xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision 5b628549)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/frame.h>
4 #include <linux/percpu.h>
5 
6 #include <asm/debugreg.h>
7 #include <asm/mmu_context.h>
8 
9 #include "cpuid.h"
10 #include "hyperv.h"
11 #include "mmu.h"
12 #include "nested.h"
13 #include "trace.h"
14 #include "x86.h"
15 
16 static bool __read_mostly enable_shadow_vmcs = 1;
17 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
18 
19 static bool __read_mostly nested_early_check = 0;
20 module_param(nested_early_check, bool, S_IRUGO);
21 
22 /*
23  * Hyper-V requires all of these, so mark them as supported even though
24  * they are just treated the same as all-context.
25  */
26 #define VMX_VPID_EXTENT_SUPPORTED_MASK		\
27 	(VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT |	\
28 	VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |	\
29 	VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT |	\
30 	VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
31 
32 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
33 
34 enum {
35 	VMX_VMREAD_BITMAP,
36 	VMX_VMWRITE_BITMAP,
37 	VMX_BITMAP_NR
38 };
39 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
40 
41 #define vmx_vmread_bitmap                    (vmx_bitmap[VMX_VMREAD_BITMAP])
42 #define vmx_vmwrite_bitmap                   (vmx_bitmap[VMX_VMWRITE_BITMAP])
43 
44 static u16 shadow_read_only_fields[] = {
45 #define SHADOW_FIELD_RO(x) x,
46 #include "vmcs_shadow_fields.h"
47 };
48 static int max_shadow_read_only_fields =
49 	ARRAY_SIZE(shadow_read_only_fields);
50 
51 static u16 shadow_read_write_fields[] = {
52 #define SHADOW_FIELD_RW(x) x,
53 #include "vmcs_shadow_fields.h"
54 };
55 static int max_shadow_read_write_fields =
56 	ARRAY_SIZE(shadow_read_write_fields);
57 
58 static void init_vmcs_shadow_fields(void)
59 {
60 	int i, j;
61 
62 	memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
63 	memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
64 
65 	for (i = j = 0; i < max_shadow_read_only_fields; i++) {
66 		u16 field = shadow_read_only_fields[i];
67 
68 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
69 		    (i + 1 == max_shadow_read_only_fields ||
70 		     shadow_read_only_fields[i + 1] != field + 1))
71 			pr_err("Missing field from shadow_read_only_field %x\n",
72 			       field + 1);
73 
74 		clear_bit(field, vmx_vmread_bitmap);
75 #ifdef CONFIG_X86_64
76 		if (field & 1)
77 			continue;
78 #endif
79 		if (j < i)
80 			shadow_read_only_fields[j] = field;
81 		j++;
82 	}
83 	max_shadow_read_only_fields = j;
84 
85 	for (i = j = 0; i < max_shadow_read_write_fields; i++) {
86 		u16 field = shadow_read_write_fields[i];
87 
88 		if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
89 		    (i + 1 == max_shadow_read_write_fields ||
90 		     shadow_read_write_fields[i + 1] != field + 1))
91 			pr_err("Missing field from shadow_read_write_field %x\n",
92 			       field + 1);
93 
94 		/*
95 		 * PML and the preemption timer can be emulated, but the
96 		 * processor cannot vmwrite to fields that don't exist
97 		 * on bare metal.
98 		 */
99 		switch (field) {
100 		case GUEST_PML_INDEX:
101 			if (!cpu_has_vmx_pml())
102 				continue;
103 			break;
104 		case VMX_PREEMPTION_TIMER_VALUE:
105 			if (!cpu_has_vmx_preemption_timer())
106 				continue;
107 			break;
108 		case GUEST_INTR_STATUS:
109 			if (!cpu_has_vmx_apicv())
110 				continue;
111 			break;
112 		default:
113 			break;
114 		}
115 
116 		clear_bit(field, vmx_vmwrite_bitmap);
117 		clear_bit(field, vmx_vmread_bitmap);
118 #ifdef CONFIG_X86_64
119 		if (field & 1)
120 			continue;
121 #endif
122 		if (j < i)
123 			shadow_read_write_fields[j] = field;
124 		j++;
125 	}
126 	max_shadow_read_write_fields = j;
127 }
128 
129 /*
130  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
131  * set the success or error code of an emulated VMX instruction (as specified
132  * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
133  * instruction.
134  */
135 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
136 {
137 	vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
138 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
139 			    X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
140 	return kvm_skip_emulated_instruction(vcpu);
141 }
142 
143 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
144 {
145 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
146 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
147 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
148 			| X86_EFLAGS_CF);
149 	return kvm_skip_emulated_instruction(vcpu);
150 }
151 
152 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
153 				u32 vm_instruction_error)
154 {
155 	struct vcpu_vmx *vmx = to_vmx(vcpu);
156 
157 	/*
158 	 * failValid writes the error number to the current VMCS, which
159 	 * can't be done if there isn't a current VMCS.
160 	 */
161 	if (vmx->nested.current_vmptr == -1ull && !vmx->nested.hv_evmcs)
162 		return nested_vmx_failInvalid(vcpu);
163 
164 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
165 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
166 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
167 			| X86_EFLAGS_ZF);
168 	get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
169 	/*
170 	 * We don't need to force a shadow sync because
171 	 * VM_INSTRUCTION_ERROR is not shadowed
172 	 */
173 	return kvm_skip_emulated_instruction(vcpu);
174 }
175 
176 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
177 {
178 	/* TODO: not to reset guest simply here. */
179 	kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
180 	pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
181 }
182 
183 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
184 {
185 	vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
186 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
187 }
188 
189 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
190 {
191 	struct vcpu_vmx *vmx = to_vmx(vcpu);
192 
193 	if (!vmx->nested.hv_evmcs)
194 		return;
195 
196 	kunmap(vmx->nested.hv_evmcs_page);
197 	kvm_release_page_dirty(vmx->nested.hv_evmcs_page);
198 	vmx->nested.hv_evmcs_vmptr = -1ull;
199 	vmx->nested.hv_evmcs_page = NULL;
200 	vmx->nested.hv_evmcs = NULL;
201 }
202 
203 /*
204  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
205  * just stops using VMX.
206  */
207 static void free_nested(struct kvm_vcpu *vcpu)
208 {
209 	struct vcpu_vmx *vmx = to_vmx(vcpu);
210 
211 	if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
212 		return;
213 
214 	vmx->nested.vmxon = false;
215 	vmx->nested.smm.vmxon = false;
216 	free_vpid(vmx->nested.vpid02);
217 	vmx->nested.posted_intr_nv = -1;
218 	vmx->nested.current_vmptr = -1ull;
219 	if (enable_shadow_vmcs) {
220 		vmx_disable_shadow_vmcs(vmx);
221 		vmcs_clear(vmx->vmcs01.shadow_vmcs);
222 		free_vmcs(vmx->vmcs01.shadow_vmcs);
223 		vmx->vmcs01.shadow_vmcs = NULL;
224 	}
225 	kfree(vmx->nested.cached_vmcs12);
226 	kfree(vmx->nested.cached_shadow_vmcs12);
227 	/* Unpin physical memory we referred to in the vmcs02 */
228 	if (vmx->nested.apic_access_page) {
229 		kvm_release_page_dirty(vmx->nested.apic_access_page);
230 		vmx->nested.apic_access_page = NULL;
231 	}
232 	if (vmx->nested.virtual_apic_page) {
233 		kvm_release_page_dirty(vmx->nested.virtual_apic_page);
234 		vmx->nested.virtual_apic_page = NULL;
235 	}
236 	if (vmx->nested.pi_desc_page) {
237 		kunmap(vmx->nested.pi_desc_page);
238 		kvm_release_page_dirty(vmx->nested.pi_desc_page);
239 		vmx->nested.pi_desc_page = NULL;
240 		vmx->nested.pi_desc = NULL;
241 	}
242 
243 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
244 
245 	nested_release_evmcs(vcpu);
246 
247 	free_loaded_vmcs(&vmx->nested.vmcs02);
248 }
249 
250 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
251 {
252 	struct vcpu_vmx *vmx = to_vmx(vcpu);
253 	int cpu;
254 
255 	if (vmx->loaded_vmcs == vmcs)
256 		return;
257 
258 	cpu = get_cpu();
259 	vmx_vcpu_put(vcpu);
260 	vmx->loaded_vmcs = vmcs;
261 	vmx_vcpu_load(vcpu, cpu);
262 	put_cpu();
263 
264 	vm_entry_controls_reset_shadow(vmx);
265 	vm_exit_controls_reset_shadow(vmx);
266 	vmx_segment_cache_clear(vmx);
267 }
268 
269 /*
270  * Ensure that the current vmcs of the logical processor is the
271  * vmcs01 of the vcpu before calling free_nested().
272  */
273 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
274 {
275 	vcpu_load(vcpu);
276 	vmx_leave_nested(vcpu);
277 	vmx_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
278 	free_nested(vcpu);
279 	vcpu_put(vcpu);
280 }
281 
282 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
283 		struct x86_exception *fault)
284 {
285 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
286 	struct vcpu_vmx *vmx = to_vmx(vcpu);
287 	u32 exit_reason;
288 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
289 
290 	if (vmx->nested.pml_full) {
291 		exit_reason = EXIT_REASON_PML_FULL;
292 		vmx->nested.pml_full = false;
293 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
294 	} else if (fault->error_code & PFERR_RSVD_MASK)
295 		exit_reason = EXIT_REASON_EPT_MISCONFIG;
296 	else
297 		exit_reason = EXIT_REASON_EPT_VIOLATION;
298 
299 	nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
300 	vmcs12->guest_physical_address = fault->address;
301 }
302 
303 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
304 {
305 	WARN_ON(mmu_is_nested(vcpu));
306 
307 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
308 	kvm_init_shadow_ept_mmu(vcpu,
309 			to_vmx(vcpu)->nested.msrs.ept_caps &
310 			VMX_EPT_EXECUTE_ONLY_BIT,
311 			nested_ept_ad_enabled(vcpu),
312 			nested_ept_get_cr3(vcpu));
313 	vcpu->arch.mmu->set_cr3           = vmx_set_cr3;
314 	vcpu->arch.mmu->get_cr3           = nested_ept_get_cr3;
315 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
316 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
317 
318 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
319 }
320 
321 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
322 {
323 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
324 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
325 }
326 
327 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
328 					    u16 error_code)
329 {
330 	bool inequality, bit;
331 
332 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
333 	inequality =
334 		(error_code & vmcs12->page_fault_error_code_mask) !=
335 		 vmcs12->page_fault_error_code_match;
336 	return inequality ^ bit;
337 }
338 
339 
340 /*
341  * KVM wants to inject page-faults which it got to the guest. This function
342  * checks whether in a nested guest, we need to inject them to L1 or L2.
343  */
344 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
345 {
346 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
347 	unsigned int nr = vcpu->arch.exception.nr;
348 	bool has_payload = vcpu->arch.exception.has_payload;
349 	unsigned long payload = vcpu->arch.exception.payload;
350 
351 	if (nr == PF_VECTOR) {
352 		if (vcpu->arch.exception.nested_apf) {
353 			*exit_qual = vcpu->arch.apf.nested_apf_token;
354 			return 1;
355 		}
356 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
357 						    vcpu->arch.exception.error_code)) {
358 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
359 			return 1;
360 		}
361 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
362 		if (nr == DB_VECTOR) {
363 			if (!has_payload) {
364 				payload = vcpu->arch.dr6;
365 				payload &= ~(DR6_FIXED_1 | DR6_BT);
366 				payload ^= DR6_RTM;
367 			}
368 			*exit_qual = payload;
369 		} else
370 			*exit_qual = 0;
371 		return 1;
372 	}
373 
374 	return 0;
375 }
376 
377 
378 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
379 		struct x86_exception *fault)
380 {
381 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
382 
383 	WARN_ON(!is_guest_mode(vcpu));
384 
385 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
386 		!to_vmx(vcpu)->nested.nested_run_pending) {
387 		vmcs12->vm_exit_intr_error_code = fault->error_code;
388 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
389 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
390 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
391 				  fault->address);
392 	} else {
393 		kvm_inject_page_fault(vcpu, fault);
394 	}
395 }
396 
397 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
398 {
399 	return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
400 }
401 
402 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
403 					       struct vmcs12 *vmcs12)
404 {
405 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
406 		return 0;
407 
408 	if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
409 	    !page_address_valid(vcpu, vmcs12->io_bitmap_b))
410 		return -EINVAL;
411 
412 	return 0;
413 }
414 
415 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
416 						struct vmcs12 *vmcs12)
417 {
418 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
419 		return 0;
420 
421 	if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
422 		return -EINVAL;
423 
424 	return 0;
425 }
426 
427 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
428 						struct vmcs12 *vmcs12)
429 {
430 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
431 		return 0;
432 
433 	if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
434 		return -EINVAL;
435 
436 	return 0;
437 }
438 
439 /*
440  * Check if MSR is intercepted for L01 MSR bitmap.
441  */
442 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
443 {
444 	unsigned long *msr_bitmap;
445 	int f = sizeof(unsigned long);
446 
447 	if (!cpu_has_vmx_msr_bitmap())
448 		return true;
449 
450 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
451 
452 	if (msr <= 0x1fff) {
453 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
454 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
455 		msr &= 0x1fff;
456 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
457 	}
458 
459 	return true;
460 }
461 
462 /*
463  * If a msr is allowed by L0, we should check whether it is allowed by L1.
464  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
465  */
466 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
467 					       unsigned long *msr_bitmap_nested,
468 					       u32 msr, int type)
469 {
470 	int f = sizeof(unsigned long);
471 
472 	/*
473 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
474 	 * have the write-low and read-high bitmap offsets the wrong way round.
475 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
476 	 */
477 	if (msr <= 0x1fff) {
478 		if (type & MSR_TYPE_R &&
479 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
480 			/* read-low */
481 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
482 
483 		if (type & MSR_TYPE_W &&
484 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
485 			/* write-low */
486 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
487 
488 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
489 		msr &= 0x1fff;
490 		if (type & MSR_TYPE_R &&
491 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
492 			/* read-high */
493 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
494 
495 		if (type & MSR_TYPE_W &&
496 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
497 			/* write-high */
498 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
499 
500 	}
501 }
502 
503 /*
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 	 * PIN CONTROLS
1984 	 */
1985 	exec_control = vmcs12->pin_based_vm_exec_control;
1986 
1987 	/* Preemption timer setting is computed directly in vmx_vcpu_run.  */
1988 	exec_control |= vmcs_config.pin_based_exec_ctrl;
1989 	exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
1990 	vmx->loaded_vmcs->hv_timer_armed = false;
1991 
1992 	/* Posted interrupts setting is only taken from vmcs12.  */
1993 	if (nested_cpu_has_posted_intr(vmcs12)) {
1994 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
1995 		vmx->nested.pi_pending = false;
1996 	} else {
1997 		exec_control &= ~PIN_BASED_POSTED_INTR;
1998 	}
1999 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
2000 
2001 	/*
2002 	 * EXEC CONTROLS
2003 	 */
2004 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2005 	exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2006 	exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2007 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2008 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2009 
2010 	/*
2011 	 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
2012 	 * nested_get_vmcs12_pages can't fix it up, the illegal value
2013 	 * will result in a VM entry failure.
2014 	 */
2015 	if (exec_control & CPU_BASED_TPR_SHADOW) {
2016 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
2017 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2018 	} else {
2019 #ifdef CONFIG_X86_64
2020 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2021 				CPU_BASED_CR8_STORE_EXITING;
2022 #endif
2023 	}
2024 
2025 	/*
2026 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2027 	 * for I/O port accesses.
