for failure.
835  */
836 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
837 {
838 	u32 i;
839 	struct vmx_msr_entry e;
840 	struct msr_data msr;
841 
842 	msr.host_initiated = false;
843 	for (i = 0; i < count; i++) {
844 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
845 					&e, sizeof(e))) {
846 			pr_debug_ratelimited(
847 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
848 				__func__, i, gpa + i * sizeof(e));
849 			goto fail;
850 		}
851 		if (nested_vmx_load_msr_check(vcpu, &e)) {
852 			pr_debug_ratelimited(
853 				"%s check failed (%u, 0x%x, 0x%x)\n",
854 				__func__, i, e.index, e.reserved);
855 			goto fail;
856 		}
857 		msr.index = e.index;
858 		msr.data = e.value;
859 		if (kvm_set_msr(vcpu, &msr)) {
860 			pr_debug_ratelimited(
861 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
862 				__func__, i, e.index, e.value);
863 			goto fail;
864 		}
865 	}
866 	return 0;
867 fail:
868 	return i + 1;
869 }
870 
871 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
872 {
873 	u32 i;
874 	struct vmx_msr_entry e;
875 
876 	for (i = 0; i < count; i++) {
877 		struct msr_data msr_info;
878 		if (kvm_vcpu_read_guest(vcpu,
879 					gpa + i * sizeof(e),
880 					&e, 2 * sizeof(u32))) {
881 			pr_debug_ratelimited(
882 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
883 				__func__, i, gpa + i * sizeof(e));
884 			return -EINVAL;
885 		}
886 		if (nested_vmx_store_msr_check(vcpu, &e)) {
887 			pr_debug_ratelimited(
888 				"%s check failed (%u, 0x%x, 0x%x)\n",
889 				__func__, i, e.index, e.reserved);
890 			return -EINVAL;
891 		}
892 		msr_info.host_initiated = false;
893 		msr_info.index = e.index;
894 		if (kvm_get_msr(vcpu, &msr_info)) {
895 			pr_debug_ratelimited(
896 				"%s cannot read MSR (%u, 0x%x)\n",
897 				__func__, i, e.index);
898 			return -EINVAL;
899 		}
900 		if (kvm_vcpu_write_guest(vcpu,
901 					 gpa + i * sizeof(e) +
902 					     offsetof(struct vmx_msr_entry, value),
903 					 &msr_info.data, sizeof(msr_info.data))) {
904 			pr_debug_ratelimited(
905 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
906 				__func__, i, e.index, msr_info.data);
907 			return -EINVAL;
908 		}
909 	}
910 	return 0;
911 }
912 
913 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
914 {
915 	unsigned long invalid_mask;
916 
917 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
918 	return (val & invalid_mask) == 0;
919 }
920 
921 /*
922  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
923  * emulating VM entry into a guest with EPT enabled.
924  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
925  * is assigned to entry_failure_code on failure.
926  */
927 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
928 			       u32 *entry_failure_code)
929 {
930 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
931 		if (!nested_cr3_valid(vcpu, cr3)) {
932 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
933 			return 1;
934 		}
935 
936 		/*
937 		 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
938 		 * must not be dereferenced.
939 		 */
940 		if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
941 		    !nested_ept) {
942 			if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
943 				*entry_failure_code = ENTRY_FAIL_PDPTE;
944 				return 1;
945 			}
946 		}
947 	}
948 
949 	if (!nested_ept)
950 		kvm_mmu_new_cr3(vcpu, cr3, false);
951 
952 	vcpu->arch.cr3 = cr3;
953 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
954 
955 	kvm_init_mmu(vcpu, false);
956 
957 	return 0;
958 }
959 
960 /*
961  * Returns if KVM is able to config CPU to tag TLB entries
962  * populated by L2 differently than TLB entries populated
963  * by L1.
964  *
965  * If L1 uses EPT, then TLB entries are tagged with different EPTP.
966  *
967  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
968  * with different VPID (L1 entries are tagged with vmx->vpid
969  * while L2 entries are tagged with vmx->nested.vpid02).
970  */
971 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
972 {
973 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
974 
975 	return nested_cpu_has_ept(vmcs12) ||
976 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
977 }
978 
979 static u16 nested_get_vpid02(struct kvm_vcpu *vcpu)
980 {
981 	struct vcpu_vmx *vmx = to_vmx(vcpu);
982 
983 	return vmx->nested.vpid02 ? vmx->nested.vpid02 : vmx->vpid;
984 }
985 
986 
987 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
988 {
989 	return fixed_bits_valid(control, low, high);
990 }
991 
992 static inline u64 vmx_control_msr(u32 low, u32 high)
993 {
994 	return low | ((u64)high << 32);
995 }
996 
997 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
998 {
999 	superset &= mask;
1000 	subset &= mask;
1001 
1002 	return (superset | subset) == superset;
1003 }
1004 
1005 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1006 {
1007 	const u64 feature_and_reserved =
1008 		/* feature (except bit 48; see below) */
1009 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1010 		/* reserved */
1011 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1012 	u64 vmx_basic = vmx->nested.msrs.basic;
1013 
1014 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1015 		return -EINVAL;
1016 
1017 	/*
1018 	 * KVM does not emulate a version of VMX that constrains physical
1019 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1020 	 */
1021 	if (data & BIT_ULL(48))
1022 		return -EINVAL;
1023 
1024 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1025 	    vmx_basic_vmcs_revision_id(data))
1026 		return -EINVAL;
1027 
1028 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1029 		return -EINVAL;
1030 
1031 	vmx->nested.msrs.basic = data;
1032 	return 0;
1033 }
1034 
1035 static int
1036 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1037 {
1038 	u64 supported;
1039 	u32 *lowp, *highp;
1040 
1041 	switch (msr_index) {
1042 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1043 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1044 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1045 		break;
1046 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1047 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1048 		highp = &vmx->nested.msrs.procbased_ctls_high;
1049 		break;
1050 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1051 		lowp = &vmx->nested.msrs.exit_ctls_low;
1052 		highp = &vmx->nested.msrs.exit_ctls_high;
1053 		break;
1054 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1055 		lowp = &vmx->nested.msrs.entry_ctls_low;
1056 		highp = &vmx->nested.msrs.entry_ctls_high;
1057 		break;
1058 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1059 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1060 		highp = &vmx->nested.msrs.secondary_ctls_high;
1061 		break;
1062 	default:
1063 		BUG();
1064 	}
1065 
1066 	supported = vmx_control_msr(*lowp, *highp);
1067 
1068 	/* Check must-be-1 bits are still 1. */
1069 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1070 		return -EINVAL;
1071 
1072 	/* Check must-be-0 bits are still 0. */
1073 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1074 		return -EINVAL;
1075 
1076 	*lowp = data;
1077 	*highp = data >> 32;
1078 	return 0;
1079 }
1080 
1081 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1082 {
1083 	const u64 feature_and_reserved_bits =
1084 		/* feature */
1085 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1086 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1087 		/* reserved */
1088 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1089 	u64 vmx_misc;
1090 
1091 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1092 				   vmx->nested.msrs.misc_high);
1093 
1094 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1095 		return -EINVAL;
1096 
1097 	if ((vmx->nested.msrs.pinbased_ctls_high &
1098 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1099 	    vmx_misc_preemption_timer_rate(data) !=
1100 	    vmx_misc_preemption_timer_rate(vmx_misc))
1101 		return -EINVAL;
1102 
1103 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1104 		return -EINVAL;
1105 
1106 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1107 		return -EINVAL;
1108 
1109 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1110 		return -EINVAL;
1111 
1112 	vmx->nested.msrs.misc_low = data;
1113 	vmx->nested.msrs.misc_high = data >> 32;
1114 
1115 	/*
1116 	 * If L1 has read-only VM-exit information fields, use the
1117 	 * less permissive vmx_vmwrite_bitmap to specify write
1118 	 * permissions for the shadow VMCS.
1119 	 */
1120 	if (enable_shadow_vmcs && !nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
1121 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
1122 
1123 	return 0;
1124 }
1125 
1126 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1127 {
1128 	u64 vmx_ept_vpid_cap;
1129 
1130 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1131 					   vmx->nested.msrs.vpid_caps);
1132 
1133 	/* Every bit is either reserved or a feature bit. */
1134 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1135 		return -EINVAL;
1136 
1137 	vmx->nested.msrs.ept_caps = data;
1138 	vmx->nested.msrs.vpid_caps = data >> 32;
1139 	return 0;
1140 }
1141 
1142 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1143 {
1144 	u64 *msr;
1145 
1146 	switch (msr_index) {
1147 	case MSR_IA32_VMX_CR0_FIXED0:
1148 		msr = &vmx->nested.msrs.cr0_fixed0;
1149 		break;
1150 	case MSR_IA32_VMX_CR4_FIXED0:
1151 		msr = &vmx->nested.msrs.cr4_fixed0;
1152 		break;
1153 	default:
1154 		BUG();
1155 	}
1156 
1157 	/*
1158 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1159 	 * must be 1 in the restored value.
1160 	 */
1161 	if (!is_bitwise_subset(data, *msr, -1ULL))
1162 		return -EINVAL;
1163 
1164 	*msr = data;
1165 	return 0;
1166 }
1167 
1168 /*
1169  * Called when userspace is restoring VMX MSRs.
1170  *
1171  * Returns 0 on success, non-0 otherwise.
1172  */
1173 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1174 {
1175 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1176 
1177 	/*
1178 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1179 	 * is in VMX operation.
1180 	 */
1181 	if (vmx->nested.vmxon)
1182 		return -EBUSY;
1183 
1184 	switch (msr_index) {
1185 	case MSR_IA32_VMX_BASIC:
1186 		return vmx_restore_vmx_basic(vmx, data);
1187 	case MSR_IA32_VMX_PINBASED_CTLS:
1188 	case MSR_IA32_VMX_PROCBASED_CTLS:
1189 	case MSR_IA32_VMX_EXIT_CTLS:
1190 	case MSR_IA32_VMX_ENTRY_CTLS:
1191 		/*
1192 		 * The "non-true" VMX capability MSRs are generated from the
1193 		 * "true" MSRs, so we do not support restoring them directly.
1194 		 *
1195 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1196 		 * should restore the "true" MSRs with the must-be-1 bits
1197 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1198 		 * DEFAULT SETTINGS".
1199 		 */
1200 		return -EINVAL;
1201 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1202 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1203 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1204 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1205 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1206 		return vmx_restore_control_msr(vmx, msr_index, data);
1207 	case MSR_IA32_VMX_MISC:
1208 		return vmx_restore_vmx_misc(vmx, data);
1209 	case MSR_IA32_VMX_CR0_FIXED0:
1210 	case MSR_IA32_VMX_CR4_FIXED0:
1211 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1212 	case MSR_IA32_VMX_CR0_FIXED1:
1213 	case MSR_IA32_VMX_CR4_FIXED1:
1214 		/*
1215 		 * These MSRs are generated based on the vCPU's CPUID, so we
1216 		 * do not support restoring them directly.
1217 		 */
1218 		return -EINVAL;
1219 	case MSR_IA32_VMX_EPT_VPID_CAP:
1220 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1221 	case MSR_IA32_VMX_VMCS_ENUM:
1222 		vmx->nested.msrs.vmcs_enum = data;
1223 		return 0;
1224 	default:
1225 		/*
1226 		 * The rest of the VMX capability MSRs do not support restore.
1227 		 */
1228 		return -EINVAL;
1229 	}
1230 }
1231 
1232 /* Returns 0 on success, non-0 otherwise. */
1233 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1234 {
1235 	switch (msr_index) {
1236 	case MSR_IA32_VMX_BASIC:
1237 		*pdata = msrs->basic;
1238 		break;
1239 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1240 	case MSR_IA32_VMX_PINBASED_CTLS:
1241 		*pdata = vmx_control_msr(
1242 			msrs->pinbased_ctls_low,
1243 			msrs->pinbased_ctls_high);
1244 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1245 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1246 		break;
1247 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1248 	case MSR_IA32_VMX_PROCBASED_CTLS:
1249 		*pdata = vmx_control_msr(
1250 			msrs->procbased_ctls_low,
1251 			msrs->procbased_ctls_high);
1252 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1253 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1254 		break;
1255 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1256 	case MSR_IA32_VMX_EXIT_CTLS:
1257 		*pdata = vmx_control_msr(
1258 			msrs->exit_ctls_low,
1259 			msrs->exit_ctls_high);
1260 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1261 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1262 		break;
1263 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1264 	case MSR_IA32_VMX_ENTRY_CTLS:
1265 		*pdata = vmx_control_msr(
1266 			msrs->entry_ctls_low,
1267 			msrs->entry_ctls_high);
1268 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1269 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1270 		break;
1271 	case MSR_IA32_VMX_MISC:
1272 		*pdata = vmx_control_msr(
1273 			msrs->misc_low,
1274 			msrs->misc_high);
1275 		break;
1276 	case MSR_IA32_VMX_CR0_FIXED0:
1277 		*pdata = msrs->cr0_fixed0;
1278 		break;
1279 	case MSR_IA32_VMX_CR0_FIXED1:
1280 		*pdata = msrs->cr0_fixed1;
1281 		break;
1282 	case MSR_IA32_VMX_CR4_FIXED0:
1283 		*pdata = msrs->cr4_fixed0;
1284 		break;
1285 	case MSR_IA32_VMX_CR4_FIXED1:
1286 		*pdata = msrs->cr4_fixed1;
1287 		break;
1288 	case MSR_IA32_VMX_VMCS_ENUM:
1289 		*pdata = msrs->vmcs_enum;
1290 		break;
1291 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1292 		*pdata = vmx_control_msr(
1293 			msrs->secondary_ctls_low,
1294 			msrs->secondary_ctls_high);
1295 		break;
1296 	case MSR_IA32_VMX_EPT_VPID_CAP:
1297 		*pdata = msrs->ept_caps |
1298 			((u64)msrs->vpid_caps << 32);
1299 		break;
1300 	case MSR_IA32_VMX_VMFUNC:
1301 		*pdata = msrs->vmfunc_controls;
1302 		break;
1303 	default:
1304 		return 1;
1305 	}
1306 
1307 	return 0;
1308 }
1309 
1310 /*
1311  * Copy the writable VMCS shadow fields back to the VMCS12, in case
1312  * they have been modified by the L1 guest. Note that the "read-only"
1313  * VM-exit information fields are actually writable if the vCPU is
1314  * configured to support "VMWRITE to any supported field in the VMCS."
1315  */
1316 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1317 {
1318 	const u16 *fields[] = {
1319 		shadow_read_write_fields,
1320 		shadow_read_only_fields
1321 	};
1322 	const int max_fields[] = {
1323 		max_shadow_read_write_fields,
1324 		max_shadow_read_only_fields
1325 	};
1326 	int i, q;
1327 	unsigned long field;
1328 	u64 field_value;
1329 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1330 
1331 	preempt_disable();
1332 
1333 	vmcs_load(shadow_vmcs);
1334 
1335 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1336 		for (i = 0; i < max_fields[q]; i++) {
1337 			field = fields[q][i];
1338 			field_value = __vmcs_readl(field);
1339 			vmcs12_write_any(get_vmcs12(&vmx->vcpu), field, field_value);
1340 		}
1341 		/*
1342 		 * Skip the VM-exit information fields if they are read-only.
1343 		 */
1344 		if (!nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
1345 			break;
1346 	}
1347 
1348 	vmcs_clear(shadow_vmcs);
1349 	vmcs_load(vmx->loaded_vmcs->vmcs);
1350 
1351 	preempt_enable();
1352 }
1353 
1354 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1355 {
1356 	const u16 *fields[] = {
1357 		shadow_read_write_fields,
1358 		shadow_read_only_fields
1359 	};
1360 	const int max_fields[] = {
1361 		max_shadow_read_write_fields,
1362 		max_shadow_read_only_fields
1363 	};
1364 	int i, q;
1365 	unsigned long field;
1366 	u64 field_value = 0;
1367 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1368 
1369 	vmcs_load(shadow_vmcs);
1370 
1371 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1372 		for (i = 0; i < max_fields[q]; i++) {
1373 			field = fields[q][i];
1374 			vmcs12_read_any(get_vmcs12(&vmx->vcpu), field, &field_value);
1375 			__vmcs_writel(field, field_value);
1376 		}
1377 	}
1378 
1379 	vmcs_clear(shadow_vmcs);
1380 	vmcs_load(vmx->loaded_vmcs->vmcs);
1381 }
1382 
1383 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1384 {
1385 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1386 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1387 
1388 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1389 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1390 	vmcs12->guest_rip = evmcs->guest_rip;
1391 
1392 	if (unlikely(!(evmcs->hv_clean_fields &
1393 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1394 		vmcs12->guest_rsp = evmcs->guest_rsp;
1395 		vmcs12->guest_rflags = evmcs->guest_rflags;
1396 		vmcs12->guest_interruptibility_info =
1397 			evmcs->guest_interruptibility_info;
1398 	}
1399 
1400 	if (unlikely(!(evmcs->hv_clean_fields &
1401 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1402 		vmcs12->cpu_based_vm_exec_control =
1403 			evmcs->cpu_based_vm_exec_control;
1404 	}
1405 
1406 	if (unlikely(!(evmcs->hv_clean_fields &
1407 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1408 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1409 	}
1410 
1411 	if (unlikely(!(evmcs->hv_clean_fields &
1412 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1413 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1414 	}
1415 
1416 	if (unlikely(!(evmcs->hv_clean_fields &
1417 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1418 		vmcs12->vm_entry_intr_info_field =
1419 			evmcs->vm_entry_intr_info_field;
1420 		vmcs12->vm_entry_exception_error_code =
1421 			evmcs->vm_entry_exception_error_code;
1422 		vmcs12->vm_entry_instruction_len =
1423 			evmcs->vm_entry_instruction_len;
1424 	}
1425 
1426 	if (unlikely(!(evmcs->hv_clean_fields &
1427 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1428 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1429 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1430 		vmcs12->host_cr0 = evmcs->host_cr0;
1431 		vmcs12->host_cr3 = evmcs->host_cr3;
1432 		vmcs12->host_cr4 = evmcs->host_cr4;
1433 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1434 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1435 		vmcs12->host_rip = evmcs->host_rip;
1436 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1437 		vmcs12->host_es_selector = evmcs->host_es_selector;
1438 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1439 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1440 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1441 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1442 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1443 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1444 	}
1445 
1446 	if (unlikely(!(evmcs->hv_clean_fields &
1447 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1448 		vmcs12->pin_based_vm_exec_control =
1449 			evmcs->pin_based_vm_exec_control;
1450 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1451 		vmcs12->secondary_vm_exec_control =
1452 			evmcs->secondary_vm_exec_control;
1453 	}
1454 
1455 	if (unlikely(!(evmcs->hv_clean_fields &
1456 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1457 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1458 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1459 	}
1460 
1461 	if (unlikely(!(evmcs->hv_clean_fields &
1462 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1463 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1464 	}
1465 
1466 	if (unlikely(!(evmcs->hv_clean_fields &
1467 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1468 		vmcs12->guest_es_base = evmcs->guest_es_base;
1469 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1470 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1471 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1472 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1473 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1474 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1475 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1476 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1477 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1478 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1479 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1480 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1481 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1482 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1483 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1484 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1485 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1486 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1487 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1488 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1489 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1490 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1491 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1492 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1493 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1494 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1495 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1496 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1497 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1498 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1499 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1500 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1501 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1502 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1503 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1504 	}
1505 
1506 	if (unlikely(!(evmcs->hv_clean_fields &
1507 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1508 		vmcs12->tsc_offset = evmcs->tsc_offset;
1509 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1510 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1511 	}
1512 
1513 	if (unlikely(!(evmcs->hv_clean_fields &
1514 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1515 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1516 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1517 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1518 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1519 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1520 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1521 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1522 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1523 	}
1524 
1525 	if (unlikely(!(evmcs->hv_clean_fields &
1526 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1527 		vmcs12->host_fs_base = evmcs->host_fs_base;
1528 		vmcs12->host_gs_base = evmcs->host_gs_base;
1529 		vmcs12->host_tr_base = evmcs->host_tr_base;
1530 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1531 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1532 		vmcs12->host_rsp = evmcs->host_rsp;
1533 	}
1534 
1535 	if (unlikely(!(evmcs->hv_clean_fields &
1536 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1537 		vmcs12->ept_pointer = evmcs->ept_pointer;
1538 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1539 	}
1540 
1541 	if (unlikely(!(evmcs->hv_clean_fields &
1542 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1543 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1544 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1545 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1546 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1547 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1548 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1549 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1550 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1551 		vmcs12->guest_pending_dbg_exceptions =
1552 			evmcs->guest_pending_dbg_exceptions;
1553 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1554 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1555 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1556 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1557 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1558 	}
1559 
1560 	/*
1561 	 * Not used?
