xref: /openbmc/linux/arch/x86/kvm/vmx/nested.c (revision c67e8ec0)
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 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_switch_vmcs(vcpu, &to_vmx(vcpu)->vmcs01);
277 	free_nested(vcpu);
278 	vcpu_put(vcpu);
279 }
280 
281 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
282 		struct x86_exception *fault)
283 {
284 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
285 	struct vcpu_vmx *vmx = to_vmx(vcpu);
286 	u32 exit_reason;
287 	unsigned long exit_qualification = vcpu->arch.exit_qualification;
288 
289 	if (vmx->nested.pml_full) {
290 		exit_reason = EXIT_REASON_PML_FULL;
291 		vmx->nested.pml_full = false;
292 		exit_qualification &= INTR_INFO_UNBLOCK_NMI;
293 	} else if (fault->error_code & PFERR_RSVD_MASK)
294 		exit_reason = EXIT_REASON_EPT_MISCONFIG;
295 	else
296 		exit_reason = EXIT_REASON_EPT_VIOLATION;
297 
298 	nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
299 	vmcs12->guest_physical_address = fault->address;
300 }
301 
302 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
303 {
304 	WARN_ON(mmu_is_nested(vcpu));
305 
306 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
307 	kvm_init_shadow_ept_mmu(vcpu,
308 			to_vmx(vcpu)->nested.msrs.ept_caps &
309 			VMX_EPT_EXECUTE_ONLY_BIT,
310 			nested_ept_ad_enabled(vcpu),
311 			nested_ept_get_cr3(vcpu));
312 	vcpu->arch.mmu->set_cr3           = vmx_set_cr3;
313 	vcpu->arch.mmu->get_cr3           = nested_ept_get_cr3;
314 	vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
315 	vcpu->arch.mmu->get_pdptr         = kvm_pdptr_read;
316 
317 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
318 }
319 
320 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
321 {
322 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
323 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
324 }
325 
326 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
327 					    u16 error_code)
328 {
329 	bool inequality, bit;
330 
331 	bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
332 	inequality =
333 		(error_code & vmcs12->page_fault_error_code_mask) !=
334 		 vmcs12->page_fault_error_code_match;
335 	return inequality ^ bit;
336 }
337 
338 
339 /*
340  * KVM wants to inject page-faults which it got to the guest. This function
341  * checks whether in a nested guest, we need to inject them to L1 or L2.
342  */
343 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
344 {
345 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
346 	unsigned int nr = vcpu->arch.exception.nr;
347 	bool has_payload = vcpu->arch.exception.has_payload;
348 	unsigned long payload = vcpu->arch.exception.payload;
349 
350 	if (nr == PF_VECTOR) {
351 		if (vcpu->arch.exception.nested_apf) {
352 			*exit_qual = vcpu->arch.apf.nested_apf_token;
353 			return 1;
354 		}
355 		if (nested_vmx_is_page_fault_vmexit(vmcs12,
356 						    vcpu->arch.exception.error_code)) {
357 			*exit_qual = has_payload ? payload : vcpu->arch.cr2;
358 			return 1;
359 		}
360 	} else if (vmcs12->exception_bitmap & (1u << nr)) {
361 		if (nr == DB_VECTOR) {
362 			if (!has_payload) {
363 				payload = vcpu->arch.dr6;
364 				payload &= ~(DR6_FIXED_1 | DR6_BT);
365 				payload ^= DR6_RTM;
366 			}
367 			*exit_qual = payload;
368 		} else
369 			*exit_qual = 0;
370 		return 1;
371 	}
372 
373 	return 0;
374 }
375 
376 
377 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
378 		struct x86_exception *fault)
379 {
380 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
381 
382 	WARN_ON(!is_guest_mode(vcpu));
383 
384 	if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
385 		!to_vmx(vcpu)->nested.nested_run_pending) {
386 		vmcs12->vm_exit_intr_error_code = fault->error_code;
387 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
388 				  PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
389 				  INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
390 				  fault->address);
391 	} else {
392 		kvm_inject_page_fault(vcpu, fault);
393 	}
394 }
395 
396 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
397 {
398 	return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
399 }
400 
401 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
402 					       struct vmcs12 *vmcs12)
403 {
404 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
405 		return 0;
406 
407 	if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
408 	    !page_address_valid(vcpu, vmcs12->io_bitmap_b))
409 		return -EINVAL;
410 
411 	return 0;
412 }
413 
414 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
415 						struct vmcs12 *vmcs12)
416 {
417 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
418 		return 0;
419 
420 	if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
421 		return -EINVAL;
422 
423 	return 0;
424 }
425 
426 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
427 						struct vmcs12 *vmcs12)
428 {
429 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
430 		return 0;
431 
432 	if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
433 		return -EINVAL;
434 
435 	return 0;
436 }
437 
438 /*
439  * Check if MSR is intercepted for L01 MSR bitmap.