2028 	 */
2029 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2030 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2031 	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2032 
2033 	/*
2034 	 * SECONDARY EXEC CONTROLS
2035 	 */
2036 	if (cpu_has_secondary_exec_ctrls()) {
2037 		exec_control = vmx->secondary_exec_control;
2038 
2039 		/* Take the following fields only from vmcs12 */
2040 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2041 				  SECONDARY_EXEC_ENABLE_INVPCID |
2042 				  SECONDARY_EXEC_RDTSCP |
2043 				  SECONDARY_EXEC_XSAVES |
2044 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2045 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2046 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2047 		if (nested_cpu_has(vmcs12,
2048 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2049 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2050 				~SECONDARY_EXEC_ENABLE_PML;
2051 			exec_control |= vmcs12_exec_ctrl;
2052 		}
2053 
2054 		/* VMCS shadowing for L2 is emulated for now */
2055 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2056 
2057 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2058 			vmcs_write16(GUEST_INTR_STATUS,
2059 				vmcs12->guest_intr_status);
2060 
2061 		/*
2062 		 * Write an illegal value to APIC_ACCESS_ADDR. Later,
2063 		 * nested_get_vmcs12_pages will either fix it up or
2064 		 * remove the VM execution control.
2065 		 */
2066 		if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
2067 			vmcs_write64(APIC_ACCESS_ADDR, -1ull);
2068 
2069 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2070 			vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2071 
2072 		vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2073 	}
2074 
2075 	/*
2076 	 * ENTRY CONTROLS
2077 	 *
2078 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2079 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2080 	 * on the related bits (if supported by the CPU) in the hope that
2081 	 * we can avoid VMWrites during vmx_set_efer().
2082 	 */
2083 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2084 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2085 	if (cpu_has_load_ia32_efer()) {
2086 		if (guest_efer & EFER_LMA)
2087 			exec_control |= VM_ENTRY_IA32E_MODE;
2088 		if (guest_efer != host_efer)
2089 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2090 	}
2091 	vm_entry_controls_init(vmx, exec_control);
2092 
2093 	/*
2094 	 * EXIT CONTROLS
2095 	 *
2096 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2097 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2098 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2099 	 */
2100 	exec_control = vmx_vmexit_ctrl();
2101 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2102 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2103 	vm_exit_controls_init(vmx, exec_control);
2104 
2105 	/*
2106 	 * Conceptually we want to copy the PML address and index from
2107 	 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
2108 	 * since we always flush the log on each vmexit and never change
2109 	 * the PML address (once set), this happens to be equivalent to
2110 	 * simply resetting the index in vmcs02.
2111 	 */
2112 	if (enable_pml)
2113 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2114 
2115 	/*
2116 	 * Interrupt/Exception Fields
2117 	 */
2118 	if (vmx->nested.nested_run_pending) {
2119 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2120 			     vmcs12->vm_entry_intr_info_field);
2121 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2122 			     vmcs12->vm_entry_exception_error_code);
2123 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2124 			     vmcs12->vm_entry_instruction_len);
2125 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2126 			     vmcs12->guest_interruptibility_info);
2127 		vmx->loaded_vmcs->nmi_known_unmasked =
2128 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2129 	} else {
2130 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2131 	}
2132 }
2133 
2134 static void prepare_vmcs02_full(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2135 {
2136 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2137 
2138 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2139 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2140 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2141 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2142 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2143 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2144 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2145 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2146 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2147 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2148 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2149 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2150 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2151 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2152 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2153 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2154 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2155 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2156 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2157 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2158 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2159 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2160 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2161 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2162 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2163 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2164 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2165 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2166 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2167 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2168 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2169 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2170 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2171 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2172 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2173 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2174 	}
2175 
2176 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2177 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2178 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2179 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2180 			    vmcs12->guest_pending_dbg_exceptions);
2181 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2182 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2183 
2184 		/*
2185 		 * L1 may access the L2's PDPTR, so save them to construct
2186 		 * vmcs12
2187 		 */
2188 		if (enable_ept) {
2189 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2190 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2191 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2192 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2193 		}
2194 	}
2195 
2196 	if (nested_cpu_has_xsaves(vmcs12))
2197 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2198 
2199 	/*
2200 	 * Whether page-faults are trapped is determined by a combination of
2201 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2202 	 * If enable_ept, L0 doesn't care about page faults and we should
2203 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2204 	 * care about (at least some) page faults, and because it is not easy
2205 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2206 	 * to exit on each and every L2 page fault. This is done by setting
2207 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2208 	 * Note that below we don't need special code to set EB.PF beyond the
2209 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2210 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2211 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2212 	 */
2213 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2214 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2215 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2216 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2217 
2218 	if (cpu_has_vmx_apicv()) {
2219 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2220 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2221 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2222 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2223 	}
2224 
2225 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2226 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2227 
2228 	set_cr4_guest_host_mask(vmx);
2229 
2230 	if (kvm_mpx_supported()) {
2231 		if (vmx->nested.nested_run_pending &&
2232 			(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2233 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2234 		else
2235 			vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2236 	}
2237 }
2238 
2239 /*
2240  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2241  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2242  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2243  * guest in a way that will both be appropriate to L1's requests, and our
2244  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2245  * function also has additional necessary side-effects, like setting various
2246  * vcpu->arch fields.
2247  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2248  * is assigned to entry_failure_code on failure.
2249  */
2250 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2251 			  u32 *entry_failure_code)
2252 {
2253 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2254 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2255 
2256 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs) {
2257 		prepare_vmcs02_full(vmx, vmcs12);
2258 		vmx->nested.dirty_vmcs12 = false;
2259 	}
2260 
2261 	/*
2262 	 * First, the fields that are shadowed.  This must be kept in sync
2263 	 * with vmcs_shadow_fields.h.
2264 	 */
2265 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2266 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2267 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2268 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2269 	}
2270 
2271 	if (vmx->nested.nested_run_pending &&
2272 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2273 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2274 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2275 	} else {
2276 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2277 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2278 	}
2279 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2280 
2281 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2282 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2283 	 * trap. Note that CR0.TS also needs updating - we do this later.
2284 	 */
2285 	update_exception_bitmap(vcpu);
2286 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2287 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2288 
2289 	if (vmx->nested.nested_run_pending &&
2290 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2291 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2292 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2293 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2294 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2295 	}
2296 
2297 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2298 
2299 	if (kvm_has_tsc_control)
2300 		decache_tsc_multiplier(vmx);
2301 
2302 	if (enable_vpid) {
2303 		/*
2304 		 * There is no direct mapping between vpid02 and vpid12, the
2305 		 * vpid02 is per-vCPU for L0 and reused while the value of
2306 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2307 		 * The vpid12 is allocated by L1 for L2, so it will not
2308 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2309 		 * even if spawn a lot of nested vCPUs.
2310 		 */
2311 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2312 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2313 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2314 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2315 			}
2316 		} else {
2317 			/*
2318 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2319 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2320 			 * flush until vmcs02->eptp is fully updated by
2321 			 * KVM_REQ_LOAD_CR3. Note that this assumes
2322 			 * KVM_REQ_TLB_FLUSH is evaluated after
2323 			 * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
2324 			 */
2325 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2326 		}
2327 	}
2328 
2329 	if (nested_cpu_has_ept(vmcs12))
2330 		nested_ept_init_mmu_context(vcpu);
2331 	else if (nested_cpu_has2(vmcs12,
2332 				 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2333 		vmx_flush_tlb(vcpu, true);
2334 
2335 	/*
2336 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2337 	 * bits which we consider mandatory enabled.
2338 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2339 	 * the specifications by L1; It's not enough to take
2340 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2341 	 * have more bits than L1 expected.
2342 	 */
2343 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2344 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2345 
2346 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2347 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2348 
2349 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2350 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2351 	vmx_set_efer(vcpu, vcpu->arch.efer);
2352 
2353 	/*
2354 	 * Guest state is invalid and unrestricted guest is disabled,
2355 	 * which means L1 attempted VMEntry to L2 with invalid state.
2356 	 * Fail the VMEntry.
2357 	 */
2358 	if (vmx->emulation_required) {
2359 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2360 		return 1;
2361 	}
2362 
2363 	/* Shadow page tables on either EPT or shadow page tables. */
2364 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2365 				entry_failure_code))
2366 		return 1;
2367 
2368 	if (!enable_ept)
2369 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2370 
2371 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
2372 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
2373 	return 0;
2374 }
2375 
2376 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2377 {
2378 	if (!nested_cpu_has_nmi_exiting(vmcs12) &&
2379 	    nested_cpu_has_virtual_nmis(vmcs12))
2380 		return -EINVAL;
2381 
2382 	if (!nested_cpu_has_virtual_nmis(vmcs12) &&
2383 	    nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING))
2384 		return -EINVAL;
2385 
2386 	return 0;
2387 }
2388 
2389 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
2390 {
2391 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2392 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2393 
2394 	/* Check for memory type validity */
2395 	switch (address & VMX_EPTP_MT_MASK) {
2396 	case VMX_EPTP_MT_UC:
2397 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT))
2398 			return false;
2399 		break;
2400 	case VMX_EPTP_MT_WB:
2401 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT))
2402 			return false;
2403 		break;
2404 	default:
2405 		return false;
2406 	}
2407 
2408 	/* only 4 levels page-walk length are valid */
2409 	if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
2410 		return false;
2411 
2412 	/* Reserved bits should not be set */
2413 	if (address >> maxphyaddr || ((address >> 7) & 0x1f))
2414 		return false;
2415 
2416 	/* AD, if set, should be supported */
2417 	if (address & VMX_EPTP_AD_ENABLE_BIT) {
2418 		if (!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT))
2419 			return false;
2420 	}
2421 
2422 	return true;
2423 }
2424 
2425 /*
2426  * Checks related to VM-Execution Control Fields
2427  */
2428 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2429                                               struct vmcs12 *vmcs12)
2430 {
2431 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2432 
2433 	if (!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2434 				vmx->nested.msrs.pinbased_ctls_low,
2435 				vmx->nested.msrs.pinbased_ctls_high) ||
2436 	    !vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2437 				vmx->nested.msrs.procbased_ctls_low,
2438 				vmx->nested.msrs.procbased_ctls_high))
2439 		return -EINVAL;
2440 
2441 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2442 	    !vmx_control_verify(vmcs12->secondary_vm_exec_control,
2443 				 vmx->nested.msrs.secondary_ctls_low,
2444 				 vmx->nested.msrs.secondary_ctls_high))
2445 		return -EINVAL;
2446 
2447 	if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu) ||
2448 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2449 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2450 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2451 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2452 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2453 	    nested_vmx_check_nmi_controls(vmcs12) ||
2454 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2455 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2456 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2457 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2458 	    (nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2459 		return -EINVAL;
2460 
2461 	if (!nested_cpu_has_preemption_timer(vmcs12) &&
2462 	    nested_cpu_has_save_preemption_timer(vmcs12))
2463 		return -EINVAL;
2464 
2465 	if (nested_cpu_has_ept(vmcs12) &&
2466 	    !valid_ept_address(vcpu, vmcs12->ept_pointer))
2467 		return -EINVAL;
2468 
2469 	if (nested_cpu_has_vmfunc(vmcs12)) {
2470 		if (vmcs12->vm_function_control &
2471 		    ~vmx->nested.msrs.vmfunc_controls)
2472 			return -EINVAL;
2473 
2474 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2475 			if (!nested_cpu_has_ept(vmcs12) ||
2476 			    !page_address_valid(vcpu, vmcs12->eptp_list_address))
2477 				return -EINVAL;
2478 		}
2479 	}
2480 
2481 	return 0;
2482 }
2483 
2484 /*
2485  * Checks related to VM-Exit Control Fields
2486  */
2487 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2488                                          struct vmcs12 *vmcs12)
2489 {
2490 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2491 
2492 	if (!vmx_control_verify(vmcs12->vm_exit_controls,
2493 				vmx->nested.msrs.exit_ctls_low,
2494 				vmx->nested.msrs.exit_ctls_high) ||
2495 	    nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12))
2496 		return -EINVAL;
2497 
2498 	return 0;
2499 }
2500 
2501 /*
2502  * Checks related to VM-Entry Control Fields
2503  */
2504 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2505 					  struct vmcs12 *vmcs12)
2506 {
2507 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2508 
2509 	if (!vmx_control_verify(vmcs12->vm_entry_controls,
2510 				vmx->nested.msrs.entry_ctls_low,
2511 				vmx->nested.msrs.entry_ctls_high))
2512 		return -EINVAL;
2513 
2514 	/*
2515 	 * From the Intel SDM, volume 3:
2516 	 * Fields relevant to VM-entry event injection must be set properly.
2517 	 * These fields are the VM-entry interruption-information field, the
2518 	 * VM-entry exception error code, and the VM-entry instruction length.
2519 	 */
2520 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2521 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2522 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2523 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2524 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2525 		bool should_have_error_code;
2526 		bool urg = nested_cpu_has2(vmcs12,
2527 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2528 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2529 
2530 		/* VM-entry interruption-info field: interruption type */
2531 		if (intr_type == INTR_TYPE_RESERVED ||
2532 		    (intr_type == INTR_TYPE_OTHER_EVENT &&
2533 		     !nested_cpu_supports_monitor_trap_flag(vcpu)))
2534 			return -EINVAL;
2535 
2536 		/* VM-entry interruption-info field: vector */
2537 		if ((intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2538 		    (intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2539 		    (intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2540 			return -EINVAL;
2541 
2542 		/* VM-entry interruption-info field: deliver error code */
2543 		should_have_error_code =
2544 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2545 			x86_exception_has_error_code(vector);
2546 		if (has_error_code != should_have_error_code)
2547 			return -EINVAL;
2548 
2549 		/* VM-entry exception error code */
2550 		if (has_error_code &&
2551 		    vmcs12->vm_entry_exception_error_code & GENMASK(31, 15))
2552 			return -EINVAL;
2553 
2554 		/* VM-entry interruption-info field: reserved bits */
2555 		if (intr_info & INTR_INFO_RESVD_BITS_MASK)
2556 			return -EINVAL;
2557 
2558 		/* VM-entry instruction length */
2559 		switch (intr_type) {
2560 		case INTR_TYPE_SOFT_EXCEPTION:
2561 		case INTR_TYPE_SOFT_INTR:
2562 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2563 			if ((vmcs12->vm_entry_instruction_len > 15) ||
2564 			    (vmcs12->vm_entry_instruction_len == 0 &&
2565 			     !nested_cpu_has_zero_length_injection(vcpu)))
2566 				return -EINVAL;
2567 		}
2568 	}
2569 
2570 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2571 		return -EINVAL;
2572 
2573 	return 0;
2574 }
2575 
2576 /*
2577  * Checks related to Host Control Registers and MSRs
2578  */
2579 static int nested_check_host_control_regs(struct kvm_vcpu *vcpu,
2580                                           struct vmcs12 *vmcs12)
2581 {
2582 	bool ia32e;
2583 
2584 	if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
2585 	    !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
2586 	    !nested_cr3_valid(vcpu, vmcs12->host_cr3))
2587 		return -EINVAL;
2588 	/*
2589 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2590 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2591 	 * the values of the LMA and LME bits in the field must each be that of
2592 	 * the host address-space size VM-exit control.
2593 	 */
2594 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2595 		ia32e = (vmcs12->vm_exit_controls &
2596 			 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
2597 		if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
2598 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
2599 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
2600 			return -EINVAL;
2601 	}
2602 
2603 	return 0;
2604 }
2605 
2606 /*
2607  * Checks related to Guest Non-register State
2608  */
2609 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2610 {
2611 	if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2612 	    vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
2613 		return -EINVAL;
2614 
2615 	return 0;
2616 }
2617 
2618 static int nested_vmx_check_vmentry_prereqs(struct kvm_vcpu *vcpu,
2619 					    struct vmcs12 *vmcs12)
2620 {
2621 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2622 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2623 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2624 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2625 
2626 	if (nested_check_host_control_regs(vcpu, vmcs12))
2627 		return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
2628 
2629 	if (nested_check_guest_non_reg_state(vmcs12))
2630 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2631 
2632 	return 0;
2633 }
2634 
2635 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2636 					  struct vmcs12 *vmcs12)
2637 {
2638 	int r;
2639 	struct page *page;
2640 	struct vmcs12 *shadow;
2641 
2642 	if (vmcs12->vmcs_link_pointer == -1ull)
2643 		return 0;
2644 
2645 	if (!page_address_valid(vcpu, vmcs12->vmcs_link_pointer))
2646 		return -EINVAL;
2647 
2648 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->vmcs_link_pointer);
2649 	if (is_error_page(page))
2650 		return -EINVAL;
2651 
2652 	r = 0;
2653 	shadow = kmap(page);
2654 	if (shadow->hdr.revision_id != VMCS12_REVISION ||
2655 	    shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12))
2656 		r = -EINVAL;
2657 	kunmap(page);
2658 	kvm_release_page_clean(page);
2659 	return r;
2660 }
2661 
2662 static int nested_vmx_check_vmentry_postreqs(struct kvm_vcpu *vcpu,
2663 					     struct vmcs12 *vmcs12,
2664 					     u32 *exit_qual)
2665 {
2666 	bool ia32e;
2667 
2668 	*exit_qual = ENTRY_FAIL_DEFAULT;
2669 
2670 	if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
2671 	    !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
2672 		return 1;
2673 
2674 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2675 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2676 		return 1;
2677 	}
2678 
2679 	/*
2680 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2681 	 * are performed on the field for the IA32_EFER MSR:
2682 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2683 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2684 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2685 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2686 	 *   CR0.PG) is 1.