1562 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1563 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1564 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1565 	 * vmcs12->cr3_target_value0 = evmcs->cr3_target_value0;
1566 	 * vmcs12->cr3_target_value1 = evmcs->cr3_target_value1;
1567 	 * vmcs12->cr3_target_value2 = evmcs->cr3_target_value2;
1568 	 * vmcs12->cr3_target_value3 = evmcs->cr3_target_value3;
1569 	 * vmcs12->page_fault_error_code_mask =
1570 	 *		evmcs->page_fault_error_code_mask;
1571 	 * vmcs12->page_fault_error_code_match =
1572 	 *		evmcs->page_fault_error_code_match;
1573 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1574 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1575 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1576 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1577 	 */
1578 
1579 	/*
1580 	 * Read only fields:
1581 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1582 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1583 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1584 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1585 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1586 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1587 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1588 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1589 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1590 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1591 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1592 	 *
1593 	 * Not present in struct vmcs12:
1594 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1595 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1596 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1597 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1598 	 */
1599 
1600 	return 0;
1601 }
1602 
1603 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1604 {
1605 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1606 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1607 
1608 	/*
1609 	 * Should not be changed by KVM:
1610 	 *
1611 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1612 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1613 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1614 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1615 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1616 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1617 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1618 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1619 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1620 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1621 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1622 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1623 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1624 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1625 	 * evmcs->host_rip = vmcs12->host_rip;
1626 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1627 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1628 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1629 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1630 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1631 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1632 	 * evmcs->host_rsp = vmcs12->host_rsp;
1633 	 * sync_vmcs12() doesn't read these:
1634 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1635 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1636 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1637 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1638 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1639 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1640 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1641 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1642 	 * evmcs->cr3_target_value0 = vmcs12->cr3_target_value0;
1643 	 * evmcs->cr3_target_value1 = vmcs12->cr3_target_value1;
1644 	 * evmcs->cr3_target_value2 = vmcs12->cr3_target_value2;
1645 	 * evmcs->cr3_target_value3 = vmcs12->cr3_target_value3;
1646 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1647 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1648 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1649 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1650 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1651 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1652 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1653 	 * evmcs->page_fault_error_code_mask =
1654 	 *		vmcs12->page_fault_error_code_mask;
1655 	 * evmcs->page_fault_error_code_match =
1656 	 *		vmcs12->page_fault_error_code_match;
1657 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1658 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1659 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1660 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1661 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1662 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1663 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1664 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1665 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1666 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1667 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1668 	 *
1669 	 * Not present in struct vmcs12:
1670 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1671 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1672 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1673 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1674 	 */
1675 
1676 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1677 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1678 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1679 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1680 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1681 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1682 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1683 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1684 
1685 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1686 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1687 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1688 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1689 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1690 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1691 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1692 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1693 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1694 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1695 
1696 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1697 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1698 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1699 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1700 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1701 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1702 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1703 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1704 
1705 	evmcs->guest_es_base = vmcs12->guest_es_base;
1706 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1707 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1708 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1709 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1710 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1711 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1712 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1713 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1714 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1715 
1716 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1717 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1718 
1719 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1720 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1721 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1722 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1723 
1724 	evmcs->guest_pending_dbg_exceptions =
1725 		vmcs12->guest_pending_dbg_exceptions;
1726 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1727 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1728 
1729 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1730 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1731 
1732 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1733 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1734 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1735 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1736 
1737 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1738 
1739 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1740 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1741 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1742 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1743 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1744 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1745 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1746 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1747 
1748 	evmcs->exit_qualification = vmcs12->exit_qualification;
1749 
1750 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1751 	evmcs->guest_rsp = vmcs12->guest_rsp;
1752 	evmcs->guest_rflags = vmcs12->guest_rflags;
1753 
1754 	evmcs->guest_interruptibility_info =
1755 		vmcs12->guest_interruptibility_info;
1756 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1757 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1758 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1759 	evmcs->vm_entry_exception_error_code =
1760 		vmcs12->vm_entry_exception_error_code;
1761 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1762 
1763 	evmcs->guest_rip = vmcs12->guest_rip;
1764 
1765 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1766 
1767 	return 0;
1768 }
1769 
1770 /*
1771  * This is an equivalent of the nested hypervisor executing the vmptrld
1772  * instruction.
1773  */
1774 static int nested_vmx_handle_enlightened_vmptrld(struct kvm_vcpu *vcpu,
1775 						 bool from_launch)
1776 {
1777 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1778 	struct hv_vp_assist_page assist_page;
1779 
1780 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1781 		return 1;
1782 
1783 	if (unlikely(!kvm_hv_get_assist_page(vcpu, &assist_page)))
1784 		return 1;
1785 
1786 	if (unlikely(!assist_page.enlighten_vmentry))
1787 		return 1;
1788 
1789 	if (unlikely(assist_page.current_nested_vmcs !=
1790 		     vmx->nested.hv_evmcs_vmptr)) {
1791 
1792 		if (!vmx->nested.hv_evmcs)
1793 			vmx->nested.current_vmptr = -1ull;
1794 
1795 		nested_release_evmcs(vcpu);
1796 
1797 		vmx->nested.hv_evmcs_page = kvm_vcpu_gpa_to_page(
1798 			vcpu, assist_page.current_nested_vmcs);
1799 
1800 		if (unlikely(is_error_page(vmx->nested.hv_evmcs_page)))
1801 			return 0;
1802 
1803 		vmx->nested.hv_evmcs = kmap(vmx->nested.hv_evmcs_page);
1804 
1805 		/*
1806 		 * Currently, KVM only supports eVMCS version 1
1807 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1808 		 * value to first u32 field of eVMCS which should specify eVMCS
1809 		 * VersionNumber.
1810 		 *
1811 		 * Guest should be aware of supported eVMCS versions by host by
1812 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1813 		 * expected to set this CPUID leaf according to the value
1814 		 * returned in vmcs_version from nested_enable_evmcs().
1815 		 *
1816 		 * However, it turns out that Microsoft Hyper-V fails to comply
1817 		 * to their own invented interface: When Hyper-V use eVMCS, it
1818 		 * just sets first u32 field of eVMCS to revision_id specified
1819 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1820 		 * which is one of the supported versions specified in
1821 		 * CPUID.0x4000000A.EAX[0:15].
1822 		 *
1823 		 * To overcome Hyper-V bug, we accept here either a supported
1824 		 * eVMCS version or VMCS12 revision_id as valid values for first
1825 		 * u32 field of eVMCS.
1826 		 */
1827 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1828 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1829 			nested_release_evmcs(vcpu);
1830 			return 0;
1831 		}
1832 
1833 		vmx->nested.dirty_vmcs12 = true;
1834 		/*
1835 		 * As we keep L2 state for one guest only 'hv_clean_fields' mask
1836 		 * can't be used when we switch between them. Reset it here for
1837 		 * simplicity.
1838 		 */
1839 		vmx->nested.hv_evmcs->hv_clean_fields &=
1840 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1841 		vmx->nested.hv_evmcs_vmptr = assist_page.current_nested_vmcs;
1842 
1843 		/*
1844 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
1845 		 * reloaded from guest's memory (read only fields, fields not
1846 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
1847 		 * are no leftovers.
1848 		 */
1849 		if (from_launch) {
1850 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1851 			memset(vmcs12, 0, sizeof(*vmcs12));
1852 			vmcs12->hdr.revision_id = VMCS12_REVISION;
1853 		}
1854 
1855 	}
1856 	return 1;
1857 }
1858 
1859 void nested_sync_from_vmcs12(struct kvm_vcpu *vcpu)
1860 {
1861 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1862 
1863 	/*
1864 	 * hv_evmcs may end up being not mapped after migration (when
1865 	 * L2 was running), map it here to make sure vmcs12 changes are
1866 	 * properly reflected.
1867 	 */
1868 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs)
1869 		nested_vmx_handle_enlightened_vmptrld(vcpu, false);
1870 
1871 	if (vmx->nested.hv_evmcs) {
1872 		copy_vmcs12_to_enlightened(vmx);
1873 		/* All fields are clean */
1874 		vmx->nested.hv_evmcs->hv_clean_fields |=
1875 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1876 	} else {
1877 		copy_vmcs12_to_shadow(vmx);
1878 	}
1879 
1880 	vmx->nested.need_vmcs12_sync = false;
1881 }
1882 
1883 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
1884 {
1885 	struct vcpu_vmx *vmx =
1886 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
1887 
1888 	vmx->nested.preemption_timer_expired = true;
1889 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
1890 	kvm_vcpu_kick(&vmx->vcpu);
1891 
1892 	return HRTIMER_NORESTART;
1893 }
1894 
1895 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
1896 {
1897 	u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
1898 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1899 
1900 	/*
1901 	 * A timer value of zero is architecturally guaranteed to cause
1902 	 * a VMExit prior to executing any instructions in the guest.
1903 	 */
1904 	if (preemption_timeout == 0) {
1905 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
1906 		return;
1907 	}
1908 
1909 	if (vcpu->arch.virtual_tsc_khz == 0)
1910 		return;
1911 
1912 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
1913 	preemption_timeout *= 1000000;
1914 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
1915 	hrtimer_start(&vmx->nested.preemption_timer,
1916 		      ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
1917 }
1918 
1919 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1920 {
1921 	if (vmx->nested.nested_run_pending &&
1922 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
1923 		return vmcs12->guest_ia32_efer;
1924 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
1925 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
1926 	else
1927 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
1928 }
1929 
1930 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
1931 {
1932 	/*
1933 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
1934 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
1935 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
1936 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
1937 	 */
1938 	if (vmx->nested.vmcs02_initialized)
1939 		return;
1940 	vmx->nested.vmcs02_initialized = true;
1941 
1942 	/*
1943 	 * We don't care what the EPTP value is we just need to guarantee
1944 	 * it's valid so we don't get a false positive when doing early
1945 	 * consistency checks.
1946 	 */
1947 	if (enable_ept && nested_early_check)
1948 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
1949 
1950 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
1951 	if (cpu_has_vmx_vmfunc())
1952 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
1953 
1954 	if (cpu_has_vmx_posted_intr())
1955 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
1956 
1957 	if (cpu_has_vmx_msr_bitmap())
1958 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
1959 
1960 	if (enable_pml)
1961 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
1962 
1963 	/*
1964 	 * Set the MSR load/store lists to match L0's settings.  Only the
1965 	 * addresses are constant (for vmcs02), the counts can change based
1966 	 * on L2's behavior, e.g. switching to/from long mode.
1967 	 */
1968 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1969 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
1970 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
1971 
1972 	vmx_set_constant_host_state(vmx);
1973 }
1974 
1975 static void prepare_vmcs02_early_full(struct vcpu_vmx *vmx,
1976 				      struct vmcs12 *vmcs12)
1977 {
1978 	prepare_vmcs02_constant_state(vmx);
1979 
1980 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
1981 
1982 	if (enable_vpid) {
1983 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
1984 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
1985 		else
1986 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
1987 	}
1988 }
1989 
1990 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1991 {
1992 	u32 exec_control, vmcs12_exec_ctrl;
1993 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
1994 
1995 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
1996 		prepare_vmcs02_early_full(vmx, vmcs12);
1997 
1998 	/*
1999 	 * PIN CONTROLS
2000 	 */
2001 	exec_control = vmcs12->pin_based_vm_exec_control;
2002 
2003 	/* Preemption timer setting is computed directly in vmx_vcpu_run.  */
2004 	exec_control |= vmcs_config.pin_based_exec_ctrl;
2005 	exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2006 	vmx->loaded_vmcs->hv_timer_armed = false;
2007 
2008 	/* Posted interrupts setting is only taken from vmcs12.  */
2009 	if (nested_cpu_has_posted_intr(vmcs12)) {
2010 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2011 		vmx->nested.pi_pending = false;
2012 	} else {
2013 		exec_control &= ~PIN_BASED_POSTED_INTR;
2014 	}
2015 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
2016 
2017 	/*
2018 	 * EXEC CONTROLS
2019 	 */
2020 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2021 	exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2022 	exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2023 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2024 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2025 
2026 	/*
2027 	 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
2028 	 * nested_get_vmcs12_pages can't fix it up, the illegal value
2029 	 * will result in a VM entry failure.
2030 	 */
2031 	if (exec_control & CPU_BASED_TPR_SHADOW) {
2032 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
2033 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2034 	} else {
2035 #ifdef CONFIG_X86_64
2036 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2037 				CPU_BASED_CR8_STORE_EXITING;
2038 #endif
2039 	}
2040 
2041 	/*
2042 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2043 	 * for I/O port accesses.
2044 	 */
2045 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2046 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2047 	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2048 
2049 	/*
2050 	 * SECONDARY EXEC CONTROLS
2051 	 */
2052 	if (cpu_has_secondary_exec_ctrls()) {
2053 		exec_control = vmx->secondary_exec_control;
2054 
2055 		/* Take the following fields only from vmcs12 */
2056 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2057 				  SECONDARY_EXEC_ENABLE_INVPCID |
2058 				  SECONDARY_EXEC_RDTSCP |
2059 				  SECONDARY_EXEC_XSAVES |
2060 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2061 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2062 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2063 		if (nested_cpu_has(vmcs12,
2064 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2065 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2066 				~SECONDARY_EXEC_ENABLE_PML;
2067 			exec_control |= vmcs12_exec_ctrl;
2068 		}
2069 
2070 		/* VMCS shadowing for L2 is emulated for now */
2071 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2072 
2073 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2074 			vmcs_write16(GUEST_INTR_STATUS,
2075 				vmcs12->guest_intr_status);
2076 
2077 		/*
2078 		 * Write an illegal value to APIC_ACCESS_ADDR. Later,
2079 		 * nested_get_vmcs12_pages will either fix it up or
2080 		 * remove the VM execution control.
2081 		 */
2082 		if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
2083 			vmcs_write64(APIC_ACCESS_ADDR, -1ull);
2084 
2085 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2086 			vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2087 
2088 		vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2089 	}
2090 
2091 	/*
2092 	 * ENTRY CONTROLS
2093 	 *
2094 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2095 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2096 	 * on the related bits (if supported by the CPU) in the hope that
2097 	 * we can avoid VMWrites during vmx_set_efer().
2098 	 */
2099 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2100 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2101 	if (cpu_has_load_ia32_efer()) {
2102 		if (guest_efer & EFER_LMA)
2103 			exec_control |= VM_ENTRY_IA32E_MODE;
2104 		if (guest_efer != host_efer)
2105 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2106 	}
2107 	vm_entry_controls_init(vmx, exec_control);
2108 
2109 	/*
2110 	 * EXIT CONTROLS
2111 	 *
2112 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2113 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2114 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2115 	 */
2116 	exec_control = vmx_vmexit_ctrl();
2117 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2118 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2119 	vm_exit_controls_init(vmx, exec_control);
2120 
2121 	/*
2122 	 * Conceptually we want to copy the PML address and index from
2123 	 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
2124 	 * since we always flush the log on each vmexit and never change
2125 	 * the PML address (once set), this happens to be equivalent to
2126 	 * simply resetting the index in vmcs02.
