xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 95777591)
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 	hrtimer_cancel(&vmx->nested.preemption_timer);
215 	vmx->nested.vmxon = false;
216 	vmx->nested.smm.vmxon = false;
217 	free_vpid(vmx->nested.vpid02);
218 	vmx->nested.posted_intr_nv = -1;
219 	vmx->nested.current_vmptr = -1ull;
220 	if (enable_shadow_vmcs) {
221 		vmx_disable_shadow_vmcs(vmx);
222 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
223 		free_vmcs(vmx->vmcs01.shadow_vmcs);
224 		vmx->vmcs01.shadow_vmcs = NULL;
225 	}
226 	kfree(vmx->nested.cached_vmcs12);
227 	kfree(vmx->nested.cached_shadow_vmcs12);
228 	/* Unpin physical memory we referred to in the vmcs02 */
229 	if (vmx->nested.apic_access_page) {
230 		kvm_release_page_dirty(vmx->nested.apic_access_page);
231 		vmx->nested.apic_access_page = NULL;
232 	}
233 	if (vmx->nested.virtual_apic_page) {
234 		kvm_release_page_dirty(vmx->nested.virtual_apic_page);
235 		vmx->nested.virtual_apic_page = NULL;
236 	}
237 	if (vmx->nested.pi_desc_page) {
238 		kunmap(vmx->nested.pi_desc_page);
239 		kvm_release_page_dirty(vmx->nested.pi_desc_page);
240 		vmx->nested.pi_desc_page = NULL;
241 		vmx->nested.pi_desc = NULL;
242 	}
243 
244 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
245 
246 	nested_release_evmcs(vcpu);
247 
248 	free_loaded_vmcs(&vmx->nested.vmcs02);
249 }
250 
251 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
252 {
253 	struct vcpu_vmx *vmx = to_vmx(vcpu);
254 	int cpu;
255 
256 	if (vmx->loaded_vmcs == vmcs)
257 		return;
258 
259 	cpu = get_cpu();
260 	vmx_vcpu_put(vcpu);
261 	vmx->loaded_vmcs = vmcs;
262 	vmx_vcpu_load(vcpu, cpu);
263 	put_cpu();
264 
265 	vm_entry_controls_reset_shadow(vmx);
266 	vm_exit_controls_reset_shadow(vmx);
267 	vmx_segment_cache_clear(vmx);
268 }
269 
270 /*
271  * Ensure that the current vmcs of the logical processor is the
272  * vmcs01 of the vcpu before calling free_nested().
273  */
274 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
275 {
276 	vcpu_load(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 /*
504  * Merge L0's and L1's MSR bitmap, return false to indicate that
505  * we do not use the hardware.
506  */
507 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
508 						 struct vmcs12 *vmcs12)
509 {
510 	int msr;
511 	struct page *page;
512 	unsigned long *msr_bitmap_l1;
513 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
514 	/*
515 	 * pred_cmd & spec_ctrl are trying to verify two things:
516 	 *
517 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
518 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
519 	 *    from the L12 MSR bitmap that is too permissive.
520 	 * 2. That L1 or L2s have actually used the MSR. This avoids
521 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
522 	 *    works properly because we only update the L01 MSR bitmap lazily.
523 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
524 	 *    updated to reflect this when L1 (or its L2s) actually write to
525 	 *    the MSR.
526 	 */
527 	bool pred_cmd = !msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD);
528 	bool spec_ctrl = !msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL);
529 
530 	/* Nothing to do if the MSR bitmap is not in use.  */
531 	if (!cpu_has_vmx_msr_bitmap() ||
532 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
533 		return false;
534 
535 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
536 	    !pred_cmd && !spec_ctrl)
537 		return false;
538 
539 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
540 	if (is_error_page(page))
541 		return false;
542 
543 	msr_bitmap_l1 = (unsigned long *)kmap(page);
544 	if (nested_cpu_has_apic_reg_virt(vmcs12)) {
545 		/*
546 		 * L0 need not intercept reads for MSRs between 0x800 and 0x8ff, it
547 		 * just lets the processor take the value from the virtual-APIC page;
548 		 * take those 256 bits directly from the L1 bitmap.
549 		 */
550 		for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
551 			unsigned word = msr / BITS_PER_LONG;
552 			msr_bitmap_l0[word] = msr_bitmap_l1[word];
553 			msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
554 		}
555 	} else {
556 		for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
557 			unsigned word = msr / BITS_PER_LONG;
558 			msr_bitmap_l0[word] = ~0;
559 			msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
560 		}
561 	}
562 
563 	nested_vmx_disable_intercept_for_msr(
564 		msr_bitmap_l1, msr_bitmap_l0,
565 		X2APIC_MSR(APIC_TASKPRI),
566 		MSR_TYPE_W);
567 
568 	if (nested_cpu_has_vid(vmcs12)) {
569 		nested_vmx_disable_intercept_for_msr(
570 			msr_bitmap_l1, msr_bitmap_l0,
571 			X2APIC_MSR(APIC_EOI),
572 			MSR_TYPE_W);
573 		nested_vmx_disable_intercept_for_msr(
574 			msr_bitmap_l1, msr_bitmap_l0,
575 			X2APIC_MSR(APIC_SELF_IPI),
576 			MSR_TYPE_W);
577 	}
578 
579 	if (spec_ctrl)
580 		nested_vmx_disable_intercept_for_msr(
581 					msr_bitmap_l1, msr_bitmap_l0,
582 					MSR_IA32_SPEC_CTRL,
583 					MSR_TYPE_R | MSR_TYPE_W);
584 
585 	if (pred_cmd)
586 		nested_vmx_disable_intercept_for_msr(
587 					msr_bitmap_l1, msr_bitmap_l0,
588 					MSR_IA32_PRED_CMD,
589 					MSR_TYPE_W);
590 
591 	kunmap(page);
592 	kvm_release_page_clean(page);
593 
594 	return true;
595 }
596 
597 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
598 				       struct vmcs12 *vmcs12)
599 {
600 	struct vmcs12 *shadow;
601 	struct page *page;
602 
603 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
604 	    vmcs12->vmcs_link_pointer == -1ull)
605 		return;
606 
607 	shadow = get_shadow_vmcs12(vcpu);
608 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->vmcs_link_pointer);
609 
610 	memcpy(shadow, kmap(page), VMCS12_SIZE);
611 
612 	kunmap(page);
613 	kvm_release_page_clean(page);
614 }
615 
616 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
617 					      struct vmcs12 *vmcs12)
618 {
619 	struct vcpu_vmx *vmx = to_vmx(vcpu);
620 
621 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
622 	    vmcs12->vmcs_link_pointer == -1ull)
623 		return;
624 
625 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
626 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
627 }
628 
629 /*
630  * In nested virtualization, check if L1 has set
631  * VM_EXIT_ACK_INTR_ON_EXIT
632  */
633 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
634 {
635 	return get_vmcs12(vcpu)->vm_exit_controls &
636 		VM_EXIT_ACK_INTR_ON_EXIT;
637 }
638 
639 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
640 {
641 	return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
642 }
643 
644 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
645 					  struct vmcs12 *vmcs12)
646 {
647 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
648 	    !page_address_valid(vcpu, vmcs12->apic_access_addr))
649 		return -EINVAL;
650 	else
651 		return 0;
652 }
653 
654 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
655 					   struct vmcs12 *vmcs12)
656 {
657 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
658 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
659 	    !nested_cpu_has_vid(vmcs12) &&
660 	    !nested_cpu_has_posted_intr(vmcs12))
661 		return 0;
662 
663 	/*
664 	 * If virtualize x2apic mode is enabled,
665 	 * virtualize apic access must be disabled.
666 	 */
667 	if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
668 	    nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
669 		return -EINVAL;
670 
671 	/*
672 	 * If virtual interrupt delivery is enabled,
673 	 * we must exit on external interrupts.
674 	 */
675 	if (nested_cpu_has_vid(vmcs12) &&
676 	   !nested_exit_on_intr(vcpu))
677 		return -EINVAL;
678 
679 	/*
680 	 * bits 15:8 should be zero in posted_intr_nv,
681 	 * the descriptor address has been already checked
682 	 * in nested_get_vmcs12_pages.
683 	 *
684 	 * bits 5:0 of posted_intr_desc_addr should be zero.
685 	 */
686 	if (nested_cpu_has_posted_intr(vmcs12) &&
687 	   (!nested_cpu_has_vid(vmcs12) ||
688 	    !nested_exit_intr_ack_set(vcpu) ||
689 	    (vmcs12->posted_intr_nv & 0xff00) ||
690 	    (vmcs12->posted_intr_desc_addr & 0x3f) ||
691 	    (vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu))))
692 		return -EINVAL;
693 
694 	/* tpr shadow is needed by all apicv features. */
695 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
696 		return -EINVAL;
697 
698 	return 0;
699 }
700 
701 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
702 				       u32 count, u64 addr)
703 {
704 	int maxphyaddr;
705 
706 	if (count == 0)
707 		return 0;
708 	maxphyaddr = cpuid_maxphyaddr(vcpu);
709 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
710 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
711 		return -EINVAL;
712 
713 	return 0;
714 }
715 
716 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
717 						     struct vmcs12 *vmcs12)
718 {
719 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_load_count,
720 					vmcs12->vm_exit_msr_load_addr) ||
721 	    nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_store_count,
722 					vmcs12->vm_exit_msr_store_addr))
723 		return -EINVAL;
724 
725 	return 0;
726 }
727 
728 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
729                                                       struct vmcs12 *vmcs12)
730 {
731 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_entry_msr_load_count,
732                                         vmcs12->vm_entry_msr_load_addr))
733                 return -EINVAL;
734 
735 	return 0;
736 }
737 
738 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
739 					 struct vmcs12 *vmcs12)
740 {
741 	if (!nested_cpu_has_pml(vmcs12))
742 		return 0;
743 
744 	if (!nested_cpu_has_ept(vmcs12) ||
745 	    !page_address_valid(vcpu, vmcs12->pml_address))
746 		return -EINVAL;
747 
748 	return 0;
749 }
750 
751 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
752 							struct vmcs12 *vmcs12)
753 {
754 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
755 	    !nested_cpu_has_ept(vmcs12))
756 		return -EINVAL;
757 	return 0;
758 }
759 
760 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
761 							 struct vmcs12 *vmcs12)
762 {
763 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
764 	    !nested_cpu_has_ept(vmcs12))
765 		return -EINVAL;
766 	return 0;
767 }
768 
769 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
770 						 struct vmcs12 *vmcs12)
771 {
772 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
773 		return 0;
774 
775 	if (!page_address_valid(vcpu, vmcs12->vmread_bitmap) ||
776 	    !page_address_valid(vcpu, vmcs12->vmwrite_bitmap))
777 		return -EINVAL;
778 
779 	return 0;
780 }
781 
782 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
783 				       struct vmx_msr_entry *e)
784 {
785 	/* x2APIC MSR accesses are not allowed */
786 	if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
787 		return -EINVAL;
788 	if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
789 	    e->index == MSR_IA32_UCODE_REV)
790 		return -EINVAL;
791 	if (e->reserved != 0)
792 		return -EINVAL;
793 	return 0;
794 }
795 
796 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
797 				     struct vmx_msr_entry *e)
798 {
799 	if (e->index == MSR_FS_BASE ||
800 	    e->index == MSR_GS_BASE ||
801 	    e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
802 	    nested_vmx_msr_check_common(vcpu, e))
803 		return -EINVAL;
804 	return 0;
805 }
806 
807 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
808 				      struct vmx_msr_entry *e)
809 {
810 	if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
811 	    nested_vmx_msr_check_common(vcpu, e))
812 		return -EINVAL;
813 	return 0;
814 }
815 
816 /*
817  * Load guest's/host's msr at nested entry/exit.
818  * return 0 for success, entry index for failure.
819  */
820 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
821 {
822 	u32 i;
823 	struct vmx_msr_entry e;
824 	struct msr_data msr;
825 
826 	msr.host_initiated = false;
827 	for (i = 0; i < count; i++) {
828 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
829 					&e, sizeof(e))) {
830 			pr_debug_ratelimited(
831 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
832 				__func__, i, gpa + i * sizeof(e));
833 			goto fail;
834 		}
835 		if (nested_vmx_load_msr_check(vcpu, &e)) {
836 			pr_debug_ratelimited(
837 				"%s check failed (%u, 0x%x, 0x%x)\n",
838 				__func__, i, e.index, e.reserved);
839 			goto fail;
840 		}
841 		msr.index = e.index;
842 		msr.data = e.value;
843 		if (kvm_set_msr(vcpu, &msr)) {
844 			pr_debug_ratelimited(
845 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
846 				__func__, i, e.index, e.value);
847 			goto fail;
848 		}
849 	}
850 	return 0;
851 fail:
852 	return i + 1;
853 }
854 
855 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
856 {
857 	u32 i;
858 	struct vmx_msr_entry e;
859 
860 	for (i = 0; i < count; i++) {
861 		struct msr_data msr_info;
862 		if (kvm_vcpu_read_guest(vcpu,
863 					gpa + i * sizeof(e),
864 					&e, 2 * sizeof(u32))) {
865 			pr_debug_ratelimited(
866 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
867 				__func__, i, gpa + i * sizeof(e));
868 			return -EINVAL;
869 		}
870 		if (nested_vmx_store_msr_check(vcpu, &e)) {
871 			pr_debug_ratelimited(
872 				"%s check failed (%u, 0x%x, 0x%x)\n",
873 				__func__, i, e.index, e.reserved);
874 			return -EINVAL;
875 		}
876 		msr_info.host_initiated = false;
877 		msr_info.index = e.index;
878 		if (kvm_get_msr(vcpu, &msr_info)) {
879 			pr_debug_ratelimited(
880 				"%s cannot read MSR (%u, 0x%x)\n",
881 				__func__, i, e.index);
882 			return -EINVAL;
883 		}
884 		if (kvm_vcpu_write_guest(vcpu,
885 					 gpa + i * sizeof(e) +
886 					     offsetof(struct vmx_msr_entry, value),
887 					 &msr_info.data, sizeof(msr_info.data))) {
888 			pr_debug_ratelimited(
889 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
890 				__func__, i, e.index, msr_info.data);
891 			return -EINVAL;
892 		}
893 	}
894 	return 0;
895 }
896 
897 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
898 {
899 	unsigned long invalid_mask;
900 
901 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
902 	return (val & invalid_mask) == 0;
903 }
904 
905 /*
906  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
907  * emulating VM entry into a guest with EPT enabled.
908  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
909  * is assigned to entry_failure_code on failure.
910  */
911 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
912 			       u32 *entry_failure_code)
913 {
914 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
915 		if (!nested_cr3_valid(vcpu, cr3)) {
916 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
917 			return 1;
918 		}
919 
920 		/*
921 		 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
922 		 * must not be dereferenced.
923 		 */
924 		if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
925 		    !nested_ept) {
926 			if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
927 				*entry_failure_code = ENTRY_FAIL_PDPTE;
928 				return 1;
929 			}
930 		}
931 	}
932 
933 	if (!nested_ept)
934 		kvm_mmu_new_cr3(vcpu, cr3, false);
935 
936 	vcpu->arch.cr3 = cr3;
937 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
938 
939 	kvm_init_mmu(vcpu, false);
940 
941 	return 0;
942 }
943 
944 /*
945  * Returns if KVM is able to config CPU to tag TLB entries
946  * populated by L2 differently than TLB entries populated
947  * by L1.
948  *
949  * If L1 uses EPT, then TLB entries are tagged with different EPTP.
950  *
951  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
952  * with different VPID (L1 entries are tagged with vmx->vpid
953  * while L2 entries are tagged with vmx->nested.vpid02).
954  */
955 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
956 {
957 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
958 
959 	return nested_cpu_has_ept(vmcs12) ||
960 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
961 }
962 
963 static u16 nested_get_vpid02(struct kvm_vcpu *vcpu)
964 {
965 	struct vcpu_vmx *vmx = to_vmx(vcpu);
966 
967 	return vmx->nested.vpid02 ? vmx->nested.vpid02 : vmx->vpid;
968 }
969 
970 
971 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
972 {
973 	return fixed_bits_valid(control, low, high);
974 }
975 
976 static inline u64 vmx_control_msr(u32 low, u32 high)
977 {
978 	return low | ((u64)high << 32);
979 }
980 
981 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
982 {
983 	superset &= mask;
984 	subset &= mask;
985 
986 	return (superset | subset) == superset;
987 }
988 
989 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
990 {
991 	const u64 feature_and_reserved =
992 		/* feature (except bit 48; see below) */
993 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
994 		/* reserved */
995 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
996 	u64 vmx_basic = vmx->nested.msrs.basic;
997 
998 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
999 		return -EINVAL;
1000 
1001 	/*
1002 	 * KVM does not emulate a version of VMX that constrains physical
1003 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1004 	 */
1005 	if (data & BIT_ULL(48))
1006 		return -EINVAL;
1007 
1008 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1009 	    vmx_basic_vmcs_revision_id(data))
1010 		return -EINVAL;
1011 
1012 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1013 		return -EINVAL;
1014 
1015 	vmx->nested.msrs.basic = data;
1016 	return 0;
1017 }
1018 
1019 static int
1020 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1021 {
1022 	u64 supported;
1023 	u32 *lowp, *highp;
1024 
1025 	switch (msr_index) {
1026 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1027 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1028 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1029 		break;
1030 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1031 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1032 		highp = &vmx->nested.msrs.procbased_ctls_high;
1033 		break;
1034 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1035 		lowp = &vmx->nested.msrs.exit_ctls_low;
1036 		highp = &vmx->nested.msrs.exit_ctls_high;
1037 		break;
1038 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1039 		lowp = &vmx->nested.msrs.entry_ctls_low;
1040 		highp = &vmx->nested.msrs.entry_ctls_high;
1041 		break;
1042 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1043 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1044 		highp = &vmx->nested.msrs.secondary_ctls_high;
1045 		break;
1046 	default:
1047 		BUG();
1048 	}
1049 
1050 	supported = vmx_control_msr(*lowp, *highp);
1051 
1052 	/* Check must-be-1 bits are still 1. */
1053 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1054 		return -EINVAL;
1055 
1056 	/* Check must-be-0 bits are still 0. */
1057 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1058 		return -EINVAL;
1059 
1060 	*lowp = data;
1061 	*highp = data >> 32;
1062 	return 0;
1063 }
1064 
1065 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1066 {
1067 	const u64 feature_and_reserved_bits =
1068 		/* feature */
1069 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1070 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1071 		/* reserved */
1072 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1073 	u64 vmx_misc;
1074 
1075 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1076 				   vmx->nested.msrs.misc_high);
1077 
1078 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1079 		return -EINVAL;
1080 
1081 	if ((vmx->nested.msrs.pinbased_ctls_high &
1082 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1083 	    vmx_misc_preemption_timer_rate(data) !=
1084 	    vmx_misc_preemption_timer_rate(vmx_misc))
1085 		return -EINVAL;
1086 
1087 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1088 		return -EINVAL;
1089 
1090 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1091 		return -EINVAL;
1092 
1093 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1094 		return -EINVAL;
1095 
1096 	vmx->nested.msrs.misc_low = data;
1097 	vmx->nested.msrs.misc_high = data >> 32;
1098 
1099 	/*
1100 	 * If L1 has read-only VM-exit information fields, use the
1101 	 * less permissive vmx_vmwrite_bitmap to specify write
1102 	 * permissions for the shadow VMCS.