440  */
441 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
442 {
443 	unsigned long *msr_bitmap;
444 	int f = sizeof(unsigned long);
445 
446 	if (!cpu_has_vmx_msr_bitmap())
447 		return true;
448 
449 	msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
450 
451 	if (msr <= 0x1fff) {
452 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
453 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
454 		msr &= 0x1fff;
455 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
456 	}
457 
458 	return true;
459 }
460 
461 /*
462  * If a msr is allowed by L0, we should check whether it is allowed by L1.
463  * The corresponding bit will be cleared unless both of L0 and L1 allow it.
464  */
465 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
466 					       unsigned long *msr_bitmap_nested,
467 					       u32 msr, int type)
468 {
469 	int f = sizeof(unsigned long);
470 
471 	/*
472 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
473 	 * have the write-low and read-high bitmap offsets the wrong way round.
474 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
475 	 */
476 	if (msr <= 0x1fff) {
477 		if (type & MSR_TYPE_R &&
478 		   !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
479 			/* read-low */
480 			__clear_bit(msr, msr_bitmap_nested + 0x000 / f);
481 
482 		if (type & MSR_TYPE_W &&
483 		   !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
484 			/* write-low */
485 			__clear_bit(msr, msr_bitmap_nested + 0x800 / f);
486 
487 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
488 		msr &= 0x1fff;
489 		if (type & MSR_TYPE_R &&
490 		   !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
491 			/* read-high */
492 			__clear_bit(msr, msr_bitmap_nested + 0x400 / f);
493 
494 		if (type & MSR_TYPE_W &&
495 		   !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
496 			/* write-high */
497 			__clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
498 
499 	}
500 }
501 
502 /*
503  * Merge L0's and L1's MSR bitmap, return false to indicate that
504  * we do not use the hardware.
505  */
506 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
507 						 struct vmcs12 *vmcs12)
508 {
509 	int msr;
510 	struct page *page;
511 	unsigned long *msr_bitmap_l1;
512 	unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
513 	/*
514 	 * pred_cmd & spec_ctrl are trying to verify two things:
515 	 *
516 	 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
517 	 *    ensures that we do not accidentally generate an L02 MSR bitmap
518 	 *    from the L12 MSR bitmap that is too permissive.
519 	 * 2. That L1 or L2s have actually used the MSR. This avoids
520 	 *    unnecessarily merging of the bitmap if the MSR is unused. This
521 	 *    works properly because we only update the L01 MSR bitmap lazily.
522 	 *    So even if L0 should pass L1 these MSRs, the L01 bitmap is only
523 	 *    updated to reflect this when L1 (or its L2s) actually write to
524 	 *    the MSR.
525 	 */
526 	bool pred_cmd = !msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD);
527 	bool spec_ctrl = !msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL);
528 
529 	/* Nothing to do if the MSR bitmap is not in use.  */
530 	if (!cpu_has_vmx_msr_bitmap() ||
531 	    !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
532 		return false;
533 
534 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
535 	    !pred_cmd && !spec_ctrl)
536 		return false;
537 
538 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
539 	if (is_error_page(page))
540 		return false;
541 
542 	msr_bitmap_l1 = (unsigned long *)kmap(page);
543 	if (nested_cpu_has_apic_reg_virt(vmcs12)) {
544 		/*
545 		 * L0 need not intercept reads for MSRs between 0x800 and 0x8ff, it
546 		 * just lets the processor take the value from the virtual-APIC page;
547 		 * take those 256 bits directly from the L1 bitmap.