2127 	 */
2128 	if (enable_pml)
2129 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2130 
2131 	/*
2132 	 * Interrupt/Exception Fields
2133 	 */
2134 	if (vmx->nested.nested_run_pending) {
2135 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2136 			     vmcs12->vm_entry_intr_info_field);
2137 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2138 			     vmcs12->vm_entry_exception_error_code);
2139 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2140 			     vmcs12->vm_entry_instruction_len);
2141 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2142 			     vmcs12->guest_interruptibility_info);
2143 		vmx->loaded_vmcs->nmi_known_unmasked =
2144 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2145 	} else {
2146 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2147 	}
2148 }
2149 
2150 static void prepare_vmcs02_full(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2151 {
2152 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2153 
2154 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2155 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2156 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2157 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2158 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2159 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2160 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2161 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2162 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2163 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2164 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2165 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2166 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2167 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2168 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2169 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2170 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2171 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2172 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2173 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2174 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2175 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2176 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2177 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2178 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2179 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2180 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2181 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2182 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2183 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2184 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2185 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2186 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2187 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2188 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2189 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2190 	}
2191 
2192 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2193 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2194 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2195 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2196 			    vmcs12->guest_pending_dbg_exceptions);
2197 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2198 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2199 
2200 		/*
2201 		 * L1 may access the L2's PDPTR, so save them to construct
2202 		 * vmcs12
2203 		 */
2204 		if (enable_ept) {
2205 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2206 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2207 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2208 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2209 		}
2210 	}
2211 
2212 	if (nested_cpu_has_xsaves(vmcs12))
2213 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2214 
2215 	/*
2216 	 * Whether page-faults are trapped is determined by a combination of
2217 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2218 	 * If enable_ept, L0 doesn't care about page faults and we should
2219 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2220 	 * care about (at least some) page faults, and because it is not easy
2221 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2222 	 * to exit on each and every L2 page fault. This is done by setting
2223 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2224 	 * Note that below we don't need special code to set EB.PF beyond the
2225 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2226 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2227 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2228 	 */
2229 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2230 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2231 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2232 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2233 
2234 	if (cpu_has_vmx_apicv()) {
2235 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2236 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2237 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2238 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2239 	}
2240 
2241 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2242 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2243 
2244 	set_cr4_guest_host_mask(vmx);
2245 
2246 	if (kvm_mpx_supported()) {
2247 		if (vmx->nested.nested_run_pending &&
2248 			(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2249 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2250 		else
2251 			vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2252 	}
2253 }
2254 
2255 /*
2256  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2257  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2258  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2259  * guest in a way that will both be appropriate to L1's requests, and our
2260  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2261  * function also has additional necessary side-effects, like setting various
2262  * vcpu->arch fields.
2263  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2264  * is assigned to entry_failure_code on failure.
2265  */
2266 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2267 			  u32 *entry_failure_code)
2268 {
2269 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2270 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2271 
2272 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs) {
2273 		prepare_vmcs02_full(vmx, vmcs12);
2274 		vmx->nested.dirty_vmcs12 = false;
2275 	}
2276 
2277 	/*
2278 	 * First, the fields that are shadowed.  This must be kept in sync
2279 	 * with vmcs_shadow_fields.h.
2280 	 */
2281 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2282 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2283 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2284 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2285 	}
2286 
2287 	if (vmx->nested.nested_run_pending &&
2288 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2289 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2290 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2291 	} else {
2292 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2293 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2294 	}
2295 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2296 
2297 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2298 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2299 	 * trap. Note that CR0.TS also needs updating - we do this later.
2300 	 */
2301 	update_exception_bitmap(vcpu);
2302 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2303 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2304 
2305 	if (vmx->nested.nested_run_pending &&
2306 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2307 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2308 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2309 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2310 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2311 	}
2312 
2313 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2314 
2315 	if (kvm_has_tsc_control)
2316 		decache_tsc_multiplier(vmx);
2317 
2318 	if (enable_vpid) {
2319 		/*
2320 		 * There is no direct mapping between vpid02 and vpid12, the
2321 		 * vpid02 is per-vCPU for L0 and reused while the value of
2322 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2323 		 * The vpid12 is allocated by L1 for L2, so it will not
2324 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2325 		 * even if spawn a lot of nested vCPUs.
2326 		 */
2327 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2328 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2329 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2330 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2331 			}
2332 		} else {
2333 			/*
2334 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2335 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2336 			 * flush until vmcs02->eptp is fully updated by
2337 			 * KVM_REQ_LOAD_CR3. Note that this assumes
2338 			 * KVM_REQ_TLB_FLUSH is evaluated after
2339 			 * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
2340 			 */
2341 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2342 		}
2343 	}
2344 
2345 	if (nested_cpu_has_ept(vmcs12))
2346 		nested_ept_init_mmu_context(vcpu);
2347 	else if (nested_cpu_has2(vmcs12,
2348 				 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2349 		vmx_flush_tlb(vcpu, true);
2350 
2351 	/*
2352 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2353 	 * bits which we consider mandatory enabled.
2354 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2355 	 * the specifications by L1; It's not enough to take
2356 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2357 	 * have more bits than L1 expected.
2358 	 */
2359 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2360 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2361 
2362 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2363 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2364 
2365 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2366 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2367 	vmx_set_efer(vcpu, vcpu->arch.efer);
2368 
2369 	/*
2370 	 * Guest state is invalid and unrestricted guest is disabled,
2371 	 * which means L1 attempted VMEntry to L2 with invalid state.
2372 	 * Fail the VMEntry.
2373 	 */
2374 	if (vmx->emulation_required) {
2375 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2376 		return 1;
2377 	}
2378 
2379 	/* Shadow page tables on either EPT or shadow page tables. */
2380 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2381 				entry_failure_code))
2382 		return 1;
2383 
2384 	if (!enable_ept)
2385 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2386 
2387 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
2388 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
2389 	return 0;
2390 }
2391 
2392 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2393 {
2394 	if (!nested_cpu_has_nmi_exiting(vmcs12) &&
2395 	    nested_cpu_has_virtual_nmis(vmcs12))
2396 		return -EINVAL;
2397 
2398 	if (!nested_cpu_has_virtual_nmis(vmcs12) &&
2399 	    nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING))
2400 		return -EINVAL;
2401 
2402 	return 0;
2403 }
2404 
2405 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
2406 {
2407 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2408 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2409 
2410 	/* Check for memory type validity */
2411 	switch (address & VMX_EPTP_MT_MASK) {
2412 	case VMX_EPTP_MT_UC:
2413 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT))
2414 			return false;
2415 		break;
2416 	case VMX_EPTP_MT_WB:
2417 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT))
2418 			return false;
2419 		break;
2420 	default:
2421 		return false;
2422 	}
2423 
2424 	/* only 4 levels page-walk length are valid */
2425 	if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
2426 		return false;
2427 
2428 	/* Reserved bits should not be set */
2429 	if (address >> maxphyaddr || ((address >> 7) & 0x1f))
2430 		return false;
2431 
2432 	/* AD, if set, should be supported */
2433 	if (address & VMX_EPTP_AD_ENABLE_BIT) {
2434 		if (!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT))
2435 			return false;
2436 	}
2437 
2438 	return true;
2439 }
2440 
2441 /*
2442  * Checks related to VM-Execution Control Fields
2443  */
2444 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2445                                               struct vmcs12 *vmcs12)
2446 {
2447 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2448 
2449 	if (!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2450 				vmx->nested.msrs.pinbased_ctls_low,
2451 				vmx->nested.msrs.pinbased_ctls_high) ||
2452 	    !vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2453 				vmx->nested.msrs.procbased_ctls_low,
2454 				vmx->nested.msrs.procbased_ctls_high))
2455 		return -EINVAL;
2456 
2457 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2458 	    !vmx_control_verify(vmcs12->secondary_vm_exec_control,
2459 				 vmx->nested.msrs.secondary_ctls_low,
2460 				 vmx->nested.msrs.secondary_ctls_high))
2461 		return -EINVAL;
2462 
2463 	if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu) ||
2464 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2465 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2466 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2467 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2468 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2469 	    nested_vmx_check_nmi_controls(vmcs12) ||
2470 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2471 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2472 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2473 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2474 	    (nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2475 		return -EINVAL;
2476 
2477 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2478 	    nested_cpu_has_save_preemption_timer(vmcs12))
2479 		return -EINVAL;
2480 
2481 	if (nested_cpu_has_ept(vmcs12) &&
2482 	    !valid_ept_address(vcpu, vmcs12->ept_pointer))
2483 		return -EINVAL;
2484 
2485 	if (nested_cpu_has_vmfunc(vmcs12)) {
2486 		if (vmcs12->vm_function_control &
2487 		    ~vmx->nested.msrs.vmfunc_controls)
2488 			return -EINVAL;
2489 
2490 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2491 			if (!nested_cpu_has_ept(vmcs12) ||
2492 			    !page_address_valid(vcpu, vmcs12->eptp_list_address))
2493 				return -EINVAL;
2494 		}
2495 	}
2496 
2497 	return 0;
2498 }
2499 
2500 /*
2501  * Checks related to VM-Exit Control Fields
2502  */
2503 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2504                                          struct vmcs12 *vmcs12)
2505 {
2506 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2507 
2508 	if (!vmx_control_verify(vmcs12->vm_exit_controls,
2509 				vmx->nested.msrs.exit_ctls_low,
2510 				vmx->nested.msrs.exit_ctls_high) ||
2511 	    nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12))
2512 		return -EINVAL;
2513 
2514 	return 0;
2515 }
2516 
2517 /*
2518  * Checks related to VM-Entry Control Fields
2519  */
2520 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2521 					  struct vmcs12 *vmcs12)
2522 {
2523 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2524 
2525 	if (!vmx_control_verify(vmcs12->vm_entry_controls,
2526 				vmx->nested.msrs.entry_ctls_low,
2527 				vmx->nested.msrs.entry_ctls_high))
2528 		return -EINVAL;
2529 
2530 	/*
2531 	 * From the Intel SDM, volume 3:
2532 	 * Fields relevant to VM-entry event injection must be set properly.
2533 	 * These fields are the VM-entry interruption-information field, the
2534 	 * VM-entry exception error code, and the VM-entry instruction length.
2535 	 */
2536 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2537 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2538 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2539 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2540 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2541 		bool should_have_error_code;
2542 		bool urg = nested_cpu_has2(vmcs12,
2543 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2544 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2545 
2546 		/* VM-entry interruption-info field: interruption type */
2547 		if (intr_type == INTR_TYPE_RESERVED ||
2548 		    (intr_type == INTR_TYPE_OTHER_EVENT &&
2549 		     !nested_cpu_supports_monitor_trap_flag(vcpu)))
2550 			return -EINVAL;
2551 
2552 		/* VM-entry interruption-info field: vector */
2553 		if ((intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2554 		    (intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2555 		    (intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2556 			return -EINVAL;
2557 
2558 		/* VM-entry interruption-info field: deliver error code */
2559 		should_have_error_code =
2560 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2561 			x86_exception_has_error_code(vector);
2562 		if (has_error_code != should_have_error_code)
2563 			return -EINVAL;
2564 
2565 		/* VM-entry exception error code */
2566 		if (has_error_code &&
2567 		    vmcs12->vm_entry_exception_error_code & GENMASK(31, 15))
2568 			return -EINVAL;
2569 
2570 		/* VM-entry interruption-info field: reserved bits */
2571 		if (intr_info & INTR_INFO_RESVD_BITS_MASK)
2572 			return -EINVAL;
2573 
2574 		/* VM-entry instruction length */
2575 		switch (intr_type) {
2576 		case INTR_TYPE_SOFT_EXCEPTION:
2577 		case INTR_TYPE_SOFT_INTR:
2578 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2579 			if ((vmcs12->vm_entry_instruction_len > 15) ||
2580 			    (vmcs12->vm_entry_instruction_len == 0 &&
2581 			     !nested_cpu_has_zero_length_injection(vcpu)))
2582 				return -EINVAL;
2583 		}
2584 	}
2585 
2586 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2587 		return -EINVAL;
2588 
2589 	return 0;
2590 }
2591 
2592 /*
2593  * Checks related to Host Control Registers and MSRs
2594  */
2595 static int nested_check_host_control_regs(struct kvm_vcpu *vcpu,
2596                                           struct vmcs12 *vmcs12)
2597 {
2598 	bool ia32e;
2599 
2600 	if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
2601 	    !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
2602 	    !nested_cr3_valid(vcpu, vmcs12->host_cr3))
2603 		return -EINVAL;
2604 
2605 	if (is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu) ||
2606 	    is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu))
2607 		return -EINVAL;
2608 
2609 	/*
2610 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2611 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2612 	 * the values of the LMA and LME bits in the field must each be that of
2613 	 * the host address-space size VM-exit control.
2614 	 */
2615 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2616 		ia32e = (vmcs12->vm_exit_controls &
2617 			 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
2618 		if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
2619 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
2620 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
2621 			return -EINVAL;
2622 	}
2623 
2624 	return 0;
2625 }
2626 
2627 /*
2628  * Checks related to Guest Non-register State
2629  */
2630 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2631 {
2632 	if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2633 	    vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
2634 		return -EINVAL;
2635 
2636 	return 0;
2637 }
2638 
2639 static int nested_vmx_check_vmentry_prereqs(struct kvm_vcpu *vcpu,
2640 					    struct vmcs12 *vmcs12)
2641 {
2642 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2643 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2644 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2645 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2646 
2647 	if (nested_check_host_control_regs(vcpu, vmcs12))
2648 		return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
2649 
2650 	if (nested_check_guest_non_reg_state(vmcs12))
2651 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2652 
2653 	return 0;
2654 }
2655 
2656 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2657 					  struct vmcs12 *vmcs12)
2658 {
2659 	int r;
2660 	struct page *page;
2661 	struct vmcs12 *shadow;
2662 
2663 	if (vmcs12->vmcs_link_pointer == -1ull)
2664 		return 0;
2665 
2666 	if (!page_address_valid(vcpu, vmcs12->vmcs_link_pointer))
2667 		return -EINVAL;
2668 
2669 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->vmcs_link_pointer);
2670 	if (is_error_page(page))
2671 		return -EINVAL;
2672 
2673 	r = 0;
2674 	shadow = kmap(page);
2675 	if (shadow->hdr.revision_id != VMCS12_REVISION ||
2676 	    shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12))
2677 		r = -EINVAL;
2678 	kunmap(page);
2679 	kvm_release_page_clean(page);
2680 	return r;
2681 }
2682 
2683 static int nested_vmx_check_vmentry_postreqs(struct kvm_vcpu *vcpu,
2684 					     struct vmcs12 *vmcs12,
2685 					     u32 *exit_qual)
2686 {
2687 	bool ia32e;
2688 
2689 	*exit_qual = ENTRY_FAIL_DEFAULT;
2690 
2691 	if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
2692 	    !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
2693 		return 1;
2694 
2695 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2696 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2697 		return 1;
2698 	}
2699 
2700 	/*
2701 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2702 	 * are performed on the field for the IA32_EFER MSR:
2703 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2704 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2705 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2706 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2707 	 *   CR0.PG) is 1.
2708 	 */
2709 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2710 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2711 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2712 		if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
2713 		    ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
2714 		    ((vmcs12->guest_cr0 & X86_CR0_PG) &&
2715 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
2716 			return 1;
2717 	}
2718 
2719 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2720 		(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
2721 		(vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
2722 			return 1;
2723 
2724 	return 0;
2725 }
2726 
2727 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2728 {
2729 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2730 	unsigned long cr3, cr4;
2731 	bool vm_fail;
2732 
2733 	if (!nested_early_check)
2734 		return 0;
2735 
2736 	if (vmx->msr_autoload.host.nr)
2737 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2738 	if (vmx->msr_autoload.guest.nr)
2739 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2740 
2741 	preempt_disable();
2742 
2743 	vmx_prepare_switch_to_guest(vcpu);
2744 
2745 	/*
2746 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2747 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2748 	 * be written (by preparve_vmcs02()) before the "real" VMEnter, i.e.
2749 	 * there is no need to preserve other bits or save/restore the field.
2750 	 */
2751 	vmcs_writel(GUEST_RFLAGS, 0);
2752 
2753 	cr3 = __get_current_cr3_fast();
2754 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2755 		vmcs_writel(HOST_CR3, cr3);
2756 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2757 	}
2758 
2759 	cr4 = cr4_read_shadow();
2760 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2761 		vmcs_writel(HOST_CR4, cr4);
2762 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2763 	}
2764 
2765 	asm(
2766 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2767 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2768 		"je 1f \n\t"
2769 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
2770 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2771 		"1: \n\t"
2772 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2773 
2774 		/* Check if vmlaunch or vmresume is needed */
2775 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
2776 
2777 		/*
2778 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
2779 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
2780 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
2781 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
2782 		 */
2783 		"call vmx_vmenter\n\t"
2784 
2785 		CC_SET(be)
2786 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
2787 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
2788 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
2789 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
2790 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
2791 		[wordsize]"i"(sizeof(ulong))
2792 	      : "cc", "memory"
2793 	);
2794 
2795 	preempt_enable();
2796 
2797 	if (vmx->msr_autoload.host.nr)
2798 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2799 	if (vmx->msr_autoload.guest.nr)
2800 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2801 
2802 	if (vm_fail) {
2803 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
2804 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
2805 		return 1;
2806 	}
2807 
2808 	/*
2809 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
2810 	 */
2811 	local_irq_enable();
2812 	if (hw_breakpoint_active())
2813 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
2814 
2815 	/*
2816 	 * A non-failing VMEntry means we somehow entered guest mode with
2817 	 * an illegal RIP, and that's just the tip of the iceberg.  There
2818 	 * is no telling what memory has been modified or what state has
2819 	 * been exposed to unknown code.  Hitting this all but guarantees
2820 	 * a (very critical) hardware issue.
2821 	 */
2822 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
2823 		VMX_EXIT_REASONS_FAILED_VMENTRY));
2824 
2825 	return 0;
2826 }
2827 
2828 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
2829 						 struct vmcs12 *vmcs12);
2830 
2831 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
2832 {
2833 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2834 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2835 	struct page *page;
2836 	u64 hpa;
2837 
2838 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2839 		/*
2840 		 * Translate L1 physical address to host physical
2841 		 * address for vmcs02. Keep the page pinned, so this
2842 		 * physical address remains valid. We keep a reference
2843 		 * to it so we can release it later.