1103 	 */
1104 	if (enable_shadow_vmcs && !nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
1105 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
1106 
1107 	return 0;
1108 }
1109 
1110 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1111 {
1112 	u64 vmx_ept_vpid_cap;
1113 
1114 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1115 					   vmx->nested.msrs.vpid_caps);
1116 
1117 	/* Every bit is either reserved or a feature bit. */
1118 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1119 		return -EINVAL;
1120 
1121 	vmx->nested.msrs.ept_caps = data;
1122 	vmx->nested.msrs.vpid_caps = data >> 32;
1123 	return 0;
1124 }
1125 
1126 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1127 {
1128 	u64 *msr;
1129 
1130 	switch (msr_index) {
1131 	case MSR_IA32_VMX_CR0_FIXED0:
1132 		msr = &vmx->nested.msrs.cr0_fixed0;
1133 		break;
1134 	case MSR_IA32_VMX_CR4_FIXED0:
1135 		msr = &vmx->nested.msrs.cr4_fixed0;
1136 		break;
1137 	default:
1138 		BUG();
1139 	}
1140 
1141 	/*
1142 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1143 	 * must be 1 in the restored value.
1144 	 */
1145 	if (!is_bitwise_subset(data, *msr, -1ULL))
1146 		return -EINVAL;
1147 
1148 	*msr = data;
1149 	return 0;
1150 }
1151 
1152 /*
1153  * Called when userspace is restoring VMX MSRs.
1154  *
1155  * Returns 0 on success, non-0 otherwise.
1156  */
1157 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1158 {
1159 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1160 
1161 	/*
1162 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1163 	 * is in VMX operation.
1164 	 */
1165 	if (vmx->nested.vmxon)
1166 		return -EBUSY;
1167 
1168 	switch (msr_index) {
1169 	case MSR_IA32_VMX_BASIC:
1170 		return vmx_restore_vmx_basic(vmx, data);
1171 	case MSR_IA32_VMX_PINBASED_CTLS:
1172 	case MSR_IA32_VMX_PROCBASED_CTLS:
1173 	case MSR_IA32_VMX_EXIT_CTLS:
1174 	case MSR_IA32_VMX_ENTRY_CTLS:
1175 		/*
1176 		 * The "non-true" VMX capability MSRs are generated from the
1177 		 * "true" MSRs, so we do not support restoring them directly.
1178 		 *
1179 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1180 		 * should restore the "true" MSRs with the must-be-1 bits
1181 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1182 		 * DEFAULT SETTINGS".
1183 		 */
1184 		return -EINVAL;
1185 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1186 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1187 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1188 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1189 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1190 		return vmx_restore_control_msr(vmx, msr_index, data);
1191 	case MSR_IA32_VMX_MISC:
1192 		return vmx_restore_vmx_misc(vmx, data);
1193 	case MSR_IA32_VMX_CR0_FIXED0:
1194 	case MSR_IA32_VMX_CR4_FIXED0:
1195 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1196 	case MSR_IA32_VMX_CR0_FIXED1:
1197 	case MSR_IA32_VMX_CR4_FIXED1:
1198 		/*
1199 		 * These MSRs are generated based on the vCPU's CPUID, so we
1200 		 * do not support restoring them directly.
1201 		 */
1202 		return -EINVAL;
1203 	case MSR_IA32_VMX_EPT_VPID_CAP:
1204 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1205 	case MSR_IA32_VMX_VMCS_ENUM:
1206 		vmx->nested.msrs.vmcs_enum = data;
1207 		return 0;
1208 	default:
1209 		/*
1210 		 * The rest of the VMX capability MSRs do not support restore.
1211 		 */
1212 		return -EINVAL;
1213 	}
1214 }
1215 
1216 /* Returns 0 on success, non-0 otherwise. */
1217 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1218 {
1219 	switch (msr_index) {
1220 	case MSR_IA32_VMX_BASIC:
1221 		*pdata = msrs->basic;
1222 		break;
1223 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1224 	case MSR_IA32_VMX_PINBASED_CTLS:
1225 		*pdata = vmx_control_msr(
1226 			msrs->pinbased_ctls_low,
1227 			msrs->pinbased_ctls_high);
1228 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1229 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1230 		break;
1231 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1232 	case MSR_IA32_VMX_PROCBASED_CTLS:
1233 		*pdata = vmx_control_msr(
1234 			msrs->procbased_ctls_low,
1235 			msrs->procbased_ctls_high);
1236 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1237 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1238 		break;
1239 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1240 	case MSR_IA32_VMX_EXIT_CTLS:
1241 		*pdata = vmx_control_msr(
1242 			msrs->exit_ctls_low,
1243 			msrs->exit_ctls_high);
1244 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1245 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1246 		break;
1247 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1248 	case MSR_IA32_VMX_ENTRY_CTLS:
1249 		*pdata = vmx_control_msr(
1250 			msrs->entry_ctls_low,
1251 			msrs->entry_ctls_high);
1252 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1253 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1254 		break;
1255 	case MSR_IA32_VMX_MISC:
1256 		*pdata = vmx_control_msr(
1257 			msrs->misc_low,
1258 			msrs->misc_high);
1259 		break;
1260 	case MSR_IA32_VMX_CR0_FIXED0:
1261 		*pdata = msrs->cr0_fixed0;
1262 		break;
1263 	case MSR_IA32_VMX_CR0_FIXED1:
1264 		*pdata = msrs->cr0_fixed1;
1265 		break;
1266 	case MSR_IA32_VMX_CR4_FIXED0:
1267 		*pdata = msrs->cr4_fixed0;
1268 		break;
1269 	case MSR_IA32_VMX_CR4_FIXED1:
1270 		*pdata = msrs->cr4_fixed1;
1271 		break;
1272 	case MSR_IA32_VMX_VMCS_ENUM:
1273 		*pdata = msrs->vmcs_enum;
1274 		break;
1275 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1276 		*pdata = vmx_control_msr(
1277 			msrs->secondary_ctls_low,
1278 			msrs->secondary_ctls_high);
1279 		break;
1280 	case MSR_IA32_VMX_EPT_VPID_CAP:
1281 		*pdata = msrs->ept_caps |
1282 			((u64)msrs->vpid_caps << 32);
1283 		break;
1284 	case MSR_IA32_VMX_VMFUNC:
1285 		*pdata = msrs->vmfunc_controls;
1286 		break;
1287 	default:
1288 		return 1;
1289 	}
1290 
1291 	return 0;
1292 }
1293 
1294 /*
1295  * Copy the writable VMCS shadow fields back to the VMCS12, in case
1296  * they have been modified by the L1 guest. Note that the "read-only"
1297  * VM-exit information fields are actually writable if the vCPU is
1298  * configured to support "VMWRITE to any supported field in the VMCS."
1299  */
1300 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1301 {
1302 	const u16 *fields[] = {
1303 		shadow_read_write_fields,
1304 		shadow_read_only_fields
1305 	};
1306 	const int max_fields[] = {
1307 		max_shadow_read_write_fields,
1308 		max_shadow_read_only_fields
1309 	};
1310 	int i, q;
1311 	unsigned long field;
1312 	u64 field_value;
1313 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1314 
1315 	preempt_disable();
1316 
1317 	vmcs_load(shadow_vmcs);
1318 
1319 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1320 		for (i = 0; i < max_fields[q]; i++) {
1321 			field = fields[q][i];
1322 			field_value = __vmcs_readl(field);
1323 			vmcs12_write_any(get_vmcs12(&vmx->vcpu), field, field_value);
1324 		}
1325 		/*
1326 		 * Skip the VM-exit information fields if they are read-only.
1327 		 */
1328 		if (!nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
1329 			break;
1330 	}
1331 
1332 	vmcs_clear(shadow_vmcs);
1333 	vmcs_load(vmx->loaded_vmcs->vmcs);
1334 
1335 	preempt_enable();
1336 }
1337 
1338 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1339 {
1340 	const u16 *fields[] = {
1341 		shadow_read_write_fields,
1342 		shadow_read_only_fields
1343 	};
1344 	const int max_fields[] = {
1345 		max_shadow_read_write_fields,
1346 		max_shadow_read_only_fields
1347 	};
1348 	int i, q;
1349 	unsigned long field;
1350 	u64 field_value = 0;
1351 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1352 
1353 	vmcs_load(shadow_vmcs);
1354 
1355 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1356 		for (i = 0; i < max_fields[q]; i++) {
1357 			field = fields[q][i];
1358 			vmcs12_read_any(get_vmcs12(&vmx->vcpu), field, &field_value);
1359 			__vmcs_writel(field, field_value);
1360 		}
1361 	}
1362 
1363 	vmcs_clear(shadow_vmcs);
1364 	vmcs_load(vmx->loaded_vmcs->vmcs);
1365 }
1366 
1367 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1368 {
1369 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1370 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1371 
1372 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1373 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1374 	vmcs12->guest_rip = evmcs->guest_rip;
1375 
1376 	if (unlikely(!(evmcs->hv_clean_fields &
1377 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1378 		vmcs12->guest_rsp = evmcs->guest_rsp;
1379 		vmcs12->guest_rflags = evmcs->guest_rflags;
1380 		vmcs12->guest_interruptibility_info =
1381 			evmcs->guest_interruptibility_info;
1382 	}
1383 
1384 	if (unlikely(!(evmcs->hv_clean_fields &
1385 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1386 		vmcs12->cpu_based_vm_exec_control =
1387 			evmcs->cpu_based_vm_exec_control;
1388 	}
1389 
1390 	if (unlikely(!(evmcs->hv_clean_fields &
1391 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1392 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1393 	}
1394 
1395 	if (unlikely(!(evmcs->hv_clean_fields &
1396 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1397 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1398 	}
1399 
1400 	if (unlikely(!(evmcs->hv_clean_fields &
1401 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1402 		vmcs12->vm_entry_intr_info_field =
1403 			evmcs->vm_entry_intr_info_field;
1404 		vmcs12->vm_entry_exception_error_code =
1405 			evmcs->vm_entry_exception_error_code;
1406 		vmcs12->vm_entry_instruction_len =
1407 			evmcs->vm_entry_instruction_len;
1408 	}
1409 
1410 	if (unlikely(!(evmcs->hv_clean_fields &
1411 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1412 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1413 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1414 		vmcs12->host_cr0 = evmcs->host_cr0;
1415 		vmcs12->host_cr3 = evmcs->host_cr3;
1416 		vmcs12->host_cr4 = evmcs->host_cr4;
1417 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1418 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1419 		vmcs12->host_rip = evmcs->host_rip;
1420 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1421 		vmcs12->host_es_selector = evmcs->host_es_selector;
1422 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1423 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1424 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1425 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1426 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1427 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1428 	}
1429 
1430 	if (unlikely(!(evmcs->hv_clean_fields &
1431 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1432 		vmcs12->pin_based_vm_exec_control =
1433 			evmcs->pin_based_vm_exec_control;
1434 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1435 		vmcs12->secondary_vm_exec_control =
1436 			evmcs->secondary_vm_exec_control;
1437 	}
1438 
1439 	if (unlikely(!(evmcs->hv_clean_fields &
1440 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1441 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1442 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1443 	}
1444 
1445 	if (unlikely(!(evmcs->hv_clean_fields &
1446 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1447 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1448 	}
1449 
1450 	if (unlikely(!(evmcs->hv_clean_fields &
1451 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1452 		vmcs12->guest_es_base = evmcs->guest_es_base;
1453 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1454 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1455 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1456 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1457 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1458 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1459 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1460 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1461 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1462 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1463 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1464 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1465 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1466 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1467 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1468 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1469 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1470 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1471 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1472 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1473 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1474 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1475 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1476 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1477 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1478 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1479 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1480 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1481 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1482 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1483 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1484 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1485 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1486 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1487 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1488 	}
1489 
1490 	if (unlikely(!(evmcs->hv_clean_fields &
1491 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1492 		vmcs12->tsc_offset = evmcs->tsc_offset;
1493 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1494 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1495 	}
1496 
1497 	if (unlikely(!(evmcs->hv_clean_fields &
1498 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1499 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1500 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1501 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1502 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1503 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1504 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1505 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1506 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1507 	}
1508 
1509 	if (unlikely(!(evmcs->hv_clean_fields &
1510 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1511 		vmcs12->host_fs_base = evmcs->host_fs_base;
1512 		vmcs12->host_gs_base = evmcs->host_gs_base;
1513 		vmcs12->host_tr_base = evmcs->host_tr_base;
1514 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1515 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1516 		vmcs12->host_rsp = evmcs->host_rsp;
1517 	}
1518 
1519 	if (unlikely(!(evmcs->hv_clean_fields &
1520 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1521 		vmcs12->ept_pointer = evmcs->ept_pointer;
1522 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1523 	}
1524 
1525 	if (unlikely(!(evmcs->hv_clean_fields &
1526 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1527 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1528 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1529 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1530 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1531 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1532 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1533 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1534 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1535 		vmcs12->guest_pending_dbg_exceptions =
1536 			evmcs->guest_pending_dbg_exceptions;
1537 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1538 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1539 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1540 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1541 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1542 	}
1543 
1544 	/*
1545 	 * Not used?
1546 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1547 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1548 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1549 	 * vmcs12->cr3_target_value0 = evmcs->cr3_target_value0;
1550 	 * vmcs12->cr3_target_value1 = evmcs->cr3_target_value1;
1551 	 * vmcs12->cr3_target_value2 = evmcs->cr3_target_value2;
1552 	 * vmcs12->cr3_target_value3 = evmcs->cr3_target_value3;
1553 	 * vmcs12->page_fault_error_code_mask =
1554 	 *		evmcs->page_fault_error_code_mask;
1555 	 * vmcs12->page_fault_error_code_match =
1556 	 *		evmcs->page_fault_error_code_match;
1557 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1558 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1559 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1560 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1561 	 */
1562 
1563 	/*
1564 	 * Read only fields:
1565 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1566 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1567 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1568 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1569 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1570 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1571 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1572 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1573 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1574 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1575 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1576 	 *
1577 	 * Not present in struct vmcs12:
1578 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1579 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1580 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1581 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1582 	 */
1583 
1584 	return 0;
1585 }
1586 
1587 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1588 {
1589 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1590 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1591 
1592 	/*
1593 	 * Should not be changed by KVM:
1594 	 *
1595 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1596 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1597 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1598 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1599 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1600 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1601 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1602 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1603 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1604 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1605 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1606 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1607 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1608 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1609 	 * evmcs->host_rip = vmcs12->host_rip;
1610 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1611 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1612 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1613 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1614 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1615 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1616 	 * evmcs->host_rsp = vmcs12->host_rsp;
1617 	 * sync_vmcs12() doesn't read these:
1618 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1619 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1620 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1621 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1622 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1623 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1624 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1625 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1626 	 * evmcs->cr3_target_value0 = vmcs12->cr3_target_value0;
1627 	 * evmcs->cr3_target_value1 = vmcs12->cr3_target_value1;
1628 	 * evmcs->cr3_target_value2 = vmcs12->cr3_target_value2;
1629 	 * evmcs->cr3_target_value3 = vmcs12->cr3_target_value3;
1630 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1631 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1632 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1633 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1634 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1635 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1636 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1637 	 * evmcs->page_fault_error_code_mask =
1638 	 *		vmcs12->page_fault_error_code_mask;
1639 	 * evmcs->page_fault_error_code_match =
1640 	 *		vmcs12->page_fault_error_code_match;
1641 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1642 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1643 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1644 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1645 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1646 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1647 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1648 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1649 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1650 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1651 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1652 	 *
1653 	 * Not present in struct vmcs12:
1654 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1655 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1656 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1657 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1658 	 */
1659 
1660 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1661 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1662 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1663 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1664 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1665 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1666 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1667 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1668 
1669 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1670 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1671 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1672 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1673 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1674 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1675 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1676 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1677 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1678 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1679 
1680 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1681 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1682 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1683 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1684 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1685 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1686 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1687 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1688 
1689 	evmcs->guest_es_base = vmcs12->guest_es_base;
1690 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1691 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1692 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1693 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1694 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1695 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1696 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1697 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1698 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1699 
1700 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1701 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1702 
1703 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1704 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1705 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1706 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1707 
1708 	evmcs->guest_pending_dbg_exceptions =
1709 		vmcs12->guest_pending_dbg_exceptions;
1710 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1711 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1712 
1713 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1714 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1715 
1716 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1717 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1718 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1719 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1720 
1721 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1722 
1723 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1724 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1725 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1726 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1727 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1728 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1729 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1730 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1731 
1732 	evmcs->exit_qualification = vmcs12->exit_qualification;
1733 
1734 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1735 	evmcs->guest_rsp = vmcs12->guest_rsp;
1736 	evmcs->guest_rflags = vmcs12->guest_rflags;
1737 
1738 	evmcs->guest_interruptibility_info =
1739 		vmcs12->guest_interruptibility_info;
1740 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1741 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1742 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1743 	evmcs->vm_entry_exception_error_code =
1744 		vmcs12->vm_entry_exception_error_code;
1745 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1746 
1747 	evmcs->guest_rip = vmcs12->guest_rip;
1748 
1749 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1750 
1751 	return 0;
1752 }
1753 
1754 /*
1755  * This is an equivalent of the nested hypervisor executing the vmptrld
1756  * instruction.
1757  */
1758 static int nested_vmx_handle_enlightened_vmptrld(struct kvm_vcpu *vcpu,
1759 						 bool from_launch)
1760 {
1761 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1762 	struct hv_vp_assist_page assist_page;
1763 
1764 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1765 		return 1;
1766 
1767 	if (unlikely(!kvm_hv_get_assist_page(vcpu, &assist_page)))
1768 		return 1;
1769 
1770 	if (unlikely(!assist_page.enlighten_vmentry))
1771 		return 1;
1772 
1773 	if (unlikely(assist_page.current_nested_vmcs !=
1774 		     vmx->nested.hv_evmcs_vmptr)) {
1775 
1776 		if (!vmx->nested.hv_evmcs)
1777 			vmx->nested.current_vmptr = -1ull;
1778 
1779 		nested_release_evmcs(vcpu);
1780 
1781 		vmx->nested.hv_evmcs_page = kvm_vcpu_gpa_to_page(
1782 			vcpu, assist_page.current_nested_vmcs);
1783 
1784 		if (unlikely(is_error_page(vmx->nested.hv_evmcs_page)))
1785 			return 0;
1786 
1787 		vmx->nested.hv_evmcs = kmap(vmx->nested.hv_evmcs_page);
1788 
1789 		/*
1790 		 * Currently, KVM only supports eVMCS version 1
1791 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1792 		 * value to first u32 field of eVMCS which should specify eVMCS
1793 		 * VersionNumber.
1794 		 *
1795 		 * Guest should be aware of supported eVMCS versions by host by
1796 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1797 		 * expected to set this CPUID leaf according to the value
1798 		 * returned in vmcs_version from nested_enable_evmcs().
1799 		 *
1800 		 * However, it turns out that Microsoft Hyper-V fails to comply
1801 		 * to their own invented interface: When Hyper-V use eVMCS, it
1802 		 * just sets first u32 field of eVMCS to revision_id specified
1803 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1804 		 * which is one of the supported versions specified in
1805 		 * CPUID.0x4000000A.EAX[0:15].
1806 		 *
1807 		 * To overcome Hyper-V bug, we accept here either a supported
1808 		 * eVMCS version or VMCS12 revision_id as valid values for first
1809 		 * u32 field of eVMCS.
1810 		 */
1811 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1812 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1813 			nested_release_evmcs(vcpu);
1814 			return 0;
1815 		}
1816 
1817 		vmx->nested.dirty_vmcs12 = true;
1818 		/*
1819 		 * As we keep L2 state for one guest only 'hv_clean_fields' mask
1820 		 * can't be used when we switch between them. Reset it here for
1821 		 * simplicity.