548 		 */
549 		for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
550 			unsigned word = msr / BITS_PER_LONG;
551 			msr_bitmap_l0[word] = msr_bitmap_l1[word];
552 			msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
553 		}
554 	} else {
555 		for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
556 			unsigned word = msr / BITS_PER_LONG;
557 			msr_bitmap_l0[word] = ~0;
558 			msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
559 		}
560 	}
561 
562 	nested_vmx_disable_intercept_for_msr(
563 		msr_bitmap_l1, msr_bitmap_l0,
564 		X2APIC_MSR(APIC_TASKPRI),
565 		MSR_TYPE_W);
566 
567 	if (nested_cpu_has_vid(vmcs12)) {
568 		nested_vmx_disable_intercept_for_msr(
569 			msr_bitmap_l1, msr_bitmap_l0,
570 			X2APIC_MSR(APIC_EOI),
571 			MSR_TYPE_W);
572 		nested_vmx_disable_intercept_for_msr(
573 			msr_bitmap_l1, msr_bitmap_l0,
574 			X2APIC_MSR(APIC_SELF_IPI),
575 			MSR_TYPE_W);
576 	}
577 
578 	if (spec_ctrl)
579 		nested_vmx_disable_intercept_for_msr(
580 					msr_bitmap_l1, msr_bitmap_l0,
581 					MSR_IA32_SPEC_CTRL,
582 					MSR_TYPE_R | MSR_TYPE_W);
583 
584 	if (pred_cmd)
585 		nested_vmx_disable_intercept_for_msr(
586 					msr_bitmap_l1, msr_bitmap_l0,
587 					MSR_IA32_PRED_CMD,
588 					MSR_TYPE_W);
589 
590 	kunmap(page);
591 	kvm_release_page_clean(page);
592 
593 	return true;
594 }
595 
596 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
597 				       struct vmcs12 *vmcs12)
598 {
599 	struct vmcs12 *shadow;
600 	struct page *page;
601 
602 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
603 	    vmcs12->vmcs_link_pointer == -1ull)
604 		return;
605 
606 	shadow = get_shadow_vmcs12(vcpu);
607 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->vmcs_link_pointer);
608 
609 	memcpy(shadow, kmap(page), VMCS12_SIZE);
610 
611 	kunmap(page);
612 	kvm_release_page_clean(page);
613 }
614 
615 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
616 					      struct vmcs12 *vmcs12)
617 {
618 	struct vcpu_vmx *vmx = to_vmx(vcpu);
619 
620 	if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
621 	    vmcs12->vmcs_link_pointer == -1ull)
622 		return;
623 
624 	kvm_write_guest(vmx->vcpu.kvm, vmcs12->vmcs_link_pointer,
625 			get_shadow_vmcs12(vcpu), VMCS12_SIZE);
626 }
627 
628 /*
629  * In nested virtualization, check if L1 has set
630  * VM_EXIT_ACK_INTR_ON_EXIT
631  */
632 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
633 {
634 	return get_vmcs12(vcpu)->vm_exit_controls &
635 		VM_EXIT_ACK_INTR_ON_EXIT;
636 }
637 
638 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
639 {
640 	return nested_cpu_has_nmi_exiting(get_vmcs12(vcpu));
641 }
642 
643 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
644 					  struct vmcs12 *vmcs12)
645 {
646 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
647 	    !page_address_valid(vcpu, vmcs12->apic_access_addr))
648 		return -EINVAL;
649 	else
650 		return 0;
651 }
652 
653 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
654 					   struct vmcs12 *vmcs12)
655 {
656 	if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
657 	    !nested_cpu_has_apic_reg_virt(vmcs12) &&
658 	    !nested_cpu_has_vid(vmcs12) &&
659 	    !nested_cpu_has_posted_intr(vmcs12))
660 		return 0;
661 
662 	/*
663 	 * If virtualize x2apic mode is enabled,
664 	 * virtualize apic access must be disabled.
665 	 */
666 	if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
667 	    nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
668 		return -EINVAL;
669 
670 	/*
671 	 * If virtual interrupt delivery is enabled,
672 	 * we must exit on external interrupts.