2844 		 */
2845 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
2846 			kvm_release_page_dirty(vmx->nested.apic_access_page);
2847 			vmx->nested.apic_access_page = NULL;
2848 		}
2849 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
2850 		/*
2851 		 * If translation failed, no matter: This feature asks
2852 		 * to exit when accessing the given address, and if it
2853 		 * can never be accessed, this feature won't do
2854 		 * anything anyway.
2855 		 */
2856 		if (!is_error_page(page)) {
2857 			vmx->nested.apic_access_page = page;
2858 			hpa = page_to_phys(vmx->nested.apic_access_page);
2859 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
2860 		} else {
2861 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
2862 					SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
2863 		}
2864 	}
2865 
2866 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
2867 		if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
2868 			kvm_release_page_dirty(vmx->nested.virtual_apic_page);
2869 			vmx->nested.virtual_apic_page = NULL;
2870 		}
2871 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
2872 
2873 		/*
2874 		 * If translation failed, VM entry will fail because
2875 		 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
2876 		 * Failing the vm entry is _not_ what the processor
2877 		 * does but it's basically the only possibility we
2878 		 * have.  We could still enter the guest if CR8 load
2879 		 * exits are enabled, CR8 store exits are enabled, and
2880 		 * virtualize APIC access is disabled; in this case
2881 		 * the processor would never use the TPR shadow and we
2882 		 * could simply clear the bit from the execution
2883 		 * control.  But such a configuration is useless, so
2884 		 * let's keep the code simple.
2885 		 */
2886 		if (!is_error_page(page)) {
2887 			vmx->nested.virtual_apic_page = page;
2888 			hpa = page_to_phys(vmx->nested.virtual_apic_page);
2889 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
2890 		}
2891 	}
2892 
2893 	if (nested_cpu_has_posted_intr(vmcs12)) {
2894 		if (vmx->nested.pi_desc_page) { /* shouldn't happen */
2895 			kunmap(vmx->nested.pi_desc_page);
2896 			kvm_release_page_dirty(vmx->nested.pi_desc_page);
2897 			vmx->nested.pi_desc_page = NULL;
2898 			vmx->nested.pi_desc = NULL;
2899 			vmcs_write64(POSTED_INTR_DESC_ADDR, -1ull);
2900 		}
2901 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
2902 		if (is_error_page(page))
2903 			return;
2904 		vmx->nested.pi_desc_page = page;
2905 		vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
2906 		vmx->nested.pi_desc =
2907 			(struct pi_desc *)((void *)vmx->nested.pi_desc +
2908 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2909 			(PAGE_SIZE - 1)));
2910 		vmcs_write64(POSTED_INTR_DESC_ADDR,
2911 			page_to_phys(vmx->nested.pi_desc_page) +
2912 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2913 			(PAGE_SIZE - 1)));
2914 	}
2915 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
2916 		vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
2917 			      CPU_BASED_USE_MSR_BITMAPS);
2918 	else
2919 		vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
2920 				CPU_BASED_USE_MSR_BITMAPS);
2921 }
2922 
2923 /*
2924  * Intel's VMX Instruction Reference specifies a common set of prerequisites
2925  * for running VMX instructions (except VMXON, whose prerequisites are
2926  * slightly different). It also specifies what exception to inject otherwise.
2927  * Note that many of these exceptions have priority over VM exits, so they
2928  * don't have to be checked again here.
2929  */
2930 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
2931 {
2932 	if (!to_vmx(vcpu)->nested.vmxon) {
2933 		kvm_queue_exception(vcpu, UD_VECTOR);
2934 		return 0;
2935 	}
2936 
2937 	if (vmx_get_cpl(vcpu)) {
2938 		kvm_inject_gp(vcpu, 0);
2939 		return 0;
2940 	}
2941 
2942 	return 1;
2943 }
2944 
2945 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
2946 {
2947 	u8 rvi = vmx_get_rvi();
2948 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
2949 
2950 	return ((rvi & 0xf0) > (vppr & 0xf0));
2951 }
2952 
2953 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
2954 				   struct vmcs12 *vmcs12);
2955 
2956 /*
2957  * If from_vmentry is false, this is being called from state restore (either RSM
2958  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
2959 + *
2960 + * Returns:
2961 + *   0 - success, i.e. proceed with actual VMEnter
2962 + *   1 - consistency check VMExit
2963 + *  -1 - consistency check VMFail
2964  */
2965 int nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
2966 {
2967 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2968 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2969 	bool evaluate_pending_interrupts;
2970 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
2971 	u32 exit_qual;
2972 
2973 	evaluate_pending_interrupts = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2974 		(CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_VIRTUAL_NMI_PENDING);
2975 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
2976 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
2977 
2978 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
2979 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
2980 	if (kvm_mpx_supported() &&
2981 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2982 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
2983 
2984 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
2985 
2986 	prepare_vmcs02_early(vmx, vmcs12);
2987 
2988 	if (from_vmentry) {
2989 		nested_get_vmcs12_pages(vcpu);
2990 
2991 		if (nested_vmx_check_vmentry_hw(vcpu)) {
2992 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
2993 			return -1;
2994 		}
2995 
2996 		if (nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
2997 			goto vmentry_fail_vmexit;
2998 	}
2999 
3000 	enter_guest_mode(vcpu);
3001 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3002 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
3003 
3004 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
3005 		goto vmentry_fail_vmexit_guest_mode;
3006 
3007 	if (from_vmentry) {
3008 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3009 		exit_qual = nested_vmx_load_msr(vcpu,
3010 						vmcs12->vm_entry_msr_load_addr,
3011 						vmcs12->vm_entry_msr_load_count);
3012 		if (exit_qual)
3013 			goto vmentry_fail_vmexit_guest_mode;
3014 	} else {
3015 		/*
3016 		 * The MMU is not initialized to point at the right entities yet and
3017 		 * "get pages" would need to read data from the guest (i.e. we will
3018 		 * need to perform gpa to hpa translation). Request a call
3019 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3020 		 * have already been set at vmentry time and should not be reset.
3021 		 */
3022 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3023 	}
3024 
3025 	/*
3026 	 * If L1 had a pending IRQ/NMI until it executed
3027 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3028 	 * disallowed (e.g. interrupts disabled), L0 needs to
3029 	 * evaluate if this pending event should cause an exit from L2
3030 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3031 	 * intercept EXTERNAL_INTERRUPT).
3032 	 *
3033 	 * Usually this would be handled by the processor noticing an
3034 	 * IRQ/NMI window request, or checking RVI during evaluation of
3035 	 * pending virtual interrupts.  However, this setting was done
3036 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3037 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3038 	 */
3039 	if (unlikely(evaluate_pending_interrupts))
3040 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3041 
3042 	/*
3043 	 * Do not start the preemption timer hrtimer until after we know
3044 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3045 	 * the timer.
3046 	 */
3047 	vmx->nested.preemption_timer_expired = false;
3048 	if (nested_cpu_has_preemption_timer(vmcs12))
3049 		vmx_start_preemption_timer(vcpu);
3050 
3051 	/*
3052 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3053 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3054 	 * returned as far as L1 is concerned. It will only return (and set
3055 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3056 	 */
3057 	return 0;
3058 
3059 	/*
3060 	 * A failed consistency check that leads to a VMExit during L1's
3061 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3062 	 * 26.7 "VM-entry failures during or after loading guest state".
3063 	 */
3064 vmentry_fail_vmexit_guest_mode:
3065 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3066 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3067 	leave_guest_mode(vcpu);
3068 
3069 vmentry_fail_vmexit:
3070 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3071 
3072 	if (!from_vmentry)
3073 		return 1;
3074 
3075 	load_vmcs12_host_state(vcpu, vmcs12);
3076 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3077 	vmcs12->exit_qualification = exit_qual;
3078 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3079 		vmx->nested.need_vmcs12_sync = true;
3080 	return 1;
3081 }
3082 
3083 /*
3084  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3085  * for running an L2 nested guest.
3086  */
3087 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3088 {
3089 	struct vmcs12 *vmcs12;
3090 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3091 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3092 	int ret;
3093 
3094 	if (!nested_vmx_check_permission(vcpu))
3095 		return 1;
3096 
3097 	if (!nested_vmx_handle_enlightened_vmptrld(vcpu, true))
3098 		return 1;
3099 
3100 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3101 		return nested_vmx_failInvalid(vcpu);
3102 
3103 	vmcs12 = get_vmcs12(vcpu);
3104 
3105 	/*
3106 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3107 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3108 	 * rather than RFLAGS.ZF, and no error number is stored to the
3109 	 * VM-instruction error field.
3110 	 */
3111 	if (vmcs12->hdr.shadow_vmcs)
3112 		return nested_vmx_failInvalid(vcpu);
3113 
3114 	if (vmx->nested.hv_evmcs) {
3115 		copy_enlightened_to_vmcs12(vmx);
3116 		/* Enlightened VMCS doesn't have launch state */
3117 		vmcs12->launch_state = !launch;
3118 	} else if (enable_shadow_vmcs) {
3119 		copy_shadow_to_vmcs12(vmx);
3120 	}
3121 
3122 	/*
3123 	 * The nested entry process starts with enforcing various prerequisites
3124 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3125 	 * they fail: As the SDM explains, some conditions should cause the
3126 	 * instruction to fail, while others will cause the instruction to seem
3127 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3128 	 * To speed up the normal (success) code path, we should avoid checking
3129 	 * for misconfigurations which will anyway be caught by the processor
3130 	 * when using the merged vmcs02.
3131 	 */
3132 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3133 		return nested_vmx_failValid(vcpu,
3134 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3135 
3136 	if (vmcs12->launch_state == launch)
3137 		return nested_vmx_failValid(vcpu,
3138 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3139 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3140 
3141 	ret = nested_vmx_check_vmentry_prereqs(vcpu, vmcs12);
3142 	if (ret)
3143 		return nested_vmx_failValid(vcpu, ret);
3144 
3145 	/*
3146 	 * We're finally done with prerequisite checking, and can start with
3147 	 * the nested entry.
3148 	 */
3149 	vmx->nested.nested_run_pending = 1;
3150 	ret = nested_vmx_enter_non_root_mode(vcpu, true);
3151 	vmx->nested.nested_run_pending = !ret;
3152 	if (ret > 0)
3153 		return 1;
3154 	else if (ret)
3155 		return nested_vmx_failValid(vcpu,
3156 			VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3157 
3158 	/* Hide L1D cache contents from the nested guest.  */
3159 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3160 
3161 	/*
3162 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3163 	 * also be used as part of restoring nVMX state for
3164 	 * snapshot restore (migration).
3165 	 *
3166 	 * In this flow, it is assumed that vmcs12 cache was
3167 	 * trasferred as part of captured nVMX state and should
3168 	 * therefore not be read from guest memory (which may not
3169 	 * exist on destination host yet).
3170 	 */
3171 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3172 
3173 	/*
3174 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3175 	 * awakened by event injection or by an NMI-window VM-exit or
3176 	 * by an interrupt-window VM-exit, halt the vcpu.
3177 	 */
3178 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3179 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3180 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_NMI_PENDING) &&
3181 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_INTR_PENDING) &&
3182 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3183 		vmx->nested.nested_run_pending = 0;
3184 		return kvm_vcpu_halt(vcpu);
3185 	}
3186 	return 1;
3187 }
3188 
3189 /*
3190  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3191  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
3192  * This function returns the new value we should put in vmcs12.guest_cr0.
3193  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3194  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3195  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3196  *     didn't trap the bit, because if L1 did, so would L0).
3197  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3198  *     been modified by L2, and L1 knows it. So just leave the old value of
3199  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3200  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3201  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3202  *     changed these bits, and therefore they need to be updated, but L0
3203  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3204  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3205  */
3206 static inline unsigned long
3207 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3208 {
3209 	return
3210 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3211 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3212 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3213 			vcpu->arch.cr0_guest_owned_bits));
3214 }
3215 
3216 static inline unsigned long
3217 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3218 {
3219 	return
3220 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3221 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3222 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3223 			vcpu->arch.cr4_guest_owned_bits));
3224 }
3225 
3226 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3227 				      struct vmcs12 *vmcs12)
3228 {
3229 	u32 idt_vectoring;
3230 	unsigned int nr;
3231 
3232 	if (vcpu->arch.exception.injected) {
3233 		nr = vcpu->arch.exception.nr;
3234 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3235 
3236 		if (kvm_exception_is_soft(nr)) {
3237 			vmcs12->vm_exit_instruction_len =
3238 				vcpu->arch.event_exit_inst_len;
3239 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3240 		} else
3241 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3242 
3243 		if (vcpu->arch.exception.has_error_code) {
3244 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3245 			vmcs12->idt_vectoring_error_code =
3246 				vcpu->arch.exception.error_code;
3247 		}
3248 
3249 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3250 	} else if (vcpu->arch.nmi_injected) {
3251 		vmcs12->idt_vectoring_info_field =
3252 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3253 	} else if (vcpu->arch.interrupt.injected) {
3254 		nr = vcpu->arch.interrupt.nr;
3255 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3256 
3257 		if (vcpu->arch.interrupt.soft) {
3258 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3259 			vmcs12->vm_entry_instruction_len =
3260 				vcpu->arch.event_exit_inst_len;
3261 		} else
3262 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3263 
3264 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3265 	}
3266 }
3267 
3268 
3269 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3270 {
3271 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3272 	gfn_t gfn;
3273 
3274 	/*
3275 	 * Don't need to mark the APIC access page dirty; it is never
3276 	 * written to by the CPU during APIC virtualization.
3277 	 */
3278 
3279 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3280 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3281 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3282 	}
3283 
3284 	if (nested_cpu_has_posted_intr(vmcs12)) {
3285 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3286 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3287 	}
3288 }
3289 
3290 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3291 {
3292 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3293 	int max_irr;
3294 	void *vapic_page;
3295 	u16 status;
3296 
3297 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3298 		return;
3299 
3300 	vmx->nested.pi_pending = false;
3301 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3302 		return;
3303 
3304 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3305 	if (max_irr != 256) {
3306 		vapic_page = kmap(vmx->nested.virtual_apic_page);
3307 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3308 			vapic_page, &max_irr);
3309 		kunmap(vmx->nested.virtual_apic_page);
3310 
3311 		status = vmcs_read16(GUEST_INTR_STATUS);
3312 		if ((u8)max_irr > ((u8)status & 0xff)) {
3313 			status &= ~0xff;
3314 			status |= (u8)max_irr;
3315 			vmcs_write16(GUEST_INTR_STATUS, status);
3316 		}
3317 	}
3318 
3319 	nested_mark_vmcs12_pages_dirty(vcpu);
3320 }
3321 
3322 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3323 					       unsigned long exit_qual)
3324 {
3325 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3326 	unsigned int nr = vcpu->arch.exception.nr;
3327 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3328 
3329 	if (vcpu->arch.exception.has_error_code) {
3330 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3331 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3332 	}
3333 
3334 	if (kvm_exception_is_soft(nr))
3335 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3336 	else
3337 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3338 
3339 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3340 	    vmx_get_nmi_mask(vcpu))
3341 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3342 
3343 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3344 }
3345 
3346 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
3347 {
3348 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3349 	unsigned long exit_qual;
3350 	bool block_nested_events =
3351 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3352 
3353 	if (vcpu->arch.exception.pending &&
3354 		nested_vmx_check_exception(vcpu, &exit_qual)) {
3355 		if (block_nested_events)
3356 			return -EBUSY;
3357 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3358 		return 0;
3359 	}
3360 
3361 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3362 	    vmx->nested.preemption_timer_expired) {
3363 		if (block_nested_events)
3364 			return -EBUSY;
3365 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3366 		return 0;
3367 	}
3368 
3369 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3370 		if (block_nested_events)
3371 			return -EBUSY;
3372 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3373 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3374 				  INTR_INFO_VALID_MASK, 0);
3375 		/*
3376 		 * The NMI-triggered VM exit counts as injection:
3377 		 * clear this one and block further NMIs.
3378 		 */
3379 		vcpu->arch.nmi_pending = 0;
3380 		vmx_set_nmi_mask(vcpu, true);
3381 		return 0;
3382 	}
3383 
3384 	if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
3385 	    nested_exit_on_intr(vcpu)) {
3386 		if (block_nested_events)
3387 			return -EBUSY;
3388 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3389 		return 0;
3390 	}
3391 
3392 	vmx_complete_nested_posted_interrupt(vcpu);
3393 	return 0;
3394 }
3395 
3396 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3397 {
3398 	ktime_t remaining =
3399 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3400 	u64 value;
3401 
3402 	if (ktime_to_ns(remaining) <= 0)
3403 		return 0;
3404 
3405 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3406 	do_div(value, 1000000);
3407 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3408 }
3409 
3410 /*
3411  * Update the guest state fields of vmcs12 to reflect changes that
3412  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3413  * VM-entry controls is also updated, since this is really a guest
3414  * state bit.)
3415  */
3416 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3417 {
3418 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3419 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3420 
3421 	vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3422 	vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
3423 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3424 
3425 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3426 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3427 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3428 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3429 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3430 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3431 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3432 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3433 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3434 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3435 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3436 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3437 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3438 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3439 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3440 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3441 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3442 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3443 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3444 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3445 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3446 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3447 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3448 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3449 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3450 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3451 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3452 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3453 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3454 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3455 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3456 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3457 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3458 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3459 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3460 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3461 
3462 	vmcs12->guest_interruptibility_info =
3463 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3464 	vmcs12->guest_pending_dbg_exceptions =
3465 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3466 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3467 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3468 	else
3469 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3470 
3471 	if (nested_cpu_has_preemption_timer(vmcs12) &&
3472 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3473 			vmcs12->vmx_preemption_timer_value =
3474 				vmx_get_preemption_timer_value(vcpu);
3475 
3476 	/*
3477 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3478 	 * own CR3 without exiting. If it has changed it, we must keep it.
3479 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3480 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3481 	 *
3482 	 * Additionally, restore L2's PDPTR to vmcs12.