1822 		 */
1823 		vmx->nested.hv_evmcs->hv_clean_fields &=
1824 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1825 		vmx->nested.hv_evmcs_vmptr = assist_page.current_nested_vmcs;
1826 
1827 		/*
1828 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
1829 		 * reloaded from guest's memory (read only fields, fields not
1830 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
1831 		 * are no leftovers.
1832 		 */
1833 		if (from_launch) {
1834 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1835 			memset(vmcs12, 0, sizeof(*vmcs12));
1836 			vmcs12->hdr.revision_id = VMCS12_REVISION;
1837 		}
1838 
1839 	}
1840 	return 1;
1841 }
1842 
1843 void nested_sync_from_vmcs12(struct kvm_vcpu *vcpu)
1844 {
1845 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1846 
1847 	/*
1848 	 * hv_evmcs may end up being not mapped after migration (when
1849 	 * L2 was running), map it here to make sure vmcs12 changes are
1850 	 * properly reflected.
1851 	 */
1852 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs)
1853 		nested_vmx_handle_enlightened_vmptrld(vcpu, false);
1854 
1855 	if (vmx->nested.hv_evmcs) {
1856 		copy_vmcs12_to_enlightened(vmx);
1857 		/* All fields are clean */
1858 		vmx->nested.hv_evmcs->hv_clean_fields |=
1859 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1860 	} else {
1861 		copy_vmcs12_to_shadow(vmx);
1862 	}
1863 
1864 	vmx->nested.need_vmcs12_sync = false;
1865 }
1866 
1867 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
1868 {
1869 	struct vcpu_vmx *vmx =
1870 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
1871 
1872 	vmx->nested.preemption_timer_expired = true;
1873 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
1874 	kvm_vcpu_kick(&vmx->vcpu);
1875 
1876 	return HRTIMER_NORESTART;
1877 }
1878 
1879 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
1880 {
1881 	u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
1882 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1883 
1884 	/*
1885 	 * A timer value of zero is architecturally guaranteed to cause
1886 	 * a VMExit prior to executing any instructions in the guest.
1887 	 */
1888 	if (preemption_timeout == 0) {
1889 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
1890 		return;
1891 	}
1892 
1893 	if (vcpu->arch.virtual_tsc_khz == 0)
1894 		return;
1895 
1896 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
1897 	preemption_timeout *= 1000000;
1898 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
1899 	hrtimer_start(&vmx->nested.preemption_timer,
1900 		      ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
1901 }
1902 
1903 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1904 {
1905 	if (vmx->nested.nested_run_pending &&
1906 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
1907 		return vmcs12->guest_ia32_efer;
1908 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
1909 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
1910 	else
1911 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
1912 }
1913 
1914 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
1915 {
1916 	/*
1917 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
1918 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
1919 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
1920 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
1921 	 */
1922 	if (vmx->nested.vmcs02_initialized)
1923 		return;
1924 	vmx->nested.vmcs02_initialized = true;
1925 
1926 	/*
1927 	 * We don't care what the EPTP value is we just need to guarantee
1928 	 * it's valid so we don't get a false positive when doing early
1929 	 * consistency checks.
1930 	 */
1931 	if (enable_ept && nested_early_check)
1932 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
1933 
1934 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
1935 	if (cpu_has_vmx_vmfunc())
1936 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
1937 
1938 	if (cpu_has_vmx_posted_intr())
1939 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
1940 
1941 	if (cpu_has_vmx_msr_bitmap())
1942 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
1943 
1944 	if (enable_pml)
1945 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
1946 
1947 	/*
1948 	 * Set the MSR load/store lists to match L0's settings.  Only the
1949 	 * addresses are constant (for vmcs02), the counts can change based
1950 	 * on L2's behavior, e.g. switching to/from long mode.
1951 	 */
1952 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1953 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
1954 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
1955 
1956 	vmx_set_constant_host_state(vmx);
1957 }
1958 
1959 static void prepare_vmcs02_early_full(struct vcpu_vmx *vmx,
1960 				      struct vmcs12 *vmcs12)
1961 {
1962 	prepare_vmcs02_constant_state(vmx);
1963 
1964 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
1965 
1966 	if (enable_vpid) {
1967 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
1968 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
1969 		else
1970 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
1971 	}
1972 }
1973 
1974 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1975 {
1976 	u32 exec_control, vmcs12_exec_ctrl;
1977 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
1978 
1979 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
1980 		prepare_vmcs02_early_full(vmx, vmcs12);
1981 
1982 	/*
1983 	 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
1984 	 * entry, but only if the current (host) sp changed from the value
1985 	 * we wrote last (vmx->host_rsp).  This cache is no longer relevant
1986 	 * if we switch vmcs, and rather than hold a separate cache per vmcs,
1987 	 * here we just force the write to happen on entry.  host_rsp will
1988 	 * also be written unconditionally by nested_vmx_check_vmentry_hw()
1989 	 * if we are doing early consistency checks via hardware.
1990 	 */
1991 	vmx->host_rsp = 0;
1992 
1993 	/*
1994 	 * PIN CONTROLS
1995 	 */
1996 	exec_control = vmcs12->pin_based_vm_exec_control;
1997 
1998 	/* Preemption timer setting is computed directly in vmx_vcpu_run.  */
1999 	exec_control |= vmcs_config.pin_based_exec_ctrl;
2000 	exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2001 	vmx->loaded_vmcs->hv_timer_armed = false;
2002 
2003 	/* Posted interrupts setting is only taken from vmcs12.  */
2004 	if (nested_cpu_has_posted_intr(vmcs12)) {
2005 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2006 		vmx->nested.pi_pending = false;
2007 	} else {
2008 		exec_control &= ~PIN_BASED_POSTED_INTR;
2009 	}
2010 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
2011 
2012 	/*
2013 	 * EXEC CONTROLS
2014 	 */
2015 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2016 	exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2017 	exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2018 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2019 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2020 
2021 	/*
2022 	 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
2023 	 * nested_get_vmcs12_pages can't fix it up, the illegal value
2024 	 * will result in a VM entry failure.
2025 	 */
2026 	if (exec_control & CPU_BASED_TPR_SHADOW) {
2027 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
2028 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2029 	} else {
2030 #ifdef CONFIG_X86_64
2031 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2032 				CPU_BASED_CR8_STORE_EXITING;
2033 #endif
2034 	}
2035 
2036 	/*
2037 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2038 	 * for I/O port accesses.
2039 	 */
2040 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2041 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2042 	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2043 
2044 	/*
2045 	 * SECONDARY EXEC CONTROLS
2046 	 */
2047 	if (cpu_has_secondary_exec_ctrls()) {
2048 		exec_control = vmx->secondary_exec_control;
2049 
2050 		/* Take the following fields only from vmcs12 */
2051 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2052 				  SECONDARY_EXEC_ENABLE_INVPCID |
2053 				  SECONDARY_EXEC_RDTSCP |
2054 				  SECONDARY_EXEC_XSAVES |
2055 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2056 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2057 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2058 		if (nested_cpu_has(vmcs12,
2059 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2060 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2061 				~SECONDARY_EXEC_ENABLE_PML;
2062 			exec_control |= vmcs12_exec_ctrl;
2063 		}
2064 
2065 		/* VMCS shadowing for L2 is emulated for now */
2066 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2067 
2068 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2069 			vmcs_write16(GUEST_INTR_STATUS,
2070 				vmcs12->guest_intr_status);
2071 
2072 		/*
2073 		 * Write an illegal value to APIC_ACCESS_ADDR. Later,
2074 		 * nested_get_vmcs12_pages will either fix it up or
2075 		 * remove the VM execution control.
2076 		 */
2077 		if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
2078 			vmcs_write64(APIC_ACCESS_ADDR, -1ull);
2079 
2080 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2081 			vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2082 
2083 		vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2084 	}
2085 
2086 	/*
2087 	 * ENTRY CONTROLS
2088 	 *
2089 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2090 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2091 	 * on the related bits (if supported by the CPU) in the hope that
2092 	 * we can avoid VMWrites during vmx_set_efer().
2093 	 */
2094 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2095 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2096 	if (cpu_has_load_ia32_efer()) {
2097 		if (guest_efer & EFER_LMA)
2098 			exec_control |= VM_ENTRY_IA32E_MODE;
2099 		if (guest_efer != host_efer)
2100 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2101 	}
2102 	vm_entry_controls_init(vmx, exec_control);
2103 
2104 	/*
2105 	 * EXIT CONTROLS
2106 	 *
2107 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2108 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2109 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2110 	 */
2111 	exec_control = vmx_vmexit_ctrl();
2112 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2113 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2114 	vm_exit_controls_init(vmx, exec_control);
2115 
2116 	/*
2117 	 * Conceptually we want to copy the PML address and index from
2118 	 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
2119 	 * since we always flush the log on each vmexit and never change
2120 	 * the PML address (once set), this happens to be equivalent to
2121 	 * simply resetting the index in vmcs02.
2122 	 */
2123 	if (enable_pml)
2124 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2125 
2126 	/*
2127 	 * Interrupt/Exception Fields
2128 	 */
2129 	if (vmx->nested.nested_run_pending) {
2130 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2131 			     vmcs12->vm_entry_intr_info_field);
2132 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2133 			     vmcs12->vm_entry_exception_error_code);
2134 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2135 			     vmcs12->vm_entry_instruction_len);
2136 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2137 			     vmcs12->guest_interruptibility_info);
2138 		vmx->loaded_vmcs->nmi_known_unmasked =
2139 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2140 	} else {
2141 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2142 	}
2143 }
2144 
2145 static void prepare_vmcs02_full(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2146 {
2147 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2148 
2149 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2150 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2151 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2152 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2153 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2154 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2155 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2156 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2157 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2158 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2159 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2160 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2161 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2162 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2163 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2164 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2165 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2166 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2167 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2168 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2169 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2170 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2171 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2172 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2173 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2174 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2175 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2176 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2177 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2178 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2179 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2180 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2181 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2182 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2183 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2184 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2185 	}
2186 
2187 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2188 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2189 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2190 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2191 			    vmcs12->guest_pending_dbg_exceptions);
2192 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2193 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2194 
2195 		/*
2196 		 * L1 may access the L2's PDPTR, so save them to construct
2197 		 * vmcs12
2198 		 */
2199 		if (enable_ept) {
2200 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2201 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2202 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2203 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2204 		}
2205 	}
2206 
2207 	if (nested_cpu_has_xsaves(vmcs12))
2208 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2209 
2210 	/*
2211 	 * Whether page-faults are trapped is determined by a combination of
2212 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2213 	 * If enable_ept, L0 doesn't care about page faults and we should
2214 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2215 	 * care about (at least some) page faults, and because it is not easy
2216 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2217 	 * to exit on each and every L2 page fault. This is done by setting
2218 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2219 	 * Note that below we don't need special code to set EB.PF beyond the
2220 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2221 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2222 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2223 	 */
2224 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2225 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2226 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2227 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2228 
2229 	if (cpu_has_vmx_apicv()) {
2230 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2231 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2232 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2233 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2234 	}
2235 
2236 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2237 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2238 
2239 	set_cr4_guest_host_mask(vmx);
2240 
2241 	if (kvm_mpx_supported()) {
2242 		if (vmx->nested.nested_run_pending &&
2243 			(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2244 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2245 		else
2246 			vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2247 	}
2248 }
2249 
2250 /*
2251  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2252  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2253  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2254  * guest in a way that will both be appropriate to L1's requests, and our
2255  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2256  * function also has additional necessary side-effects, like setting various
2257  * vcpu->arch fields.
2258  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2259  * is assigned to entry_failure_code on failure.
2260  */
2261 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2262 			  u32 *entry_failure_code)
2263 {
2264 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2265 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2266 
2267 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs) {
2268 		prepare_vmcs02_full(vmx, vmcs12);
2269 		vmx->nested.dirty_vmcs12 = false;
2270 	}
2271 
2272 	/*
2273 	 * First, the fields that are shadowed.  This must be kept in sync
2274 	 * with vmcs_shadow_fields.h.
2275 	 */
2276 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2277 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2278 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2279 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2280 	}
2281 
2282 	if (vmx->nested.nested_run_pending &&
2283 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2284 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2285 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2286 	} else {
2287 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2288 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2289 	}
2290 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2291 
2292 	vmx->nested.preemption_timer_expired = false;
2293 	if (nested_cpu_has_preemption_timer(vmcs12))
2294 		vmx_start_preemption_timer(vcpu);
2295 
2296 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2297 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2298 	 * trap. Note that CR0.TS also needs updating - we do this later.
2299 	 */
2300 	update_exception_bitmap(vcpu);
2301 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2302 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2303 
2304 	if (vmx->nested.nested_run_pending &&
2305 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2306 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2307 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2308 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2309 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2310 	}
2311 
2312 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2313 
2314 	if (kvm_has_tsc_control)
2315 		decache_tsc_multiplier(vmx);
2316 
2317 	if (enable_vpid) {
2318 		/*
2319 		 * There is no direct mapping between vpid02 and vpid12, the
2320 		 * vpid02 is per-vCPU for L0 and reused while the value of
2321 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2322 		 * The vpid12 is allocated by L1 for L2, so it will not
2323 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2324 		 * even if spawn a lot of nested vCPUs.
2325 		 */
2326 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2327 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2328 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2329 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2330 			}
2331 		} else {
2332 			/*
2333 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2334 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2335 			 * flush until vmcs02->eptp is fully updated by
2336 			 * KVM_REQ_LOAD_CR3. Note that this assumes
2337 			 * KVM_REQ_TLB_FLUSH is evaluated after
2338 			 * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
2339 			 */
2340 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2341 		}
2342 	}
2343 
2344 	if (nested_cpu_has_ept(vmcs12))
2345 		nested_ept_init_mmu_context(vcpu);
2346 	else if (nested_cpu_has2(vmcs12,
2347 				 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2348 		vmx_flush_tlb(vcpu, true);
2349 
2350 	/*
2351 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2352 	 * bits which we consider mandatory enabled.
2353 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2354 	 * the specifications by L1; It's not enough to take
2355 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2356 	 * have more bits than L1 expected.
2357 	 */
2358 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2359 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2360 
2361 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2362 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2363 
2364 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2365 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2366 	vmx_set_efer(vcpu, vcpu->arch.efer);
2367 
2368 	/*
2369 	 * Guest state is invalid and unrestricted guest is disabled,
2370 	 * which means L1 attempted VMEntry to L2 with invalid state.
2371 	 * Fail the VMEntry.
2372 	 */
2373 	if (vmx->emulation_required) {
2374 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2375 		return 1;
2376 	}
2377 
2378 	/* Shadow page tables on either EPT or shadow page tables. */
2379 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2380 				entry_failure_code))
2381 		return 1;
2382 
2383 	if (!enable_ept)
2384 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2385 
2386 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
2387 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
2388 	return 0;
2389 }
2390 
2391 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2392 {
2393 	if (!nested_cpu_has_nmi_exiting(vmcs12) &&
2394 	    nested_cpu_has_virtual_nmis(vmcs12))
2395 		return -EINVAL;
2396 
2397 	if (!nested_cpu_has_virtual_nmis(vmcs12) &&
2398 	    nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING))
2399 		return -EINVAL;
2400 
2401 	return 0;
2402 }
2403 
2404 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
2405 {
2406 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2407 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2408 
2409 	/* Check for memory type validity */
2410 	switch (address & VMX_EPTP_MT_MASK) {
2411 	case VMX_EPTP_MT_UC:
2412 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT))
2413 			return false;
2414 		break;
2415 	case VMX_EPTP_MT_WB:
2416 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT))
2417 			return false;
2418 		break;
2419 	default:
2420 		return false;
2421 	}
2422 
2423 	/* only 4 levels page-walk length are valid */
2424 	if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
2425 		return false;
2426 
2427 	/* Reserved bits should not be set */
2428 	if (address >> maxphyaddr || ((address >> 7) & 0x1f))
2429 		return false;
2430 
2431 	/* AD, if set, should be supported */
2432 	if (address & VMX_EPTP_AD_ENABLE_BIT) {
2433 		if (!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT))
2434 			return false;
2435 	}
2436 
2437 	return true;
2438 }
2439 
2440 /*
2441  * Checks related to VM-Execution Control Fields
2442  */
2443 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2444                                               struct vmcs12 *vmcs12)
2445 {
2446 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2447 
2448 	if (!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2449 				vmx->nested.msrs.pinbased_ctls_low,
2450 				vmx->nested.msrs.pinbased_ctls_high) ||
2451 	    !vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2452 				vmx->nested.msrs.procbased_ctls_low,
2453 				vmx->nested.msrs.procbased_ctls_high))
2454 		return -EINVAL;
2455 
2456 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2457 	    !vmx_control_verify(vmcs12->secondary_vm_exec_control,
2458 				 vmx->nested.msrs.secondary_ctls_low,
2459 				 vmx->nested.msrs.secondary_ctls_high))
2460 		return -EINVAL;
2461 
2462 	if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu) ||
2463 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2464 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2465 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2466 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2467 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2468 	    nested_vmx_check_nmi_controls(vmcs12) ||
2469 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2470 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2471 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2472 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2473 	    (nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2474 		return -EINVAL;
2475 
2476 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2477 	    nested_cpu_has_save_preemption_timer(vmcs12))
2478 		return -EINVAL;
2479 
2480 	if (nested_cpu_has_ept(vmcs12) &&
2481 	    !valid_ept_address(vcpu, vmcs12->ept_pointer))
2482 		return -EINVAL;
2483 
2484 	if (nested_cpu_has_vmfunc(vmcs12)) {
2485 		if (vmcs12->vm_function_control &
2486 		    ~vmx->nested.msrs.vmfunc_controls)
2487 			return -EINVAL;
2488 
2489 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2490 			if (!nested_cpu_has_ept(vmcs12) ||
2491 			    !page_address_valid(vcpu, vmcs12->eptp_list_address))
2492 				return -EINVAL;
2493 		}
2494 	}
2495 
2496 	return 0;
2497 }
2498 
2499 /*
2500  * Checks related to VM-Exit Control Fields
2501  */
2502 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2503                                          struct vmcs12 *vmcs12)
2504 {
2505 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2506 
2507 	if (!vmx_control_verify(vmcs12->vm_exit_controls,
2508 				vmx->nested.msrs.exit_ctls_low,
2509 				vmx->nested.msrs.exit_ctls_high) ||
2510 	    nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12))
2511 		return -EINVAL;
2512 
2513 	return 0;
2514 }
2515 
2516 /*
2517  * Checks related to VM-Entry Control Fields
2518  */
2519 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2520 					  struct vmcs12 *vmcs12)
2521 {
2522 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2523 
2524 	if (!vmx_control_verify(vmcs12->vm_entry_controls,
2525 				vmx->nested.msrs.entry_ctls_low,
2526 				vmx->nested.msrs.entry_ctls_high))
2527 		return -EINVAL;
2528 
2529 	/*
2530 	 * From the Intel SDM, volume 3:
2531 	 * Fields relevant to VM-entry event injection must be set properly.
2532 	 * These fields are the VM-entry interruption-information field, the
2533 	 * VM-entry exception error code, and the VM-entry instruction length.