2687 	 */
2688 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2689 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2690 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2691 		if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
2692 		    ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
2693 		    ((vmcs12->guest_cr0 & X86_CR0_PG) &&
2694 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
2695 			return 1;
2696 	}
2697 
2698 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2699 		(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
2700 		(vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
2701 			return 1;
2702 
2703 	return 0;
2704 }
2705 
2706 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2707 {
2708 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2709 	unsigned long cr3, cr4;
2710 	bool vm_fail;
2711 
2712 	if (!nested_early_check)
2713 		return 0;
2714 
2715 	if (vmx->msr_autoload.host.nr)
2716 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2717 	if (vmx->msr_autoload.guest.nr)
2718 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2719 
2720 	preempt_disable();
2721 
2722 	vmx_prepare_switch_to_guest(vcpu);
2723 
2724 	/*
2725 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2726 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2727 	 * be written (by preparve_vmcs02()) before the "real" VMEnter, i.e.
2728 	 * there is no need to preserve other bits or save/restore the field.
2729 	 */
2730 	vmcs_writel(GUEST_RFLAGS, 0);
2731 
2732 	cr3 = __get_current_cr3_fast();
2733 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2734 		vmcs_writel(HOST_CR3, cr3);
2735 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2736 	}
2737 
2738 	cr4 = cr4_read_shadow();
2739 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2740 		vmcs_writel(HOST_CR4, cr4);
2741 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2742 	}
2743 
2744 	asm(
2745 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2746 		"cmp %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2747 		"je 1f \n\t"
2748 		__ex("vmwrite %%" _ASM_SP ", %[HOST_RSP]") "\n\t"
2749 		"mov %%" _ASM_SP ", %c[host_state_rsp](%[loaded_vmcs]) \n\t"
2750 		"1: \n\t"
2751 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2752 
2753 		/* Check if vmlaunch or vmresume is needed */
2754 		"cmpb $0, %c[launched](%[loaded_vmcs])\n\t"
2755 
2756 		/*
2757 		 * VMLAUNCH and VMRESUME clear RFLAGS.{CF,ZF} on VM-Exit, set
2758 		 * RFLAGS.CF on VM-Fail Invalid and set RFLAGS.ZF on VM-Fail
2759 		 * Valid.  vmx_vmenter() directly "returns" RFLAGS, and so the
2760 		 * results of VM-Enter is captured via CC_{SET,OUT} to vm_fail.
2761 		 */
2762 		"call vmx_vmenter\n\t"
2763 
2764 		CC_SET(be)
2765 	      : ASM_CALL_CONSTRAINT, CC_OUT(be) (vm_fail)
2766 	      :	[HOST_RSP]"r"((unsigned long)HOST_RSP),
2767 		[loaded_vmcs]"r"(vmx->loaded_vmcs),
2768 		[launched]"i"(offsetof(struct loaded_vmcs, launched)),
2769 		[host_state_rsp]"i"(offsetof(struct loaded_vmcs, host_state.rsp)),
2770 		[wordsize]"i"(sizeof(ulong))
2771 	      : "cc", "memory"
2772 	);
2773 
2774 	preempt_enable();
2775 
2776 	if (vmx->msr_autoload.host.nr)
2777 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2778 	if (vmx->msr_autoload.guest.nr)
2779 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2780 
2781 	if (vm_fail) {
2782 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
2783 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
2784 		return 1;
2785 	}
2786 
2787 	/*
2788 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
2789 	 */
2790 	local_irq_enable();
2791 	if (hw_breakpoint_active())
2792 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
2793 
2794 	/*
2795 	 * A non-failing VMEntry means we somehow entered guest mode with
2796 	 * an illegal RIP, and that's just the tip of the iceberg.  There
2797 	 * is no telling what memory has been modified or what state has
2798 	 * been exposed to unknown code.  Hitting this all but guarantees
2799 	 * a (very critical) hardware issue.
2800 	 */
2801 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
2802 		VMX_EXIT_REASONS_FAILED_VMENTRY));
2803 
2804 	return 0;
2805 }
2806 
2807 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
2808 						 struct vmcs12 *vmcs12);
2809 
2810 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
2811 {
2812 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2813 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2814 	struct page *page;
2815 	u64 hpa;
2816 
2817 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2818 		/*
2819 		 * Translate L1 physical address to host physical
2820 		 * address for vmcs02. Keep the page pinned, so this
2821 		 * physical address remains valid. We keep a reference
2822 		 * to it so we can release it later.
2823 		 */
2824 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
2825 			kvm_release_page_dirty(vmx->nested.apic_access_page);
2826 			vmx->nested.apic_access_page = NULL;
2827 		}
2828 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
2829 		/*
2830 		 * If translation failed, no matter: This feature asks
2831 		 * to exit when accessing the given address, and if it
2832 		 * can never be accessed, this feature won't do
2833 		 * anything anyway.
2834 		 */
2835 		if (!is_error_page(page)) {
2836 			vmx->nested.apic_access_page = page;
2837 			hpa = page_to_phys(vmx->nested.apic_access_page);
2838 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
2839 		} else {
2840 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
2841 					SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
2842 		}
2843 	}
2844 
2845 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
2846 		if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
2847 			kvm_release_page_dirty(vmx->nested.virtual_apic_page);
2848 			vmx->nested.virtual_apic_page = NULL;
2849 		}
2850 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
2851 
2852 		/*
2853 		 * If translation failed, VM entry will fail because
2854 		 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
2855 		 * Failing the vm entry is _not_ what the processor
2856 		 * does but it's basically the only possibility we
2857 		 * have.  We could still enter the guest if CR8 load
2858 		 * exits are enabled, CR8 store exits are enabled, and
2859 		 * virtualize APIC access is disabled; in this case
2860 		 * the processor would never use the TPR shadow and we
2861 		 * could simply clear the bit from the execution
2862 		 * control.  But such a configuration is useless, so
2863 		 * let's keep the code simple.
2864 		 */
2865 		if (!is_error_page(page)) {
2866 			vmx->nested.virtual_apic_page = page;
2867 			hpa = page_to_phys(vmx->nested.virtual_apic_page);
2868 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
2869 		}
2870 	}
2871 
2872 	if (nested_cpu_has_posted_intr(vmcs12)) {
2873 		if (vmx->nested.pi_desc_page) { /* shouldn't happen */
2874 			kunmap(vmx->nested.pi_desc_page);
2875 			kvm_release_page_dirty(vmx->nested.pi_desc_page);
2876 			vmx->nested.pi_desc_page = NULL;
2877 			vmx->nested.pi_desc = NULL;
2878 			vmcs_write64(POSTED_INTR_DESC_ADDR, -1ull);
2879 		}
2880 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
2881 		if (is_error_page(page))
2882 			return;
2883 		vmx->nested.pi_desc_page = page;
2884 		vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
2885 		vmx->nested.pi_desc =
2886 			(struct pi_desc *)((void *)vmx->nested.pi_desc +
2887 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2888 			(PAGE_SIZE - 1)));
2889 		vmcs_write64(POSTED_INTR_DESC_ADDR,
2890 			page_to_phys(vmx->nested.pi_desc_page) +
2891 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2892 			(PAGE_SIZE - 1)));
2893 	}
2894 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
2895 		vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
2896 			      CPU_BASED_USE_MSR_BITMAPS);
2897 	else
2898 		vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
2899 				CPU_BASED_USE_MSR_BITMAPS);
2900 }
2901 
2902 /*
2903  * Intel's VMX Instruction Reference specifies a common set of prerequisites
2904  * for running VMX instructions (except VMXON, whose prerequisites are
2905  * slightly different). It also specifies what exception to inject otherwise.
2906  * Note that many of these exceptions have priority over VM exits, so they
2907  * don't have to be checked again here.
2908  */
2909 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
2910 {
2911 	if (!to_vmx(vcpu)->nested.vmxon) {
2912 		kvm_queue_exception(vcpu, UD_VECTOR);
2913 		return 0;
2914 	}
2915 
2916 	if (vmx_get_cpl(vcpu)) {
2917 		kvm_inject_gp(vcpu, 0);
2918 		return 0;
2919 	}
2920 
2921 	return 1;
2922 }
2923 
2924 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
2925 {
2926 	u8 rvi = vmx_get_rvi();
2927 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
2928 
2929 	return ((rvi & 0xf0) > (vppr & 0xf0));
2930 }
2931 
2932 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
2933 				   struct vmcs12 *vmcs12);
2934 
2935 /*
2936  * If from_vmentry is false, this is being called from state restore (either RSM
2937  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
2938 + *
2939 + * Returns:
2940 + *   0 - success, i.e. proceed with actual VMEnter
2941 + *   1 - consistency check VMExit
2942 + *  -1 - consistency check VMFail
2943  */
2944 int nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
2945 {
2946 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2947 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2948 	bool evaluate_pending_interrupts;
2949 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
2950 	u32 exit_qual;
2951 
2952 	evaluate_pending_interrupts = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2953 		(CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_VIRTUAL_NMI_PENDING);
2954 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
2955 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
2956 
2957 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
2958 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
2959 	if (kvm_mpx_supported() &&
2960 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2961 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
2962 
2963 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
2964 
2965 	prepare_vmcs02_early(vmx, vmcs12);
2966 
2967 	if (from_vmentry) {
2968 		nested_get_vmcs12_pages(vcpu);
2969 
2970 		if (nested_vmx_check_vmentry_hw(vcpu)) {
2971 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
2972 			return -1;
2973 		}
2974 
2975 		if (nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
2976 			goto vmentry_fail_vmexit;
2977 	}
2978 
2979 	enter_guest_mode(vcpu);
2980 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
2981 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
2982 
2983 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
2984 		goto vmentry_fail_vmexit_guest_mode;
2985 
2986 	if (from_vmentry) {
2987 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
2988 		exit_qual = nested_vmx_load_msr(vcpu,
2989 						vmcs12->vm_entry_msr_load_addr,
2990 						vmcs12->vm_entry_msr_load_count);
2991 		if (exit_qual)
2992 			goto vmentry_fail_vmexit_guest_mode;
2993 	} else {
2994 		/*
2995 		 * The MMU is not initialized to point at the right entities yet and
2996 		 * "get pages" would need to read data from the guest (i.e. we will
2997 		 * need to perform gpa to hpa translation). Request a call
2998 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
2999 		 * have already been set at vmentry time and should not be reset.
3000 		 */
3001 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3002 	}
3003 
3004 	/*
3005 	 * If L1 had a pending IRQ/NMI until it executed
3006 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3007 	 * disallowed (e.g. interrupts disabled), L0 needs to
3008 	 * evaluate if this pending event should cause an exit from L2
3009 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3010 	 * intercept EXTERNAL_INTERRUPT).
3011 	 *
3012 	 * Usually this would be handled by the processor noticing an
3013 	 * IRQ/NMI window request, or checking RVI during evaluation of
3014 	 * pending virtual interrupts.  However, this setting was done
3015 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3016 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3017 	 */
3018 	if (unlikely(evaluate_pending_interrupts))
3019 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3020 
3021 	/*
3022 	 * Do not start the preemption timer hrtimer until after we know
3023 	 * we are successful, so that only nested_vmx_vmexit needs to cancel
3024 	 * the timer.
3025 	 */
3026 	vmx->nested.preemption_timer_expired = false;
3027 	if (nested_cpu_has_preemption_timer(vmcs12))
3028 		vmx_start_preemption_timer(vcpu);
3029 
3030 	/*
3031 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3032 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3033 	 * returned as far as L1 is concerned. It will only return (and set
3034 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3035 	 */
3036 	return 0;
3037 
3038 	/*
3039 	 * A failed consistency check that leads to a VMExit during L1's
3040 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3041 	 * 26.7 "VM-entry failures during or after loading guest state".
3042 	 */
3043 vmentry_fail_vmexit_guest_mode:
3044 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3045 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3046 	leave_guest_mode(vcpu);
3047 
3048 vmentry_fail_vmexit:
3049 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3050 
3051 	if (!from_vmentry)
3052 		return 1;
3053 
3054 	load_vmcs12_host_state(vcpu, vmcs12);
3055 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3056 	vmcs12->exit_qualification = exit_qual;
3057 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3058 		vmx->nested.need_vmcs12_sync = true;
3059 	return 1;
3060 }
3061 
3062 /*
3063  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3064  * for running an L2 nested guest.
3065  */
3066 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3067 {
3068 	struct vmcs12 *vmcs12;
3069 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3070 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3071 	int ret;
3072 
3073 	if (!nested_vmx_check_permission(vcpu))
3074 		return 1;
3075 
3076 	if (!nested_vmx_handle_enlightened_vmptrld(vcpu, true))
3077 		return 1;
3078 
3079 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3080 		return nested_vmx_failInvalid(vcpu);
3081 
3082 	vmcs12 = get_vmcs12(vcpu);
3083 
3084 	/*
3085 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3086 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3087 	 * rather than RFLAGS.ZF, and no error number is stored to the
3088 	 * VM-instruction error field.
3089 	 */
3090 	if (vmcs12->hdr.shadow_vmcs)
3091 		return nested_vmx_failInvalid(vcpu);
3092 
3093 	if (vmx->nested.hv_evmcs) {
3094 		copy_enlightened_to_vmcs12(vmx);
3095 		/* Enlightened VMCS doesn't have launch state */
3096 		vmcs12->launch_state = !launch;
3097 	} else if (enable_shadow_vmcs) {
3098 		copy_shadow_to_vmcs12(vmx);
3099 	}
3100 
3101 	/*
3102 	 * The nested entry process starts with enforcing various prerequisites
3103 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3104 	 * they fail: As the SDM explains, some conditions should cause the
3105 	 * instruction to fail, while others will cause the instruction to seem
3106 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3107 	 * To speed up the normal (success) code path, we should avoid checking
3108 	 * for misconfigurations which will anyway be caught by the processor
3109 	 * when using the merged vmcs02.
3110 	 */
3111 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3112 		return nested_vmx_failValid(vcpu,
3113 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3114 
3115 	if (vmcs12->launch_state == launch)
3116 		return nested_vmx_failValid(vcpu,
3117 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3118 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3119 
3120 	ret = nested_vmx_check_vmentry_prereqs(vcpu, vmcs12);
3121 	if (ret)
3122 		return nested_vmx_failValid(vcpu, ret);
3123 
3124 	/*
3125 	 * We're finally done with prerequisite checking, and can start with
3126 	 * the nested entry.
3127 	 */
3128 	vmx->nested.nested_run_pending = 1;
3129 	ret = nested_vmx_enter_non_root_mode(vcpu, true);
3130 	vmx->nested.nested_run_pending = !ret;
3131 	if (ret > 0)
3132 		return 1;
3133 	else if (ret)
3134 		return nested_vmx_failValid(vcpu,
3135 			VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3136 
3137 	/* Hide L1D cache contents from the nested guest.  */
3138 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3139 
3140 	/*
3141 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3142 	 * also be used as part of restoring nVMX state for
3143 	 * snapshot restore (migration).