3483 	 */
3484 	if (enable_ept) {
3485 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3486 		vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3487 		vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3488 		vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3489 		vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3490 	}
3491 
3492 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3493 
3494 	if (nested_cpu_has_vid(vmcs12))
3495 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3496 
3497 	vmcs12->vm_entry_controls =
3498 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3499 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3500 
3501 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
3502 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3503 		vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3504 	}
3505 
3506 	/* TODO: These cannot have changed unless we have MSR bitmaps and
3507 	 * the relevant bit asks not to trap the change */
3508 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
3509 		vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
3510 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3511 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3512 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3513 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3514 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3515 	if (kvm_mpx_supported())
3516 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3517 }
3518 
3519 /*
3520  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3521  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3522  * and this function updates it to reflect the changes to the guest state while
3523  * L2 was running (and perhaps made some exits which were handled directly by L0
3524  * without going back to L1), and to reflect the exit reason.
3525  * Note that we do not have to copy here all VMCS fields, just those that
3526  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3527  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3528  * which already writes to vmcs12 directly.
3529  */
3530 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3531 			   u32 exit_reason, u32 exit_intr_info,
3532 			   unsigned long exit_qualification)
3533 {
3534 	/* update guest state fields: */
3535 	sync_vmcs12(vcpu, vmcs12);
3536 
3537 	/* update exit information fields: */
3538 
3539 	vmcs12->vm_exit_reason = exit_reason;
3540 	vmcs12->exit_qualification = exit_qualification;
3541 	vmcs12->vm_exit_intr_info = exit_intr_info;
3542 
3543 	vmcs12->idt_vectoring_info_field = 0;
3544 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3545 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3546 
3547 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3548 		vmcs12->launch_state = 1;
3549 
3550 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3551 		 * instead of reading the real value. */
3552 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3553 
3554 		/*
3555 		 * Transfer the event that L0 or L1 may wanted to inject into
3556 		 * L2 to IDT_VECTORING_INFO_FIELD.
3557 		 */
3558 		vmcs12_save_pending_event(vcpu, vmcs12);
3559 
3560 		/*
3561 		 * According to spec, there's no need to store the guest's
3562 		 * MSRs if the exit is due to a VM-entry failure that occurs
3563 		 * during or after loading the guest state. Since this exit
3564 		 * does not fall in that category, we need to save the MSRs.
3565 		 */
3566 		if (nested_vmx_store_msr(vcpu,
3567 					 vmcs12->vm_exit_msr_store_addr,
3568 					 vmcs12->vm_exit_msr_store_count))
3569 			nested_vmx_abort(vcpu,
3570 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3571 	}
3572 
3573 	/*
3574 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3575 	 * preserved above and would only end up incorrectly in L1.
3576 	 */
3577 	vcpu->arch.nmi_injected = false;
3578 	kvm_clear_exception_queue(vcpu);
3579 	kvm_clear_interrupt_queue(vcpu);
3580 }
3581 
3582 /*
3583  * A part of what we need to when the nested L2 guest exits and we want to
3584  * run its L1 parent, is to reset L1's guest state to the host state specified
3585  * in vmcs12.
3586  * This function is to be called not only on normal nested exit, but also on
3587  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3588  * Failures During or After Loading Guest State").
3589  * This function should be called when the active VMCS is L1's (vmcs01).
3590  */
3591 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3592 				   struct vmcs12 *vmcs12)
3593 {
3594 	struct kvm_segment seg;
3595 	u32 entry_failure_code;
3596 
3597 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
3598 		vcpu->arch.efer = vmcs12->host_ia32_efer;
3599 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3600 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
3601 	else
3602 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
3603 	vmx_set_efer(vcpu, vcpu->arch.efer);
3604 
3605 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
3606 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
3607 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
3608 	vmx_set_interrupt_shadow(vcpu, 0);
3609 
3610 	/*
3611 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
3612 	 * actually changed, because vmx_set_cr0 refers to efer set above.
3613 	 *
3614 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
3615 	 * (KVM doesn't change it);
3616 	 */
3617 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3618 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
3619 
3620 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
3621 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3622 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
3623 
3624 	nested_ept_uninit_mmu_context(vcpu);
3625 
3626 	/*
3627 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
3628 	 * couldn't have changed.
3629 	 */
3630 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
3631 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
3632 
3633 	if (!enable_ept)
3634 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
3635 
3636 	/*
3637 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
3638 	 * VMEntry/VMExit. Thus, no need to flush TLB.
3639 	 *
3640 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
3641 	 * flushed on every VMEntry/VMExit.
3642 	 *
3643 	 * Otherwise, we can preserve TLB entries as long as we are
3644 	 * able to tag L1 TLB entries differently than L2 TLB entries.
3645 	 *
3646 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
3647 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
3648 	 * has been set by KVM_REQ_LOAD_CR3.
3649 	 */
3650 	if (enable_vpid &&
3651 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
3652 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3653 	}
3654 
3655 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
3656 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
3657 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
3658 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
3659 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
3660 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
3661 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
3662 
3663 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
3664 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
3665 		vmcs_write64(GUEST_BNDCFGS, 0);
3666 
3667 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
3668 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
3669 		vcpu->arch.pat = vmcs12->host_ia32_pat;
3670 	}
3671 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
3672 		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
3673 			vmcs12->host_ia32_perf_global_ctrl);
3674 
3675 	/* Set L1 segment info according to Intel SDM
3676 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
3677 	seg = (struct kvm_segment) {
3678 		.base = 0,
3679 		.limit = 0xFFFFFFFF,
3680 		.selector = vmcs12->host_cs_selector,
3681 		.type = 11,
3682 		.present = 1,
3683 		.s = 1,
3684 		.g = 1
3685 	};
3686 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3687 		seg.l = 1;
3688 	else
3689 		seg.db = 1;
3690 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
3691 	seg = (struct kvm_segment) {
3692 		.base = 0,
3693 		.limit = 0xFFFFFFFF,
3694 		.type = 3,
3695 		.present = 1,
3696 		.s = 1,
3697 		.db = 1,
3698 		.g = 1
3699 	};
3700 	seg.selector = vmcs12->host_ds_selector;
3701 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
3702 	seg.selector = vmcs12->host_es_selector;
3703 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
3704 	seg.selector = vmcs12->host_ss_selector;
3705 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
3706 	seg.selector = vmcs12->host_fs_selector;
3707 	seg.base = vmcs12->host_fs_base;
3708 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
3709 	seg.selector = vmcs12->host_gs_selector;
3710 	seg.base = vmcs12->host_gs_base;
3711 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
3712 	seg = (struct kvm_segment) {
3713 		.base = vmcs12->host_tr_base,
3714 		.limit = 0x67,
3715 		.selector = vmcs12->host_tr_selector,
3716 		.type = 11,
3717 		.present = 1
3718 	};
3719 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
3720 
3721 	kvm_set_dr(vcpu, 7, 0x400);
3722 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3723 
3724 	if (cpu_has_vmx_msr_bitmap())
3725 		vmx_update_msr_bitmap(vcpu);
3726 
3727 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
3728 				vmcs12->vm_exit_msr_load_count))
3729 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3730 }
3731 
3732 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
3733 {
3734 	struct shared_msr_entry *efer_msr;
3735 	unsigned int i;
3736 
3737 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
3738 		return vmcs_read64(GUEST_IA32_EFER);
3739 
3740 	if (cpu_has_load_ia32_efer())
3741 		return host_efer;
3742 
3743 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
3744 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
3745 			return vmx->msr_autoload.guest.val[i].value;
3746 	}
3747 
3748 	efer_msr = find_msr_entry(vmx, MSR_EFER);
3749 	if (efer_msr)
3750 		return efer_msr->data;
3751 
3752 	return host_efer;
3753 }
3754 
3755 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
3756 {
3757 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3758 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3759 	struct vmx_msr_entry g, h;
3760 	struct msr_data msr;
3761 	gpa_t gpa;
3762 	u32 i, j;
3763 
3764 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
3765 
3766 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
3767 		/*
3768 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
3769 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
3770 		 * and vcpu->arch.dr7 is not squirreled away before the
3771 		 * nested VMENTER (not worth adding a variable in nested_vmx).
3772 		 */
3773 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
3774 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
3775 		else
3776 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
3777 	}
3778 
3779 	/*
3780 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
3781 	 * handle a variety of side effects to KVM's software model.
3782 	 */
3783 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
3784 
3785 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3786 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
3787 
3788 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3789 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
3790 
3791 	nested_ept_uninit_mmu_context(vcpu);
3792 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3793 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3794 
3795 	/*
3796 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
3797 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
3798 	 * VMFail, like everything else we just need to ensure our
3799 	 * software model is up-to-date.
3800 	 */
3801 	ept_save_pdptrs(vcpu);
3802 
3803 	kvm_mmu_reset_context(vcpu);
3804 
3805 	if (cpu_has_vmx_msr_bitmap())
3806 		vmx_update_msr_bitmap(vcpu);
3807 
3808 	/*
3809 	 * This nasty bit of open coding is a compromise between blindly
3810 	 * loading L1's MSRs using the exit load lists (incorrect emulation
3811 	 * of VMFail), leaving the nested VM's MSRs in the software model
3812 	 * (incorrect behavior) and snapshotting the modified MSRs (too
3813 	 * expensive since the lists are unbound by hardware).  For each
3814 	 * MSR that was (prematurely) loaded from the nested VMEntry load
3815 	 * list, reload it from the exit load list if it exists and differs
3816 	 * from the guest value.  The intent is to stuff host state as
3817 	 * silently as possible, not to fully process the exit load list.
3818 	 */
3819 	msr.host_initiated = false;
3820 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
3821 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
3822 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
3823 			pr_debug_ratelimited(
3824 				"%s read MSR index failed (%u, 0x%08llx)\n",
3825 				__func__, i, gpa);
3826 			goto vmabort;
3827 		}
3828 
3829 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
3830 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
3831 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
3832 				pr_debug_ratelimited(
3833 					"%s read MSR failed (%u, 0x%08llx)\n",
3834 					__func__, j, gpa);
3835 				goto vmabort;
3836 			}
3837 			if (h.index != g.index)
3838 				continue;
3839 			if (h.value == g.value)
3840 				break;
3841 
3842 			if (nested_vmx_load_msr_check(vcpu, &h)) {
3843 				pr_debug_ratelimited(
3844 					"%s check failed (%u, 0x%x, 0x%x)\n",
3845 					__func__, j, h.index, h.reserved);
3846 				goto vmabort;
3847 			}
3848 
3849 			msr.index = h.index;
3850 			msr.data = h.value;
3851 			if (kvm_set_msr(vcpu, &msr)) {
3852 				pr_debug_ratelimited(
3853 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
3854 					__func__, j, h.index, h.value);
3855 				goto vmabort;
3856 			}
3857 		}
3858 	}
3859 
3860 	return;
3861 
3862 vmabort:
3863 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3864 }
3865 
3866 /*
3867  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
3868  * and modify vmcs12 to make it see what it would expect to see there if
3869  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
3870  */
3871 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
3872 		       u32 exit_intr_info, unsigned long exit_qualification)
3873 {
3874 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3875 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3876 
3877 	/* trying to cancel vmlaunch/vmresume is a bug */
3878 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
3879 
3880 	leave_guest_mode(vcpu);
3881 
3882 	if (nested_cpu_has_preemption_timer(vmcs12))
3883 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
3884 
3885 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3886 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3887 
3888 	if (likely(!vmx->fail)) {
3889 		if (exit_reason == -1)
3890 			sync_vmcs12(vcpu, vmcs12);
3891 		else
3892 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
3893 				       exit_qualification);
3894 
3895 		/*
3896 		 * Must happen outside of sync_vmcs12() as it will
3897 		 * also be used to capture vmcs12 cache as part of
3898 		 * capturing nVMX state for snapshot (migration).
3899 		 *
3900 		 * Otherwise, this flush will dirty guest memory at a
3901 		 * point it is already assumed by user-space to be
3902 		 * immutable.
3903 		 */
3904 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
3905 	} else {
3906 		/*
3907 		 * The only expected VM-instruction error is "VM entry with
3908 		 * invalid control field(s)." Anything else indicates a
3909 		 * problem with L0.  And we should never get here with a
3910 		 * VMFail of any type if early consistency checks are enabled.
3911 		 */
3912 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
3913 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3914 		WARN_ON_ONCE(nested_early_check);
3915 	}
3916 
3917 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3918 
3919 	/* Update any VMCS fields that might have changed while L2 ran */
3920 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3921 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3922 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
3923 
3924 	if (kvm_has_tsc_control)
3925 		decache_tsc_multiplier(vmx);
3926 
3927 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
3928 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
3929 		vmx_set_virtual_apic_mode(vcpu);
3930 	} else if (!nested_cpu_has_ept(vmcs12) &&
3931 		   nested_cpu_has2(vmcs12,
3932 				   SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3933 		vmx_flush_tlb(vcpu, true);
3934 	}
3935 
3936 	/* Unpin physical memory we referred to in vmcs02 */
3937 	if (vmx->nested.apic_access_page) {
3938 		kvm_release_page_dirty(vmx->nested.apic_access_page);
3939 		vmx->nested.apic_access_page = NULL;
3940 	}
3941 	if (vmx->nested.virtual_apic_page) {
3942 		kvm_release_page_dirty(vmx->nested.virtual_apic_page);
3943 		vmx->nested.virtual_apic_page = NULL;
3944 	}
3945 	if (vmx->nested.pi_desc_page) {
3946 		kunmap(vmx->nested.pi_desc_page);
3947 		kvm_release_page_dirty(vmx->nested.pi_desc_page);
3948 		vmx->nested.pi_desc_page = NULL;
3949 		vmx->nested.pi_desc = NULL;
3950 	}
3951 
3952 	/*
3953 	 * We are now running in L2, mmu_notifier will force to reload the
3954 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
3955 	 */
3956 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
3957 
3958 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
3959 		vmx->nested.need_vmcs12_sync = true;
3960 
3961 	/* in case we halted in L2 */
3962 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3963 
3964 	if (likely(!vmx->fail)) {
3965 		/*
3966 		 * TODO: SDM says that with acknowledge interrupt on
3967 		 * exit, bit 31 of the VM-exit interrupt information
3968 		 * (valid interrupt) is always set to 1 on
3969 		 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
3970 		 * need kvm_cpu_has_interrupt().  See the commit
3971 		 * message for details.
3972 		 */
3973 		if (nested_exit_intr_ack_set(vcpu) &&
3974 		    exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
3975 		    kvm_cpu_has_interrupt(vcpu)) {
3976 			int irq = kvm_cpu_get_interrupt(vcpu);
3977 			WARN_ON(irq < 0);
3978 			vmcs12->vm_exit_intr_info = irq |
3979 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
3980 		}
3981 
3982 		if (exit_reason != -1)
3983 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
3984 						       vmcs12->exit_qualification,
3985 						       vmcs12->idt_vectoring_info_field,
3986 						       vmcs12->vm_exit_intr_info,
3987 						       vmcs12->vm_exit_intr_error_code,
3988 						       KVM_ISA_VMX);
3989 
3990 		load_vmcs12_host_state(vcpu, vmcs12);
3991 
3992 		return;
3993 	}
3994 
3995 	/*
3996 	 * After an early L2 VM-entry failure, we're now back
3997 	 * in L1 which thinks it just finished a VMLAUNCH or
3998 	 * VMRESUME instruction, so we need to set the failure
3999 	 * flag and the VM-instruction error field of the VMCS
4000 	 * accordingly, and skip the emulated instruction.
4001 	 */
4002 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4003 
4004 	/*
4005 	 * Restore L1's host state to KVM's software model.  We're here
4006 	 * because a consistency check was caught by hardware, which
4007 	 * means some amount of guest state has been propagated to KVM's
4008 	 * model and needs to be unwound to the host's state.
4009 	 */
4010 	nested_vmx_restore_host_state(vcpu);
4011 
4012 	vmx->fail = 0;
4013 }
4014 
4015 /*
4016  * Decode the memory-address operand of a vmx instruction, as recorded on an
4017  * exit caused by such an instruction (run by a guest hypervisor).
4018  * On success, returns 0. When the operand is invalid, returns 1 and throws
4019  * #UD or #GP.
4020  */
4021 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4022 			u32 vmx_instruction_info, bool wr, gva_t *ret)
4023 {
4024 	gva_t off;
4025 	bool exn;
4026 	struct kvm_segment s;
4027 
4028 	/*
4029 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4030 	 * Execution", on an exit, vmx_instruction_info holds most of the
4031 	 * addressing components of the operand. Only the displacement part
4032 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4033 	 * For how an actual address is calculated from all these components,
4034 	 * refer to Vol. 1, "Operand Addressing".
4035 	 */
4036 	int  scaling = vmx_instruction_info & 3;
4037 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4038 	bool is_reg = vmx_instruction_info & (1u << 10);
4039 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4040 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4041 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4042 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4043 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4044 
4045 	if (is_reg) {
4046 		kvm_queue_exception(vcpu, UD_VECTOR);
4047 		return 1;
4048 	}
4049 
4050 	/* Addr = segment_base + offset */
4051 	/* offset = base + [index * scale] + displacement */
4052 	off = exit_qualification; /* holds the displacement */
4053 	if (addr_size == 1)
4054 		off = (gva_t)sign_extend64(off, 31);
4055 	else if (addr_size == 0)
4056 		off = (gva_t)sign_extend64(off, 15);
4057 	if (base_is_valid)
4058 		off += kvm_register_read(vcpu, base_reg);
4059 	if (index_is_valid)
4060 		off += kvm_register_read(vcpu, index_reg)<<scaling;
4061 	vmx_get_segment(vcpu, &s, seg_reg);
4062 
4063 	/*
4064 	 * The effective address, i.e. @off, of a memory operand is truncated
4065 	 * based on the address size of the instruction.  Note that this is
4066 	 * the *effective address*, i.e. the address prior to accounting for
4067 	 * the segment's base.
4068 	 */
4069 	if (addr_size == 1) /* 32 bit */
4070 		off &= 0xffffffff;
4071 	else if (addr_size == 0) /* 16 bit */
4072 		off &= 0xffff;
4073 
4074 	/* Checks for #GP/#SS exceptions. */
4075 	exn = false;
4076 	if (is_long_mode(vcpu)) {
4077 		/*
4078 		 * The virtual/linear address is never truncated in 64-bit
4079 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4080 		 * address when using FS/GS with a non-zero base.