2534 	 */
2535 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2536 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2537 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2538 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2539 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2540 		bool should_have_error_code;
2541 		bool urg = nested_cpu_has2(vmcs12,
2542 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2543 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2544 
2545 		/* VM-entry interruption-info field: interruption type */
2546 		if (intr_type == INTR_TYPE_RESERVED ||
2547 		    (intr_type == INTR_TYPE_OTHER_EVENT &&
2548 		     !nested_cpu_supports_monitor_trap_flag(vcpu)))
2549 			return -EINVAL;
2550 
2551 		/* VM-entry interruption-info field: vector */
2552 		if ((intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2553 		    (intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2554 		    (intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2555 			return -EINVAL;
2556 
2557 		/* VM-entry interruption-info field: deliver error code */
2558 		should_have_error_code =
2559 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2560 			x86_exception_has_error_code(vector);
2561 		if (has_error_code != should_have_error_code)
2562 			return -EINVAL;
2563 
2564 		/* VM-entry exception error code */
2565 		if (has_error_code &&
2566 		    vmcs12->vm_entry_exception_error_code & GENMASK(31, 15))
2567 			return -EINVAL;
2568 
2569 		/* VM-entry interruption-info field: reserved bits */
2570 		if (intr_info & INTR_INFO_RESVD_BITS_MASK)
2571 			return -EINVAL;
2572 
2573 		/* VM-entry instruction length */
2574 		switch (intr_type) {
2575 		case INTR_TYPE_SOFT_EXCEPTION:
2576 		case INTR_TYPE_SOFT_INTR:
2577 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2578 			if ((vmcs12->vm_entry_instruction_len > 15) ||
2579 			    (vmcs12->vm_entry_instruction_len == 0 &&
2580 			     !nested_cpu_has_zero_length_injection(vcpu)))
2581 				return -EINVAL;
2582 		}
2583 	}
2584 
2585 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2586 		return -EINVAL;
2587 
2588 	return 0;
2589 }
2590 
2591 /*
2592  * Checks related to Host Control Registers and MSRs
2593  */
2594 static int nested_check_host_control_regs(struct kvm_vcpu *vcpu,
2595                                           struct vmcs12 *vmcs12)
2596 {
2597 	bool ia32e;
2598 
2599 	if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
2600 	    !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
2601 	    !nested_cr3_valid(vcpu, vmcs12->host_cr3))
2602 		return -EINVAL;
2603 	/*
2604 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2605 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2606 	 * the values of the LMA and LME bits in the field must each be that of
2607 	 * the host address-space size VM-exit control.
2608 	 */
2609 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2610 		ia32e = (vmcs12->vm_exit_controls &
2611 			 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
2612 		if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
2613 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
2614 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
2615 			return -EINVAL;
2616 	}
2617 
2618 	return 0;
2619 }
2620 
2621 /*
2622  * Checks related to Guest Non-register State
2623  */
2624 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2625 {
2626 	if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2627 	    vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
2628 		return -EINVAL;
2629 
2630 	return 0;
2631 }
2632 
2633 static int nested_vmx_check_vmentry_prereqs(struct kvm_vcpu *vcpu,
2634 					    struct vmcs12 *vmcs12)
2635 {
2636 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2637 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2638 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2639 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2640 
2641 	if (nested_check_host_control_regs(vcpu, vmcs12))
2642 		return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
2643 
2644 	if (nested_check_guest_non_reg_state(vmcs12))
2645 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2646 
2647 	return 0;
2648 }
2649 
2650 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2651 					  struct vmcs12 *vmcs12)
2652 {
2653 	int r;
2654 	struct page *page;
2655 	struct vmcs12 *shadow;
2656 
2657 	if (vmcs12->vmcs_link_pointer == -1ull)
2658 		return 0;
2659 
2660 	if (!page_address_valid(vcpu, vmcs12->vmcs_link_pointer))
2661 		return -EINVAL;
2662 
2663 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->vmcs_link_pointer);
2664 	if (is_error_page(page))
2665 		return -EINVAL;
2666 
2667 	r = 0;
2668 	shadow = kmap(page);
2669 	if (shadow->hdr.revision_id != VMCS12_REVISION ||
2670 	    shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12))
2671 		r = -EINVAL;
2672 	kunmap(page);
2673 	kvm_release_page_clean(page);
2674 	return r;
2675 }
2676 
2677 static int nested_vmx_check_vmentry_postreqs(struct kvm_vcpu *vcpu,
2678 					     struct vmcs12 *vmcs12,
2679 					     u32 *exit_qual)
2680 {
2681 	bool ia32e;
2682 
2683 	*exit_qual = ENTRY_FAIL_DEFAULT;
2684 
2685 	if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
2686 	    !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
2687 		return 1;
2688 
2689 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2690 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2691 		return 1;
2692 	}
2693 
2694 	/*
2695 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2696 	 * are performed on the field for the IA32_EFER MSR:
2697 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2698 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2699 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2700 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2701 	 *   CR0.PG) is 1.
2702 	 */
2703 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2704 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2705 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2706 		if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
2707 		    ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
2708 		    ((vmcs12->guest_cr0 & X86_CR0_PG) &&
2709 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
2710 			return 1;
2711 	}
2712 
2713 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2714 		(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
2715 		(vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
2716 			return 1;
2717 
2718 	return 0;
2719 }
2720 
2721 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2722 {
2723 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2724 	unsigned long cr3, cr4;
2725 
2726 	if (!nested_early_check)
2727 		return 0;
2728 
2729 	if (vmx->msr_autoload.host.nr)
2730 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2731 	if (vmx->msr_autoload.guest.nr)
2732 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2733 
2734 	preempt_disable();
2735 
2736 	vmx_prepare_switch_to_guest(vcpu);
2737 
2738 	/*
2739 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2740 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2741 	 * be written (by preparve_vmcs02()) before the "real" VMEnter, i.e.
2742 	 * there is no need to preserve other bits or save/restore the field.
2743 	 */
2744 	vmcs_writel(GUEST_RFLAGS, 0);
2745 
2746 	cr3 = __get_current_cr3_fast();
2747 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2748 		vmcs_writel(HOST_CR3, cr3);
2749 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2750 	}
2751 
2752 	cr4 = cr4_read_shadow();
2753 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2754 		vmcs_writel(HOST_CR4, cr4);
2755 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2756 	}
2757 
2758 	vmx->__launched = vmx->loaded_vmcs->launched;
2759 
2760 	asm(
2761 		/* Set HOST_RSP */
2762 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2763 		__ex("vmwrite %%" _ASM_SP ", %%" _ASM_DX) "\n\t"
2764 		"mov %%" _ASM_SP ", %c[host_rsp](%1)\n\t"
2765 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2766 
2767 		/* Check if vmlaunch or vmresume is needed */
2768 		"cmpl $0, %c[launched](%% " _ASM_CX")\n\t"
2769 
2770 		"call vmx_vmenter\n\t"
2771 
2772 		/* Set vmx->fail accordingly */
2773 		"setbe %c[fail](%% " _ASM_CX")\n\t"
2774 	      : ASM_CALL_CONSTRAINT
2775 	      : "c"(vmx), "d"((unsigned long)HOST_RSP),
2776 		[launched]"i"(offsetof(struct vcpu_vmx, __launched)),
2777 		[fail]"i"(offsetof(struct vcpu_vmx, fail)),
2778 		[host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
2779 		[wordsize]"i"(sizeof(ulong))
2780 	      : "rax", "cc", "memory"
2781 	);
2782 
2783 	preempt_enable();
2784 
2785 	if (vmx->msr_autoload.host.nr)
2786 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2787 	if (vmx->msr_autoload.guest.nr)
2788 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2789 
2790 	if (vmx->fail) {
2791 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
2792 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
2793 		vmx->fail = 0;
2794 		return 1;
2795 	}
2796 
2797 	/*
2798 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
2799 	 */
2800 	local_irq_enable();
2801 	if (hw_breakpoint_active())
2802 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
2803 
2804 	/*
2805 	 * A non-failing VMEntry means we somehow entered guest mode with
2806 	 * an illegal RIP, and that's just the tip of the iceberg.  There
2807 	 * is no telling what memory has been modified or what state has
2808 	 * been exposed to unknown code.  Hitting this all but guarantees
2809 	 * a (very critical) hardware issue.
2810 	 */
2811 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
2812 		VMX_EXIT_REASONS_FAILED_VMENTRY));
2813 
2814 	return 0;
2815 }
2816 STACK_FRAME_NON_STANDARD(nested_vmx_check_vmentry_hw);
2817 
2818 
2819 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
2820 						 struct vmcs12 *vmcs12);
2821 
2822 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
2823 {
2824 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2825 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2826 	struct page *page;
2827 	u64 hpa;
2828 
2829 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2830 		/*
2831 		 * Translate L1 physical address to host physical
2832 		 * address for vmcs02. Keep the page pinned, so this
2833 		 * physical address remains valid. We keep a reference
2834 		 * to it so we can release it later.
2835 		 */
2836 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
2837 			kvm_release_page_dirty(vmx->nested.apic_access_page);
2838 			vmx->nested.apic_access_page = NULL;
2839 		}
2840 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
2841 		/*
2842 		 * If translation failed, no matter: This feature asks
2843 		 * to exit when accessing the given address, and if it
2844 		 * can never be accessed, this feature won't do
2845 		 * anything anyway.
2846 		 */
2847 		if (!is_error_page(page)) {
2848 			vmx->nested.apic_access_page = page;
2849 			hpa = page_to_phys(vmx->nested.apic_access_page);
2850 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
2851 		} else {
2852 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
2853 					SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
2854 		}
2855 	}
2856 
2857 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
2858 		if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
2859 			kvm_release_page_dirty(vmx->nested.virtual_apic_page);
2860 			vmx->nested.virtual_apic_page = NULL;
2861 		}
2862 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
2863 
2864 		/*
2865 		 * If translation failed, VM entry will fail because
2866 		 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
2867 		 * Failing the vm entry is _not_ what the processor
2868 		 * does but it's basically the only possibility we
2869 		 * have.  We could still enter the guest if CR8 load
2870 		 * exits are enabled, CR8 store exits are enabled, and
2871 		 * virtualize APIC access is disabled; in this case
2872 		 * the processor would never use the TPR shadow and we
2873 		 * could simply clear the bit from the execution
2874 		 * control.  But such a configuration is useless, so
2875 		 * let's keep the code simple.
2876 		 */
2877 		if (!is_error_page(page)) {
2878 			vmx->nested.virtual_apic_page = page;
2879 			hpa = page_to_phys(vmx->nested.virtual_apic_page);
2880 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
2881 		}
2882 	}
2883 
2884 	if (nested_cpu_has_posted_intr(vmcs12)) {
2885 		if (vmx->nested.pi_desc_page) { /* shouldn't happen */
2886 			kunmap(vmx->nested.pi_desc_page);
2887 			kvm_release_page_dirty(vmx->nested.pi_desc_page);
2888 			vmx->nested.pi_desc_page = NULL;
2889 			vmx->nested.pi_desc = NULL;
2890 			vmcs_write64(POSTED_INTR_DESC_ADDR, -1ull);
2891 		}
2892 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
2893 		if (is_error_page(page))
2894 			return;
2895 		vmx->nested.pi_desc_page = page;
2896 		vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
2897 		vmx->nested.pi_desc =
2898 			(struct pi_desc *)((void *)vmx->nested.pi_desc +
2899 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2900 			(PAGE_SIZE - 1)));
2901 		vmcs_write64(POSTED_INTR_DESC_ADDR,
2902 			page_to_phys(vmx->nested.pi_desc_page) +
2903 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2904 			(PAGE_SIZE - 1)));
2905 	}
2906 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
2907 		vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
2908 			      CPU_BASED_USE_MSR_BITMAPS);
2909 	else
2910 		vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
2911 				CPU_BASED_USE_MSR_BITMAPS);
2912 }
2913 
2914 /*
2915  * Intel's VMX Instruction Reference specifies a common set of prerequisites
2916  * for running VMX instructions (except VMXON, whose prerequisites are
2917  * slightly different). It also specifies what exception to inject otherwise.
2918  * Note that many of these exceptions have priority over VM exits, so they
2919  * don't have to be checked again here.
2920  */
2921 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
2922 {
2923 	if (!to_vmx(vcpu)->nested.vmxon) {
2924 		kvm_queue_exception(vcpu, UD_VECTOR);
2925 		return 0;
2926 	}
2927 
2928 	if (vmx_get_cpl(vcpu)) {
2929 		kvm_inject_gp(vcpu, 0);
2930 		return 0;
2931 	}
2932 
2933 	return 1;
2934 }
2935 
2936 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
2937 {
2938 	u8 rvi = vmx_get_rvi();
2939 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
2940 
2941 	return ((rvi & 0xf0) > (vppr & 0xf0));
2942 }
2943 
2944 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
2945 				   struct vmcs12 *vmcs12);
2946 
2947 /*
2948  * If from_vmentry is false, this is being called from state restore (either RSM
2949  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
2950 + *
2951 + * Returns:
2952 + *   0 - success, i.e. proceed with actual VMEnter
2953 + *   1 - consistency check VMExit
2954 + *  -1 - consistency check VMFail
2955  */
2956 int nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
2957 {
2958 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2959 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2960 	bool evaluate_pending_interrupts;
2961 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
2962 	u32 exit_qual;
2963 
2964 	evaluate_pending_interrupts = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2965 		(CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_VIRTUAL_NMI_PENDING);
2966 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
2967 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
2968 
2969 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
2970 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
2971 	if (kvm_mpx_supported() &&
2972 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2973 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
2974 
2975 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
2976 
2977 	prepare_vmcs02_early(vmx, vmcs12);
2978 
2979 	if (from_vmentry) {
2980 		nested_get_vmcs12_pages(vcpu);
2981 
2982 		if (nested_vmx_check_vmentry_hw(vcpu)) {
2983 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
2984 			return -1;
2985 		}
2986 
2987 		if (nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
2988 			goto vmentry_fail_vmexit;
2989 	}
2990 
2991 	enter_guest_mode(vcpu);
2992 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
2993 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
2994 
2995 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
2996 		goto vmentry_fail_vmexit_guest_mode;
2997 
2998 	if (from_vmentry) {
2999 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
3000 		exit_qual = nested_vmx_load_msr(vcpu,
3001 						vmcs12->vm_entry_msr_load_addr,
3002 						vmcs12->vm_entry_msr_load_count);
3003 		if (exit_qual)
3004 			goto vmentry_fail_vmexit_guest_mode;
3005 	} else {
3006 		/*
3007 		 * The MMU is not initialized to point at the right entities yet and
3008 		 * "get pages" would need to read data from the guest (i.e. we will
3009 		 * need to perform gpa to hpa translation). Request a call
3010 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3011 		 * have already been set at vmentry time and should not be reset.
3012 		 */
3013 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3014 	}
3015 
3016 	/*
3017 	 * If L1 had a pending IRQ/NMI until it executed
3018 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3019 	 * disallowed (e.g. interrupts disabled), L0 needs to
3020 	 * evaluate if this pending event should cause an exit from L2
3021 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3022 	 * intercept EXTERNAL_INTERRUPT).
3023 	 *
3024 	 * Usually this would be handled by the processor noticing an
3025 	 * IRQ/NMI window request, or checking RVI during evaluation of
3026 	 * pending virtual interrupts.  However, this setting was done
3027 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3028 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3029 	 */
3030 	if (unlikely(evaluate_pending_interrupts))
3031 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3032 
3033 	/*
3034 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3035 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3036 	 * returned as far as L1 is concerned. It will only return (and set
3037 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3038 	 */
3039 	return 0;
3040 
3041 	/*
3042 	 * A failed consistency check that leads to a VMExit during L1's
3043 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3044 	 * 26.7 "VM-entry failures during or after loading guest state".
3045 	 */
3046 vmentry_fail_vmexit_guest_mode:
3047 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3048 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3049 	leave_guest_mode(vcpu);
3050 
3051 vmentry_fail_vmexit:
3052 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3053 
3054 	if (!from_vmentry)
3055 		return 1;
3056 
3057 	load_vmcs12_host_state(vcpu, vmcs12);
3058 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3059 	vmcs12->exit_qualification = exit_qual;
3060 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3061 		vmx->nested.need_vmcs12_sync = true;
3062 	return 1;
3063 }
3064 
3065 /*
3066  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3067  * for running an L2 nested guest.
3068  */
3069 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3070 {
3071 	struct vmcs12 *vmcs12;
3072 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3073 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3074 	int ret;
3075 
3076 	if (!nested_vmx_check_permission(vcpu))
3077 		return 1;
3078 
3079 	if (!nested_vmx_handle_enlightened_vmptrld(vcpu, true))
3080 		return 1;
3081 
3082 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3083 		return nested_vmx_failInvalid(vcpu);
3084 
3085 	vmcs12 = get_vmcs12(vcpu);
3086 
3087 	/*
3088 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3089 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3090 	 * rather than RFLAGS.ZF, and no error number is stored to the
3091 	 * VM-instruction error field.
3092 	 */
3093 	if (vmcs12->hdr.shadow_vmcs)
3094 		return nested_vmx_failInvalid(vcpu);
3095 
3096 	if (vmx->nested.hv_evmcs) {
3097 		copy_enlightened_to_vmcs12(vmx);
3098 		/* Enlightened VMCS doesn't have launch state */
3099 		vmcs12->launch_state = !launch;
3100 	} else if (enable_shadow_vmcs) {
3101 		copy_shadow_to_vmcs12(vmx);
3102 	}
3103 
3104 	/*
3105 	 * The nested entry process starts with enforcing various prerequisites
3106 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3107 	 * they fail: As the SDM explains, some conditions should cause the
3108 	 * instruction to fail, while others will cause the instruction to seem
3109 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3110 	 * To speed up the normal (success) code path, we should avoid checking
3111 	 * for misconfigurations which will anyway be caught by the processor
3112 	 * when using the merged vmcs02.
3113 	 */
3114 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3115 		return nested_vmx_failValid(vcpu,
3116 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3117 
3118 	if (vmcs12->launch_state == launch)
3119 		return nested_vmx_failValid(vcpu,
3120 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3121 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3122 
3123 	ret = nested_vmx_check_vmentry_prereqs(vcpu, vmcs12);
3124 	if (ret)
3125 		return nested_vmx_failValid(vcpu, ret);
3126 
3127 	/*
3128 	 * We're finally done with prerequisite checking, and can start with
3129 	 * the nested entry.
3130 	 */
3131 	vmx->nested.nested_run_pending = 1;
3132 	ret = nested_vmx_enter_non_root_mode(vcpu, true);
3133 	vmx->nested.nested_run_pending = !ret;
3134 	if (ret > 0)
3135 		return 1;
3136 	else if (ret)
3137 		return nested_vmx_failValid(vcpu,
3138 			VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3139 
3140 	/* Hide L1D cache contents from the nested guest.  */
3141 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3142 
3143 	/*
3144 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3145 	 * also be used as part of restoring nVMX state for
3146 	 * snapshot restore (migration).
3147 	 *
3148 	 * In this flow, it is assumed that vmcs12 cache was
3149 	 * trasferred as part of captured nVMX state and should
3150 	 * therefore not be read from guest memory (which may not
3151 	 * exist on destination host yet).
3152 	 */
3153 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3154 
3155 	/*
3156 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3157 	 * awakened by event injection or by an NMI-window VM-exit or
3158 	 * by an interrupt-window VM-exit, halt the vcpu.
3159 	 */
3160 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3161 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3162 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_NMI_PENDING) &&
3163 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_INTR_PENDING) &&
3164 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3165 		vmx->nested.nested_run_pending = 0;
3166 		return kvm_vcpu_halt(vcpu);
3167 	}
3168 	return 1;
3169 }
3170 
3171 /*
3172  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3173  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
3174  * This function returns the new value we should put in vmcs12.guest_cr0.