3144 	 *
3145 	 * In this flow, it is assumed that vmcs12 cache was
3146 	 * trasferred as part of captured nVMX state and should
3147 	 * therefore not be read from guest memory (which may not
3148 	 * exist on destination host yet).
3149 	 */
3150 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3151 
3152 	/*
3153 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3154 	 * awakened by event injection or by an NMI-window VM-exit or
3155 	 * by an interrupt-window VM-exit, halt the vcpu.
3156 	 */
3157 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3158 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3159 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_NMI_PENDING) &&
3160 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_INTR_PENDING) &&
3161 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3162 		vmx->nested.nested_run_pending = 0;
3163 		return kvm_vcpu_halt(vcpu);
3164 	}
3165 	return 1;
3166 }
3167 
3168 /*
3169  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3170  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
3171  * This function returns the new value we should put in vmcs12.guest_cr0.
3172  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3173  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3174  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3175  *     didn't trap the bit, because if L1 did, so would L0).
3176  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3177  *     been modified by L2, and L1 knows it. So just leave the old value of
3178  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3179  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3180  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3181  *     changed these bits, and therefore they need to be updated, but L0
3182  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3183  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3184  */
3185 static inline unsigned long
3186 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3187 {
3188 	return
3189 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3190 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3191 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3192 			vcpu->arch.cr0_guest_owned_bits));
3193 }
3194 
3195 static inline unsigned long
3196 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3197 {
3198 	return
3199 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3200 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3201 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3202 			vcpu->arch.cr4_guest_owned_bits));
3203 }
3204 
3205 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3206 				      struct vmcs12 *vmcs12)
3207 {
3208 	u32 idt_vectoring;
3209 	unsigned int nr;
3210 
3211 	if (vcpu->arch.exception.injected) {
3212 		nr = vcpu->arch.exception.nr;
3213 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3214 
3215 		if (kvm_exception_is_soft(nr)) {
3216 			vmcs12->vm_exit_instruction_len =
3217 				vcpu->arch.event_exit_inst_len;
3218 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3219 		} else
3220 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3221 
3222 		if (vcpu->arch.exception.has_error_code) {
3223 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3224 			vmcs12->idt_vectoring_error_code =
3225 				vcpu->arch.exception.error_code;
3226 		}
3227 
3228 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3229 	} else if (vcpu->arch.nmi_injected) {
3230 		vmcs12->idt_vectoring_info_field =
3231 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3232 	} else if (vcpu->arch.interrupt.injected) {
3233 		nr = vcpu->arch.interrupt.nr;
3234 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3235 
3236 		if (vcpu->arch.interrupt.soft) {
3237 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3238 			vmcs12->vm_entry_instruction_len =
3239 				vcpu->arch.event_exit_inst_len;
3240 		} else
3241 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3242 
3243 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3244 	}
3245 }
3246 
3247 
3248 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3249 {
3250 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3251 	gfn_t gfn;
3252 
3253 	/*
3254 	 * Don't need to mark the APIC access page dirty; it is never
3255 	 * written to by the CPU during APIC virtualization.
3256 	 */
3257 
3258 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3259 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3260 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3261 	}
3262 
3263 	if (nested_cpu_has_posted_intr(vmcs12)) {
3264 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3265 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3266 	}
3267 }
3268 
3269 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3270 {
3271 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3272 	int max_irr;
3273 	void *vapic_page;
3274 	u16 status;
3275 
3276 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3277 		return;
3278 
3279 	vmx->nested.pi_pending = false;
3280 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3281 		return;
3282 
3283 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3284 	if (max_irr != 256) {
3285 		vapic_page = kmap(vmx->nested.virtual_apic_page);
3286 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3287 			vapic_page, &max_irr);
3288 		kunmap(vmx->nested.virtual_apic_page);
3289 
3290 		status = vmcs_read16(GUEST_INTR_STATUS);
3291 		if ((u8)max_irr > ((u8)status & 0xff)) {
3292 			status &= ~0xff;
3293 			status |= (u8)max_irr;
3294 			vmcs_write16(GUEST_INTR_STATUS, status);
3295 		}
3296 	}
3297 
3298 	nested_mark_vmcs12_pages_dirty(vcpu);
3299 }
3300 
3301 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3302 					       unsigned long exit_qual)
3303 {
3304 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3305 	unsigned int nr = vcpu->arch.exception.nr;
3306 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3307 
3308 	if (vcpu->arch.exception.has_error_code) {
3309 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3310 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3311 	}
3312 
3313 	if (kvm_exception_is_soft(nr))
3314 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3315 	else
3316 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3317 
3318 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3319 	    vmx_get_nmi_mask(vcpu))
3320 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3321 
3322 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3323 }
3324 
3325 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
3326 {
3327 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3328 	unsigned long exit_qual;
3329 	bool block_nested_events =
3330 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3331 
3332 	if (vcpu->arch.exception.pending &&
3333 		nested_vmx_check_exception(vcpu, &exit_qual)) {
3334 		if (block_nested_events)
3335 			return -EBUSY;
3336 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3337 		return 0;
3338 	}
3339 
3340 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3341 	    vmx->nested.preemption_timer_expired) {
3342 		if (block_nested_events)
3343 			return -EBUSY;
3344 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3345 		return 0;
3346 	}
3347 
3348 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3349 		if (block_nested_events)
3350 			return -EBUSY;
3351 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3352 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3353 				  INTR_INFO_VALID_MASK, 0);
3354 		/*
3355 		 * The NMI-triggered VM exit counts as injection:
3356 		 * clear this one and block further NMIs.
3357 		 */
3358 		vcpu->arch.nmi_pending = 0;
3359 		vmx_set_nmi_mask(vcpu, true);
3360 		return 0;
3361 	}
3362 
3363 	if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
3364 	    nested_exit_on_intr(vcpu)) {
3365 		if (block_nested_events)
3366 			return -EBUSY;
3367 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3368 		return 0;
3369 	}
3370 
3371 	vmx_complete_nested_posted_interrupt(vcpu);
3372 	return 0;
3373 }
3374 
3375 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3376 {
3377 	ktime_t remaining =
3378 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3379 	u64 value;
3380 
3381 	if (ktime_to_ns(remaining) <= 0)
3382 		return 0;
3383 
3384 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3385 	do_div(value, 1000000);
3386 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3387 }
3388 
3389 /*
3390  * Update the guest state fields of vmcs12 to reflect changes that
3391  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3392  * VM-entry controls is also updated, since this is really a guest
3393  * state bit.)
3394  */
3395 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3396 {
3397 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3398 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3399 
3400 	vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3401 	vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
3402 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3403 
3404 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3405 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3406 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3407 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3408 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3409 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3410 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3411 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3412 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3413 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3414 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3415 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3416 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3417 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3418 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3419 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3420 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3421 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3422 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3423 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3424 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3425 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3426 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3427 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3428 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3429 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3430 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3431 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3432 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3433 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3434 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3435 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3436 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3437 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3438 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3439 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3440 
3441 	vmcs12->guest_interruptibility_info =
3442 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3443 	vmcs12->guest_pending_dbg_exceptions =
3444 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3445 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3446 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3447 	else
3448 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3449 
3450 	if (nested_cpu_has_preemption_timer(vmcs12) &&
3451 	    vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3452 			vmcs12->vmx_preemption_timer_value =
3453 				vmx_get_preemption_timer_value(vcpu);
3454 
3455 	/*
3456 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3457 	 * own CR3 without exiting. If it has changed it, we must keep it.
3458 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3459 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3460 	 *
3461 	 * Additionally, restore L2's PDPTR to vmcs12.
3462 	 */
3463 	if (enable_ept) {
3464 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3465 		vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3466 		vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3467 		vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3468 		vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3469 	}
3470 
3471 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3472 
3473 	if (nested_cpu_has_vid(vmcs12))
3474 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3475 
3476 	vmcs12->vm_entry_controls =
3477 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3478 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3479 
3480 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
3481 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3482 		vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3483 	}
3484 
3485 	/* TODO: These cannot have changed unless we have MSR bitmaps and
3486 	 * the relevant bit asks not to trap the change */
3487 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
3488 		vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
3489 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3490 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3491 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3492 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3493 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3494 	if (kvm_mpx_supported())
3495 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3496 }
3497 
3498 /*
3499  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3500  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3501  * and this function updates it to reflect the changes to the guest state while
3502  * L2 was running (and perhaps made some exits which were handled directly by L0
3503  * without going back to L1), and to reflect the exit reason.
3504  * Note that we do not have to copy here all VMCS fields, just those that
3505  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3506  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3507  * which already writes to vmcs12 directly.
3508  */
3509 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3510 			   u32 exit_reason, u32 exit_intr_info,
3511 			   unsigned long exit_qualification)
3512 {
3513 	/* update guest state fields: */
3514 	sync_vmcs12(vcpu, vmcs12);
3515 
3516 	/* update exit information fields: */
3517 
3518 	vmcs12->vm_exit_reason = exit_reason;
3519 	vmcs12->exit_qualification = exit_qualification;
3520 	vmcs12->vm_exit_intr_info = exit_intr_info;
3521 
3522 	vmcs12->idt_vectoring_info_field = 0;
3523 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3524 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3525 
3526 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3527 		vmcs12->launch_state = 1;
3528 
3529 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3530 		 * instead of reading the real value. */
3531 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3532 
3533 		/*
3534 		 * Transfer the event that L0 or L1 may wanted to inject into
3535 		 * L2 to IDT_VECTORING_INFO_FIELD.
3536 		 */
3537 		vmcs12_save_pending_event(vcpu, vmcs12);
3538 
3539 		/*
3540 		 * According to spec, there's no need to store the guest's
3541 		 * MSRs if the exit is due to a VM-entry failure that occurs
3542 		 * during or after loading the guest state. Since this exit
3543 		 * does not fall in that category, we need to save the MSRs.
3544 		 */
3545 		if (nested_vmx_store_msr(vcpu,
3546 					 vmcs12->vm_exit_msr_store_addr,
3547 					 vmcs12->vm_exit_msr_store_count))
3548 			nested_vmx_abort(vcpu,
3549 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3550 	}
3551 
3552 	/*
3553 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3554 	 * preserved above and would only end up incorrectly in L1.
3555 	 */
3556 	vcpu->arch.nmi_injected = false;
3557 	kvm_clear_exception_queue(vcpu);
3558 	kvm_clear_interrupt_queue(vcpu);
3559 }
3560 
3561 /*
3562  * A part of what we need to when the nested L2 guest exits and we want to
3563  * run its L1 parent, is to reset L1's guest state to the host state specified
3564  * in vmcs12.
3565  * This function is to be called not only on normal nested exit, but also on
3566  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3567  * Failures During or After Loading Guest State").
3568  * This function should be called when the active VMCS is L1's (vmcs01).
3569  */
3570 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3571 				   struct vmcs12 *vmcs12)
3572 {
3573 	struct kvm_segment seg;
3574 	u32 entry_failure_code;
3575 
3576 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
3577 		vcpu->arch.efer = vmcs12->host_ia32_efer;
3578 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3579 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
3580 	else
3581 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
3582 	vmx_set_efer(vcpu, vcpu->arch.efer);
3583 
3584 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
3585 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
3586 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
3587 	vmx_set_interrupt_shadow(vcpu, 0);
3588 
3589 	/*
3590 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
3591 	 * actually changed, because vmx_set_cr0 refers to efer set above.
3592 	 *
3593 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
3594 	 * (KVM doesn't change it);
3595 	 */
3596 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3597 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
3598 
3599 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
3600 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3601 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
3602 
3603 	nested_ept_uninit_mmu_context(vcpu);
3604 
3605 	/*
3606 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
3607 	 * couldn't have changed.
3608 	 */
3609 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
3610 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
3611 
3612 	if (!enable_ept)
3613 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
3614 
3615 	/*
3616 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
3617 	 * VMEntry/VMExit. Thus, no need to flush TLB.
3618 	 *
3619 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
3620 	 * flushed on every VMEntry/VMExit.
3621 	 *
3622 	 * Otherwise, we can preserve TLB entries as long as we are
3623 	 * able to tag L1 TLB entries differently than L2 TLB entries.
3624 	 *
3625 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
3626 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
3627 	 * has been set by KVM_REQ_LOAD_CR3.
3628 	 */
3629 	if (enable_vpid &&
3630 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
3631 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3632 	}
3633 
3634 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
3635 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
3636 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
3637 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
3638 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
3639 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
3640 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
3641 
3642 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
3643 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
3644 		vmcs_write64(GUEST_BNDCFGS, 0);
3645 
3646 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
3647 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
3648 		vcpu->arch.pat = vmcs12->host_ia32_pat;
3649 	}
3650 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
3651 		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
3652 			vmcs12->host_ia32_perf_global_ctrl);
3653 
3654 	/* Set L1 segment info according to Intel SDM
3655 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
3656 	seg = (struct kvm_segment) {
3657 		.base = 0,
3658 		.limit = 0xFFFFFFFF,
3659 		.selector = vmcs12->host_cs_selector,
3660 		.type = 11,
3661 		.present = 1,
3662 		.s = 1,
3663 		.g = 1
3664 	};
3665 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3666 		seg.l = 1;
3667 	else
3668 		seg.db = 1;
3669 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
3670 	seg = (struct kvm_segment) {
3671 		.base = 0,
3672 		.limit = 0xFFFFFFFF,
3673 		.type = 3,
3674 		.present = 1,
3675 		.s = 1,
3676 		.db = 1,
3677 		.g = 1
3678 	};
3679 	seg.selector = vmcs12->host_ds_selector;
3680 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
3681 	seg.selector = vmcs12->host_es_selector;
3682 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
3683 	seg.selector = vmcs12->host_ss_selector;
3684 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
3685 	seg.selector = vmcs12->host_fs_selector;
3686 	seg.base = vmcs12->host_fs_base;
3687 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
3688 	seg.selector = vmcs12->host_gs_selector;
3689 	seg.base = vmcs12->host_gs_base;
3690 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
3691 	seg = (struct kvm_segment) {
3692 		.base = vmcs12->host_tr_base,
3693 		.limit = 0x67,
3694 		.selector = vmcs12->host_tr_selector,
3695 		.type = 11,
3696 		.present = 1
3697 	};
3698 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
3699 
3700 	kvm_set_dr(vcpu, 7, 0x400);
3701 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3702 
3703 	if (cpu_has_vmx_msr_bitmap())
3704 		vmx_update_msr_bitmap(vcpu);
3705 
3706 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
3707 				vmcs12->vm_exit_msr_load_count))
3708 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3709 }
3710 
3711 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
3712 {
3713 	struct shared_msr_entry *efer_msr;
3714 	unsigned int i;
3715 
3716 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
3717 		return vmcs_read64(GUEST_IA32_EFER);
3718 
3719 	if (cpu_has_load_ia32_efer())
3720 		return host_efer;
3721 
3722 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
3723 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
3724 			return vmx->msr_autoload.guest.val[i].value;
3725 	}
3726 
3727 	efer_msr = find_msr_entry(vmx, MSR_EFER);
3728 	if (efer_msr)
3729 		return efer_msr->data;
3730 
3731 	return host_efer;
3732 }
3733 
3734 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
3735 {
3736 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3737 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3738 	struct vmx_msr_entry g, h;
3739 	struct msr_data msr;
3740 	gpa_t gpa;
3741 	u32 i, j;
3742 
3743 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
3744 
3745 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
3746 		/*
3747 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
3748 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
3749 		 * and vcpu->arch.dr7 is not squirreled away before the
3750 		 * nested VMENTER (not worth adding a variable in nested_vmx).
3751 		 */
3752 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
3753 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
3754 		else
3755 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
3756 	}
3757 
3758 	/*
3759 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
3760 	 * handle a variety of side effects to KVM's software model.
3761 	 */
3762 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
3763 
3764 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3765 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
3766 
3767 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3768 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
3769 
3770 	nested_ept_uninit_mmu_context(vcpu);
3771 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3772 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3773 
3774 	/*
3775 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
3776 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
3777 	 * VMFail, like everything else we just need to ensure our
3778 	 * software model is up-to-date.