4081 		 */
4082 		*ret = s.base + off;
4083 
4084 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4085 		 * non-canonical form. This is the only check on the memory
4086 		 * destination for long mode!
4087 		 */
4088 		exn = is_noncanonical_address(*ret, vcpu);
4089 	} else {
4090 		/*
4091 		 * When not in long mode, the virtual/linear address is
4092 		 * unconditionally truncated to 32 bits regardless of the
4093 		 * address size.
4094 		 */
4095 		*ret = (s.base + off) & 0xffffffff;
4096 
4097 		/* Protected mode: apply checks for segment validity in the
4098 		 * following order:
4099 		 * - segment type check (#GP(0) may be thrown)
4100 		 * - usability check (#GP(0)/#SS(0))
4101 		 * - limit check (#GP(0)/#SS(0))
4102 		 */
4103 		if (wr)
4104 			/* #GP(0) if the destination operand is located in a
4105 			 * read-only data segment or any code segment.
4106 			 */
4107 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4108 		else
4109 			/* #GP(0) if the source operand is located in an
4110 			 * execute-only code segment
4111 			 */
4112 			exn = ((s.type & 0xa) == 8);
4113 		if (exn) {
4114 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4115 			return 1;
4116 		}
4117 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4118 		 */
4119 		exn = (s.unusable != 0);
4120 
4121 		/*
4122 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4123 		 * outside the segment limit.  All CPUs that support VMX ignore
4124 		 * limit checks for flat segments, i.e. segments with base==0,
4125 		 * limit==0xffffffff and of type expand-up data or code.
4126 		 */
4127 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4128 		     ((s.type & 8) || !(s.type & 4))))
4129 			exn = exn || (off + sizeof(u64) > s.limit);
4130 	}
4131 	if (exn) {
4132 		kvm_queue_exception_e(vcpu,
4133 				      seg_reg == VCPU_SREG_SS ?
4134 						SS_VECTOR : GP_VECTOR,
4135 				      0);
4136 		return 1;
4137 	}
4138 
4139 	return 0;
4140 }
4141 
4142 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4143 {
4144 	gva_t gva;
4145 	struct x86_exception e;
4146 
4147 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4148 			vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
4149 		return 1;
4150 
4151 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4152 		kvm_inject_page_fault(vcpu, &e);
4153 		return 1;
4154 	}
4155 
4156 	return 0;
4157 }
4158 
4159 /*
4160  * Allocate a shadow VMCS and associate it with the currently loaded
4161  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4162  * VMCS is also VMCLEARed, so that it is ready for use.
4163  */
4164 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4165 {
4166 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4167 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4168 
4169 	/*
4170 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4171 	 * executes VMXON and free it when L1 executes VMXOFF.
4172 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4173 	 * here when vmcs01 already have an allocated shadow vmcs.
4174 	 */
4175 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4176 
4177 	if (!loaded_vmcs->shadow_vmcs) {
4178 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4179 		if (loaded_vmcs->shadow_vmcs)
4180 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4181 	}
4182 	return loaded_vmcs->shadow_vmcs;
4183 }
4184 
4185 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4186 {
4187 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4188 	int r;
4189 
4190 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4191 	if (r < 0)
4192 		goto out_vmcs02;
4193 
4194 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4195 	if (!vmx->nested.cached_vmcs12)
4196 		goto out_cached_vmcs12;
4197 
4198 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4199 	if (!vmx->nested.cached_shadow_vmcs12)
4200 		goto out_cached_shadow_vmcs12;
4201 
4202 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4203 		goto out_shadow_vmcs;
4204 
4205 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4206 		     HRTIMER_MODE_REL_PINNED);
4207 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4208 
4209 	vmx->nested.vpid02 = allocate_vpid();
4210 
4211 	vmx->nested.vmcs02_initialized = false;
4212 	vmx->nested.vmxon = true;
4213 
4214 	if (pt_mode == PT_MODE_HOST_GUEST) {
4215 		vmx->pt_desc.guest.ctl = 0;
4216 		pt_update_intercept_for_msr(vmx);
4217 	}
4218 
4219 	return 0;
4220 
4221 out_shadow_vmcs:
4222 	kfree(vmx->nested.cached_shadow_vmcs12);
4223 
4224 out_cached_shadow_vmcs12:
4225 	kfree(vmx->nested.cached_vmcs12);
4226 
4227 out_cached_vmcs12:
4228 	free_loaded_vmcs(&vmx->nested.vmcs02);
4229 
4230 out_vmcs02:
4231 	return -ENOMEM;
4232 }
4233 
4234 /*
4235  * Emulate the VMXON instruction.
4236  * Currently, we just remember that VMX is active, and do not save or even
4237  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4238  * do not currently need to store anything in that guest-allocated memory
4239  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4240  * argument is different from the VMXON pointer (which the spec says they do).
4241  */
4242 static int handle_vmon(struct kvm_vcpu *vcpu)
4243 {
4244 	int ret;
4245 	gpa_t vmptr;
4246 	struct page *page;
4247 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4248 	const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
4249 		| FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
4250 
4251 	/*
4252 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4253 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4254 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4255 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4256 	 * have already been checked by hardware, prior to the VM-exit for
4257 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4258 	 * that bit set to 1 in non-root mode.
4259 	 */
4260 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4261 		kvm_queue_exception(vcpu, UD_VECTOR);
4262 		return 1;
4263 	}
4264 
4265 	/* CPL=0 must be checked manually. */
4266 	if (vmx_get_cpl(vcpu)) {
4267 		kvm_inject_gp(vcpu, 0);
4268 		return 1;
4269 	}
4270 
4271 	if (vmx->nested.vmxon)
4272 		return nested_vmx_failValid(vcpu,
4273 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4274 
4275 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4276 			!= VMXON_NEEDED_FEATURES) {
4277 		kvm_inject_gp(vcpu, 0);
4278 		return 1;
4279 	}
4280 
4281 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4282 		return 1;
4283 
4284 	/*
4285 	 * SDM 3: 24.11.5
4286 	 * The first 4 bytes of VMXON region contain the supported
4287 	 * VMCS revision identifier
4288 	 *
4289 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4290 	 * which replaces physical address width with 32
4291 	 */
4292 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4293 		return nested_vmx_failInvalid(vcpu);
4294 
4295 	page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4296 	if (is_error_page(page))
4297 		return nested_vmx_failInvalid(vcpu);
4298 
4299 	if (*(u32 *)kmap(page) != VMCS12_REVISION) {
4300 		kunmap(page);
4301 		kvm_release_page_clean(page);
4302 		return nested_vmx_failInvalid(vcpu);
4303 	}
4304 	kunmap(page);
4305 	kvm_release_page_clean(page);
4306 
4307 	vmx->nested.vmxon_ptr = vmptr;
4308 	ret = enter_vmx_operation(vcpu);
4309 	if (ret)
4310 		return ret;
4311 
4312 	return nested_vmx_succeed(vcpu);
4313 }
4314 
4315 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4316 {
4317 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4318 
4319 	if (vmx->nested.current_vmptr == -1ull)
4320 		return;
4321 
4322 	if (enable_shadow_vmcs) {
4323 		/* copy to memory all shadowed fields in case
4324 		   they were modified */
4325 		copy_shadow_to_vmcs12(vmx);
4326 		vmx->nested.need_vmcs12_sync = false;
4327 		vmx_disable_shadow_vmcs(vmx);
4328 	}
4329 	vmx->nested.posted_intr_nv = -1;
4330 
4331 	/* Flush VMCS12 to guest memory */
4332 	kvm_vcpu_write_guest_page(vcpu,
4333 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4334 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4335 
4336 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4337 
4338 	vmx->nested.current_vmptr = -1ull;
4339 }
4340 
4341 /* Emulate the VMXOFF instruction */
4342 static int handle_vmoff(struct kvm_vcpu *vcpu)
4343 {
4344 	if (!nested_vmx_check_permission(vcpu))
4345 		return 1;
4346 	free_nested(vcpu);
4347 	return nested_vmx_succeed(vcpu);
4348 }
4349 
4350 /* Emulate the VMCLEAR instruction */
4351 static int handle_vmclear(struct kvm_vcpu *vcpu)
4352 {
4353 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4354 	u32 zero = 0;
4355 	gpa_t vmptr;
4356 
4357 	if (!nested_vmx_check_permission(vcpu))
4358 		return 1;
4359 
4360 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4361 		return 1;
4362 
4363 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4364 		return nested_vmx_failValid(vcpu,
4365 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4366 
4367 	if (vmptr == vmx->nested.vmxon_ptr)
4368 		return nested_vmx_failValid(vcpu,
4369 			VMXERR_VMCLEAR_VMXON_POINTER);
4370 
4371 	if (vmx->nested.hv_evmcs_page) {
4372 		if (vmptr == vmx->nested.hv_evmcs_vmptr)
4373 			nested_release_evmcs(vcpu);
4374 	} else {
4375 		if (vmptr == vmx->nested.current_vmptr)
4376 			nested_release_vmcs12(vcpu);
4377 
4378 		kvm_vcpu_write_guest(vcpu,
4379 				     vmptr + offsetof(struct vmcs12,
4380 						      launch_state),
4381 				     &zero, sizeof(zero));
4382 	}
4383 
4384 	return nested_vmx_succeed(vcpu);
4385 }
4386 
4387 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4388 
4389 /* Emulate the VMLAUNCH instruction */
4390 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4391 {
4392 	return nested_vmx_run(vcpu, true);
4393 }
4394 
4395 /* Emulate the VMRESUME instruction */
4396 static int handle_vmresume(struct kvm_vcpu *vcpu)
4397 {
4398 
4399 	return nested_vmx_run(vcpu, false);
4400 }
4401 
4402 static int handle_vmread(struct kvm_vcpu *vcpu)
4403 {
4404 	unsigned long field;
4405 	u64 field_value;
4406 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4407 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4408 	gva_t gva = 0;
4409 	struct vmcs12 *vmcs12;
4410 
4411 	if (!nested_vmx_check_permission(vcpu))
4412 		return 1;
4413 
4414 	if (to_vmx(vcpu)->nested.current_vmptr == -1ull)
4415 		return nested_vmx_failInvalid(vcpu);
4416 
4417 	if (!is_guest_mode(vcpu))
4418 		vmcs12 = get_vmcs12(vcpu);
4419 	else {
4420 		/*
4421 		 * When vmcs->vmcs_link_pointer is -1ull, any VMREAD
4422 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4423 		 */
4424 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4425 			return nested_vmx_failInvalid(vcpu);
4426 		vmcs12 = get_shadow_vmcs12(vcpu);
4427 	}
4428 
4429 	/* Decode instruction info and find the field to read */
4430 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4431 	/* Read the field, zero-extended to a u64 field_value */
4432 	if (vmcs12_read_any(vmcs12, field, &field_value) < 0)
4433 		return nested_vmx_failValid(vcpu,
4434 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4435 
4436 	/*
4437 	 * Now copy part of this value to register or memory, as requested.
4438 	 * Note that the number of bits actually copied is 32 or 64 depending
4439 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4440 	 */
4441 	if (vmx_instruction_info & (1u << 10)) {
4442 		kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
4443 			field_value);
4444 	} else {
4445 		if (get_vmx_mem_address(vcpu, exit_qualification,
4446 				vmx_instruction_info, true, &gva))
4447 			return 1;
4448 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4449 		kvm_write_guest_virt_system(vcpu, gva, &field_value,
4450 					    (is_long_mode(vcpu) ? 8 : 4), NULL);
4451 	}
4452 
4453 	return nested_vmx_succeed(vcpu);
4454 }
4455 
4456 
4457 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4458 {
4459 	unsigned long field;
4460 	gva_t gva;
4461 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4462 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4463 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4464 
4465 	/* The value to write might be 32 or 64 bits, depending on L1's long
4466 	 * mode, and eventually we need to write that into a field of several
4467 	 * possible lengths. The code below first zero-extends the value to 64
4468 	 * bit (field_value), and then copies only the appropriate number of
4469 	 * bits into the vmcs12 field.
4470 	 */
4471 	u64 field_value = 0;
4472 	struct x86_exception e;
4473 	struct vmcs12 *vmcs12;
4474 
4475 	if (!nested_vmx_check_permission(vcpu))
4476 		return 1;
4477 
4478 	if (vmx->nested.current_vmptr == -1ull)
4479 		return nested_vmx_failInvalid(vcpu);
4480 
4481 	if (vmx_instruction_info & (1u << 10))
4482 		field_value = kvm_register_readl(vcpu,
4483 			(((vmx_instruction_info) >> 3) & 0xf));
4484 	else {
4485 		if (get_vmx_mem_address(vcpu, exit_qualification,
4486 				vmx_instruction_info, false, &gva))
4487 			return 1;
4488 		if (kvm_read_guest_virt(vcpu, gva, &field_value,
4489 					(is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
4490 			kvm_inject_page_fault(vcpu, &e);
4491 			return 1;
4492 		}
4493 	}
4494 
4495 
4496 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4497 	/*
4498 	 * If the vCPU supports "VMWRITE to any supported field in the
4499 	 * VMCS," then the "read-only" fields are actually read/write.
4500 	 */
4501 	if (vmcs_field_readonly(field) &&
4502 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4503 		return nested_vmx_failValid(vcpu,
4504 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4505 
4506 	if (!is_guest_mode(vcpu))
4507 		vmcs12 = get_vmcs12(vcpu);
4508 	else {
4509 		/*
4510 		 * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
4511 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4512 		 */
4513 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4514 			return nested_vmx_failInvalid(vcpu);
4515 		vmcs12 = get_shadow_vmcs12(vcpu);
4516 	}
4517 
4518 	if (vmcs12_write_any(vmcs12, field, field_value) < 0)
4519 		return nested_vmx_failValid(vcpu,
4520 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4521 
4522 	/*
4523 	 * Do not track vmcs12 dirty-state if in guest-mode
4524 	 * as we actually dirty shadow vmcs12 instead of vmcs12.
4525 	 */
4526 	if (!is_guest_mode(vcpu)) {
4527 		switch (field) {
4528 #define SHADOW_FIELD_RW(x) case x:
4529 #include "vmcs_shadow_fields.h"
4530 			/*
4531 			 * The fields that can be updated by L1 without a vmexit are
4532 			 * always updated in the vmcs02, the others go down the slow
4533 			 * path of prepare_vmcs02.
4534 			 */
4535 			break;
4536 		default:
4537 			vmx->nested.dirty_vmcs12 = true;
4538 			break;
4539 		}
4540 	}
4541 
4542 	return nested_vmx_succeed(vcpu);
4543 }
4544 
4545 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
4546 {
4547 	vmx->nested.current_vmptr = vmptr;
4548 	if (enable_shadow_vmcs) {
4549 		vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4550 			      SECONDARY_EXEC_SHADOW_VMCS);
4551 		vmcs_write64(VMCS_LINK_POINTER,
4552 			     __pa(vmx->vmcs01.shadow_vmcs));
4553 		vmx->nested.need_vmcs12_sync = true;
4554 	}
4555 	vmx->nested.dirty_vmcs12 = true;
4556 }
4557 
4558 /* Emulate the VMPTRLD instruction */
4559 static int handle_vmptrld(struct kvm_vcpu *vcpu)
4560 {
4561 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4562 	gpa_t vmptr;
4563 
4564 	if (!nested_vmx_check_permission(vcpu))
4565 		return 1;
4566 
4567 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4568 		return 1;
4569 
4570 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4571 		return nested_vmx_failValid(vcpu,
4572 			VMXERR_VMPTRLD_INVALID_ADDRESS);
4573 
4574 	if (vmptr == vmx->nested.vmxon_ptr)
4575 		return nested_vmx_failValid(vcpu,
4576 			VMXERR_VMPTRLD_VMXON_POINTER);
4577 
4578 	/* Forbid normal VMPTRLD if Enlightened version was used */
4579 	if (vmx->nested.hv_evmcs)
4580 		return 1;
4581 
4582 	if (vmx->nested.current_vmptr != vmptr) {
4583 		struct vmcs12 *new_vmcs12;
4584 		struct page *page;
4585 
4586 		page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4587 		if (is_error_page(page)) {
4588 			/*
4589 			 * Reads from an unbacked page return all 1s,
4590 			 * which means that the 32 bits located at the
4591 			 * given physical address won't match the required
4592 			 * VMCS12_REVISION identifier.
4593 			 */
4594 			return nested_vmx_failValid(vcpu,
4595 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4596 		}
4597 		new_vmcs12 = kmap(page);
4598 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
4599 		    (new_vmcs12->hdr.shadow_vmcs &&
4600 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
4601 			kunmap(page);
4602 			kvm_release_page_clean(page);
4603 			return nested_vmx_failValid(vcpu,
4604 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4605 		}
4606 
4607 		nested_release_vmcs12(vcpu);
4608 
4609 		/*
4610 		 * Load VMCS12 from guest memory since it is not already
4611 		 * cached.