3175  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3176  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3177  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3178  *     didn't trap the bit, because if L1 did, so would L0).
3179  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3180  *     been modified by L2, and L1 knows it. So just leave the old value of
3181  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3182  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3183  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3184  *     changed these bits, and therefore they need to be updated, but L0
3185  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3186  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3187  */
3188 static inline unsigned long
3189 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3190 {
3191 	return
3192 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3193 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3194 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3195 			vcpu->arch.cr0_guest_owned_bits));
3196 }
3197 
3198 static inline unsigned long
3199 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3200 {
3201 	return
3202 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3203 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3204 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3205 			vcpu->arch.cr4_guest_owned_bits));
3206 }
3207 
3208 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3209 				      struct vmcs12 *vmcs12)
3210 {
3211 	u32 idt_vectoring;
3212 	unsigned int nr;
3213 
3214 	if (vcpu->arch.exception.injected) {
3215 		nr = vcpu->arch.exception.nr;
3216 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3217 
3218 		if (kvm_exception_is_soft(nr)) {
3219 			vmcs12->vm_exit_instruction_len =
3220 				vcpu->arch.event_exit_inst_len;
3221 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3222 		} else
3223 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3224 
3225 		if (vcpu->arch.exception.has_error_code) {
3226 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3227 			vmcs12->idt_vectoring_error_code =
3228 				vcpu->arch.exception.error_code;
3229 		}
3230 
3231 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3232 	} else if (vcpu->arch.nmi_injected) {
3233 		vmcs12->idt_vectoring_info_field =
3234 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3235 	} else if (vcpu->arch.interrupt.injected) {
3236 		nr = vcpu->arch.interrupt.nr;
3237 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3238 
3239 		if (vcpu->arch.interrupt.soft) {
3240 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3241 			vmcs12->vm_entry_instruction_len =
3242 				vcpu->arch.event_exit_inst_len;
3243 		} else
3244 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3245 
3246 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3247 	}
3248 }
3249 
3250 
3251 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3252 {
3253 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3254 	gfn_t gfn;
3255 
3256 	/*
3257 	 * Don't need to mark the APIC access page dirty; it is never
3258 	 * written to by the CPU during APIC virtualization.
3259 	 */
3260 
3261 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3262 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3263 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3264 	}
3265 
3266 	if (nested_cpu_has_posted_intr(vmcs12)) {
3267 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3268 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3269 	}
3270 }
3271 
3272 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3273 {
3274 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3275 	int max_irr;
3276 	void *vapic_page;
3277 	u16 status;
3278 
3279 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3280 		return;
3281 
3282 	vmx->nested.pi_pending = false;
3283 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3284 		return;
3285 
3286 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3287 	if (max_irr != 256) {
3288 		vapic_page = kmap(vmx->nested.virtual_apic_page);
3289 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3290 			vapic_page, &max_irr);
3291 		kunmap(vmx->nested.virtual_apic_page);
3292 
3293 		status = vmcs_read16(GUEST_INTR_STATUS);
3294 		if ((u8)max_irr > ((u8)status & 0xff)) {
3295 			status &= ~0xff;
3296 			status |= (u8)max_irr;
3297 			vmcs_write16(GUEST_INTR_STATUS, status);
3298 		}
3299 	}
3300 
3301 	nested_mark_vmcs12_pages_dirty(vcpu);
3302 }
3303 
3304 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3305 					       unsigned long exit_qual)
3306 {
3307 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3308 	unsigned int nr = vcpu->arch.exception.nr;
3309 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3310 
3311 	if (vcpu->arch.exception.has_error_code) {
3312 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3313 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3314 	}
3315 
3316 	if (kvm_exception_is_soft(nr))
3317 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3318 	else
3319 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3320 
3321 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3322 	    vmx_get_nmi_mask(vcpu))
3323 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3324 
3325 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3326 }
3327 
3328 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
3329 {
3330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3331 	unsigned long exit_qual;
3332 	bool block_nested_events =
3333 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3334 
3335 	if (vcpu->arch.exception.pending &&
3336 		nested_vmx_check_exception(vcpu, &exit_qual)) {
3337 		if (block_nested_events)
3338 			return -EBUSY;
3339 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3340 		return 0;
3341 	}
3342 
3343 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3344 	    vmx->nested.preemption_timer_expired) {
3345 		if (block_nested_events)
3346 			return -EBUSY;
3347 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3348 		return 0;
3349 	}
3350 
3351 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3352 		if (block_nested_events)
3353 			return -EBUSY;
3354 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3355 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3356 				  INTR_INFO_VALID_MASK, 0);
3357 		/*
3358 		 * The NMI-triggered VM exit counts as injection:
3359 		 * clear this one and block further NMIs.
3360 		 */
3361 		vcpu->arch.nmi_pending = 0;
3362 		vmx_set_nmi_mask(vcpu, true);
3363 		return 0;
3364 	}
3365 
3366 	if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
3367 	    nested_exit_on_intr(vcpu)) {
3368 		if (block_nested_events)
3369 			return -EBUSY;
3370 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3371 		return 0;
3372 	}
3373 
3374 	vmx_complete_nested_posted_interrupt(vcpu);
3375 	return 0;
3376 }
3377 
3378 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3379 {
3380 	ktime_t remaining =
3381 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3382 	u64 value;
3383 
3384 	if (ktime_to_ns(remaining) <= 0)
3385 		return 0;
3386 
3387 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3388 	do_div(value, 1000000);
3389 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3390 }
3391 
3392 /*
3393  * Update the guest state fields of vmcs12 to reflect changes that
3394  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3395  * VM-entry controls is also updated, since this is really a guest
3396  * state bit.)
3397  */
3398 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3399 {
3400 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3401 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3402 
3403 	vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3404 	vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
3405 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3406 
3407 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3408 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3409 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3410 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3411 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3412 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3413 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3414 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3415 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3416 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3417 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3418 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3419 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3420 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3421 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3422 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3423 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3424 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3425 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3426 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3427 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3428 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3429 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3430 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3431 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3432 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3433 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3434 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3435 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3436 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3437 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3438 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3439 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3440 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3441 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3442 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3443 
3444 	vmcs12->guest_interruptibility_info =
3445 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3446 	vmcs12->guest_pending_dbg_exceptions =
3447 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3448 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3449 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3450 	else
3451 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3452 
3453 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3454 		if (vmcs12->vm_exit_controls &
3455 		    VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3456 			vmcs12->vmx_preemption_timer_value =
3457 				vmx_get_preemption_timer_value(vcpu);
3458 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
3459 	}
3460 
3461 	/*
3462 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3463 	 * own CR3 without exiting. If it has changed it, we must keep it.
3464 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3465 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3466 	 *
3467 	 * Additionally, restore L2's PDPTR to vmcs12.
3468 	 */
3469 	if (enable_ept) {
3470 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3471 		vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3472 		vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3473 		vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3474 		vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3475 	}
3476 
3477 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3478 
3479 	if (nested_cpu_has_vid(vmcs12))
3480 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3481 
3482 	vmcs12->vm_entry_controls =
3483 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3484 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3485 
3486 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
3487 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3488 		vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3489 	}
3490 
3491 	/* TODO: These cannot have changed unless we have MSR bitmaps and
3492 	 * the relevant bit asks not to trap the change */
3493 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
3494 		vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
3495 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3496 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3497 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3498 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3499 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3500 	if (kvm_mpx_supported())
3501 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3502 }
3503 
3504 /*
3505  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3506  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3507  * and this function updates it to reflect the changes to the guest state while
3508  * L2 was running (and perhaps made some exits which were handled directly by L0
3509  * without going back to L1), and to reflect the exit reason.
3510  * Note that we do not have to copy here all VMCS fields, just those that
3511  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3512  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3513  * which already writes to vmcs12 directly.
3514  */
3515 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3516 			   u32 exit_reason, u32 exit_intr_info,
3517 			   unsigned long exit_qualification)
3518 {
3519 	/* update guest state fields: */
3520 	sync_vmcs12(vcpu, vmcs12);
3521 
3522 	/* update exit information fields: */
3523 
3524 	vmcs12->vm_exit_reason = exit_reason;
3525 	vmcs12->exit_qualification = exit_qualification;
3526 	vmcs12->vm_exit_intr_info = exit_intr_info;
3527 
3528 	vmcs12->idt_vectoring_info_field = 0;
3529 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3530 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3531 
3532 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3533 		vmcs12->launch_state = 1;
3534 
3535 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3536 		 * instead of reading the real value. */
3537 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3538 
3539 		/*
3540 		 * Transfer the event that L0 or L1 may wanted to inject into
3541 		 * L2 to IDT_VECTORING_INFO_FIELD.
3542 		 */
3543 		vmcs12_save_pending_event(vcpu, vmcs12);
3544 
3545 		/*
3546 		 * According to spec, there's no need to store the guest's
3547 		 * MSRs if the exit is due to a VM-entry failure that occurs
3548 		 * during or after loading the guest state. Since this exit
3549 		 * does not fall in that category, we need to save the MSRs.
3550 		 */
3551 		if (nested_vmx_store_msr(vcpu,
3552 					 vmcs12->vm_exit_msr_store_addr,
3553 					 vmcs12->vm_exit_msr_store_count))
3554 			nested_vmx_abort(vcpu,
3555 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3556 	}
3557 
3558 	/*
3559 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3560 	 * preserved above and would only end up incorrectly in L1.
3561 	 */
3562 	vcpu->arch.nmi_injected = false;
3563 	kvm_clear_exception_queue(vcpu);
3564 	kvm_clear_interrupt_queue(vcpu);
3565 }
3566 
3567 /*
3568  * A part of what we need to when the nested L2 guest exits and we want to
3569  * run its L1 parent, is to reset L1's guest state to the host state specified
3570  * in vmcs12.
3571  * This function is to be called not only on normal nested exit, but also on
3572  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3573  * Failures During or After Loading Guest State").
3574  * This function should be called when the active VMCS is L1's (vmcs01).
3575  */
3576 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3577 				   struct vmcs12 *vmcs12)
3578 {
3579 	struct kvm_segment seg;
3580 	u32 entry_failure_code;
3581 
3582 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
3583 		vcpu->arch.efer = vmcs12->host_ia32_efer;
3584 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3585 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
3586 	else
3587 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
3588 	vmx_set_efer(vcpu, vcpu->arch.efer);
3589 
3590 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
3591 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
3592 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
3593 	vmx_set_interrupt_shadow(vcpu, 0);
3594 
3595 	/*
3596 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
3597 	 * actually changed, because vmx_set_cr0 refers to efer set above.
3598 	 *
3599 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
3600 	 * (KVM doesn't change it);
3601 	 */
3602 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3603 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
3604 
3605 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
3606 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3607 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
3608 
3609 	nested_ept_uninit_mmu_context(vcpu);
3610 
3611 	/*
3612 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
3613 	 * couldn't have changed.
3614 	 */
3615 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
3616 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
3617 
3618 	if (!enable_ept)
3619 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
3620 
3621 	/*
3622 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
3623 	 * VMEntry/VMExit. Thus, no need to flush TLB.
3624 	 *
3625 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
3626 	 * flushed on every VMEntry/VMExit.
3627 	 *
3628 	 * Otherwise, we can preserve TLB entries as long as we are
3629 	 * able to tag L1 TLB entries differently than L2 TLB entries.
3630 	 *
3631 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
3632 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
3633 	 * has been set by KVM_REQ_LOAD_CR3.
3634 	 */
3635 	if (enable_vpid &&
3636 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
3637 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3638 	}
3639 
3640 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
3641 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
3642 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
3643 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
3644 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
3645 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
3646 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
3647 
3648 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
3649 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
3650 		vmcs_write64(GUEST_BNDCFGS, 0);
3651 
3652 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
3653 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
3654 		vcpu->arch.pat = vmcs12->host_ia32_pat;
3655 	}
3656 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
3657 		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
3658 			vmcs12->host_ia32_perf_global_ctrl);
3659 
3660 	/* Set L1 segment info according to Intel SDM
3661 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
3662 	seg = (struct kvm_segment) {
3663 		.base = 0,
3664 		.limit = 0xFFFFFFFF,
3665 		.selector = vmcs12->host_cs_selector,
3666 		.type = 11,
3667 		.present = 1,
3668 		.s = 1,
3669 		.g = 1
3670 	};
3671 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3672 		seg.l = 1;
3673 	else
3674 		seg.db = 1;
3675 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
3676 	seg = (struct kvm_segment) {
3677 		.base = 0,
3678 		.limit = 0xFFFFFFFF,
3679 		.type = 3,
3680 		.present = 1,
3681 		.s = 1,
3682 		.db = 1,
3683 		.g = 1
3684 	};
3685 	seg.selector = vmcs12->host_ds_selector;
3686 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
3687 	seg.selector = vmcs12->host_es_selector;
3688 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
3689 	seg.selector = vmcs12->host_ss_selector;
3690 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
3691 	seg.selector = vmcs12->host_fs_selector;
3692 	seg.base = vmcs12->host_fs_base;
3693 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
3694 	seg.selector = vmcs12->host_gs_selector;
3695 	seg.base = vmcs12->host_gs_base;
3696 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
3697 	seg = (struct kvm_segment) {
3698 		.base = vmcs12->host_tr_base,
3699 		.limit = 0x67,
3700 		.selector = vmcs12->host_tr_selector,
3701 		.type = 11,
3702 		.present = 1
3703 	};
3704 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
3705 
3706 	kvm_set_dr(vcpu, 7, 0x400);
3707 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3708 
3709 	if (cpu_has_vmx_msr_bitmap())
3710 		vmx_update_msr_bitmap(vcpu);
3711 
3712 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
3713 				vmcs12->vm_exit_msr_load_count))
3714 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3715 }
3716 
3717 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
3718 {
3719 	struct shared_msr_entry *efer_msr;
3720 	unsigned int i;
3721 
3722 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
3723 		return vmcs_read64(GUEST_IA32_EFER);
3724 
3725 	if (cpu_has_load_ia32_efer())
3726 		return host_efer;
3727 
3728 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
3729 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
3730 			return vmx->msr_autoload.guest.val[i].value;
3731 	}
3732 
3733 	efer_msr = find_msr_entry(vmx, MSR_EFER);
3734 	if (efer_msr)
3735 		return efer_msr->data;
3736 
3737 	return host_efer;
3738 }
3739 
3740 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
3741 {
3742 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3743 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3744 	struct vmx_msr_entry g, h;
3745 	struct msr_data msr;
3746 	gpa_t gpa;
3747 	u32 i, j;
3748 
3749 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
3750 
3751 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
3752 		/*
3753 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
3754 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
3755 		 * and vcpu->arch.dr7 is not squirreled away before the
3756 		 * nested VMENTER (not worth adding a variable in nested_vmx).
3757 		 */
3758 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
3759 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
3760 		else
3761 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
3762 	}
3763 
3764 	/*
3765 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
3766 	 * handle a variety of side effects to KVM's software model.
3767 	 */
3768 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
3769 
3770 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3771 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
3772 
3773 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3774 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
3775 
3776 	nested_ept_uninit_mmu_context(vcpu);
3777 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3778 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3779 
3780 	/*
3781 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
3782 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
3783 	 * VMFail, like everything else we just need to ensure our
3784 	 * software model is up-to-date.
3785 	 */
3786 	ept_save_pdptrs(vcpu);
3787 
3788 	kvm_mmu_reset_context(vcpu);
3789 
3790 	if (cpu_has_vmx_msr_bitmap())
3791 		vmx_update_msr_bitmap(vcpu);
3792 
3793 	/*
3794 	 * This nasty bit of open coding is a compromise between blindly
3795 	 * loading L1's MSRs using the exit load lists (incorrect emulation
3796 	 * of VMFail), leaving the nested VM's MSRs in the software model
3797 	 * (incorrect behavior) and snapshotting the modified MSRs (too
3798 	 * expensive since the lists are unbound by hardware).  For each
3799 	 * MSR that was (prematurely) loaded from the nested VMEntry load
3800 	 * list, reload it from the exit load list if it exists and differs
3801 	 * from the guest value.  The intent is to stuff host state as
3802 	 * silently as possible, not to fully process the exit load list.
3803 	 */
3804 	msr.host_initiated = false;
3805 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
3806 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
3807 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
3808 			pr_debug_ratelimited(
3809 				"%s read MSR index failed (%u, 0x%08llx)\n",
3810 				__func__, i, gpa);
3811 			goto vmabort;
3812 		}
3813 
3814 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
3815 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
3816 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
3817 				pr_debug_ratelimited(
3818 					"%s read MSR failed (%u, 0x%08llx)\n",
3819 					__func__, j, gpa);
3820 				goto vmabort;
3821 			}
3822 			if (h.index != g.index)
3823 				continue;
3824 			if (h.value == g.value)
3825 				break;
3826 
3827 			if (nested_vmx_load_msr_check(vcpu, &h)) {
3828 				pr_debug_ratelimited(
3829 					"%s check failed (%u, 0x%x, 0x%x)\n",
3830 					__func__, j, h.index, h.reserved);
3831 				goto vmabort;
3832 			}
3833 
3834 			msr.index = h.index;
3835 			msr.data = h.value;
3836 			if (kvm_set_msr(vcpu, &msr)) {
3837 				pr_debug_ratelimited(
3838 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
3839 					__func__, j, h.index, h.value);
3840 				goto vmabort;
3841 			}
3842 		}
3843 	}
3844 
3845 	return;
3846 
3847 vmabort:
3848 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3849 }
3850 
3851 /*
3852  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
3853  * and modify vmcs12 to make it see what it would expect to see there if
3854  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
3855  */
3856 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
3857 		       u32 exit_intr_info, unsigned long exit_qualification)
3858 {
3859 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3860 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3861 
3862 	/* trying to cancel vmlaunch/vmresume is a bug */
3863 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
3864 
3865 	leave_guest_mode(vcpu);
3866 
3867 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3868 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3869 
3870 	if (likely(!vmx->fail)) {
3871 		if (exit_reason == -1)
3872 			sync_vmcs12(vcpu, vmcs12);
3873 		else
3874 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
3875 				       exit_qualification);
3876 
3877 		/*
3878 		 * Must happen outside of sync_vmcs12() as it will
3879 		 * also be used to capture vmcs12 cache as part of
3880 		 * capturing nVMX state for snapshot (migration).
3881 		 *
3882 		 * Otherwise, this flush will dirty guest memory at a
3883 		 * point it is already assumed by user-space to be
3884 		 * immutable.
3885 		 */
3886 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
3887 	} else {
3888 		/*
3889 		 * The only expected VM-instruction error is "VM entry with
3890 		 * invalid control field(s)." Anything else indicates a
3891 		 * problem with L0.  And we should never get here with a
3892 		 * VMFail of any type if early consistency checks are enabled.