3779 	 */
3780 	ept_save_pdptrs(vcpu);
3781 
3782 	kvm_mmu_reset_context(vcpu);
3783 
3784 	if (cpu_has_vmx_msr_bitmap())
3785 		vmx_update_msr_bitmap(vcpu);
3786 
3787 	/*
3788 	 * This nasty bit of open coding is a compromise between blindly
3789 	 * loading L1's MSRs using the exit load lists (incorrect emulation
3790 	 * of VMFail), leaving the nested VM's MSRs in the software model
3791 	 * (incorrect behavior) and snapshotting the modified MSRs (too
3792 	 * expensive since the lists are unbound by hardware).  For each
3793 	 * MSR that was (prematurely) loaded from the nested VMEntry load
3794 	 * list, reload it from the exit load list if it exists and differs
3795 	 * from the guest value.  The intent is to stuff host state as
3796 	 * silently as possible, not to fully process the exit load list.
3797 	 */
3798 	msr.host_initiated = false;
3799 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
3800 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
3801 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
3802 			pr_debug_ratelimited(
3803 				"%s read MSR index failed (%u, 0x%08llx)\n",
3804 				__func__, i, gpa);
3805 			goto vmabort;
3806 		}
3807 
3808 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
3809 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
3810 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
3811 				pr_debug_ratelimited(
3812 					"%s read MSR failed (%u, 0x%08llx)\n",
3813 					__func__, j, gpa);
3814 				goto vmabort;
3815 			}
3816 			if (h.index != g.index)
3817 				continue;
3818 			if (h.value == g.value)
3819 				break;
3820 
3821 			if (nested_vmx_load_msr_check(vcpu, &h)) {
3822 				pr_debug_ratelimited(
3823 					"%s check failed (%u, 0x%x, 0x%x)\n",
3824 					__func__, j, h.index, h.reserved);
3825 				goto vmabort;
3826 			}
3827 
3828 			msr.index = h.index;
3829 			msr.data = h.value;
3830 			if (kvm_set_msr(vcpu, &msr)) {
3831 				pr_debug_ratelimited(
3832 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
3833 					__func__, j, h.index, h.value);
3834 				goto vmabort;
3835 			}
3836 		}
3837 	}
3838 
3839 	return;
3840 
3841 vmabort:
3842 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3843 }
3844 
3845 /*
3846  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
3847  * and modify vmcs12 to make it see what it would expect to see there if
3848  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
3849  */
3850 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
3851 		       u32 exit_intr_info, unsigned long exit_qualification)
3852 {
3853 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3854 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3855 
3856 	/* trying to cancel vmlaunch/vmresume is a bug */
3857 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
3858 
3859 	leave_guest_mode(vcpu);
3860 
3861 	if (nested_cpu_has_preemption_timer(vmcs12))
3862 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
3863 
3864 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3865 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3866 
3867 	if (likely(!vmx->fail)) {
3868 		if (exit_reason == -1)
3869 			sync_vmcs12(vcpu, vmcs12);
3870 		else
3871 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
3872 				       exit_qualification);
3873 
3874 		/*
3875 		 * Must happen outside of sync_vmcs12() as it will
3876 		 * also be used to capture vmcs12 cache as part of
3877 		 * capturing nVMX state for snapshot (migration).
3878 		 *
3879 		 * Otherwise, this flush will dirty guest memory at a
3880 		 * point it is already assumed by user-space to be
3881 		 * immutable.
3882 		 */
3883 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
3884 	} else {
3885 		/*
3886 		 * The only expected VM-instruction error is "VM entry with
3887 		 * invalid control field(s)." Anything else indicates a
3888 		 * problem with L0.  And we should never get here with a
3889 		 * VMFail of any type if early consistency checks are enabled.
3890 		 */
3891 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
3892 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3893 		WARN_ON_ONCE(nested_early_check);
3894 	}
3895 
3896 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3897 
3898 	/* Update any VMCS fields that might have changed while L2 ran */
3899 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3900 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3901 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
3902 
3903 	if (kvm_has_tsc_control)
3904 		decache_tsc_multiplier(vmx);
3905 
3906 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
3907 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
3908 		vmx_set_virtual_apic_mode(vcpu);
3909 	} else if (!nested_cpu_has_ept(vmcs12) &&
3910 		   nested_cpu_has2(vmcs12,
3911 				   SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3912 		vmx_flush_tlb(vcpu, true);
3913 	}
3914 
3915 	/* Unpin physical memory we referred to in vmcs02 */
3916 	if (vmx->nested.apic_access_page) {
3917 		kvm_release_page_dirty(vmx->nested.apic_access_page);
3918 		vmx->nested.apic_access_page = NULL;
3919 	}
3920 	if (vmx->nested.virtual_apic_page) {
3921 		kvm_release_page_dirty(vmx->nested.virtual_apic_page);
3922 		vmx->nested.virtual_apic_page = NULL;
3923 	}
3924 	if (vmx->nested.pi_desc_page) {
3925 		kunmap(vmx->nested.pi_desc_page);
3926 		kvm_release_page_dirty(vmx->nested.pi_desc_page);
3927 		vmx->nested.pi_desc_page = NULL;
3928 		vmx->nested.pi_desc = NULL;
3929 	}
3930 
3931 	/*
3932 	 * We are now running in L2, mmu_notifier will force to reload the
3933 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
3934 	 */
3935 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
3936 
3937 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
3938 		vmx->nested.need_vmcs12_sync = true;
3939 
3940 	/* in case we halted in L2 */
3941 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3942 
3943 	if (likely(!vmx->fail)) {
3944 		/*
3945 		 * TODO: SDM says that with acknowledge interrupt on
3946 		 * exit, bit 31 of the VM-exit interrupt information
3947 		 * (valid interrupt) is always set to 1 on
3948 		 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
3949 		 * need kvm_cpu_has_interrupt().  See the commit
3950 		 * message for details.
3951 		 */
3952 		if (nested_exit_intr_ack_set(vcpu) &&
3953 		    exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
3954 		    kvm_cpu_has_interrupt(vcpu)) {
3955 			int irq = kvm_cpu_get_interrupt(vcpu);
3956 			WARN_ON(irq < 0);
3957 			vmcs12->vm_exit_intr_info = irq |
3958 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
3959 		}
3960 
3961 		if (exit_reason != -1)
3962 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
3963 						       vmcs12->exit_qualification,
3964 						       vmcs12->idt_vectoring_info_field,
3965 						       vmcs12->vm_exit_intr_info,
3966 						       vmcs12->vm_exit_intr_error_code,
3967 						       KVM_ISA_VMX);
3968 
3969 		load_vmcs12_host_state(vcpu, vmcs12);
3970 
3971 		return;
3972 	}
3973 
3974 	/*
3975 	 * After an early L2 VM-entry failure, we're now back
3976 	 * in L1 which thinks it just finished a VMLAUNCH or
3977 	 * VMRESUME instruction, so we need to set the failure
3978 	 * flag and the VM-instruction error field of the VMCS
3979 	 * accordingly, and skip the emulated instruction.
3980 	 */
3981 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3982 
3983 	/*
3984 	 * Restore L1's host state to KVM's software model.  We're here
3985 	 * because a consistency check was caught by hardware, which
3986 	 * means some amount of guest state has been propagated to KVM's
3987 	 * model and needs to be unwound to the host's state.
3988 	 */
3989 	nested_vmx_restore_host_state(vcpu);
3990 
3991 	vmx->fail = 0;
3992 }
3993 
3994 /*
3995  * Decode the memory-address operand of a vmx instruction, as recorded on an
3996  * exit caused by such an instruction (run by a guest hypervisor).
3997  * On success, returns 0. When the operand is invalid, returns 1 and throws
3998  * #UD or #GP.
3999  */
4000 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4001 			u32 vmx_instruction_info, bool wr, gva_t *ret)
4002 {
4003 	gva_t off;
4004 	bool exn;
4005 	struct kvm_segment s;
4006 
4007 	/*
4008 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4009 	 * Execution", on an exit, vmx_instruction_info holds most of the
4010 	 * addressing components of the operand. Only the displacement part
4011 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4012 	 * For how an actual address is calculated from all these components,
4013 	 * refer to Vol. 1, "Operand Addressing".
4014 	 */
4015 	int  scaling = vmx_instruction_info & 3;
4016 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4017 	bool is_reg = vmx_instruction_info & (1u << 10);
4018 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4019 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4020 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4021 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4022 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4023 
4024 	if (is_reg) {
4025 		kvm_queue_exception(vcpu, UD_VECTOR);
4026 		return 1;
4027 	}
4028 
4029 	/* Addr = segment_base + offset */
4030 	/* offset = base + [index * scale] + displacement */
4031 	off = exit_qualification; /* holds the displacement */
4032 	if (addr_size == 1)
4033 		off = (gva_t)sign_extend64(off, 31);
4034 	else if (addr_size == 0)
4035 		off = (gva_t)sign_extend64(off, 15);
4036 	if (base_is_valid)
4037 		off += kvm_register_read(vcpu, base_reg);
4038 	if (index_is_valid)
4039 		off += kvm_register_read(vcpu, index_reg)<<scaling;
4040 	vmx_get_segment(vcpu, &s, seg_reg);
4041 
4042 	/*
4043 	 * The effective address, i.e. @off, of a memory operand is truncated
4044 	 * based on the address size of the instruction.  Note that this is
4045 	 * the *effective address*, i.e. the address prior to accounting for
4046 	 * the segment's base.
4047 	 */
4048 	if (addr_size == 1) /* 32 bit */
4049 		off &= 0xffffffff;
4050 	else if (addr_size == 0) /* 16 bit */
4051 		off &= 0xffff;
4052 
4053 	/* Checks for #GP/#SS exceptions. */
4054 	exn = false;
4055 	if (is_long_mode(vcpu)) {
4056 		/*
4057 		 * The virtual/linear address is never truncated in 64-bit
4058 		 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
4059 		 * address when using FS/GS with a non-zero base.
4060 		 */
4061 		*ret = s.base + off;
4062 
4063 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4064 		 * non-canonical form. This is the only check on the memory
4065 		 * destination for long mode!
4066 		 */
4067 		exn = is_noncanonical_address(*ret, vcpu);
4068 	} else {
4069 		/*
4070 		 * When not in long mode, the virtual/linear address is
4071 		 * unconditionally truncated to 32 bits regardless of the
4072 		 * address size.
4073 		 */
4074 		*ret = (s.base + off) & 0xffffffff;
4075 
4076 		/* Protected mode: apply checks for segment validity in the
4077 		 * following order:
4078 		 * - segment type check (#GP(0) may be thrown)
4079 		 * - usability check (#GP(0)/#SS(0))
4080 		 * - limit check (#GP(0)/#SS(0))
4081 		 */
4082 		if (wr)
4083 			/* #GP(0) if the destination operand is located in a
4084 			 * read-only data segment or any code segment.
4085 			 */
4086 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4087 		else
4088 			/* #GP(0) if the source operand is located in an
4089 			 * execute-only code segment
4090 			 */
4091 			exn = ((s.type & 0xa) == 8);
4092 		if (exn) {
4093 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4094 			return 1;
4095 		}
4096 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4097 		 */
4098 		exn = (s.unusable != 0);
4099 
4100 		/*
4101 		 * Protected mode: #GP(0)/#SS(0) if the memory operand is
4102 		 * outside the segment limit.  All CPUs that support VMX ignore
4103 		 * limit checks for flat segments, i.e. segments with base==0,
4104 		 * limit==0xffffffff and of type expand-up data or code.
4105 		 */
4106 		if (!(s.base == 0 && s.limit == 0xffffffff &&
4107 		     ((s.type & 8) || !(s.type & 4))))
4108 			exn = exn || (off + sizeof(u64) > s.limit);
4109 	}
4110 	if (exn) {
4111 		kvm_queue_exception_e(vcpu,
4112 				      seg_reg == VCPU_SREG_SS ?
4113 						SS_VECTOR : GP_VECTOR,
4114 				      0);
4115 		return 1;
4116 	}
4117 
4118 	return 0;
4119 }
4120 
4121 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4122 {
4123 	gva_t gva;
4124 	struct x86_exception e;
4125 
4126 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4127 			vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
4128 		return 1;
4129 
4130 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4131 		kvm_inject_page_fault(vcpu, &e);
4132 		return 1;
4133 	}
4134 
4135 	return 0;
4136 }
4137 
4138 /*
4139  * Allocate a shadow VMCS and associate it with the currently loaded
4140  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4141  * VMCS is also VMCLEARed, so that it is ready for use.
4142  */
4143 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4144 {
4145 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4146 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4147 
4148 	/*
4149 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4150 	 * executes VMXON and free it when L1 executes VMXOFF.
4151 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4152 	 * here when vmcs01 already have an allocated shadow vmcs.
4153 	 */
4154 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4155 
4156 	if (!loaded_vmcs->shadow_vmcs) {
4157 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4158 		if (loaded_vmcs->shadow_vmcs)
4159 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4160 	}
4161 	return loaded_vmcs->shadow_vmcs;
4162 }
4163 
4164 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4165 {
4166 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4167 	int r;
4168 
4169 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4170 	if (r < 0)
4171 		goto out_vmcs02;
4172 
4173 	vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4174 	if (!vmx->nested.cached_vmcs12)
4175 		goto out_cached_vmcs12;
4176 
4177 	vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
4178 	if (!vmx->nested.cached_shadow_vmcs12)
4179 		goto out_cached_shadow_vmcs12;
4180 
4181 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4182 		goto out_shadow_vmcs;
4183 
4184 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4185 		     HRTIMER_MODE_REL_PINNED);
4186 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4187 
4188 	vmx->nested.vpid02 = allocate_vpid();
4189 
4190 	vmx->nested.vmcs02_initialized = false;
4191 	vmx->nested.vmxon = true;
4192 
4193 	if (pt_mode == PT_MODE_HOST_GUEST) {
4194 		vmx->pt_desc.guest.ctl = 0;
4195 		pt_update_intercept_for_msr(vmx);
4196 	}
4197 
4198 	return 0;
4199 
4200 out_shadow_vmcs:
4201 	kfree(vmx->nested.cached_shadow_vmcs12);
4202 
4203 out_cached_shadow_vmcs12:
4204 	kfree(vmx->nested.cached_vmcs12);
4205 
4206 out_cached_vmcs12:
4207 	free_loaded_vmcs(&vmx->nested.vmcs02);
4208 
4209 out_vmcs02:
4210 	return -ENOMEM;
4211 }
4212 
4213 /*
4214  * Emulate the VMXON instruction.
4215  * Currently, we just remember that VMX is active, and do not save or even
4216  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4217  * do not currently need to store anything in that guest-allocated memory
4218  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4219  * argument is different from the VMXON pointer (which the spec says they do).
4220  */
4221 static int handle_vmon(struct kvm_vcpu *vcpu)
4222 {
4223 	int ret;
4224 	gpa_t vmptr;
4225 	struct page *page;
4226 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4227 	const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
4228 		| FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
4229 
4230 	/*
4231 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4232 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4233 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4234 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4235 	 * have already been checked by hardware, prior to the VM-exit for
4236 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4237 	 * that bit set to 1 in non-root mode.