4612 		 */
4613 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
4614 		kunmap(page);
4615 		kvm_release_page_clean(page);
4616 
4617 		set_current_vmptr(vmx, vmptr);
4618 	}
4619 
4620 	return nested_vmx_succeed(vcpu);
4621 }
4622 
4623 /* Emulate the VMPTRST instruction */
4624 static int handle_vmptrst(struct kvm_vcpu *vcpu)
4625 {
4626 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
4627 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4628 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
4629 	struct x86_exception e;
4630 	gva_t gva;
4631 
4632 	if (!nested_vmx_check_permission(vcpu))
4633 		return 1;
4634 
4635 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
4636 		return 1;
4637 
4638 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info, true, &gva))
4639 		return 1;
4640 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
4641 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
4642 					sizeof(gpa_t), &e)) {
4643 		kvm_inject_page_fault(vcpu, &e);
4644 		return 1;
4645 	}
4646 	return nested_vmx_succeed(vcpu);
4647 }
4648 
4649 /* Emulate the INVEPT instruction */
4650 static int handle_invept(struct kvm_vcpu *vcpu)
4651 {
4652 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4653 	u32 vmx_instruction_info, types;
4654 	unsigned long type;
4655 	gva_t gva;
4656 	struct x86_exception e;
4657 	struct {
4658 		u64 eptp, gpa;
4659 	} operand;
4660 
4661 	if (!(vmx->nested.msrs.secondary_ctls_high &
4662 	      SECONDARY_EXEC_ENABLE_EPT) ||
4663 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
4664 		kvm_queue_exception(vcpu, UD_VECTOR);
4665 		return 1;
4666 	}
4667 
4668 	if (!nested_vmx_check_permission(vcpu))
4669 		return 1;
4670 
4671 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4672 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4673 
4674 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
4675 
4676 	if (type >= 32 || !(types & (1 << type)))
4677 		return nested_vmx_failValid(vcpu,
4678 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4679 
4680 	/* According to the Intel VMX instruction reference, the memory
4681 	 * operand is read even if it isn't needed (e.g., for type==global)
4682 	 */
4683 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4684 			vmx_instruction_info, false, &gva))
4685 		return 1;
4686 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4687 		kvm_inject_page_fault(vcpu, &e);
4688 		return 1;
4689 	}
4690 
4691 	switch (type) {
4692 	case VMX_EPT_EXTENT_GLOBAL:
4693 	/*
4694 	 * TODO: track mappings and invalidate
4695 	 * single context requests appropriately
4696 	 */
4697 	case VMX_EPT_EXTENT_CONTEXT:
4698 		kvm_mmu_sync_roots(vcpu);
4699 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4700 		break;
4701 	default:
4702 		BUG_ON(1);
4703 		break;
4704 	}
4705 
4706 	return nested_vmx_succeed(vcpu);
4707 }
4708 
4709 static int handle_invvpid(struct kvm_vcpu *vcpu)
4710 {
4711 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4712 	u32 vmx_instruction_info;
4713 	unsigned long type, types;
4714 	gva_t gva;
4715 	struct x86_exception e;
4716 	struct {
4717 		u64 vpid;
4718 		u64 gla;
4719 	} operand;
4720 	u16 vpid02;
4721 
4722 	if (!(vmx->nested.msrs.secondary_ctls_high &
4723 	      SECONDARY_EXEC_ENABLE_VPID) ||
4724 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
4725 		kvm_queue_exception(vcpu, UD_VECTOR);
4726 		return 1;
4727 	}
4728 
4729 	if (!nested_vmx_check_permission(vcpu))
4730 		return 1;
4731 
4732 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4733 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4734 
4735 	types = (vmx->nested.msrs.vpid_caps &
4736 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
4737 
4738 	if (type >= 32 || !(types & (1 << type)))
4739 		return nested_vmx_failValid(vcpu,
4740 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4741 
4742 	/* according to the intel vmx instruction reference, the memory
4743 	 * operand is read even if it isn't needed (e.g., for type==global)
4744 	 */
4745 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4746 			vmx_instruction_info, false, &gva))
4747 		return 1;
4748 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4749 		kvm_inject_page_fault(vcpu, &e);
4750 		return 1;
4751 	}
4752 	if (operand.vpid >> 16)
4753 		return nested_vmx_failValid(vcpu,
4754 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4755 
4756 	vpid02 = nested_get_vpid02(vcpu);
4757 	switch (type) {
4758 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
4759 		if (!operand.vpid ||
4760 		    is_noncanonical_address(operand.gla, vcpu))
4761 			return nested_vmx_failValid(vcpu,
4762 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4763 		if (cpu_has_vmx_invvpid_individual_addr()) {
4764 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
4765 				vpid02, operand.gla);
4766 		} else
4767 			__vmx_flush_tlb(vcpu, vpid02, false);
4768 		break;
4769 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
4770 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
4771 		if (!operand.vpid)
4772 			return nested_vmx_failValid(vcpu,
4773 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4774 		__vmx_flush_tlb(vcpu, vpid02, false);
4775 		break;
4776 	case VMX_VPID_EXTENT_ALL_CONTEXT:
4777 		__vmx_flush_tlb(vcpu, vpid02, false);
4778 		break;
4779 	default:
4780 		WARN_ON_ONCE(1);
4781 		return kvm_skip_emulated_instruction(vcpu);
4782 	}
4783 
4784 	return nested_vmx_succeed(vcpu);
4785 }
4786 
4787 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
4788 				     struct vmcs12 *vmcs12)
4789 {
4790 	u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
4791 	u64 address;
4792 	bool accessed_dirty;
4793 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4794 
4795 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
4796 	    !nested_cpu_has_ept(vmcs12))
4797 		return 1;
4798 
4799 	if (index >= VMFUNC_EPTP_ENTRIES)
4800 		return 1;
4801 
4802 
4803 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
4804 				     &address, index * 8, 8))
4805 		return 1;
4806 
4807 	accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
4808 
4809 	/*
4810 	 * If the (L2) guest does a vmfunc to the currently
4811 	 * active ept pointer, we don't have to do anything else
4812 	 */
4813 	if (vmcs12->ept_pointer != address) {
4814 		if (!valid_ept_address(vcpu, address))
4815 			return 1;
4816 
4817 		kvm_mmu_unload(vcpu);
4818 		mmu->ept_ad = accessed_dirty;
4819 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
4820 		vmcs12->ept_pointer = address;
4821 		/*
4822 		 * TODO: Check what's the correct approach in case
4823 		 * mmu reload fails. Currently, we just let the next
4824 		 * reload potentially fail
4825 		 */
4826 		kvm_mmu_reload(vcpu);
4827 	}
4828 
4829 	return 0;
4830 }
4831 
4832 static int handle_vmfunc(struct kvm_vcpu *vcpu)
4833 {
4834 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4835 	struct vmcs12 *vmcs12;
4836 	u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
4837 
4838 	/*
4839 	 * VMFUNC is only supported for nested guests, but we always enable the
4840 	 * secondary control for simplicity; for non-nested mode, fake that we
4841 	 * didn't by injecting #UD.
4842 	 */
4843 	if (!is_guest_mode(vcpu)) {
4844 		kvm_queue_exception(vcpu, UD_VECTOR);
4845 		return 1;
4846 	}
4847 
4848 	vmcs12 = get_vmcs12(vcpu);
4849 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
4850 		goto fail;
4851 
4852 	switch (function) {
4853 	case 0:
4854 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
4855 			goto fail;
4856 		break;
4857 	default:
4858 		goto fail;
4859 	}
4860 	return kvm_skip_emulated_instruction(vcpu);
4861 
4862 fail:
4863 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
4864 			  vmcs_read32(VM_EXIT_INTR_INFO),
4865 			  vmcs_readl(EXIT_QUALIFICATION));
4866 	return 1;
4867 }
4868 
4869 
4870 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
4871 				       struct vmcs12 *vmcs12)
4872 {
4873 	unsigned long exit_qualification;
4874 	gpa_t bitmap, last_bitmap;
4875 	unsigned int port;
4876 	int size;
4877 	u8 b;
4878 
4879 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
4880 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
4881 
4882 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4883 
4884 	port = exit_qualification >> 16;
4885 	size = (exit_qualification & 7) + 1;
4886 
4887 	last_bitmap = (gpa_t)-1;
4888 	b = -1;
4889 
4890 	while (size > 0) {
4891 		if (port < 0x8000)
4892 			bitmap = vmcs12->io_bitmap_a;
4893 		else if (port < 0x10000)
4894 			bitmap = vmcs12->io_bitmap_b;
4895 		else
4896 			return true;
4897 		bitmap += (port & 0x7fff) / 8;
4898 
4899 		if (last_bitmap != bitmap)
4900 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
4901 				return true;
4902 		if (b & (1 << (port & 7)))
4903 			return true;
4904 
4905 		port++;
4906 		size--;
4907 		last_bitmap = bitmap;
4908 	}
4909 
4910 	return false;
4911 }
4912 
4913 /*
4914  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
4915  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
4916  * disinterest in the current event (read or write a specific MSR) by using an
4917  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
4918  */
4919 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
4920 	struct vmcs12 *vmcs12, u32 exit_reason)
4921 {
4922 	u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
4923 	gpa_t bitmap;
4924 
4925 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
4926 		return true;
4927 
4928 	/*
4929 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
4930 	 * for the four combinations of read/write and low/high MSR numbers.
4931 	 * First we need to figure out which of the four to use:
4932 	 */
4933 	bitmap = vmcs12->msr_bitmap;
4934 	if (exit_reason == EXIT_REASON_MSR_WRITE)
4935 		bitmap += 2048;
4936 	if (msr_index >= 0xc0000000) {
4937 		msr_index -= 0xc0000000;
4938 		bitmap += 1024;
4939 	}
4940 
4941 	/* Then read the msr_index'th bit from this bitmap: */
4942 	if (msr_index < 1024*8) {
4943 		unsigned char b;
4944 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
4945 			return true;
4946 		return 1 & (b >> (msr_index & 7));
4947 	} else
4948 		return true; /* let L1 handle the wrong parameter */
4949 }
4950 
4951 /*
4952  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
4953  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
4954  * intercept (via guest_host_mask etc.) the current event.
4955  */
4956 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
4957 	struct vmcs12 *vmcs12)
4958 {
4959 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4960 	int cr = exit_qualification & 15;
4961 	int reg;
4962 	unsigned long val;
4963 
4964 	switch ((exit_qualification >> 4) & 3) {
4965 	case 0: /* mov to cr */
4966 		reg = (exit_qualification >> 8) & 15;
4967 		val = kvm_register_readl(vcpu, reg);
4968 		switch (cr) {
4969 		case 0:
4970 			if (vmcs12->cr0_guest_host_mask &
4971 			    (val ^ vmcs12->cr0_read_shadow))
4972 				return true;
4973 			break;
4974 		case 3:
4975 			if ((vmcs12->cr3_target_count >= 1 &&
4976 					vmcs12->cr3_target_value0 == val) ||
4977 				(vmcs12->cr3_target_count >= 2 &&
4978 					vmcs12->cr3_target_value1 == val) ||
4979 				(vmcs12->cr3_target_count >= 3 &&
4980 					vmcs12->cr3_target_value2 == val) ||
4981 				(vmcs12->cr3_target_count >= 4 &&
4982 					vmcs12->cr3_target_value3 == val))
4983 				return false;
4984 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
4985 				return true;
4986 			break;
4987 		case 4:
4988 			if (vmcs12->cr4_guest_host_mask &
4989 			    (vmcs12->cr4_read_shadow ^ val))
4990 				return true;
4991 			break;
4992 		case 8:
4993 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
4994 				return true;
4995 			break;
4996 		}
4997 		break;
4998 	case 2: /* clts */
4999 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5000 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
5001 			return true;
5002 		break;
5003 	case 1: /* mov from cr */
5004 		switch (cr) {
5005 		case 3:
5006 			if (vmcs12->cpu_based_vm_exec_control &
5007 			    CPU_BASED_CR3_STORE_EXITING)
5008 				return true;
5009 			break;
5010 		case 8:
5011 			if (vmcs12->cpu_based_vm_exec_control &
5012 			    CPU_BASED_CR8_STORE_EXITING)
5013 				return true;
5014 			break;
5015 		}
5016 		break;
5017 	case 3: /* lmsw */
5018 		/*
5019 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5020 		 * cr0. Other attempted changes are ignored, with no exit.
5021 		 */
5022 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5023 		if (vmcs12->cr0_guest_host_mask & 0xe &
5024 		    (val ^ vmcs12->cr0_read_shadow))
5025 			return true;
5026 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5027 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5028 		    (val & 0x1))
5029 			return true;
5030 		break;
5031 	}
5032 	return false;
5033 }
5034 
5035 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5036 	struct vmcs12 *vmcs12, gpa_t bitmap)
5037 {
5038 	u32 vmx_instruction_info;
5039 	unsigned long field;
5040 	u8 b;
5041 
5042 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5043 		return true;
5044 
5045 	/* Decode instruction info and find the field to access */
5046 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5047 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5048 
5049 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5050 	if (field >> 15)
5051 		return true;
5052 
5053 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5054 		return true;
5055 
5056 	return 1 & (b >> (field & 7));
5057 }
5058 
5059 /*
5060  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5061  * should handle it ourselves in L0 (and then continue L2). Only call this
5062  * when in is_guest_mode (L2).
5063  */
5064 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5065 {
5066 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5067 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5068 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5069 
5070 	if (vmx->nested.nested_run_pending)
5071 		return false;
5072 
5073 	if (unlikely(vmx->fail)) {
5074 		pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5075 				    vmcs_read32(VM_INSTRUCTION_ERROR));
5076 		return true;
5077 	}
5078 
5079 	/*
5080 	 * The host physical addresses of some pages of guest memory
5081 	 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5082 	 * Page). The CPU may write to these pages via their host
5083 	 * physical address while L2 is running, bypassing any
5084 	 * address-translation-based dirty tracking (e.g. EPT write
5085 	 * protection).
5086 	 *
5087 	 * Mark them dirty on every exit from L2 to prevent them from
5088 	 * getting out of sync with dirty tracking.
5089 	 */
5090 	nested_mark_vmcs12_pages_dirty(vcpu);
5091 
5092 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5093 				vmcs_readl(EXIT_QUALIFICATION),
5094 				vmx->idt_vectoring_info,
5095 				intr_info,
5096 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5097 				KVM_ISA_VMX);
5098 
5099 	switch (exit_reason) {
5100 	case EXIT_REASON_EXCEPTION_NMI:
5101 		if (is_nmi(intr_info))
5102 			return false;
5103 		else if (is_page_fault(intr_info))
5104 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5105 		else if (is_debug(intr_info) &&
5106 			 vcpu->guest_debug &
5107 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5108 			return false;
5109 		else if (is_breakpoint(intr_info) &&
5110 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5111 			return false;
5112 		return vmcs12->exception_bitmap &
5113 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5114 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5115 		return false;
5116 	case EXIT_REASON_TRIPLE_FAULT:
5117 		return true;
5118 	case EXIT_REASON_PENDING_INTERRUPT:
5119 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
5120 	case EXIT_REASON_NMI_WINDOW:
5121 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
5122 	case EXIT_REASON_TASK_SWITCH:
5123 		return true;
5124 	case EXIT_REASON_CPUID:
5125 		return true;
5126 	case EXIT_REASON_HLT:
5127 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5128 	case EXIT_REASON_INVD:
5129 		return true;
5130 	case EXIT_REASON_INVLPG:
5131 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5132 	case EXIT_REASON_RDPMC:
5133 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5134 	case EXIT_REASON_RDRAND:
5135 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5136 	case EXIT_REASON_RDSEED:
5137 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5138 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5139 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5140 	case EXIT_REASON_VMREAD:
5141 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5142 			vmcs12->vmread_bitmap);
5143 	case EXIT_REASON_VMWRITE:
5144 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5145 			vmcs12->vmwrite_bitmap);
5146 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5147 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5148 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5149 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5150 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5151 		/*
5152 		 * VMX instructions trap unconditionally. This allows L1 to
5153 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5154 		 */
5155 		return true;
5156 	case EXIT_REASON_CR_ACCESS:
5157 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5158 	case EXIT_REASON_DR_ACCESS:
5159 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5160 	case EXIT_REASON_IO_INSTRUCTION:
5161 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5162 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5163 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5164 	case EXIT_REASON_MSR_READ:
5165 	case EXIT_REASON_MSR_WRITE:
5166 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5167 	case EXIT_REASON_INVALID_STATE:
5168 		return true;
5169 	case EXIT_REASON_MWAIT_INSTRUCTION:
5170 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5171 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5172 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
5173 	case EXIT_REASON_MONITOR_INSTRUCTION:
5174 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5175 	case EXIT_REASON_PAUSE_INSTRUCTION:
5176 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5177 			nested_cpu_has2(vmcs12,
5178 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5179 	case EXIT_REASON_MCE_DURING_VMENTRY:
5180 		return false;
5181 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5182 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5183 	case EXIT_REASON_APIC_ACCESS:
5184 	case EXIT_REASON_APIC_WRITE:
5185 	case EXIT_REASON_EOI_INDUCED:
5186 		/*
5187 		 * The controls for "virtualize APIC accesses," "APIC-
5188 		 * register virtualization," and "virtual-interrupt
5189 		 * delivery" only come from vmcs12.
5190 		 */
5191 		return true;
5192 	case EXIT_REASON_EPT_VIOLATION:
5193 		/*
5194 		 * L0 always deals with the EPT violation. If nested EPT is
5195 		 * used, and the nested mmu code discovers that the address is
5196 		 * missing in the guest EPT table (EPT12), the EPT violation
5197 		 * will be injected with nested_ept_inject_page_fault()
5198 		 */
5199 		return false;
5200 	case EXIT_REASON_EPT_MISCONFIG:
5201 		/*
5202 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5203 		 * table (shadow on EPT) or a merged EPT table that L0 built
5204 		 * (EPT on EPT). So any problems with the structure of the
5205 		 * table is L0's fault.
5206 		 */
5207 		return false;
5208 	case EXIT_REASON_INVPCID:
5209 		return
5210 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5211 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5212 	case EXIT_REASON_WBINVD:
5213 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5214 	case EXIT_REASON_XSETBV:
5215 		return true;
5216 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5217 		/*
5218 		 * This should never happen, since it is not possible to
5219 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5220 		 * If if it were, XSS would have to be checked against
5221 		 * the XSS exit bitmap in vmcs12.