3893 		 */
3894 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
3895 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3896 		WARN_ON_ONCE(nested_early_check);
3897 	}
3898 
3899 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3900 
3901 	/* Update any VMCS fields that might have changed while L2 ran */
3902 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3903 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3904 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
3905 
3906 	if (kvm_has_tsc_control)
3907 		decache_tsc_multiplier(vmx);
3908 
3909 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
3910 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
3911 		vmx_set_virtual_apic_mode(vcpu);
3912 	} else if (!nested_cpu_has_ept(vmcs12) &&
3913 		   nested_cpu_has2(vmcs12,
3914 				   SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3915 		vmx_flush_tlb(vcpu, true);
3916 	}
3917 
3918 	/* This is needed for same reason as it was needed in prepare_vmcs02 */
3919 	vmx->host_rsp = 0;
3920 
3921 	/* Unpin physical memory we referred to in vmcs02 */
3922 	if (vmx->nested.apic_access_page) {
3923 		kvm_release_page_dirty(vmx->nested.apic_access_page);
3924 		vmx->nested.apic_access_page = NULL;
3925 	}
3926 	if (vmx->nested.virtual_apic_page) {
3927 		kvm_release_page_dirty(vmx->nested.virtual_apic_page);
3928 		vmx->nested.virtual_apic_page = NULL;
3929 	}
3930 	if (vmx->nested.pi_desc_page) {
3931 		kunmap(vmx->nested.pi_desc_page);
3932 		kvm_release_page_dirty(vmx->nested.pi_desc_page);
3933 		vmx->nested.pi_desc_page = NULL;
3934 		vmx->nested.pi_desc = NULL;
3935 	}
3936 
3937 	/*
3938 	 * We are now running in L2, mmu_notifier will force to reload the
3939 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
3940 	 */
3941 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
3942 
3943 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
3944 		vmx->nested.need_vmcs12_sync = true;
3945 
3946 	/* in case we halted in L2 */
3947 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3948 
3949 	if (likely(!vmx->fail)) {
3950 		/*
3951 		 * TODO: SDM says that with acknowledge interrupt on
3952 		 * exit, bit 31 of the VM-exit interrupt information
3953 		 * (valid interrupt) is always set to 1 on
3954 		 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
3955 		 * need kvm_cpu_has_interrupt().  See the commit
3956 		 * message for details.
3957 		 */
3958 		if (nested_exit_intr_ack_set(vcpu) &&
3959 		    exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
3960 		    kvm_cpu_has_interrupt(vcpu)) {
3961 			int irq = kvm_cpu_get_interrupt(vcpu);
3962 			WARN_ON(irq < 0);
3963 			vmcs12->vm_exit_intr_info = irq |
3964 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
3965 		}
3966 
3967 		if (exit_reason != -1)
3968 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
3969 						       vmcs12->exit_qualification,
3970 						       vmcs12->idt_vectoring_info_field,
3971 						       vmcs12->vm_exit_intr_info,
3972 						       vmcs12->vm_exit_intr_error_code,
3973 						       KVM_ISA_VMX);
3974 
3975 		load_vmcs12_host_state(vcpu, vmcs12);
3976 
3977 		return;
3978 	}
3979 
3980 	/*
3981 	 * After an early L2 VM-entry failure, we're now back
3982 	 * in L1 which thinks it just finished a VMLAUNCH or
3983 	 * VMRESUME instruction, so we need to set the failure
3984 	 * flag and the VM-instruction error field of the VMCS
3985 	 * accordingly, and skip the emulated instruction.
3986 	 */
3987 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3988 
3989 	/*
3990 	 * Restore L1's host state to KVM's software model.  We're here
3991 	 * because a consistency check was caught by hardware, which
3992 	 * means some amount of guest state has been propagated to KVM's
3993 	 * model and needs to be unwound to the host's state.
3994 	 */
3995 	nested_vmx_restore_host_state(vcpu);
3996 
3997 	vmx->fail = 0;
3998 }
3999 
4000 /*
4001  * Decode the memory-address operand of a vmx instruction, as recorded on an
4002  * exit caused by such an instruction (run by a guest hypervisor).
4003  * On success, returns 0. When the operand is invalid, returns 1 and throws
4004  * #UD or #GP.
4005  */
4006 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4007 			u32 vmx_instruction_info, bool wr, gva_t *ret)
4008 {
4009 	gva_t off;
4010 	bool exn;
4011 	struct kvm_segment s;
4012 
4013 	/*
4014 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4015 	 * Execution", on an exit, vmx_instruction_info holds most of the
4016 	 * addressing components of the operand. Only the displacement part
4017 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4018 	 * For how an actual address is calculated from all these components,
4019 	 * refer to Vol. 1, "Operand Addressing".
4020 	 */
4021 	int  scaling = vmx_instruction_info & 3;
4022 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4023 	bool is_reg = vmx_instruction_info & (1u << 10);
4024 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4025 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4026 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4027 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4028 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4029 
4030 	if (is_reg) {
4031 		kvm_queue_exception(vcpu, UD_VECTOR);
4032 		return 1;
4033 	}
4034 
4035 	/* Addr = segment_base + offset */
4036 	/* offset = base + [index * scale] + displacement */
4037 	off = exit_qualification; /* holds the displacement */
4038 	if (base_is_valid)
4039 		off += kvm_register_read(vcpu, base_reg);
4040 	if (index_is_valid)
4041 		off += kvm_register_read(vcpu, index_reg)<<scaling;
4042 	vmx_get_segment(vcpu, &s, seg_reg);
4043 	*ret = s.base + off;
4044 
4045 	if (addr_size == 1) /* 32 bit */
4046 		*ret &= 0xffffffff;
4047 
4048 	/* Checks for #GP/#SS exceptions. */
4049 	exn = false;
4050 	if (is_long_mode(vcpu)) {
4051 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4052 		 * non-canonical form. This is the only check on the memory
4053 		 * destination for long mode!
4054 		 */
4055 		exn = is_noncanonical_address(*ret, vcpu);
4056 	} else if (is_protmode(vcpu)) {
4057 		/* Protected mode: apply checks for segment validity in the
4058 		 * following order:
4059 		 * - segment type check (#GP(0) may be thrown)
4060 		 * - usability check (#GP(0)/#SS(0))
4061 		 * - limit check (#GP(0)/#SS(0))
4062 		 */
4063 		if (wr)
4064 			/* #GP(0) if the destination operand is located in a
4065 			 * read-only data segment or any code segment.
4066 			 */
4067 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4068 		else
4069 			/* #GP(0) if the source operand is located in an
4070 			 * execute-only code segment
4071 			 */
4072 			exn = ((s.type & 0xa) == 8);
4073 		if (exn) {
4074 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4075 			return 1;
4076 		}
4077 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4078 		 */
4079 		exn = (s.unusable != 0);
4080 		/* Protected mode: #GP(0)/#SS(0) if the memory
4081 		 * operand is outside the segment limit.
4082 		 */
4083 		exn = exn || (off + sizeof(u64) > s.limit);
4084 	}
4085 	if (exn) {
4086 		kvm_queue_exception_e(vcpu,
4087 				      seg_reg == VCPU_SREG_SS ?
4088 						SS_VECTOR : GP_VECTOR,
4089 				      0);
4090 		return 1;
4091 	}
4092 
4093 	return 0;
4094 }
4095 
4096 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4097 {
4098 	gva_t gva;
4099 	struct x86_exception e;
4100 
4101 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4102 			vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
4103 		return 1;
4104 
4105 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4106 		kvm_inject_page_fault(vcpu, &e);
4107 		return 1;
4108 	}
4109 
4110 	return 0;
4111 }
4112 
4113 /*
4114  * Allocate a shadow VMCS and associate it with the currently loaded
4115  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4116  * VMCS is also VMCLEARed, so that it is ready for use.
4117  */
4118 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4119 {
4120 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4121 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4122 
4123 	/*
4124 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4125 	 * executes VMXON and free it when L1 executes VMXOFF.
4126 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4127 	 * here when vmcs01 already have an allocated shadow vmcs.
4128 	 */
4129 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4130 
4131 	if (!loaded_vmcs->shadow_vmcs) {
4132 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4133 		if (loaded_vmcs->shadow_vmcs)
4134 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4135 	}
4136 	return loaded_vmcs->shadow_vmcs;
4137 }
4138 
4139 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4140 {
4141 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4142 	int r;
4143 
4144 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4145 	if (r < 0)
4146 		goto out_vmcs02;
4147 
4148 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL);
4149 	if (!vmx->nested.cached_vmcs12)
4150 		goto out_cached_vmcs12;
4151 
4152 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL);
4153 	if (!vmx->nested.cached_shadow_vmcs12)
4154 		goto out_cached_shadow_vmcs12;
4155 
4156 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4157 		goto out_shadow_vmcs;
4158 
4159 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4160 		     HRTIMER_MODE_REL_PINNED);
4161 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4162 
4163 	vmx->nested.vpid02 = allocate_vpid();
4164 
4165 	vmx->nested.vmcs02_initialized = false;
4166 	vmx->nested.vmxon = true;
4167 
4168 	if (pt_mode == PT_MODE_HOST_GUEST) {
4169 		vmx->pt_desc.guest.ctl = 0;
4170 		pt_update_intercept_for_msr(vmx);
4171 	}
4172 
4173 	return 0;
4174 
4175 out_shadow_vmcs:
4176 	kfree(vmx->nested.cached_shadow_vmcs12);
4177 
4178 out_cached_shadow_vmcs12:
4179 	kfree(vmx->nested.cached_vmcs12);
4180 
4181 out_cached_vmcs12:
4182 	free_loaded_vmcs(&vmx->nested.vmcs02);
4183 
4184 out_vmcs02:
4185 	return -ENOMEM;
4186 }
4187 
4188 /*
4189  * Emulate the VMXON instruction.
4190  * Currently, we just remember that VMX is active, and do not save or even
4191  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4192  * do not currently need to store anything in that guest-allocated memory
4193  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4194  * argument is different from the VMXON pointer (which the spec says they do).
4195  */
4196 static int handle_vmon(struct kvm_vcpu *vcpu)
4197 {
4198 	int ret;
4199 	gpa_t vmptr;
4200 	struct page *page;
4201 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4202 	const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
4203 		| FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
4204 
4205 	/*
4206 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4207 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4208 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4209 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4210 	 * have already been checked by hardware, prior to the VM-exit for
4211 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4212 	 * that bit set to 1 in non-root mode.
4213 	 */
4214 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4215 		kvm_queue_exception(vcpu, UD_VECTOR);
4216 		return 1;
4217 	}
4218 
4219 	/* CPL=0 must be checked manually. */
4220 	if (vmx_get_cpl(vcpu)) {
4221 		kvm_inject_gp(vcpu, 0);
4222 		return 1;
4223 	}
4224 
4225 	if (vmx->nested.vmxon)
4226 		return nested_vmx_failValid(vcpu,
4227 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4228 
4229 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4230 			!= VMXON_NEEDED_FEATURES) {
4231 		kvm_inject_gp(vcpu, 0);
4232 		return 1;
4233 	}
4234 
4235 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4236 		return 1;
4237 
4238 	/*
4239 	 * SDM 3: 24.11.5
4240 	 * The first 4 bytes of VMXON region contain the supported
4241 	 * VMCS revision identifier
4242 	 *
4243 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4244 	 * which replaces physical address width with 32
4245 	 */
4246 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4247 		return nested_vmx_failInvalid(vcpu);
4248 
4249 	page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4250 	if (is_error_page(page))
4251 		return nested_vmx_failInvalid(vcpu);
4252 
4253 	if (*(u32 *)kmap(page) != VMCS12_REVISION) {
4254 		kunmap(page);
4255 		kvm_release_page_clean(page);
4256 		return nested_vmx_failInvalid(vcpu);
4257 	}
4258 	kunmap(page);
4259 	kvm_release_page_clean(page);
4260 
4261 	vmx->nested.vmxon_ptr = vmptr;
4262 	ret = enter_vmx_operation(vcpu);
4263 	if (ret)
4264 		return ret;
4265 
4266 	return nested_vmx_succeed(vcpu);
4267 }
4268 
4269 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4270 {
4271 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4272 
4273 	if (vmx->nested.current_vmptr == -1ull)
4274 		return;
4275 
4276 	if (enable_shadow_vmcs) {
4277 		/* copy to memory all shadowed fields in case
4278 		   they were modified */
4279 		copy_shadow_to_vmcs12(vmx);
4280 		vmx->nested.need_vmcs12_sync = false;
4281 		vmx_disable_shadow_vmcs(vmx);
4282 	}
4283 	vmx->nested.posted_intr_nv = -1;
4284 
4285 	/* Flush VMCS12 to guest memory */
4286 	kvm_vcpu_write_guest_page(vcpu,
4287 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4288 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4289 
4290 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4291 
4292 	vmx->nested.current_vmptr = -1ull;
4293 }
4294 
4295 /* Emulate the VMXOFF instruction */
4296 static int handle_vmoff(struct kvm_vcpu *vcpu)
4297 {
4298 	if (!nested_vmx_check_permission(vcpu))
4299 		return 1;
4300 	free_nested(vcpu);
4301 	return nested_vmx_succeed(vcpu);
4302 }
4303 
4304 /* Emulate the VMCLEAR instruction */
4305 static int handle_vmclear(struct kvm_vcpu *vcpu)
4306 {
4307 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4308 	u32 zero = 0;
4309 	gpa_t vmptr;
4310 
4311 	if (!nested_vmx_check_permission(vcpu))
4312 		return 1;
4313 
4314 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4315 		return 1;
4316 
4317 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4318 		return nested_vmx_failValid(vcpu,
4319 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4320 
4321 	if (vmptr == vmx->nested.vmxon_ptr)
4322 		return nested_vmx_failValid(vcpu,
4323 			VMXERR_VMCLEAR_VMXON_POINTER);
4324 
4325 	if (vmx->nested.hv_evmcs_page) {
4326 		if (vmptr == vmx->nested.hv_evmcs_vmptr)
4327 			nested_release_evmcs(vcpu);
4328 	} else {
4329 		if (vmptr == vmx->nested.current_vmptr)
4330 			nested_release_vmcs12(vcpu);
4331 
4332 		kvm_vcpu_write_guest(vcpu,
4333 				     vmptr + offsetof(struct vmcs12,
4334 						      launch_state),
4335 				     &zero, sizeof(zero));
4336 	}
4337 
4338 	return nested_vmx_succeed(vcpu);
4339 }
4340 
4341 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4342 
4343 /* Emulate the VMLAUNCH instruction */
4344 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4345 {
4346 	return nested_vmx_run(vcpu, true);
4347 }
4348 
4349 /* Emulate the VMRESUME instruction */
4350 static int handle_vmresume(struct kvm_vcpu *vcpu)
4351 {
4352 
4353 	return nested_vmx_run(vcpu, false);
4354 }
4355 
4356 static int handle_vmread(struct kvm_vcpu *vcpu)
4357 {
4358 	unsigned long field;
4359 	u64 field_value;
4360 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4361 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4362 	gva_t gva = 0;
4363 	struct vmcs12 *vmcs12;
4364 
4365 	if (!nested_vmx_check_permission(vcpu))
4366 		return 1;
4367 
4368 	if (to_vmx(vcpu)->nested.current_vmptr == -1ull)
4369 		return nested_vmx_failInvalid(vcpu);
4370 
4371 	if (!is_guest_mode(vcpu))
4372 		vmcs12 = get_vmcs12(vcpu);
4373 	else {
4374 		/*
4375 		 * When vmcs->vmcs_link_pointer is -1ull, any VMREAD
4376 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4377 		 */
4378 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4379 			return nested_vmx_failInvalid(vcpu);
4380 		vmcs12 = get_shadow_vmcs12(vcpu);
4381 	}
4382 
4383 	/* Decode instruction info and find the field to read */
4384 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4385 	/* Read the field, zero-extended to a u64 field_value */
4386 	if (vmcs12_read_any(vmcs12, field, &field_value) < 0)
4387 		return nested_vmx_failValid(vcpu,
4388 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4389 
4390 	/*
4391 	 * Now copy part of this value to register or memory, as requested.
4392 	 * Note that the number of bits actually copied is 32 or 64 depending
4393 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4394 	 */
4395 	if (vmx_instruction_info & (1u << 10)) {
4396 		kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
4397 			field_value);
4398 	} else {
4399 		if (get_vmx_mem_address(vcpu, exit_qualification,
4400 				vmx_instruction_info, true, &gva))
4401 			return 1;
4402 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4403 		kvm_write_guest_virt_system(vcpu, gva, &field_value,
4404 					    (is_long_mode(vcpu) ? 8 : 4), NULL);
4405 	}
4406 
4407 	return nested_vmx_succeed(vcpu);
4408 }
4409 
4410 
4411 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4412 {
4413 	unsigned long field;
4414 	gva_t gva;
4415 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4416 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4417 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4418 
4419 	/* The value to write might be 32 or 64 bits, depending on L1's long
4420 	 * mode, and eventually we need to write that into a field of several
4421 	 * possible lengths. The code below first zero-extends the value to 64
4422 	 * bit (field_value), and then copies only the appropriate number of
4423 	 * bits into the vmcs12 field.
4424 	 */
4425 	u64 field_value = 0;
4426 	struct x86_exception e;
4427 	struct vmcs12 *vmcs12;
4428 
4429 	if (!nested_vmx_check_permission(vcpu))
4430 		return 1;
4431 
4432 	if (vmx->nested.current_vmptr == -1ull)
4433 		return nested_vmx_failInvalid(vcpu);
4434 
4435 	if (vmx_instruction_info & (1u << 10))
4436 		field_value = kvm_register_readl(vcpu,
4437 			(((vmx_instruction_info) >> 3) & 0xf));
4438 	else {
4439 		if (get_vmx_mem_address(vcpu, exit_qualification,
4440 				vmx_instruction_info, false, &gva))
4441 			return 1;
4442 		if (kvm_read_guest_virt(vcpu, gva, &field_value,
4443 					(is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
4444 			kvm_inject_page_fault(vcpu, &e);
4445 			return 1;
4446 		}
4447 	}
4448 
4449 
4450 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4451 	/*
4452 	 * If the vCPU supports "VMWRITE to any supported field in the
4453 	 * VMCS," then the "read-only" fields are actually read/write.
4454 	 */
4455 	if (vmcs_field_readonly(field) &&
4456 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4457 		return nested_vmx_failValid(vcpu,
4458 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4459 
4460 	if (!is_guest_mode(vcpu))
4461 		vmcs12 = get_vmcs12(vcpu);
4462 	else {
4463 		/*
4464 		 * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
4465 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4466 		 */
4467 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4468 			return nested_vmx_failInvalid(vcpu);
4469 		vmcs12 = get_shadow_vmcs12(vcpu);
4470 	}
4471 
4472 	if (vmcs12_write_any(vmcs12, field, field_value) < 0)
4473 		return nested_vmx_failValid(vcpu,
4474 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4475 
4476 	/*
4477 	 * Do not track vmcs12 dirty-state if in guest-mode
4478 	 * as we actually dirty shadow vmcs12 instead of vmcs12.
4479 	 */
4480 	if (!is_guest_mode(vcpu)) {
4481 		switch (field) {
4482 #define SHADOW_FIELD_RW(x) case x:
4483 #include "vmcs_shadow_fields.h"
4484 			/*
4485 			 * The fields that can be updated by L1 without a vmexit are
4486 			 * always updated in the vmcs02, the others go down the slow
4487 			 * path of prepare_vmcs02.