4238 	 */
4239 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4240 		kvm_queue_exception(vcpu, UD_VECTOR);
4241 		return 1;
4242 	}
4243 
4244 	/* CPL=0 must be checked manually. */
4245 	if (vmx_get_cpl(vcpu)) {
4246 		kvm_inject_gp(vcpu, 0);
4247 		return 1;
4248 	}
4249 
4250 	if (vmx->nested.vmxon)
4251 		return nested_vmx_failValid(vcpu,
4252 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4253 
4254 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4255 			!= VMXON_NEEDED_FEATURES) {
4256 		kvm_inject_gp(vcpu, 0);
4257 		return 1;
4258 	}
4259 
4260 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4261 		return 1;
4262 
4263 	/*
4264 	 * SDM 3: 24.11.5
4265 	 * The first 4 bytes of VMXON region contain the supported
4266 	 * VMCS revision identifier
4267 	 *
4268 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4269 	 * which replaces physical address width with 32
4270 	 */
4271 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4272 		return nested_vmx_failInvalid(vcpu);
4273 
4274 	page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4275 	if (is_error_page(page))
4276 		return nested_vmx_failInvalid(vcpu);
4277 
4278 	if (*(u32 *)kmap(page) != VMCS12_REVISION) {
4279 		kunmap(page);
4280 		kvm_release_page_clean(page);
4281 		return nested_vmx_failInvalid(vcpu);
4282 	}
4283 	kunmap(page);
4284 	kvm_release_page_clean(page);
4285 
4286 	vmx->nested.vmxon_ptr = vmptr;
4287 	ret = enter_vmx_operation(vcpu);
4288 	if (ret)
4289 		return ret;
4290 
4291 	return nested_vmx_succeed(vcpu);
4292 }
4293 
4294 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4295 {
4296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4297 
4298 	if (vmx->nested.current_vmptr == -1ull)
4299 		return;
4300 
4301 	if (enable_shadow_vmcs) {
4302 		/* copy to memory all shadowed fields in case
4303 		   they were modified */
4304 		copy_shadow_to_vmcs12(vmx);
4305 		vmx->nested.need_vmcs12_sync = false;
4306 		vmx_disable_shadow_vmcs(vmx);
4307 	}
4308 	vmx->nested.posted_intr_nv = -1;
4309 
4310 	/* Flush VMCS12 to guest memory */
4311 	kvm_vcpu_write_guest_page(vcpu,
4312 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4313 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4314 
4315 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4316 
4317 	vmx->nested.current_vmptr = -1ull;
4318 }
4319 
4320 /* Emulate the VMXOFF instruction */
4321 static int handle_vmoff(struct kvm_vcpu *vcpu)
4322 {
4323 	if (!nested_vmx_check_permission(vcpu))
4324 		return 1;
4325 	free_nested(vcpu);
4326 	return nested_vmx_succeed(vcpu);
4327 }
4328 
4329 /* Emulate the VMCLEAR instruction */
4330 static int handle_vmclear(struct kvm_vcpu *vcpu)
4331 {
4332 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4333 	u32 zero = 0;
4334 	gpa_t vmptr;
4335 
4336 	if (!nested_vmx_check_permission(vcpu))
4337 		return 1;
4338 
4339 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4340 		return 1;
4341 
4342 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4343 		return nested_vmx_failValid(vcpu,
4344 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4345 
4346 	if (vmptr == vmx->nested.vmxon_ptr)
4347 		return nested_vmx_failValid(vcpu,
4348 			VMXERR_VMCLEAR_VMXON_POINTER);
4349 
4350 	if (vmx->nested.hv_evmcs_page) {
4351 		if (vmptr == vmx->nested.hv_evmcs_vmptr)
4352 			nested_release_evmcs(vcpu);
4353 	} else {
4354 		if (vmptr == vmx->nested.current_vmptr)
4355 			nested_release_vmcs12(vcpu);
4356 
4357 		kvm_vcpu_write_guest(vcpu,
4358 				     vmptr + offsetof(struct vmcs12,
4359 						      launch_state),
4360 				     &zero, sizeof(zero));
4361 	}
4362 
4363 	return nested_vmx_succeed(vcpu);
4364 }
4365 
4366 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4367 
4368 /* Emulate the VMLAUNCH instruction */
4369 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4370 {
4371 	return nested_vmx_run(vcpu, true);
4372 }
4373 
4374 /* Emulate the VMRESUME instruction */
4375 static int handle_vmresume(struct kvm_vcpu *vcpu)
4376 {
4377 
4378 	return nested_vmx_run(vcpu, false);
4379 }
4380 
4381 static int handle_vmread(struct kvm_vcpu *vcpu)
4382 {
4383 	unsigned long field;
4384 	u64 field_value;
4385 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4386 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4387 	gva_t gva = 0;
4388 	struct vmcs12 *vmcs12;
4389 
4390 	if (!nested_vmx_check_permission(vcpu))
4391 		return 1;
4392 
4393 	if (to_vmx(vcpu)->nested.current_vmptr == -1ull)
4394 		return nested_vmx_failInvalid(vcpu);
4395 
4396 	if (!is_guest_mode(vcpu))
4397 		vmcs12 = get_vmcs12(vcpu);
4398 	else {
4399 		/*
4400 		 * When vmcs->vmcs_link_pointer is -1ull, any VMREAD
4401 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4402 		 */
4403 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4404 			return nested_vmx_failInvalid(vcpu);
4405 		vmcs12 = get_shadow_vmcs12(vcpu);
4406 	}
4407 
4408 	/* Decode instruction info and find the field to read */
4409 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4410 	/* Read the field, zero-extended to a u64 field_value */
4411 	if (vmcs12_read_any(vmcs12, field, &field_value) < 0)
4412 		return nested_vmx_failValid(vcpu,
4413 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4414 
4415 	/*
4416 	 * Now copy part of this value to register or memory, as requested.
4417 	 * Note that the number of bits actually copied is 32 or 64 depending
4418 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4419 	 */
4420 	if (vmx_instruction_info & (1u << 10)) {
4421 		kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
4422 			field_value);
4423 	} else {
4424 		if (get_vmx_mem_address(vcpu, exit_qualification,
4425 				vmx_instruction_info, true, &gva))
4426 			return 1;
4427 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4428 		kvm_write_guest_virt_system(vcpu, gva, &field_value,
4429 					    (is_long_mode(vcpu) ? 8 : 4), NULL);
4430 	}
4431 
4432 	return nested_vmx_succeed(vcpu);
4433 }
4434 
4435 
4436 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4437 {
4438 	unsigned long field;
4439 	gva_t gva;
4440 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4441 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4442 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4443 
4444 	/* The value to write might be 32 or 64 bits, depending on L1's long
4445 	 * mode, and eventually we need to write that into a field of several
4446 	 * possible lengths. The code below first zero-extends the value to 64
4447 	 * bit (field_value), and then copies only the appropriate number of
4448 	 * bits into the vmcs12 field.
4449 	 */
4450 	u64 field_value = 0;
4451 	struct x86_exception e;
4452 	struct vmcs12 *vmcs12;
4453 
4454 	if (!nested_vmx_check_permission(vcpu))
4455 		return 1;
4456 
4457 	if (vmx->nested.current_vmptr == -1ull)
4458 		return nested_vmx_failInvalid(vcpu);
4459 
4460 	if (vmx_instruction_info & (1u << 10))
4461 		field_value = kvm_register_readl(vcpu,
4462 			(((vmx_instruction_info) >> 3) & 0xf));
4463 	else {
4464 		if (get_vmx_mem_address(vcpu, exit_qualification,
4465 				vmx_instruction_info, false, &gva))
4466 			return 1;
4467 		if (kvm_read_guest_virt(vcpu, gva, &field_value,
4468 					(is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
4469 			kvm_inject_page_fault(vcpu, &e);
4470 			return 1;
4471 		}
4472 	}
4473 
4474 
4475 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4476 	/*
4477 	 * If the vCPU supports "VMWRITE to any supported field in the
4478 	 * VMCS," then the "read-only" fields are actually read/write.
4479 	 */
4480 	if (vmcs_field_readonly(field) &&
4481 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4482 		return nested_vmx_failValid(vcpu,
4483 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4484 
4485 	if (!is_guest_mode(vcpu))
4486 		vmcs12 = get_vmcs12(vcpu);
4487 	else {
4488 		/*
4489 		 * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
4490 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4491 		 */
4492 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4493 			return nested_vmx_failInvalid(vcpu);
4494 		vmcs12 = get_shadow_vmcs12(vcpu);
4495 	}
4496 
4497 	if (vmcs12_write_any(vmcs12, field, field_value) < 0)
4498 		return nested_vmx_failValid(vcpu,
4499 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4500 
4501 	/*
4502 	 * Do not track vmcs12 dirty-state if in guest-mode
4503 	 * as we actually dirty shadow vmcs12 instead of vmcs12.
4504 	 */
4505 	if (!is_guest_mode(vcpu)) {
4506 		switch (field) {
4507 #define SHADOW_FIELD_RW(x) case x:
4508 #include "vmcs_shadow_fields.h"
4509 			/*
4510 			 * The fields that can be updated by L1 without a vmexit are
4511 			 * always updated in the vmcs02, the others go down the slow
4512 			 * path of prepare_vmcs02.
4513 			 */
4514 			break;
4515 		default:
4516 			vmx->nested.dirty_vmcs12 = true;
4517 			break;
4518 		}
4519 	}
4520 
4521 	return nested_vmx_succeed(vcpu);
4522 }
4523 
4524 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
4525 {
4526 	vmx->nested.current_vmptr = vmptr;
4527 	if (enable_shadow_vmcs) {
4528 		vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4529 			      SECONDARY_EXEC_SHADOW_VMCS);
4530 		vmcs_write64(VMCS_LINK_POINTER,
4531 			     __pa(vmx->vmcs01.shadow_vmcs));
4532 		vmx->nested.need_vmcs12_sync = true;
4533 	}
4534 	vmx->nested.dirty_vmcs12 = true;
4535 }
4536 
4537 /* Emulate the VMPTRLD instruction */
4538 static int handle_vmptrld(struct kvm_vcpu *vcpu)
4539 {
4540 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4541 	gpa_t vmptr;
4542 
4543 	if (!nested_vmx_check_permission(vcpu))
4544 		return 1;
4545 
4546 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4547 		return 1;
4548 
4549 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4550 		return nested_vmx_failValid(vcpu,
4551 			VMXERR_VMPTRLD_INVALID_ADDRESS);
4552 
4553 	if (vmptr == vmx->nested.vmxon_ptr)
4554 		return nested_vmx_failValid(vcpu,
4555 			VMXERR_VMPTRLD_VMXON_POINTER);
4556 
4557 	/* Forbid normal VMPTRLD if Enlightened version was used */
4558 	if (vmx->nested.hv_evmcs)
4559 		return 1;
4560 
4561 	if (vmx->nested.current_vmptr != vmptr) {
4562 		struct vmcs12 *new_vmcs12;
4563 		struct page *page;
4564 
4565 		page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4566 		if (is_error_page(page)) {
4567 			/*
4568 			 * Reads from an unbacked page return all 1s,
4569 			 * which means that the 32 bits located at the
4570 			 * given physical address won't match the required
4571 			 * VMCS12_REVISION identifier.
4572 			 */
4573 			return nested_vmx_failValid(vcpu,
4574 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4575 		}
4576 		new_vmcs12 = kmap(page);
4577 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
4578 		    (new_vmcs12->hdr.shadow_vmcs &&
4579 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
4580 			kunmap(page);
4581 			kvm_release_page_clean(page);
4582 			return nested_vmx_failValid(vcpu,
4583 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4584 		}
4585 
4586 		nested_release_vmcs12(vcpu);
4587 
4588 		/*
4589 		 * Load VMCS12 from guest memory since it is not already
4590 		 * cached.
4591 		 */
4592 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
4593 		kunmap(page);
4594 		kvm_release_page_clean(page);
4595 
4596 		set_current_vmptr(vmx, vmptr);
4597 	}
4598 
4599 	return nested_vmx_succeed(vcpu);
4600 }
4601 
4602 /* Emulate the VMPTRST instruction */
4603 static int handle_vmptrst(struct kvm_vcpu *vcpu)
4604 {
4605 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
4606 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4607 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
4608 	struct x86_exception e;
4609 	gva_t gva;
4610 
4611 	if (!nested_vmx_check_permission(vcpu))
4612 		return 1;
4613 
4614 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
4615 		return 1;
4616 
4617 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info, true, &gva))
4618 		return 1;
4619 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
4620 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
4621 					sizeof(gpa_t), &e)) {
4622 		kvm_inject_page_fault(vcpu, &e);
4623 		return 1;
4624 	}
4625 	return nested_vmx_succeed(vcpu);
4626 }
4627 
4628 /* Emulate the INVEPT instruction */
4629 static int handle_invept(struct kvm_vcpu *vcpu)
4630 {
4631 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4632 	u32 vmx_instruction_info, types;
4633 	unsigned long type;
4634 	gva_t gva;
4635 	struct x86_exception e;
4636 	struct {
4637 		u64 eptp, gpa;
4638 	} operand;
4639 
4640 	if (!(vmx->nested.msrs.secondary_ctls_high &
4641 	      SECONDARY_EXEC_ENABLE_EPT) ||
4642 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
4643 		kvm_queue_exception(vcpu, UD_VECTOR);
4644 		return 1;
4645 	}
4646 
4647 	if (!nested_vmx_check_permission(vcpu))
4648 		return 1;
4649 
4650 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4651 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4652 
4653 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
4654 
4655 	if (type >= 32 || !(types & (1 << type)))
4656 		return nested_vmx_failValid(vcpu,
4657 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4658 
4659 	/* According to the Intel VMX instruction reference, the memory
4660 	 * operand is read even if it isn't needed (e.g., for type==global)
4661 	 */
4662 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4663 			vmx_instruction_info, false, &gva))
4664 		return 1;
4665 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4666 		kvm_inject_page_fault(vcpu, &e);
4667 		return 1;
4668 	}
4669 
4670 	switch (type) {
4671 	case VMX_EPT_EXTENT_GLOBAL:
4672 	/*
4673 	 * TODO: track mappings and invalidate
4674 	 * single context requests appropriately
4675 	 */
4676 	case VMX_EPT_EXTENT_CONTEXT:
4677 		kvm_mmu_sync_roots(vcpu);
4678 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4679 		break;
4680 	default:
4681 		BUG_ON(1);
4682 		break;
4683 	}
4684 
4685 	return nested_vmx_succeed(vcpu);
4686 }
4687 
4688 static int handle_invvpid(struct kvm_vcpu *vcpu)
4689 {
4690 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4691 	u32 vmx_instruction_info;
4692 	unsigned long type, types;
4693 	gva_t gva;
4694 	struct x86_exception e;
4695 	struct {
4696 		u64 vpid;
4697 		u64 gla;
4698 	} operand;
4699 	u16 vpid02;
4700 
4701 	if (!(vmx->nested.msrs.secondary_ctls_high &
4702 	      SECONDARY_EXEC_ENABLE_VPID) ||
4703 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
4704 		kvm_queue_exception(vcpu, UD_VECTOR);
4705 		return 1;
4706 	}
4707 
4708 	if (!nested_vmx_check_permission(vcpu))
4709 		return 1;
4710 
4711 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4712 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4713 
4714 	types = (vmx->nested.msrs.vpid_caps &
4715 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
4716 
4717 	if (type >= 32 || !(types & (1 << type)))
4718 		return nested_vmx_failValid(vcpu,
4719 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4720 
4721 	/* according to the intel vmx instruction reference, the memory
4722 	 * operand is read even if it isn't needed (e.g., for type==global)
4723 	 */
4724 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4725 			vmx_instruction_info, false, &gva))
4726 		return 1;
4727 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4728 		kvm_inject_page_fault(vcpu, &e);
4729 		return 1;
4730 	}
4731 	if (operand.vpid >> 16)
4732 		return nested_vmx_failValid(vcpu,
4733 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4734 
4735 	vpid02 = nested_get_vpid02(vcpu);
4736 	switch (type) {
4737 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
4738 		if (!operand.vpid ||
4739 		    is_noncanonical_address(operand.gla, vcpu))
4740 			return nested_vmx_failValid(vcpu,
4741 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4742 		if (cpu_has_vmx_invvpid_individual_addr()) {
4743 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
4744 				vpid02, operand.gla);
4745 		} else
4746 			__vmx_flush_tlb(vcpu, vpid02, false);
4747 		break;
4748 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
4749 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
4750 		if (!operand.vpid)
4751 			return nested_vmx_failValid(vcpu,
4752 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4753 		__vmx_flush_tlb(vcpu, vpid02, false);
4754 		break;
4755 	case VMX_VPID_EXTENT_ALL_CONTEXT:
4756 		__vmx_flush_tlb(vcpu, vpid02, false);
4757 		break;
4758 	default:
4759 		WARN_ON_ONCE(1);
4760 		return kvm_skip_emulated_instruction(vcpu);
4761 	}
4762 
4763 	return nested_vmx_succeed(vcpu);
4764 }
4765 
4766 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
4767 				     struct vmcs12 *vmcs12)
4768 {
4769 	u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
4770 	u64 address;
4771 	bool accessed_dirty;
4772 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4773 
4774 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
4775 	    !nested_cpu_has_ept(vmcs12))
4776 		return 1;
4777 
4778 	if (index >= VMFUNC_EPTP_ENTRIES)
4779 		return 1;
4780 
4781 
4782 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
4783 				     &address, index * 8, 8))
4784 		return 1;
4785 
4786 	accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
4787 
4788 	/*
4789 	 * If the (L2) guest does a vmfunc to the currently
4790 	 * active ept pointer, we don't have to do anything else
4791 	 */
4792 	if (vmcs12->ept_pointer != address) {
4793 		if (!valid_ept_address(vcpu, address))
4794 			return 1;
4795 
4796 		kvm_mmu_unload(vcpu);
4797 		mmu->ept_ad = accessed_dirty;
4798 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
4799 		vmcs12->ept_pointer = address;
4800 		/*
4801 		 * TODO: Check what's the correct approach in case
4802 		 * mmu reload fails. Currently, we just let the next
4803 		 * reload potentially fail
4804 		 */
4805 		kvm_mmu_reload(vcpu);
4806 	}
4807 
4808 	return 0;
4809 }
4810 
4811 static int handle_vmfunc(struct kvm_vcpu *vcpu)
4812 {
4813 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4814 	struct vmcs12 *vmcs12;
4815 	u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
4816 
4817 	/*
4818 	 * VMFUNC is only supported for nested guests, but we always enable the
4819 	 * secondary control for simplicity; for non-nested mode, fake that we
4820 	 * didn't by injecting #UD.