5222 		 */
5223 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5224 	case EXIT_REASON_PREEMPTION_TIMER:
5225 		return false;
5226 	case EXIT_REASON_PML_FULL:
5227 		/* We emulate PML support to L1. */
5228 		return false;
5229 	case EXIT_REASON_VMFUNC:
5230 		/* VM functions are emulated through L2->L0 vmexits. */
5231 		return false;
5232 	case EXIT_REASON_ENCLS:
5233 		/* SGX is never exposed to L1 */
5234 		return false;
5235 	default:
5236 		return true;
5237 	}
5238 }
5239 
5240 
5241 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5242 				struct kvm_nested_state __user *user_kvm_nested_state,
5243 				u32 user_data_size)
5244 {
5245 	struct vcpu_vmx *vmx;
5246 	struct vmcs12 *vmcs12;
5247 	struct kvm_nested_state kvm_state = {
5248 		.flags = 0,
5249 		.format = 0,
5250 		.size = sizeof(kvm_state),
5251 		.vmx.vmxon_pa = -1ull,
5252 		.vmx.vmcs_pa = -1ull,
5253 	};
5254 
5255 	if (!vcpu)
5256 		return kvm_state.size + 2 * VMCS12_SIZE;
5257 
5258 	vmx = to_vmx(vcpu);
5259 	vmcs12 = get_vmcs12(vcpu);
5260 
5261 	if (nested_vmx_allowed(vcpu) && vmx->nested.enlightened_vmcs_enabled)
5262 		kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5263 
5264 	if (nested_vmx_allowed(vcpu) &&
5265 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5266 		kvm_state.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5267 		kvm_state.vmx.vmcs_pa = vmx->nested.current_vmptr;
5268 
5269 		if (vmx_has_valid_vmcs12(vcpu)) {
5270 			kvm_state.size += VMCS12_SIZE;
5271 
5272 			if (is_guest_mode(vcpu) &&
5273 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5274 			    vmcs12->vmcs_link_pointer != -1ull)
5275 				kvm_state.size += VMCS12_SIZE;
5276 		}
5277 
5278 		if (vmx->nested.smm.vmxon)
5279 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5280 
5281 		if (vmx->nested.smm.guest_mode)
5282 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5283 
5284 		if (is_guest_mode(vcpu)) {
5285 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5286 
5287 			if (vmx->nested.nested_run_pending)
5288 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5289 		}
5290 	}
5291 
5292 	if (user_data_size < kvm_state.size)
5293 		goto out;
5294 
5295 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5296 		return -EFAULT;
5297 
5298 	if (!vmx_has_valid_vmcs12(vcpu))
5299 		goto out;
5300 
5301 	/*
5302 	 * When running L2, the authoritative vmcs12 state is in the
5303 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5304 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5305 	 * need_vmcs12_sync is set, in which case, the authoritative
5306 	 * vmcs12 state is in the vmcs12 already.
5307 	 */
5308 	if (is_guest_mode(vcpu)) {
5309 		sync_vmcs12(vcpu, vmcs12);
5310 	} else if (!vmx->nested.need_vmcs12_sync) {
5311 		if (vmx->nested.hv_evmcs)
5312 			copy_enlightened_to_vmcs12(vmx);
5313 		else if (enable_shadow_vmcs)
5314 			copy_shadow_to_vmcs12(vmx);
5315 	}
5316 
5317 	/*
5318 	 * Copy over the full allocated size of vmcs12 rather than just the size
5319 	 * of the struct.
5320 	 */
5321 	if (copy_to_user(user_kvm_nested_state->data, vmcs12, VMCS12_SIZE))
5322 		return -EFAULT;
5323 
5324 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5325 	    vmcs12->vmcs_link_pointer != -1ull) {
5326 		if (copy_to_user(user_kvm_nested_state->data + VMCS12_SIZE,
5327 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
5328 			return -EFAULT;
5329 	}
5330 
5331 out:
5332 	return kvm_state.size;
5333 }
5334 
5335 /*
5336  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5337  */
5338 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5339 {
5340 	if (is_guest_mode(vcpu)) {
5341 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5342 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5343 	}
5344 	free_nested(vcpu);
5345 }
5346 
5347 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5348 				struct kvm_nested_state __user *user_kvm_nested_state,
5349 				struct kvm_nested_state *kvm_state)
5350 {
5351 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5352 	struct vmcs12 *vmcs12;
5353 	u32 exit_qual;
5354 	int ret;
5355 
5356 	if (kvm_state->format != 0)
5357 		return -EINVAL;
5358 
5359 	if (kvm_state->flags & KVM_STATE_NESTED_EVMCS)
5360 		nested_enable_evmcs(vcpu, NULL);
5361 
5362 	if (!nested_vmx_allowed(vcpu))
5363 		return kvm_state->vmx.vmxon_pa == -1ull ? 0 : -EINVAL;
5364 
5365 	if (kvm_state->vmx.vmxon_pa == -1ull) {
5366 		if (kvm_state->vmx.smm.flags)
5367 			return -EINVAL;
5368 
5369 		if (kvm_state->vmx.vmcs_pa != -1ull)
5370 			return -EINVAL;
5371 
5372 		vmx_leave_nested(vcpu);
5373 		return 0;
5374 	}
5375 
5376 	if (!page_address_valid(vcpu, kvm_state->vmx.vmxon_pa))
5377 		return -EINVAL;
5378 
5379 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5380 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5381 		return -EINVAL;
5382 
5383 	if (kvm_state->vmx.smm.flags &
5384 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5385 		return -EINVAL;
5386 
5387 	/*
5388 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5389 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5390 	 * must be zero.
5391 	 */
5392 	if (is_smm(vcpu) ? kvm_state->flags : kvm_state->vmx.smm.flags)
5393 		return -EINVAL;
5394 
5395 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5396 	    !(kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5397 		return -EINVAL;
5398 
5399 	vmx_leave_nested(vcpu);
5400 	if (kvm_state->vmx.vmxon_pa == -1ull)
5401 		return 0;
5402 
5403 	vmx->nested.vmxon_ptr = kvm_state->vmx.vmxon_pa;
5404 	ret = enter_vmx_operation(vcpu);
5405 	if (ret)
5406 		return ret;
5407 
5408 	/* Empty 'VMXON' state is permitted */
5409 	if (kvm_state->size < sizeof(kvm_state) + sizeof(*vmcs12))
5410 		return 0;
5411 
5412 	if (kvm_state->vmx.vmcs_pa != -1ull) {
5413 		if (kvm_state->vmx.vmcs_pa == kvm_state->vmx.vmxon_pa ||
5414 		    !page_address_valid(vcpu, kvm_state->vmx.vmcs_pa))
5415 			return -EINVAL;
5416 
5417 		set_current_vmptr(vmx, kvm_state->vmx.vmcs_pa);
5418 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5419 		/*
5420 		 * Sync eVMCS upon entry as we may not have
5421 		 * HV_X64_MSR_VP_ASSIST_PAGE set up yet.
5422 		 */
5423 		vmx->nested.need_vmcs12_sync = true;
5424 	} else {
5425 		return -EINVAL;
5426 	}
5427 
5428 	if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5429 		vmx->nested.smm.vmxon = true;
5430 		vmx->nested.vmxon = false;
5431 
5432 		if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5433 			vmx->nested.smm.guest_mode = true;
5434 	}
5435 
5436 	vmcs12 = get_vmcs12(vcpu);
5437 	if (copy_from_user(vmcs12, user_kvm_nested_state->data, sizeof(*vmcs12)))
5438 		return -EFAULT;
5439 
5440 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5441 		return -EINVAL;
5442 
5443 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5444 		return 0;
5445 
5446 	vmx->nested.nested_run_pending =
5447 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5448 
5449 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5450 	    vmcs12->vmcs_link_pointer != -1ull) {
5451 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5452 
5453 		if (kvm_state->size < sizeof(kvm_state) + 2 * sizeof(*vmcs12))
5454 			return -EINVAL;
5455 
5456 		if (copy_from_user(shadow_vmcs12,
5457 				   user_kvm_nested_state->data + VMCS12_SIZE,
5458 				   sizeof(*vmcs12)))
5459 			return -EFAULT;
5460 
5461 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5462 		    !shadow_vmcs12->hdr.shadow_vmcs)
5463 			return -EINVAL;
5464 	}
5465 
5466 	if (nested_vmx_check_vmentry_prereqs(vcpu, vmcs12) ||
5467 	    nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
5468 		return -EINVAL;
5469 
5470 	vmx->nested.dirty_vmcs12 = true;
5471 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5472 	if (ret)
5473 		return -EINVAL;
5474 
5475 	return 0;
5476 }
5477 
5478 void nested_vmx_vcpu_setup(void)
5479 {
5480 	if (enable_shadow_vmcs) {
5481 		/*
5482 		 * At vCPU creation, "VMWRITE to any supported field
5483 		 * in the VMCS" is supported, so use the more
5484 		 * permissive vmx_vmread_bitmap to specify both read
5485 		 * and write permissions for the shadow VMCS.
5486 		 */
5487 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5488 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmread_bitmap));
5489 	}
5490 }
5491 
5492 /*
5493  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5494  * returned for the various VMX controls MSRs when nested VMX is enabled.
5495  * The same values should also be used to verify that vmcs12 control fields are
5496  * valid during nested entry from L1 to L2.
5497  * Each of these control msrs has a low and high 32-bit half: A low bit is on
5498  * if the corresponding bit in the (32-bit) control field *must* be on, and a
5499  * bit in the high half is on if the corresponding bit in the control field
5500  * may be on. See also vmx_control_verify().
5501  */
5502 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps,
5503 				bool apicv)
5504 {
5505 	/*
5506 	 * Note that as a general rule, the high half of the MSRs (bits in
5507 	 * the control fields which may be 1) should be initialized by the
5508 	 * intersection of the underlying hardware's MSR (i.e., features which
5509 	 * can be supported) and the list of features we want to expose -
5510 	 * because they are known to be properly supported in our code.
5511 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
5512 	 * be set to 0, meaning that L1 may turn off any of these bits. The
5513 	 * reason is that if one of these bits is necessary, it will appear
5514 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
5515 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
5516 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
5517 	 * These rules have exceptions below.
5518 	 */
5519 
5520 	/* pin-based controls */
5521 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
5522 		msrs->pinbased_ctls_low,
5523 		msrs->pinbased_ctls_high);
5524 	msrs->pinbased_ctls_low |=
5525 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5526 	msrs->pinbased_ctls_high &=
5527 		PIN_BASED_EXT_INTR_MASK |
5528 		PIN_BASED_NMI_EXITING |
5529 		PIN_BASED_VIRTUAL_NMIS |
5530 		(apicv ? PIN_BASED_POSTED_INTR : 0);
5531 	msrs->pinbased_ctls_high |=
5532 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5533 		PIN_BASED_VMX_PREEMPTION_TIMER;
5534 
5535 	/* exit controls */
5536 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
5537 		msrs->exit_ctls_low,
5538 		msrs->exit_ctls_high);
5539 	msrs->exit_ctls_low =
5540 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
5541 
5542 	msrs->exit_ctls_high &=
5543 #ifdef CONFIG_X86_64
5544 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
5545 #endif
5546 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
5547 	msrs->exit_ctls_high |=
5548 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
5549 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
5550 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
5551 
5552 	/* We support free control of debug control saving. */
5553 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
5554 
5555 	/* entry controls */
5556 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
5557 		msrs->entry_ctls_low,
5558 		msrs->entry_ctls_high);
5559 	msrs->entry_ctls_low =
5560 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
5561 	msrs->entry_ctls_high &=
5562 #ifdef CONFIG_X86_64
5563 		VM_ENTRY_IA32E_MODE |
5564 #endif
5565 		VM_ENTRY_LOAD_IA32_PAT;
5566 	msrs->entry_ctls_high |=
5567 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
5568 
5569 	/* We support free control of debug control loading. */
5570 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
5571 
5572 	/* cpu-based controls */
5573 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
5574 		msrs->procbased_ctls_low,
5575 		msrs->procbased_ctls_high);
5576 	msrs->procbased_ctls_low =
5577 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5578 	msrs->procbased_ctls_high &=
5579 		CPU_BASED_VIRTUAL_INTR_PENDING |
5580 		CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
5581 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
5582 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
5583 		CPU_BASED_CR3_STORE_EXITING |
5584 #ifdef CONFIG_X86_64
5585 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
5586 #endif
5587 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
5588 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
5589 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
5590 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
5591 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
5592 	/*
5593 	 * We can allow some features even when not supported by the
5594 	 * hardware. For example, L1 can specify an MSR bitmap - and we
5595 	 * can use it to avoid exits to L1 - even when L0 runs L2
5596 	 * without MSR bitmaps.
5597 	 */
5598 	msrs->procbased_ctls_high |=
5599 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5600 		CPU_BASED_USE_MSR_BITMAPS;
5601 
5602 	/* We support free control of CR3 access interception. */
5603 	msrs->procbased_ctls_low &=
5604 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
5605 
5606 	/*
5607 	 * secondary cpu-based controls.  Do not include those that
5608 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
5609 	 */
5610 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
5611 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
5612 		      msrs->secondary_ctls_low,
5613 		      msrs->secondary_ctls_high);
5614 
5615 	msrs->secondary_ctls_low = 0;
5616 	msrs->secondary_ctls_high &=
5617 		SECONDARY_EXEC_DESC |
5618 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
5619 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
5620 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
5621 		SECONDARY_EXEC_WBINVD_EXITING;
5622 
5623 	/*
5624 	 * We can emulate "VMCS shadowing," even if the hardware
5625 	 * doesn't support it.
5626 	 */
5627 	msrs->secondary_ctls_high |=
5628 		SECONDARY_EXEC_SHADOW_VMCS;
5629 
5630 	if (enable_ept) {
5631 		/* nested EPT: emulate EPT also to L1 */
5632 		msrs->secondary_ctls_high |=
5633 			SECONDARY_EXEC_ENABLE_EPT;
5634 		msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
5635 			 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
5636 		if (cpu_has_vmx_ept_execute_only())
5637 			msrs->ept_caps |=
5638 				VMX_EPT_EXECUTE_ONLY_BIT;
5639 		msrs->ept_caps &= ept_caps;
5640 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
5641 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
5642 			VMX_EPT_1GB_PAGE_BIT;
5643 		if (enable_ept_ad_bits) {
5644 			msrs->secondary_ctls_high |=
5645 				SECONDARY_EXEC_ENABLE_PML;
5646 			msrs->ept_caps |= VMX_EPT_AD_BIT;
5647 		}
5648 	}
5649 
5650 	if (cpu_has_vmx_vmfunc()) {
5651 		msrs->secondary_ctls_high |=
5652 			SECONDARY_EXEC_ENABLE_VMFUNC;
5653 		/*
5654 		 * Advertise EPTP switching unconditionally
5655 		 * since we emulate it
5656 		 */
5657 		if (enable_ept)
5658 			msrs->vmfunc_controls =
5659 				VMX_VMFUNC_EPTP_SWITCHING;
5660 	}
5661 
5662 	/*
5663 	 * Old versions of KVM use the single-context version without
5664 	 * checking for support, so declare that it is supported even
5665 	 * though it is treated as global context.  The alternative is
5666 	 * not failing the single-context invvpid, and it is worse.
5667 	 */
5668 	if (enable_vpid) {
5669 		msrs->secondary_ctls_high |=
5670 			SECONDARY_EXEC_ENABLE_VPID;
5671 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
5672 			VMX_VPID_EXTENT_SUPPORTED_MASK;
5673 	}
5674 
5675 	if (enable_unrestricted_guest)
5676 		msrs->secondary_ctls_high |=
5677 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
5678 
5679 	if (flexpriority_enabled)
5680 		msrs->secondary_ctls_high |=
5681 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5682 
5683 	/* miscellaneous data */
5684 	rdmsr(MSR_IA32_VMX_MISC,
5685 		msrs->misc_low,
5686 		msrs->misc_high);
5687 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
5688 	msrs->misc_low |=
5689 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
5690 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
5691 		VMX_MISC_ACTIVITY_HLT;
5692 	msrs->misc_high = 0;
5693 
5694 	/*
5695 	 * This MSR reports some information about VMX support. We
5696 	 * should return information about the VMX we emulate for the
5697 	 * guest, and the VMCS structure we give it - not about the
5698 	 * VMX support of the underlying hardware.
5699 	 */
5700 	msrs->basic =
5701 		VMCS12_REVISION |
5702 		VMX_BASIC_TRUE_CTLS |
5703 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
5704 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
5705 
5706 	if (cpu_has_vmx_basic_inout())
5707 		msrs->basic |= VMX_BASIC_INOUT;
5708 
5709 	/*
5710 	 * These MSRs specify bits which the guest must keep fixed on
5711 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
5712 	 * We picked the standard core2 setting.
5713 	 */
5714 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
5715 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
5716 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
5717 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
5718 
5719 	/* These MSRs specify bits which the guest must keep fixed off. */
5720 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
5721 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
5722 
5723 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
5724 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
5725 }
5726 
5727 void nested_vmx_hardware_unsetup(void)
5728 {
5729 	int i;
5730 
5731 	if (enable_shadow_vmcs) {
5732 		for (i = 0; i < VMX_BITMAP_NR; i++)
5733 			free_page((unsigned long)vmx_bitmap[i]);
5734 	}
5735 }
5736 
5737 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
5738 {
5739 	int i;
5740 
5741 	if (!cpu_has_vmx_shadow_vmcs())
5742 		enable_shadow_vmcs = 0;
5743 	if (enable_shadow_vmcs) {
5744 		for (i = 0; i < VMX_BITMAP_NR; i++) {
5745 			/*
5746 			 * The vmx_bitmap is not tied to a VM and so should
5747 			 * not be charged to a memcg.
5748 			 */
5749 			vmx_bitmap[i] = (unsigned long *)
5750 				__get_free_page(GFP_KERNEL);
5751 			if (!vmx_bitmap[i]) {
5752 				nested_vmx_hardware_unsetup();
5753 				return -ENOMEM;
5754 			}
5755 		}
5756 
5757 		init_vmcs_shadow_fields();
5758 	}
5759 
5760 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear,
5761 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch,
5762 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld,
5763 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst,
5764 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread,
5765 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume,
5766 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite,
5767 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff,
5768 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon,
5769 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept,
5770 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid,
5771 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc,
5772 
5773 	kvm_x86_ops->check_nested_events = vmx_check_nested_events;
5774 	kvm_x86_ops->get_nested_state = vmx_get_nested_state;
5775 	kvm_x86_ops->set_nested_state = vmx_set_nested_state;
5776 	kvm_x86_ops->get_vmcs12_pages = nested_get_vmcs12_pages,
5777 	kvm_x86_ops->nested_enable_evmcs = nested_enable_evmcs;
5778 	kvm_x86_ops->nested_get_evmcs_version = nested_get_evmcs_version;
5779 
5780 	return 0;
5781 }
5782