4488 			 */
4489 			break;
4490 		default:
4491 			vmx->nested.dirty_vmcs12 = true;
4492 			break;
4493 		}
4494 	}
4495 
4496 	return nested_vmx_succeed(vcpu);
4497 }
4498 
4499 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
4500 {
4501 	vmx->nested.current_vmptr = vmptr;
4502 	if (enable_shadow_vmcs) {
4503 		vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4504 			      SECONDARY_EXEC_SHADOW_VMCS);
4505 		vmcs_write64(VMCS_LINK_POINTER,
4506 			     __pa(vmx->vmcs01.shadow_vmcs));
4507 		vmx->nested.need_vmcs12_sync = true;
4508 	}
4509 	vmx->nested.dirty_vmcs12 = true;
4510 }
4511 
4512 /* Emulate the VMPTRLD instruction */
4513 static int handle_vmptrld(struct kvm_vcpu *vcpu)
4514 {
4515 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4516 	gpa_t vmptr;
4517 
4518 	if (!nested_vmx_check_permission(vcpu))
4519 		return 1;
4520 
4521 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4522 		return 1;
4523 
4524 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4525 		return nested_vmx_failValid(vcpu,
4526 			VMXERR_VMPTRLD_INVALID_ADDRESS);
4527 
4528 	if (vmptr == vmx->nested.vmxon_ptr)
4529 		return nested_vmx_failValid(vcpu,
4530 			VMXERR_VMPTRLD_VMXON_POINTER);
4531 
4532 	/* Forbid normal VMPTRLD if Enlightened version was used */
4533 	if (vmx->nested.hv_evmcs)
4534 		return 1;
4535 
4536 	if (vmx->nested.current_vmptr != vmptr) {
4537 		struct vmcs12 *new_vmcs12;
4538 		struct page *page;
4539 
4540 		page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4541 		if (is_error_page(page)) {
4542 			/*
4543 			 * Reads from an unbacked page return all 1s,
4544 			 * which means that the 32 bits located at the
4545 			 * given physical address won't match the required
4546 			 * VMCS12_REVISION identifier.
4547 			 */
4548 			return nested_vmx_failValid(vcpu,
4549 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4550 		}
4551 		new_vmcs12 = kmap(page);
4552 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
4553 		    (new_vmcs12->hdr.shadow_vmcs &&
4554 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
4555 			kunmap(page);
4556 			kvm_release_page_clean(page);
4557 			return nested_vmx_failValid(vcpu,
4558 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4559 		}
4560 
4561 		nested_release_vmcs12(vcpu);
4562 
4563 		/*
4564 		 * Load VMCS12 from guest memory since it is not already
4565 		 * cached.
4566 		 */
4567 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
4568 		kunmap(page);
4569 		kvm_release_page_clean(page);
4570 
4571 		set_current_vmptr(vmx, vmptr);
4572 	}
4573 
4574 	return nested_vmx_succeed(vcpu);
4575 }
4576 
4577 /* Emulate the VMPTRST instruction */
4578 static int handle_vmptrst(struct kvm_vcpu *vcpu)
4579 {
4580 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
4581 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4582 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
4583 	struct x86_exception e;
4584 	gva_t gva;
4585 
4586 	if (!nested_vmx_check_permission(vcpu))
4587 		return 1;
4588 
4589 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
4590 		return 1;
4591 
4592 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info, true, &gva))
4593 		return 1;
4594 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
4595 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
4596 					sizeof(gpa_t), &e)) {
4597 		kvm_inject_page_fault(vcpu, &e);
4598 		return 1;
4599 	}
4600 	return nested_vmx_succeed(vcpu);
4601 }
4602 
4603 /* Emulate the INVEPT instruction */
4604 static int handle_invept(struct kvm_vcpu *vcpu)
4605 {
4606 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4607 	u32 vmx_instruction_info, types;
4608 	unsigned long type;
4609 	gva_t gva;
4610 	struct x86_exception e;
4611 	struct {
4612 		u64 eptp, gpa;
4613 	} operand;
4614 
4615 	if (!(vmx->nested.msrs.secondary_ctls_high &
4616 	      SECONDARY_EXEC_ENABLE_EPT) ||
4617 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
4618 		kvm_queue_exception(vcpu, UD_VECTOR);
4619 		return 1;
4620 	}
4621 
4622 	if (!nested_vmx_check_permission(vcpu))
4623 		return 1;
4624 
4625 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4626 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4627 
4628 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
4629 
4630 	if (type >= 32 || !(types & (1 << type)))
4631 		return nested_vmx_failValid(vcpu,
4632 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4633 
4634 	/* According to the Intel VMX instruction reference, the memory
4635 	 * operand is read even if it isn't needed (e.g., for type==global)
4636 	 */
4637 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4638 			vmx_instruction_info, false, &gva))
4639 		return 1;
4640 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4641 		kvm_inject_page_fault(vcpu, &e);
4642 		return 1;
4643 	}
4644 
4645 	switch (type) {
4646 	case VMX_EPT_EXTENT_GLOBAL:
4647 	/*
4648 	 * TODO: track mappings and invalidate
4649 	 * single context requests appropriately
4650 	 */
4651 	case VMX_EPT_EXTENT_CONTEXT:
4652 		kvm_mmu_sync_roots(vcpu);
4653 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4654 		break;
4655 	default:
4656 		BUG_ON(1);
4657 		break;
4658 	}
4659 
4660 	return nested_vmx_succeed(vcpu);
4661 }
4662 
4663 static int handle_invvpid(struct kvm_vcpu *vcpu)
4664 {
4665 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4666 	u32 vmx_instruction_info;
4667 	unsigned long type, types;
4668 	gva_t gva;
4669 	struct x86_exception e;
4670 	struct {
4671 		u64 vpid;
4672 		u64 gla;
4673 	} operand;
4674 	u16 vpid02;
4675 
4676 	if (!(vmx->nested.msrs.secondary_ctls_high &
4677 	      SECONDARY_EXEC_ENABLE_VPID) ||
4678 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
4679 		kvm_queue_exception(vcpu, UD_VECTOR);
4680 		return 1;
4681 	}
4682 
4683 	if (!nested_vmx_check_permission(vcpu))
4684 		return 1;
4685 
4686 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4687 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4688 
4689 	types = (vmx->nested.msrs.vpid_caps &
4690 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
4691 
4692 	if (type >= 32 || !(types & (1 << type)))
4693 		return nested_vmx_failValid(vcpu,
4694 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4695 
4696 	/* according to the intel vmx instruction reference, the memory
4697 	 * operand is read even if it isn't needed (e.g., for type==global)
4698 	 */
4699 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4700 			vmx_instruction_info, false, &gva))
4701 		return 1;
4702 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4703 		kvm_inject_page_fault(vcpu, &e);
4704 		return 1;
4705 	}
4706 	if (operand.vpid >> 16)
4707 		return nested_vmx_failValid(vcpu,
4708 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4709 
4710 	vpid02 = nested_get_vpid02(vcpu);
4711 	switch (type) {
4712 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
4713 		if (!operand.vpid ||
4714 		    is_noncanonical_address(operand.gla, vcpu))
4715 			return nested_vmx_failValid(vcpu,
4716 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4717 		if (cpu_has_vmx_invvpid_individual_addr()) {
4718 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
4719 				vpid02, operand.gla);
4720 		} else
4721 			__vmx_flush_tlb(vcpu, vpid02, false);
4722 		break;
4723 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
4724 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
4725 		if (!operand.vpid)
4726 			return nested_vmx_failValid(vcpu,
4727 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4728 		__vmx_flush_tlb(vcpu, vpid02, false);
4729 		break;
4730 	case VMX_VPID_EXTENT_ALL_CONTEXT:
4731 		__vmx_flush_tlb(vcpu, vpid02, false);
4732 		break;
4733 	default:
4734 		WARN_ON_ONCE(1);
4735 		return kvm_skip_emulated_instruction(vcpu);
4736 	}
4737 
4738 	return nested_vmx_succeed(vcpu);
4739 }
4740 
4741 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
4742 				     struct vmcs12 *vmcs12)
4743 {
4744 	u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
4745 	u64 address;
4746 	bool accessed_dirty;
4747 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4748 
4749 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
4750 	    !nested_cpu_has_ept(vmcs12))
4751 		return 1;
4752 
4753 	if (index >= VMFUNC_EPTP_ENTRIES)
4754 		return 1;
4755 
4756 
4757 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
4758 				     &address, index * 8, 8))
4759 		return 1;
4760 
4761 	accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
4762 
4763 	/*
4764 	 * If the (L2) guest does a vmfunc to the currently
4765 	 * active ept pointer, we don't have to do anything else
4766 	 */
4767 	if (vmcs12->ept_pointer != address) {
4768 		if (!valid_ept_address(vcpu, address))
4769 			return 1;
4770 
4771 		kvm_mmu_unload(vcpu);
4772 		mmu->ept_ad = accessed_dirty;
4773 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
4774 		vmcs12->ept_pointer = address;
4775 		/*
4776 		 * TODO: Check what's the correct approach in case
4777 		 * mmu reload fails. Currently, we just let the next
4778 		 * reload potentially fail
4779 		 */
4780 		kvm_mmu_reload(vcpu);
4781 	}
4782 
4783 	return 0;
4784 }
4785 
4786 static int handle_vmfunc(struct kvm_vcpu *vcpu)
4787 {
4788 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4789 	struct vmcs12 *vmcs12;
4790 	u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
4791 
4792 	/*
4793 	 * VMFUNC is only supported for nested guests, but we always enable the
4794 	 * secondary control for simplicity; for non-nested mode, fake that we
4795 	 * didn't by injecting #UD.
4796 	 */
4797 	if (!is_guest_mode(vcpu)) {
4798 		kvm_queue_exception(vcpu, UD_VECTOR);
4799 		return 1;
4800 	}
4801 
4802 	vmcs12 = get_vmcs12(vcpu);
4803 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
4804 		goto fail;
4805 
4806 	switch (function) {
4807 	case 0:
4808 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
4809 			goto fail;
4810 		break;
4811 	default:
4812 		goto fail;
4813 	}
4814 	return kvm_skip_emulated_instruction(vcpu);
4815 
4816 fail:
4817 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
4818 			  vmcs_read32(VM_EXIT_INTR_INFO),
4819 			  vmcs_readl(EXIT_QUALIFICATION));
4820 	return 1;
4821 }
4822 
4823 
4824 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
4825 				       struct vmcs12 *vmcs12)
4826 {
4827 	unsigned long exit_qualification;
4828 	gpa_t bitmap, last_bitmap;
4829 	unsigned int port;
4830 	int size;
4831 	u8 b;
4832 
4833 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
4834 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
4835 
4836 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4837 
4838 	port = exit_qualification >> 16;
4839 	size = (exit_qualification & 7) + 1;
4840 
4841 	last_bitmap = (gpa_t)-1;
4842 	b = -1;
4843 
4844 	while (size > 0) {
4845 		if (port < 0x8000)
4846 			bitmap = vmcs12->io_bitmap_a;
4847 		else if (port < 0x10000)
4848 			bitmap = vmcs12->io_bitmap_b;
4849 		else
4850 			return true;
4851 		bitmap += (port & 0x7fff) / 8;
4852 
4853 		if (last_bitmap != bitmap)
4854 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
4855 				return true;
4856 		if (b & (1 << (port & 7)))
4857 			return true;
4858 
4859 		port++;
4860 		size--;
4861 		last_bitmap = bitmap;
4862 	}
4863 
4864 	return false;
4865 }
4866 
4867 /*
4868  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
4869  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
4870  * disinterest in the current event (read or write a specific MSR) by using an
4871  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
4872  */
4873 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
4874 	struct vmcs12 *vmcs12, u32 exit_reason)
4875 {
4876 	u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
4877 	gpa_t bitmap;
4878 
4879 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
4880 		return true;
4881 
4882 	/*
4883 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
4884 	 * for the four combinations of read/write and low/high MSR numbers.
4885 	 * First we need to figure out which of the four to use:
4886 	 */
4887 	bitmap = vmcs12->msr_bitmap;
4888 	if (exit_reason == EXIT_REASON_MSR_WRITE)
4889 		bitmap += 2048;
4890 	if (msr_index >= 0xc0000000) {
4891 		msr_index -= 0xc0000000;
4892 		bitmap += 1024;
4893 	}
4894 
4895 	/* Then read the msr_index'th bit from this bitmap: */
4896 	if (msr_index < 1024*8) {
4897 		unsigned char b;
4898 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
4899 			return true;
4900 		return 1 & (b >> (msr_index & 7));
4901 	} else
4902 		return true; /* let L1 handle the wrong parameter */
4903 }
4904 
4905 /*
4906  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
4907  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
4908  * intercept (via guest_host_mask etc.) the current event.
4909  */
4910 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
4911 	struct vmcs12 *vmcs12)
4912 {
4913 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4914 	int cr = exit_qualification & 15;
4915 	int reg;
4916 	unsigned long val;
4917 
4918 	switch ((exit_qualification >> 4) & 3) {
4919 	case 0: /* mov to cr */
4920 		reg = (exit_qualification >> 8) & 15;
4921 		val = kvm_register_readl(vcpu, reg);
4922 		switch (cr) {
4923 		case 0:
4924 			if (vmcs12->cr0_guest_host_mask &
4925 			    (val ^ vmcs12->cr0_read_shadow))
4926 				return true;
4927 			break;
4928 		case 3:
4929 			if ((vmcs12->cr3_target_count >= 1 &&
4930 					vmcs12->cr3_target_value0 == val) ||
4931 				(vmcs12->cr3_target_count >= 2 &&
4932 					vmcs12->cr3_target_value1 == val) ||
4933 				(vmcs12->cr3_target_count >= 3 &&
4934 					vmcs12->cr3_target_value2 == val) ||
4935 				(vmcs12->cr3_target_count >= 4 &&
4936 					vmcs12->cr3_target_value3 == val))
4937 				return false;
4938 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
4939 				return true;
4940 			break;
4941 		case 4:
4942 			if (vmcs12->cr4_guest_host_mask &
4943 			    (vmcs12->cr4_read_shadow ^ val))
4944 				return true;
4945 			break;
4946 		case 8:
4947 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
4948 				return true;
4949 			break;
4950 		}
4951 		break;
4952 	case 2: /* clts */
4953 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
4954 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
4955 			return true;
4956 		break;
4957 	case 1: /* mov from cr */
4958 		switch (cr) {
4959 		case 3:
4960 			if (vmcs12->cpu_based_vm_exec_control &
4961 			    CPU_BASED_CR3_STORE_EXITING)
4962 				return true;
4963 			break;
4964 		case 8:
4965 			if (vmcs12->cpu_based_vm_exec_control &
4966 			    CPU_BASED_CR8_STORE_EXITING)
4967 				return true;
4968 			break;
4969 		}
4970 		break;
4971 	case 3: /* lmsw */
4972 		/*
4973 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
4974 		 * cr0. Other attempted changes are ignored, with no exit.
4975 		 */
4976 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4977 		if (vmcs12->cr0_guest_host_mask & 0xe &
4978 		    (val ^ vmcs12->cr0_read_shadow))
4979 			return true;
4980 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
4981 		    !(vmcs12->cr0_read_shadow & 0x1) &&
4982 		    (val & 0x1))
4983 			return true;
4984 		break;
4985 	}
4986 	return false;
4987 }
4988 
4989 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
4990 	struct vmcs12 *vmcs12, gpa_t bitmap)
4991 {
4992 	u32 vmx_instruction_info;
4993 	unsigned long field;
4994 	u8 b;
4995 
4996 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
4997 		return true;
4998 
4999 	/* Decode instruction info and find the field to access */
5000 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5001 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5002 
5003 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5004 	if (field >> 15)
5005 		return true;
5006 
5007 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5008 		return true;
5009 
5010 	return 1 & (b >> (field & 7));
5011 }
5012 
5013 /*
5014  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5015  * should handle it ourselves in L0 (and then continue L2). Only call this
5016  * when in is_guest_mode (L2).
5017  */
5018 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5019 {
5020 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5021 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5022 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5023 
5024 	if (vmx->nested.nested_run_pending)
5025 		return false;
5026 
5027 	if (unlikely(vmx->fail)) {
5028 		pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5029 				    vmcs_read32(VM_INSTRUCTION_ERROR));
5030 		return true;
5031 	}
5032 
5033 	/*
5034 	 * The host physical addresses of some pages of guest memory
5035 	 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5036 	 * Page). The CPU may write to these pages via their host
5037 	 * physical address while L2 is running, bypassing any
5038 	 * address-translation-based dirty tracking (e.g. EPT write
5039 	 * protection).
5040 	 *
5041 	 * Mark them dirty on every exit from L2 to prevent them from
5042 	 * getting out of sync with dirty tracking.
5043 	 */
5044 	nested_mark_vmcs12_pages_dirty(vcpu);
5045 
5046 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5047 				vmcs_readl(EXIT_QUALIFICATION),
5048 				vmx->idt_vectoring_info,
5049 				intr_info,
5050 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5051 				KVM_ISA_VMX);
5052 
5053 	switch (exit_reason) {
5054 	case EXIT_REASON_EXCEPTION_NMI:
5055 		if (is_nmi(intr_info))
5056 			return false;
5057 		else if (is_page_fault(intr_info))
5058 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5059 		else if (is_debug(intr_info) &&
5060 			 vcpu->guest_debug &
5061 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5062 			return false;
5063 		else if (is_breakpoint(intr_info) &&
5064 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5065 			return false;
5066 		return vmcs12->exception_bitmap &
5067 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5068 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5069 		return false;
5070 	case EXIT_REASON_TRIPLE_FAULT:
5071 		return true;
5072 	case EXIT_REASON_PENDING_INTERRUPT:
5073 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
5074 	case EXIT_REASON_NMI_WINDOW:
5075 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
5076 	case EXIT_REASON_TASK_SWITCH:
5077 		return true;
5078 	case EXIT_REASON_CPUID:
5079 		return true;
5080 	case EXIT_REASON_HLT:
5081 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5082 	case EXIT_REASON_INVD:
5083 		return true;
5084 	case EXIT_REASON_INVLPG:
5085 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5086 	case EXIT_REASON_RDPMC:
5087 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5088 	case EXIT_REASON_RDRAND:
5089 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5090 	case EXIT_REASON_RDSEED:
5091 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5092 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5093 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5094 	case EXIT_REASON_VMREAD:
5095 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5096 			vmcs12->vmread_bitmap);
5097 	case EXIT_REASON_VMWRITE:
5098 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5099 			vmcs12->vmwrite_bitmap);
5100 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5101 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5102 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5103 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5104 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5105 		/*
5106 		 * VMX instructions trap unconditionally. This allows L1 to
5107 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5108 		 */
5109 		return true;
5110 	case EXIT_REASON_CR_ACCESS:
5111 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5112 	case EXIT_REASON_DR_ACCESS:
5113 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5114 	case EXIT_REASON_IO_INSTRUCTION:
5115 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5116 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5117 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5118 	case EXIT_REASON_MSR_READ:
5119 	case EXIT_REASON_MSR_WRITE:
5120 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5121 	case EXIT_REASON_INVALID_STATE:
5122 		return true;
5123 	case EXIT_REASON_MWAIT_INSTRUCTION:
5124 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5125 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5126 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
5127 	case EXIT_REASON_MONITOR_INSTRUCTION:
5128 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5129 	case EXIT_REASON_PAUSE_INSTRUCTION:
5130 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5131 			nested_cpu_has2(vmcs12,
5132 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5133 	case EXIT_REASON_MCE_DURING_VMENTRY:
5134 		return false;
5135 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5136 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5137 	case EXIT_REASON_APIC_ACCESS:
5138 	case EXIT_REASON_APIC_WRITE:
5139 	case EXIT_REASON_EOI_INDUCED:
5140 		/*
5141 		 * The controls for "virtualize APIC accesses," "APIC-
5142 		 * register virtualization," and "virtual-interrupt
5143 		 * delivery" only come from vmcs12.
5144 		 */
5145 		return true;
5146 	case EXIT_REASON_EPT_VIOLATION:
5147 		/*
5148 		 * L0 always deals with the EPT violation. If nested EPT is
5149 		 * used, and the nested mmu code discovers that the address is
5150 		 * missing in the guest EPT table (EPT12), the EPT violation
5151 		 * will be injected with nested_ept_inject_page_fault()
5152 		 */
5153 		return false;
5154 	case EXIT_REASON_EPT_MISCONFIG:
5155 		/*
5156 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5157 		 * table (shadow on EPT) or a merged EPT table that L0 built
5158 		 * (EPT on EPT). So any problems with the structure of the
5159 		 * table is L0's fault.