4821 	 */
4822 	if (!is_guest_mode(vcpu)) {
4823 		kvm_queue_exception(vcpu, UD_VECTOR);
4824 		return 1;
4825 	}
4826 
4827 	vmcs12 = get_vmcs12(vcpu);
4828 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
4829 		goto fail;
4830 
4831 	switch (function) {
4832 	case 0:
4833 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
4834 			goto fail;
4835 		break;
4836 	default:
4837 		goto fail;
4838 	}
4839 	return kvm_skip_emulated_instruction(vcpu);
4840 
4841 fail:
4842 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
4843 			  vmcs_read32(VM_EXIT_INTR_INFO),
4844 			  vmcs_readl(EXIT_QUALIFICATION));
4845 	return 1;
4846 }
4847 
4848 
4849 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
4850 				       struct vmcs12 *vmcs12)
4851 {
4852 	unsigned long exit_qualification;
4853 	gpa_t bitmap, last_bitmap;
4854 	unsigned int port;
4855 	int size;
4856 	u8 b;
4857 
4858 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
4859 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
4860 
4861 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4862 
4863 	port = exit_qualification >> 16;
4864 	size = (exit_qualification & 7) + 1;
4865 
4866 	last_bitmap = (gpa_t)-1;
4867 	b = -1;
4868 
4869 	while (size > 0) {
4870 		if (port < 0x8000)
4871 			bitmap = vmcs12->io_bitmap_a;
4872 		else if (port < 0x10000)
4873 			bitmap = vmcs12->io_bitmap_b;
4874 		else
4875 			return true;
4876 		bitmap += (port & 0x7fff) / 8;
4877 
4878 		if (last_bitmap != bitmap)
4879 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
4880 				return true;
4881 		if (b & (1 << (port & 7)))
4882 			return true;
4883 
4884 		port++;
4885 		size--;
4886 		last_bitmap = bitmap;
4887 	}
4888 
4889 	return false;
4890 }
4891 
4892 /*
4893  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
4894  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
4895  * disinterest in the current event (read or write a specific MSR) by using an
4896  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
4897  */
4898 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
4899 	struct vmcs12 *vmcs12, u32 exit_reason)
4900 {
4901 	u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
4902 	gpa_t bitmap;
4903 
4904 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
4905 		return true;
4906 
4907 	/*
4908 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
4909 	 * for the four combinations of read/write and low/high MSR numbers.
4910 	 * First we need to figure out which of the four to use:
4911 	 */
4912 	bitmap = vmcs12->msr_bitmap;
4913 	if (exit_reason == EXIT_REASON_MSR_WRITE)
4914 		bitmap += 2048;
4915 	if (msr_index >= 0xc0000000) {
4916 		msr_index -= 0xc0000000;
4917 		bitmap += 1024;
4918 	}
4919 
4920 	/* Then read the msr_index'th bit from this bitmap: */
4921 	if (msr_index < 1024*8) {
4922 		unsigned char b;
4923 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
4924 			return true;
4925 		return 1 & (b >> (msr_index & 7));
4926 	} else
4927 		return true; /* let L1 handle the wrong parameter */
4928 }
4929 
4930 /*
4931  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
4932  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
4933  * intercept (via guest_host_mask etc.) the current event.
4934  */
4935 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
4936 	struct vmcs12 *vmcs12)
4937 {
4938 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4939 	int cr = exit_qualification & 15;
4940 	int reg;
4941 	unsigned long val;
4942 
4943 	switch ((exit_qualification >> 4) & 3) {
4944 	case 0: /* mov to cr */
4945 		reg = (exit_qualification >> 8) & 15;
4946 		val = kvm_register_readl(vcpu, reg);
4947 		switch (cr) {
4948 		case 0:
4949 			if (vmcs12->cr0_guest_host_mask &
4950 			    (val ^ vmcs12->cr0_read_shadow))
4951 				return true;
4952 			break;
4953 		case 3:
4954 			if ((vmcs12->cr3_target_count >= 1 &&
4955 					vmcs12->cr3_target_value0 == val) ||
4956 				(vmcs12->cr3_target_count >= 2 &&
4957 					vmcs12->cr3_target_value1 == val) ||
4958 				(vmcs12->cr3_target_count >= 3 &&
4959 					vmcs12->cr3_target_value2 == val) ||
4960 				(vmcs12->cr3_target_count >= 4 &&
4961 					vmcs12->cr3_target_value3 == val))
4962 				return false;
4963 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
4964 				return true;
4965 			break;
4966 		case 4:
4967 			if (vmcs12->cr4_guest_host_mask &
4968 			    (vmcs12->cr4_read_shadow ^ val))
4969 				return true;
4970 			break;
4971 		case 8:
4972 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
4973 				return true;
4974 			break;
4975 		}
4976 		break;
4977 	case 2: /* clts */
4978 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
4979 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
4980 			return true;
4981 		break;
4982 	case 1: /* mov from cr */
4983 		switch (cr) {
4984 		case 3:
4985 			if (vmcs12->cpu_based_vm_exec_control &
4986 			    CPU_BASED_CR3_STORE_EXITING)
4987 				return true;
4988 			break;
4989 		case 8:
4990 			if (vmcs12->cpu_based_vm_exec_control &
4991 			    CPU_BASED_CR8_STORE_EXITING)
4992 				return true;
4993 			break;
4994 		}
4995 		break;
4996 	case 3: /* lmsw */
4997 		/*
4998 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
4999 		 * cr0. Other attempted changes are ignored, with no exit.
5000 		 */
5001 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5002 		if (vmcs12->cr0_guest_host_mask & 0xe &
5003 		    (val ^ vmcs12->cr0_read_shadow))
5004 			return true;
5005 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5006 		    !(vmcs12->cr0_read_shadow & 0x1) &&
5007 		    (val & 0x1))
5008 			return true;
5009 		break;
5010 	}
5011 	return false;
5012 }
5013 
5014 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
5015 	struct vmcs12 *vmcs12, gpa_t bitmap)
5016 {
5017 	u32 vmx_instruction_info;
5018 	unsigned long field;
5019 	u8 b;
5020 
5021 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
5022 		return true;
5023 
5024 	/* Decode instruction info and find the field to access */
5025 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5026 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5027 
5028 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
5029 	if (field >> 15)
5030 		return true;
5031 
5032 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5033 		return true;
5034 
5035 	return 1 & (b >> (field & 7));
5036 }
5037 
5038 /*
5039  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5040  * should handle it ourselves in L0 (and then continue L2). Only call this
5041  * when in is_guest_mode (L2).
5042  */
5043 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5044 {
5045 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5046 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5047 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5048 
5049 	if (vmx->nested.nested_run_pending)
5050 		return false;
5051 
5052 	if (unlikely(vmx->fail)) {
5053 		pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5054 				    vmcs_read32(VM_INSTRUCTION_ERROR));
5055 		return true;
5056 	}
5057 
5058 	/*
5059 	 * The host physical addresses of some pages of guest memory
5060 	 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5061 	 * Page). The CPU may write to these pages via their host
5062 	 * physical address while L2 is running, bypassing any
5063 	 * address-translation-based dirty tracking (e.g. EPT write
5064 	 * protection).
5065 	 *
5066 	 * Mark them dirty on every exit from L2 to prevent them from
5067 	 * getting out of sync with dirty tracking.
5068 	 */
5069 	nested_mark_vmcs12_pages_dirty(vcpu);
5070 
5071 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5072 				vmcs_readl(EXIT_QUALIFICATION),
5073 				vmx->idt_vectoring_info,
5074 				intr_info,
5075 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5076 				KVM_ISA_VMX);
5077 
5078 	switch (exit_reason) {
5079 	case EXIT_REASON_EXCEPTION_NMI:
5080 		if (is_nmi(intr_info))
5081 			return false;
5082 		else if (is_page_fault(intr_info))
5083 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5084 		else if (is_debug(intr_info) &&
5085 			 vcpu->guest_debug &
5086 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5087 			return false;
5088 		else if (is_breakpoint(intr_info) &&
5089 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5090 			return false;
5091 		return vmcs12->exception_bitmap &
5092 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5093 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5094 		return false;
5095 	case EXIT_REASON_TRIPLE_FAULT:
5096 		return true;
5097 	case EXIT_REASON_PENDING_INTERRUPT:
5098 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
5099 	case EXIT_REASON_NMI_WINDOW:
5100 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
5101 	case EXIT_REASON_TASK_SWITCH:
5102 		return true;
5103 	case EXIT_REASON_CPUID:
5104 		return true;
5105 	case EXIT_REASON_HLT:
5106 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5107 	case EXIT_REASON_INVD:
5108 		return true;
5109 	case EXIT_REASON_INVLPG:
5110 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5111 	case EXIT_REASON_RDPMC:
5112 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5113 	case EXIT_REASON_RDRAND:
5114 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5115 	case EXIT_REASON_RDSEED:
5116 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5117 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5118 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5119 	case EXIT_REASON_VMREAD:
5120 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5121 			vmcs12->vmread_bitmap);
5122 	case EXIT_REASON_VMWRITE:
5123 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5124 			vmcs12->vmwrite_bitmap);
5125 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5126 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5127 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5128 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5129 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5130 		/*
5131 		 * VMX instructions trap unconditionally. This allows L1 to
5132 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5133 		 */
5134 		return true;
5135 	case EXIT_REASON_CR_ACCESS:
5136 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5137 	case EXIT_REASON_DR_ACCESS:
5138 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5139 	case EXIT_REASON_IO_INSTRUCTION:
5140 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5141 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5142 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5143 	case EXIT_REASON_MSR_READ:
5144 	case EXIT_REASON_MSR_WRITE:
5145 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5146 	case EXIT_REASON_INVALID_STATE:
5147 		return true;
5148 	case EXIT_REASON_MWAIT_INSTRUCTION:
5149 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5150 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5151 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
5152 	case EXIT_REASON_MONITOR_INSTRUCTION:
5153 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5154 	case EXIT_REASON_PAUSE_INSTRUCTION:
5155 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5156 			nested_cpu_has2(vmcs12,
5157 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5158 	case EXIT_REASON_MCE_DURING_VMENTRY:
5159 		return false;
5160 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5161 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5162 	case EXIT_REASON_APIC_ACCESS:
5163 	case EXIT_REASON_APIC_WRITE:
5164 	case EXIT_REASON_EOI_INDUCED:
5165 		/*
5166 		 * The controls for "virtualize APIC accesses," "APIC-
5167 		 * register virtualization," and "virtual-interrupt
5168 		 * delivery" only come from vmcs12.
5169 		 */
5170 		return true;
5171 	case EXIT_REASON_EPT_VIOLATION:
5172 		/*
5173 		 * L0 always deals with the EPT violation. If nested EPT is
5174 		 * used, and the nested mmu code discovers that the address is
5175 		 * missing in the guest EPT table (EPT12), the EPT violation
5176 		 * will be injected with nested_ept_inject_page_fault()
5177 		 */
5178 		return false;
5179 	case EXIT_REASON_EPT_MISCONFIG:
5180 		/*
5181 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5182 		 * table (shadow on EPT) or a merged EPT table that L0 built
5183 		 * (EPT on EPT). So any problems with the structure of the
5184 		 * table is L0's fault.
5185 		 */
5186 		return false;
5187 	case EXIT_REASON_INVPCID:
5188 		return
5189 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5190 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5191 	case EXIT_REASON_WBINVD:
5192 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5193 	case EXIT_REASON_XSETBV:
5194 		return true;
5195 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5196 		/*
5197 		 * This should never happen, since it is not possible to
5198 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5199 		 * If if it were, XSS would have to be checked against
5200 		 * the XSS exit bitmap in vmcs12.
5201 		 */
5202 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5203 	case EXIT_REASON_PREEMPTION_TIMER:
5204 		return false;
5205 	case EXIT_REASON_PML_FULL:
5206 		/* We emulate PML support to L1. */
5207 		return false;
5208 	case EXIT_REASON_VMFUNC:
5209 		/* VM functions are emulated through L2->L0 vmexits. */
5210 		return false;
5211 	case EXIT_REASON_ENCLS:
5212 		/* SGX is never exposed to L1 */
5213 		return false;
5214 	default:
5215 		return true;
5216 	}
5217 }
5218 
5219 
5220 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5221 				struct kvm_nested_state __user *user_kvm_nested_state,
5222 				u32 user_data_size)
5223 {
5224 	struct vcpu_vmx *vmx;
5225 	struct vmcs12 *vmcs12;
5226 	struct kvm_nested_state kvm_state = {
5227 		.flags = 0,
5228 		.format = 0,
5229 		.size = sizeof(kvm_state),
5230 		.vmx.vmxon_pa = -1ull,
5231 		.vmx.vmcs_pa = -1ull,
5232 	};
5233 
5234 	if (!vcpu)
5235 		return kvm_state.size + 2 * VMCS12_SIZE;
5236 
5237 	vmx = to_vmx(vcpu);
5238 	vmcs12 = get_vmcs12(vcpu);
5239 
5240 	if (nested_vmx_allowed(vcpu) && vmx->nested.enlightened_vmcs_enabled)
5241 		kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5242 
5243 	if (nested_vmx_allowed(vcpu) &&
5244 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5245 		kvm_state.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5246 		kvm_state.vmx.vmcs_pa = vmx->nested.current_vmptr;
5247 
5248 		if (vmx_has_valid_vmcs12(vcpu)) {
5249 			kvm_state.size += VMCS12_SIZE;
5250 
5251 			if (is_guest_mode(vcpu) &&
5252 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5253 			    vmcs12->vmcs_link_pointer != -1ull)
5254 				kvm_state.size += VMCS12_SIZE;
5255 		}
5256 
5257 		if (vmx->nested.smm.vmxon)
5258 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5259 
5260 		if (vmx->nested.smm.guest_mode)
5261 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5262 
5263 		if (is_guest_mode(vcpu)) {
5264 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5265 
5266 			if (vmx->nested.nested_run_pending)
5267 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5268 		}
5269 	}
5270 
5271 	if (user_data_size < kvm_state.size)
5272 		goto out;
5273 
5274 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5275 		return -EFAULT;
5276 
5277 	if (!vmx_has_valid_vmcs12(vcpu))
5278 		goto out;
5279 
5280 	/*
5281 	 * When running L2, the authoritative vmcs12 state is in the
5282 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5283 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5284 	 * need_vmcs12_sync is set, in which case, the authoritative
5285 	 * vmcs12 state is in the vmcs12 already.