5160 		 */
5161 		return false;
5162 	case EXIT_REASON_INVPCID:
5163 		return
5164 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5165 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5166 	case EXIT_REASON_WBINVD:
5167 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5168 	case EXIT_REASON_XSETBV:
5169 		return true;
5170 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5171 		/*
5172 		 * This should never happen, since it is not possible to
5173 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5174 		 * If if it were, XSS would have to be checked against
5175 		 * the XSS exit bitmap in vmcs12.
5176 		 */
5177 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5178 	case EXIT_REASON_PREEMPTION_TIMER:
5179 		return false;
5180 	case EXIT_REASON_PML_FULL:
5181 		/* We emulate PML support to L1. */
5182 		return false;
5183 	case EXIT_REASON_VMFUNC:
5184 		/* VM functions are emulated through L2->L0 vmexits. */
5185 		return false;
5186 	case EXIT_REASON_ENCLS:
5187 		/* SGX is never exposed to L1 */
5188 		return false;
5189 	default:
5190 		return true;
5191 	}
5192 }
5193 
5194 
5195 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5196 				struct kvm_nested_state __user *user_kvm_nested_state,
5197 				u32 user_data_size)
5198 {
5199 	struct vcpu_vmx *vmx;
5200 	struct vmcs12 *vmcs12;
5201 	struct kvm_nested_state kvm_state = {
5202 		.flags = 0,
5203 		.format = 0,
5204 		.size = sizeof(kvm_state),
5205 		.vmx.vmxon_pa = -1ull,
5206 		.vmx.vmcs_pa = -1ull,
5207 	};
5208 
5209 	if (!vcpu)
5210 		return kvm_state.size + 2 * VMCS12_SIZE;
5211 
5212 	vmx = to_vmx(vcpu);
5213 	vmcs12 = get_vmcs12(vcpu);
5214 
5215 	if (nested_vmx_allowed(vcpu) && vmx->nested.enlightened_vmcs_enabled)
5216 		kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5217 
5218 	if (nested_vmx_allowed(vcpu) &&
5219 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5220 		kvm_state.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5221 		kvm_state.vmx.vmcs_pa = vmx->nested.current_vmptr;
5222 
5223 		if (vmx_has_valid_vmcs12(vcpu)) {
5224 			kvm_state.size += VMCS12_SIZE;
5225 
5226 			if (is_guest_mode(vcpu) &&
5227 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5228 			    vmcs12->vmcs_link_pointer != -1ull)
5229 				kvm_state.size += VMCS12_SIZE;
5230 		}
5231 
5232 		if (vmx->nested.smm.vmxon)
5233 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5234 
5235 		if (vmx->nested.smm.guest_mode)
5236 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5237 
5238 		if (is_guest_mode(vcpu)) {
5239 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5240 
5241 			if (vmx->nested.nested_run_pending)
5242 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5243 		}
5244 	}
5245 
5246 	if (user_data_size < kvm_state.size)
5247 		goto out;
5248 
5249 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5250 		return -EFAULT;
5251 
5252 	if (!vmx_has_valid_vmcs12(vcpu))
5253 		goto out;
5254 
5255 	/*
5256 	 * When running L2, the authoritative vmcs12 state is in the
5257 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5258 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5259 	 * need_vmcs12_sync is set, in which case, the authoritative
5260 	 * vmcs12 state is in the vmcs12 already.
5261 	 */
5262 	if (is_guest_mode(vcpu)) {
5263 		sync_vmcs12(vcpu, vmcs12);
5264 	} else if (!vmx->nested.need_vmcs12_sync) {
5265 		if (vmx->nested.hv_evmcs)
5266 			copy_enlightened_to_vmcs12(vmx);
5267 		else if (enable_shadow_vmcs)
5268 			copy_shadow_to_vmcs12(vmx);
5269 	}
5270 
5271 	/*
5272 	 * Copy over the full allocated size of vmcs12 rather than just the size
5273 	 * of the struct.
5274 	 */
5275 	if (copy_to_user(user_kvm_nested_state->data, vmcs12, VMCS12_SIZE))
5276 		return -EFAULT;
5277 
5278 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5279 	    vmcs12->vmcs_link_pointer != -1ull) {
5280 		if (copy_to_user(user_kvm_nested_state->data + VMCS12_SIZE,
5281 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
5282 			return -EFAULT;
5283 	}
5284 
5285 out:
5286 	return kvm_state.size;
5287 }
5288 
5289 /*
5290  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5291  */
5292 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5293 {
5294 	if (is_guest_mode(vcpu)) {
5295 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5296 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5297 	}
5298 	free_nested(vcpu);
5299 }
5300 
5301 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5302 				struct kvm_nested_state __user *user_kvm_nested_state,
5303 				struct kvm_nested_state *kvm_state)
5304 {
5305 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5306 	struct vmcs12 *vmcs12;
5307 	u32 exit_qual;
5308 	int ret;
5309 
5310 	if (kvm_state->format != 0)
5311 		return -EINVAL;
5312 
5313 	if (kvm_state->flags & KVM_STATE_NESTED_EVMCS)
5314 		nested_enable_evmcs(vcpu, NULL);
5315 
5316 	if (!nested_vmx_allowed(vcpu))
5317 		return kvm_state->vmx.vmxon_pa == -1ull ? 0 : -EINVAL;
5318 
5319 	if (kvm_state->vmx.vmxon_pa == -1ull) {
5320 		if (kvm_state->vmx.smm.flags)
5321 			return -EINVAL;
5322 
5323 		if (kvm_state->vmx.vmcs_pa != -1ull)
5324 			return -EINVAL;
5325 
5326 		vmx_leave_nested(vcpu);
5327 		return 0;
5328 	}
5329 
5330 	if (!page_address_valid(vcpu, kvm_state->vmx.vmxon_pa))
5331 		return -EINVAL;
5332 
5333 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5334 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5335 		return -EINVAL;
5336 
5337 	if (kvm_state->vmx.smm.flags &
5338 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5339 		return -EINVAL;
5340 
5341 	/*
5342 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5343 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5344 	 * must be zero.
5345 	 */
5346 	if (is_smm(vcpu) ? kvm_state->flags : kvm_state->vmx.smm.flags)
5347 		return -EINVAL;
5348 
5349 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5350 	    !(kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5351 		return -EINVAL;
5352 
5353 	vmx_leave_nested(vcpu);
5354 	if (kvm_state->vmx.vmxon_pa == -1ull)
5355 		return 0;
5356 
5357 	vmx->nested.vmxon_ptr = kvm_state->vmx.vmxon_pa;
5358 	ret = enter_vmx_operation(vcpu);
5359 	if (ret)
5360 		return ret;
5361 
5362 	/* Empty 'VMXON' state is permitted */
5363 	if (kvm_state->size < sizeof(kvm_state) + sizeof(*vmcs12))
5364 		return 0;
5365 
5366 	if (kvm_state->vmx.vmcs_pa != -1ull) {
5367 		if (kvm_state->vmx.vmcs_pa == kvm_state->vmx.vmxon_pa ||
5368 		    !page_address_valid(vcpu, kvm_state->vmx.vmcs_pa))
5369 			return -EINVAL;
5370 
5371 		set_current_vmptr(vmx, kvm_state->vmx.vmcs_pa);
5372 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5373 		/*
5374 		 * Sync eVMCS upon entry as we may not have
5375 		 * HV_X64_MSR_VP_ASSIST_PAGE set up yet.
5376 		 */
5377 		vmx->nested.need_vmcs12_sync = true;
5378 	} else {
5379 		return -EINVAL;
5380 	}
5381 
5382 	if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5383 		vmx->nested.smm.vmxon = true;
5384 		vmx->nested.vmxon = false;
5385 
5386 		if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5387 			vmx->nested.smm.guest_mode = true;
5388 	}
5389 
5390 	vmcs12 = get_vmcs12(vcpu);
5391 	if (copy_from_user(vmcs12, user_kvm_nested_state->data, sizeof(*vmcs12)))
5392 		return -EFAULT;
5393 
5394 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5395 		return -EINVAL;
5396 
5397 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5398 		return 0;
5399 
5400 	vmx->nested.nested_run_pending =
5401 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5402 
5403 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5404 	    vmcs12->vmcs_link_pointer != -1ull) {
5405 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5406 
5407 		if (kvm_state->size < sizeof(kvm_state) + 2 * sizeof(*vmcs12))
5408 			return -EINVAL;
5409 
5410 		if (copy_from_user(shadow_vmcs12,
5411 				   user_kvm_nested_state->data + VMCS12_SIZE,
5412 				   sizeof(*vmcs12)))
5413 			return -EFAULT;
5414 
5415 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5416 		    !shadow_vmcs12->hdr.shadow_vmcs)
5417 			return -EINVAL;
5418 	}
5419 
5420 	if (nested_vmx_check_vmentry_prereqs(vcpu, vmcs12) ||
5421 	    nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
5422 		return -EINVAL;
5423 
5424 	vmx->nested.dirty_vmcs12 = true;
5425 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5426 	if (ret)
5427 		return -EINVAL;
5428 
5429 	return 0;
5430 }
5431 
5432 void nested_vmx_vcpu_setup(void)
5433 {
5434 	if (enable_shadow_vmcs) {
5435 		/*
5436 		 * At vCPU creation, "VMWRITE to any supported field
5437 		 * in the VMCS" is supported, so use the more
5438 		 * permissive vmx_vmread_bitmap to specify both read
5439 		 * and write permissions for the shadow VMCS.
5440 		 */
5441 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5442 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmread_bitmap));
5443 	}
5444 }
5445 
5446 /*
5447  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5448  * returned for the various VMX controls MSRs when nested VMX is enabled.
5449  * The same values should also be used to verify that vmcs12 control fields are
5450  * valid during nested entry from L1 to L2.
5451  * Each of these control msrs has a low and high 32-bit half: A low bit is on
5452  * if the corresponding bit in the (32-bit) control field *must* be on, and a
5453  * bit in the high half is on if the corresponding bit in the control field
5454  * may be on. See also vmx_control_verify().
5455  */
5456 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps,
5457 				bool apicv)
5458 {
5459 	/*
5460 	 * Note that as a general rule, the high half of the MSRs (bits in
5461 	 * the control fields which may be 1) should be initialized by the
5462 	 * intersection of the underlying hardware's MSR (i.e., features which
5463 	 * can be supported) and the list of features we want to expose -
5464 	 * because they are known to be properly supported in our code.
5465 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
5466 	 * be set to 0, meaning that L1 may turn off any of these bits. The
5467 	 * reason is that if one of these bits is necessary, it will appear
5468 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
5469 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
5470 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
5471 	 * These rules have exceptions below.
5472 	 */
5473 
5474 	/* pin-based controls */
5475 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
5476 		msrs->pinbased_ctls_low,
5477 		msrs->pinbased_ctls_high);
5478 	msrs->pinbased_ctls_low |=
5479 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5480 	msrs->pinbased_ctls_high &=
5481 		PIN_BASED_EXT_INTR_MASK |
5482 		PIN_BASED_NMI_EXITING |
5483 		PIN_BASED_VIRTUAL_NMIS |
5484 		(apicv ? PIN_BASED_POSTED_INTR : 0);
5485 	msrs->pinbased_ctls_high |=
5486 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5487 		PIN_BASED_VMX_PREEMPTION_TIMER;
5488 
5489 	/* exit controls */
5490 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
5491 		msrs->exit_ctls_low,
5492 		msrs->exit_ctls_high);
5493 	msrs->exit_ctls_low =
5494 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
5495 
5496 	msrs->exit_ctls_high &=
5497 #ifdef CONFIG_X86_64
5498 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
5499 #endif
5500 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
5501 	msrs->exit_ctls_high |=
5502 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
5503 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
5504 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
5505 
5506 	/* We support free control of debug control saving. */
5507 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
5508 
5509 	/* entry controls */
5510 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
5511 		msrs->entry_ctls_low,
5512 		msrs->entry_ctls_high);
5513 	msrs->entry_ctls_low =
5514 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
5515 	msrs->entry_ctls_high &=
5516 #ifdef CONFIG_X86_64
5517 		VM_ENTRY_IA32E_MODE |
5518 #endif
5519 		VM_ENTRY_LOAD_IA32_PAT;
5520 	msrs->entry_ctls_high |=
5521 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
5522 
5523 	/* We support free control of debug control loading. */
5524 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
5525 
5526 	/* cpu-based controls */
5527 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
5528 		msrs->procbased_ctls_low,
5529 		msrs->procbased_ctls_high);
5530 	msrs->procbased_ctls_low =
5531 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5532 	msrs->procbased_ctls_high &=
5533 		CPU_BASED_VIRTUAL_INTR_PENDING |
5534 		CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
5535 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
5536 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
5537 		CPU_BASED_CR3_STORE_EXITING |
5538 #ifdef CONFIG_X86_64
5539 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
5540 #endif
5541 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
5542 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
5543 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
5544 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
5545 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
5546 	/*
5547 	 * We can allow some features even when not supported by the
5548 	 * hardware. For example, L1 can specify an MSR bitmap - and we
5549 	 * can use it to avoid exits to L1 - even when L0 runs L2
5550 	 * without MSR bitmaps.
5551 	 */
5552 	msrs->procbased_ctls_high |=
5553 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5554 		CPU_BASED_USE_MSR_BITMAPS;
5555 
5556 	/* We support free control of CR3 access interception. */
5557 	msrs->procbased_ctls_low &=
5558 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
5559 
5560 	/*
5561 	 * secondary cpu-based controls.  Do not include those that
5562 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
5563 	 */
5564 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
5565 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
5566 		      msrs->secondary_ctls_low,
5567 		      msrs->secondary_ctls_high);
5568 
5569 	msrs->secondary_ctls_low = 0;
5570 	msrs->secondary_ctls_high &=
5571 		SECONDARY_EXEC_DESC |
5572 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
5573 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
5574 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
5575 		SECONDARY_EXEC_WBINVD_EXITING;
5576 
5577 	/*
5578 	 * We can emulate "VMCS shadowing," even if the hardware
5579 	 * doesn't support it.
5580 	 */
5581 	msrs->secondary_ctls_high |=
5582 		SECONDARY_EXEC_SHADOW_VMCS;
5583 
5584 	if (enable_ept) {
5585 		/* nested EPT: emulate EPT also to L1 */
5586 		msrs->secondary_ctls_high |=
5587 			SECONDARY_EXEC_ENABLE_EPT;
5588 		msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
5589 			 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
5590 		if (cpu_has_vmx_ept_execute_only())
5591 			msrs->ept_caps |=
5592 				VMX_EPT_EXECUTE_ONLY_BIT;
5593 		msrs->ept_caps &= ept_caps;
5594 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
5595 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
5596 			VMX_EPT_1GB_PAGE_BIT;
5597 		if (enable_ept_ad_bits) {
5598 			msrs->secondary_ctls_high |=
5599 				SECONDARY_EXEC_ENABLE_PML;
5600 			msrs->ept_caps |= VMX_EPT_AD_BIT;
5601 		}
5602 	}
5603 
5604 	if (cpu_has_vmx_vmfunc()) {
5605 		msrs->secondary_ctls_high |=
5606 			SECONDARY_EXEC_ENABLE_VMFUNC;
5607 		/*
5608 		 * Advertise EPTP switching unconditionally
5609 		 * since we emulate it
5610 		 */
5611 		if (enable_ept)
5612 			msrs->vmfunc_controls =
5613 				VMX_VMFUNC_EPTP_SWITCHING;
5614 	}
5615 
5616 	/*
5617 	 * Old versions of KVM use the single-context version without
5618 	 * checking for support, so declare that it is supported even
5619 	 * though it is treated as global context.  The alternative is
5620 	 * not failing the single-context invvpid, and it is worse.
5621 	 */
5622 	if (enable_vpid) {
5623 		msrs->secondary_ctls_high |=
5624 			SECONDARY_EXEC_ENABLE_VPID;
5625 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
5626 			VMX_VPID_EXTENT_SUPPORTED_MASK;
5627 	}
5628 
5629 	if (enable_unrestricted_guest)
5630 		msrs->secondary_ctls_high |=
5631 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
5632 
5633 	if (flexpriority_enabled)
5634 		msrs->secondary_ctls_high |=
5635 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5636 
5637 	/* miscellaneous data */
5638 	rdmsr(MSR_IA32_VMX_MISC,
5639 		msrs->misc_low,
5640 		msrs->misc_high);
5641 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
5642 	msrs->misc_low |=
5643 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
5644 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
5645 		VMX_MISC_ACTIVITY_HLT;
5646 	msrs->misc_high = 0;
5647 
5648 	/*
5649 	 * This MSR reports some information about VMX support. We
5650 	 * should return information about the VMX we emulate for the
5651 	 * guest, and the VMCS structure we give it - not about the
5652 	 * VMX support of the underlying hardware.
5653 	 */
5654 	msrs->basic =
5655 		VMCS12_REVISION |
5656 		VMX_BASIC_TRUE_CTLS |
5657 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
5658 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
5659 
5660 	if (cpu_has_vmx_basic_inout())
5661 		msrs->basic |= VMX_BASIC_INOUT;
5662 
5663 	/*
5664 	 * These MSRs specify bits which the guest must keep fixed on
5665 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
5666 	 * We picked the standard core2 setting.
5667 	 */
5668 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
5669 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
5670 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
5671 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
5672 
5673 	/* These MSRs specify bits which the guest must keep fixed off. */
5674 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
5675 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
5676 
5677 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
5678 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
5679 }
5680 
5681 void nested_vmx_hardware_unsetup(void)
5682 {
5683 	int i;
5684 
5685 	if (enable_shadow_vmcs) {
5686 		for (i = 0; i < VMX_BITMAP_NR; i++)
5687 			free_page((unsigned long)vmx_bitmap[i]);
5688 	}
5689 }
5690 
5691 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
5692 {
5693 	int i;
5694 
5695 	if (!cpu_has_vmx_shadow_vmcs())
5696 		enable_shadow_vmcs = 0;
5697 	if (enable_shadow_vmcs) {
5698 		for (i = 0; i < VMX_BITMAP_NR; i++) {
5699 			vmx_bitmap[i] = (unsigned long *)
5700 				__get_free_page(GFP_KERNEL);
5701 			if (!vmx_bitmap[i]) {
5702 				nested_vmx_hardware_unsetup();
5703 				return -ENOMEM;
5704 			}
5705 		}
5706 
5707 		init_vmcs_shadow_fields();
5708 	}
5709 
5710 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear,
5711 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch,
5712 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld,
5713 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst,
5714 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread,
5715 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume,
5716 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite,
5717 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff,
5718 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon,
5719 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept,
5720 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid,
5721 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc,
5722 
5723 	kvm_x86_ops->check_nested_events = vmx_check_nested_events;
5724 	kvm_x86_ops->get_nested_state = vmx_get_nested_state;
5725 	kvm_x86_ops->set_nested_state = vmx_set_nested_state;
5726 	kvm_x86_ops->get_vmcs12_pages = nested_get_vmcs12_pages,
5727 	kvm_x86_ops->nested_enable_evmcs = nested_enable_evmcs;
5728 	kvm_x86_ops->nested_get_evmcs_version = nested_get_evmcs_version;
5729 
5730 	return 0;
5731 }
5732