5286 	 */
5287 	if (is_guest_mode(vcpu)) {
5288 		sync_vmcs12(vcpu, vmcs12);
5289 	} else if (!vmx->nested.need_vmcs12_sync) {
5290 		if (vmx->nested.hv_evmcs)
5291 			copy_enlightened_to_vmcs12(vmx);
5292 		else if (enable_shadow_vmcs)
5293 			copy_shadow_to_vmcs12(vmx);
5294 	}
5295 
5296 	/*
5297 	 * Copy over the full allocated size of vmcs12 rather than just the size
5298 	 * of the struct.
5299 	 */
5300 	if (copy_to_user(user_kvm_nested_state->data, vmcs12, VMCS12_SIZE))
5301 		return -EFAULT;
5302 
5303 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5304 	    vmcs12->vmcs_link_pointer != -1ull) {
5305 		if (copy_to_user(user_kvm_nested_state->data + VMCS12_SIZE,
5306 				 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
5307 			return -EFAULT;
5308 	}
5309 
5310 out:
5311 	return kvm_state.size;
5312 }
5313 
5314 /*
5315  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5316  */
5317 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5318 {
5319 	if (is_guest_mode(vcpu)) {
5320 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5321 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5322 	}
5323 	free_nested(vcpu);
5324 }
5325 
5326 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5327 				struct kvm_nested_state __user *user_kvm_nested_state,
5328 				struct kvm_nested_state *kvm_state)
5329 {
5330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5331 	struct vmcs12 *vmcs12;
5332 	u32 exit_qual;
5333 	int ret;
5334 
5335 	if (kvm_state->format != 0)
5336 		return -EINVAL;
5337 
5338 	if (kvm_state->flags & KVM_STATE_NESTED_EVMCS)
5339 		nested_enable_evmcs(vcpu, NULL);
5340 
5341 	if (!nested_vmx_allowed(vcpu))
5342 		return kvm_state->vmx.vmxon_pa == -1ull ? 0 : -EINVAL;
5343 
5344 	if (kvm_state->vmx.vmxon_pa == -1ull) {
5345 		if (kvm_state->vmx.smm.flags)
5346 			return -EINVAL;
5347 
5348 		if (kvm_state->vmx.vmcs_pa != -1ull)
5349 			return -EINVAL;
5350 
5351 		vmx_leave_nested(vcpu);
5352 		return 0;
5353 	}
5354 
5355 	if (!page_address_valid(vcpu, kvm_state->vmx.vmxon_pa))
5356 		return -EINVAL;
5357 
5358 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5359 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5360 		return -EINVAL;
5361 
5362 	if (kvm_state->vmx.smm.flags &
5363 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5364 		return -EINVAL;
5365 
5366 	/*
5367 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5368 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5369 	 * must be zero.
5370 	 */
5371 	if (is_smm(vcpu) ? kvm_state->flags : kvm_state->vmx.smm.flags)
5372 		return -EINVAL;
5373 
5374 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5375 	    !(kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5376 		return -EINVAL;
5377 
5378 	vmx_leave_nested(vcpu);
5379 	if (kvm_state->vmx.vmxon_pa == -1ull)
5380 		return 0;
5381 
5382 	vmx->nested.vmxon_ptr = kvm_state->vmx.vmxon_pa;
5383 	ret = enter_vmx_operation(vcpu);
5384 	if (ret)
5385 		return ret;
5386 
5387 	/* Empty 'VMXON' state is permitted */
5388 	if (kvm_state->size < sizeof(kvm_state) + sizeof(*vmcs12))
5389 		return 0;
5390 
5391 	if (kvm_state->vmx.vmcs_pa != -1ull) {
5392 		if (kvm_state->vmx.vmcs_pa == kvm_state->vmx.vmxon_pa ||
5393 		    !page_address_valid(vcpu, kvm_state->vmx.vmcs_pa))
5394 			return -EINVAL;
5395 
5396 		set_current_vmptr(vmx, kvm_state->vmx.vmcs_pa);
5397 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5398 		/*
5399 		 * Sync eVMCS upon entry as we may not have
5400 		 * HV_X64_MSR_VP_ASSIST_PAGE set up yet.
5401 		 */
5402 		vmx->nested.need_vmcs12_sync = true;
5403 	} else {
5404 		return -EINVAL;
5405 	}
5406 
5407 	if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5408 		vmx->nested.smm.vmxon = true;
5409 		vmx->nested.vmxon = false;
5410 
5411 		if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5412 			vmx->nested.smm.guest_mode = true;
5413 	}
5414 
5415 	vmcs12 = get_vmcs12(vcpu);
5416 	if (copy_from_user(vmcs12, user_kvm_nested_state->data, sizeof(*vmcs12)))
5417 		return -EFAULT;
5418 
5419 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5420 		return -EINVAL;
5421 
5422 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5423 		return 0;
5424 
5425 	vmx->nested.nested_run_pending =
5426 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5427 
5428 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5429 	    vmcs12->vmcs_link_pointer != -1ull) {
5430 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5431 
5432 		if (kvm_state->size < sizeof(kvm_state) + 2 * sizeof(*vmcs12))
5433 			return -EINVAL;
5434 
5435 		if (copy_from_user(shadow_vmcs12,
5436 				   user_kvm_nested_state->data + VMCS12_SIZE,
5437 				   sizeof(*vmcs12)))
5438 			return -EFAULT;
5439 
5440 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5441 		    !shadow_vmcs12->hdr.shadow_vmcs)
5442 			return -EINVAL;
5443 	}
5444 
5445 	if (nested_vmx_check_vmentry_prereqs(vcpu, vmcs12) ||
5446 	    nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
5447 		return -EINVAL;
5448 
5449 	vmx->nested.dirty_vmcs12 = true;
5450 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5451 	if (ret)
5452 		return -EINVAL;
5453 
5454 	return 0;
5455 }
5456 
5457 void nested_vmx_vcpu_setup(void)
5458 {
5459 	if (enable_shadow_vmcs) {
5460 		/*
5461 		 * At vCPU creation, "VMWRITE to any supported field
5462 		 * in the VMCS" is supported, so use the more
5463 		 * permissive vmx_vmread_bitmap to specify both read
5464 		 * and write permissions for the shadow VMCS.
5465 		 */
5466 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5467 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmread_bitmap));
5468 	}
5469 }
5470 
5471 /*
5472  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5473  * returned for the various VMX controls MSRs when nested VMX is enabled.
5474  * The same values should also be used to verify that vmcs12 control fields are
5475  * valid during nested entry from L1 to L2.
5476  * Each of these control msrs has a low and high 32-bit half: A low bit is on
5477  * if the corresponding bit in the (32-bit) control field *must* be on, and a
5478  * bit in the high half is on if the corresponding bit in the control field
5479  * may be on. See also vmx_control_verify().
5480  */
5481 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps,
5482 				bool apicv)
5483 {
5484 	/*
5485 	 * Note that as a general rule, the high half of the MSRs (bits in
5486 	 * the control fields which may be 1) should be initialized by the
5487 	 * intersection of the underlying hardware's MSR (i.e., features which
5488 	 * can be supported) and the list of features we want to expose -
5489 	 * because they are known to be properly supported in our code.
5490 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
5491 	 * be set to 0, meaning that L1 may turn off any of these bits. The
5492 	 * reason is that if one of these bits is necessary, it will appear
5493 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
5494 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
5495 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
5496 	 * These rules have exceptions below.
5497 	 */
5498 
5499 	/* pin-based controls */
5500 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
5501 		msrs->pinbased_ctls_low,
5502 		msrs->pinbased_ctls_high);
5503 	msrs->pinbased_ctls_low |=
5504 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5505 	msrs->pinbased_ctls_high &=
5506 		PIN_BASED_EXT_INTR_MASK |
5507 		PIN_BASED_NMI_EXITING |
5508 		PIN_BASED_VIRTUAL_NMIS |
5509 		(apicv ? PIN_BASED_POSTED_INTR : 0);
5510 	msrs->pinbased_ctls_high |=
5511 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5512 		PIN_BASED_VMX_PREEMPTION_TIMER;
5513 
5514 	/* exit controls */
5515 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
5516 		msrs->exit_ctls_low,
5517 		msrs->exit_ctls_high);
5518 	msrs->exit_ctls_low =
5519 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
5520 
5521 	msrs->exit_ctls_high &=
5522 #ifdef CONFIG_X86_64
5523 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
5524 #endif
5525 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
5526 	msrs->exit_ctls_high |=
5527 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
5528 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
5529 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
5530 
5531 	/* We support free control of debug control saving. */
5532 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
5533 
5534 	/* entry controls */
5535 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
5536 		msrs->entry_ctls_low,
5537 		msrs->entry_ctls_high);
5538 	msrs->entry_ctls_low =
5539 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
5540 	msrs->entry_ctls_high &=
5541 #ifdef CONFIG_X86_64
5542 		VM_ENTRY_IA32E_MODE |
5543 #endif
5544 		VM_ENTRY_LOAD_IA32_PAT;
5545 	msrs->entry_ctls_high |=
5546 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
5547 
5548 	/* We support free control of debug control loading. */
5549 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
5550 
5551 	/* cpu-based controls */
5552 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
5553 		msrs->procbased_ctls_low,
5554 		msrs->procbased_ctls_high);
5555 	msrs->procbased_ctls_low =
5556 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5557 	msrs->procbased_ctls_high &=
5558 		CPU_BASED_VIRTUAL_INTR_PENDING |
5559 		CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
5560 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
5561 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
5562 		CPU_BASED_CR3_STORE_EXITING |
5563 #ifdef CONFIG_X86_64
5564 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
5565 #endif
5566 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
5567 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
5568 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
5569 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
5570 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
5571 	/*
5572 	 * We can allow some features even when not supported by the
5573 	 * hardware. For example, L1 can specify an MSR bitmap - and we
5574 	 * can use it to avoid exits to L1 - even when L0 runs L2
5575 	 * without MSR bitmaps.
5576 	 */
5577 	msrs->procbased_ctls_high |=
5578 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5579 		CPU_BASED_USE_MSR_BITMAPS;
5580 
5581 	/* We support free control of CR3 access interception. */
5582 	msrs->procbased_ctls_low &=
5583 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
5584 
5585 	/*
5586 	 * secondary cpu-based controls.  Do not include those that
5587 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
5588 	 */
5589 	if (msrs->procbased_ctls_high & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)
5590 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
5591 		      msrs->secondary_ctls_low,
5592 		      msrs->secondary_ctls_high);
5593 
5594 	msrs->secondary_ctls_low = 0;
5595 	msrs->secondary_ctls_high &=
5596 		SECONDARY_EXEC_DESC |
5597 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
5598 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
5599 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
5600 		SECONDARY_EXEC_WBINVD_EXITING;
5601 
5602 	/*
5603 	 * We can emulate "VMCS shadowing," even if the hardware
5604 	 * doesn't support it.
5605 	 */
5606 	msrs->secondary_ctls_high |=
5607 		SECONDARY_EXEC_SHADOW_VMCS;
5608 
5609 	if (enable_ept) {
5610 		/* nested EPT: emulate EPT also to L1 */
5611 		msrs->secondary_ctls_high |=
5612 			SECONDARY_EXEC_ENABLE_EPT;
5613 		msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
5614 			 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
5615 		if (cpu_has_vmx_ept_execute_only())
5616 			msrs->ept_caps |=
5617 				VMX_EPT_EXECUTE_ONLY_BIT;
5618 		msrs->ept_caps &= ept_caps;
5619 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
5620 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
5621 			VMX_EPT_1GB_PAGE_BIT;
5622 		if (enable_ept_ad_bits) {
5623 			msrs->secondary_ctls_high |=
5624 				SECONDARY_EXEC_ENABLE_PML;
5625 			msrs->ept_caps |= VMX_EPT_AD_BIT;
5626 		}
5627 	}
5628 
5629 	if (cpu_has_vmx_vmfunc()) {
5630 		msrs->secondary_ctls_high |=
5631 			SECONDARY_EXEC_ENABLE_VMFUNC;
5632 		/*
5633 		 * Advertise EPTP switching unconditionally
5634 		 * since we emulate it
5635 		 */
5636 		if (enable_ept)
5637 			msrs->vmfunc_controls =
5638 				VMX_VMFUNC_EPTP_SWITCHING;
5639 	}
5640 
5641 	/*
5642 	 * Old versions of KVM use the single-context version without
5643 	 * checking for support, so declare that it is supported even
5644 	 * though it is treated as global context.  The alternative is
5645 	 * not failing the single-context invvpid, and it is worse.
5646 	 */
5647 	if (enable_vpid) {
5648 		msrs->secondary_ctls_high |=
5649 			SECONDARY_EXEC_ENABLE_VPID;
5650 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
5651 			VMX_VPID_EXTENT_SUPPORTED_MASK;
5652 	}
5653 
5654 	if (enable_unrestricted_guest)
5655 		msrs->secondary_ctls_high |=
5656 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
5657 
5658 	if (flexpriority_enabled)
5659 		msrs->secondary_ctls_high |=
5660 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5661 
5662 	/* miscellaneous data */
5663 	rdmsr(MSR_IA32_VMX_MISC,
5664 		msrs->misc_low,
5665 		msrs->misc_high);
5666 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
5667 	msrs->misc_low |=
5668 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
5669 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
5670 		VMX_MISC_ACTIVITY_HLT;
5671 	msrs->misc_high = 0;
5672 
5673 	/*
5674 	 * This MSR reports some information about VMX support. We
5675 	 * should return information about the VMX we emulate for the
5676 	 * guest, and the VMCS structure we give it - not about the
5677 	 * VMX support of the underlying hardware.
5678 	 */
5679 	msrs->basic =
5680 		VMCS12_REVISION |
5681 		VMX_BASIC_TRUE_CTLS |
5682 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
5683 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
5684 
5685 	if (cpu_has_vmx_basic_inout())
5686 		msrs->basic |= VMX_BASIC_INOUT;
5687 
5688 	/*
5689 	 * These MSRs specify bits which the guest must keep fixed on
5690 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
5691 	 * We picked the standard core2 setting.
5692 	 */
5693 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
5694 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
5695 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
5696 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
5697 
5698 	/* These MSRs specify bits which the guest must keep fixed off. */
5699 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
5700 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
5701 
5702 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
5703 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
5704 }
5705 
5706 void nested_vmx_hardware_unsetup(void)
5707 {
5708 	int i;
5709 
5710 	if (enable_shadow_vmcs) {
5711 		for (i = 0; i < VMX_BITMAP_NR; i++)
5712 			free_page((unsigned long)vmx_bitmap[i]);
5713 	}
5714 }
5715 
5716 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
5717 {
5718 	int i;
5719 
5720 	if (!cpu_has_vmx_shadow_vmcs())
5721 		enable_shadow_vmcs = 0;
5722 	if (enable_shadow_vmcs) {
5723 		for (i = 0; i < VMX_BITMAP_NR; i++) {
5724 			/*
5725 			 * The vmx_bitmap is not tied to a VM and so should
5726 			 * not be charged to a memcg.
5727 			 */
5728 			vmx_bitmap[i] = (unsigned long *)
5729 				__get_free_page(GFP_KERNEL);
5730 			if (!vmx_bitmap[i]) {
5731 				nested_vmx_hardware_unsetup();
5732 				return -ENOMEM;
5733 			}
5734 		}
5735 
5736 		init_vmcs_shadow_fields();
5737 	}
5738 
5739 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear,
5740 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch,
5741 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld,
5742 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst,
5743 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread,
5744 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume,
5745 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite,
5746 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff,
5747 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon,
5748 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept,
5749 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid,
5750 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc,
5751 
5752 	kvm_x86_ops->check_nested_events = vmx_check_nested_events;
5753 	kvm_x86_ops->get_nested_state = vmx_get_nested_state;
5754 	kvm_x86_ops->set_nested_state = vmx_set_nested_state;
5755 	kvm_x86_ops->get_vmcs12_pages = nested_get_vmcs12_pages,
5756 	kvm_x86_ops->nested_enable_evmcs = nested_enable_evmcs;
5757 	kvm_x86_ops->nested_get_evmcs_version = nested_get_evmcs_version;
5758 
5759 	return 0;
5760 }
5761