673 	 */
674 	if (nested_cpu_has_vid(vmcs12) &&
675 	   !nested_exit_on_intr(vcpu))
676 		return -EINVAL;
677 
678 	/*
679 	 * bits 15:8 should be zero in posted_intr_nv,
680 	 * the descriptor address has been already checked
681 	 * in nested_get_vmcs12_pages.
682 	 *
683 	 * bits 5:0 of posted_intr_desc_addr should be zero.
684 	 */
685 	if (nested_cpu_has_posted_intr(vmcs12) &&
686 	   (!nested_cpu_has_vid(vmcs12) ||
687 	    !nested_exit_intr_ack_set(vcpu) ||
688 	    (vmcs12->posted_intr_nv & 0xff00) ||
689 	    (vmcs12->posted_intr_desc_addr & 0x3f) ||
690 	    (vmcs12->posted_intr_desc_addr >> cpuid_maxphyaddr(vcpu))))
691 		return -EINVAL;
692 
693 	/* tpr shadow is needed by all apicv features. */
694 	if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
695 		return -EINVAL;
696 
697 	return 0;
698 }
699 
700 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
701 				       u32 count, u64 addr)
702 {
703 	int maxphyaddr;
704 
705 	if (count == 0)
706 		return 0;
707 	maxphyaddr = cpuid_maxphyaddr(vcpu);
708 	if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
709 	    (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr)
710 		return -EINVAL;
711 
712 	return 0;
713 }
714 
715 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
716 						     struct vmcs12 *vmcs12)
717 {
718 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_load_count,
719 					vmcs12->vm_exit_msr_load_addr) ||
720 	    nested_vmx_check_msr_switch(vcpu, vmcs12->vm_exit_msr_store_count,
721 					vmcs12->vm_exit_msr_store_addr))
722 		return -EINVAL;
723 
724 	return 0;
725 }
726 
727 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
728                                                       struct vmcs12 *vmcs12)
729 {
730 	if (nested_vmx_check_msr_switch(vcpu, vmcs12->vm_entry_msr_load_count,
731                                         vmcs12->vm_entry_msr_load_addr))
732                 return -EINVAL;
733 
734 	return 0;
735 }
736 
737 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
738 					 struct vmcs12 *vmcs12)
739 {
740 	if (!nested_cpu_has_pml(vmcs12))
741 		return 0;
742 
743 	if (!nested_cpu_has_ept(vmcs12) ||
744 	    !page_address_valid(vcpu, vmcs12->pml_address))
745 		return -EINVAL;
746 
747 	return 0;
748 }
749 
750 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
751 							struct vmcs12 *vmcs12)
752 {
753 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
754 	    !nested_cpu_has_ept(vmcs12))
755 		return -EINVAL;
756 	return 0;
757 }
758 
759 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
760 							 struct vmcs12 *vmcs12)
761 {
762 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
763 	    !nested_cpu_has_ept(vmcs12))
764 		return -EINVAL;
765 	return 0;
766 }
767 
768 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
769 						 struct vmcs12 *vmcs12)
770 {
771 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
772 		return 0;
773 
774 	if (!page_address_valid(vcpu, vmcs12->vmread_bitmap) ||
775 	    !page_address_valid(vcpu, vmcs12->vmwrite_bitmap))
776 		return -EINVAL;
777 
778 	return 0;
779 }
780 
781 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
782 				       struct vmx_msr_entry *e)
783 {
784 	/* x2APIC MSR accesses are not allowed */
785 	if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
786 		return -EINVAL;
787 	if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
788 	    e->index == MSR_IA32_UCODE_REV)
789 		return -EINVAL;
790 	if (e->reserved != 0)
791 		return -EINVAL;
792 	return 0;
793 }
794 
795 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
796 				     struct vmx_msr_entry *e)
797 {
798 	if (e->index == MSR_FS_BASE ||
799 	    e->index == MSR_GS_BASE ||
800 	    e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
801 	    nested_vmx_msr_check_common(vcpu, e))
802 		return -EINVAL;
803 	return 0;
804 }
805 
806 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
807 				      struct vmx_msr_entry *e)
808 {
809 	if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
810 	    nested_vmx_msr_check_common(vcpu, e))
811 		return -EINVAL;
812 	return 0;
813 }
814 
815 /*
816  * Load guest's/host's msr at nested entry/exit.
817  * return 0 for success, entry index for failure.
818  */
819 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
820 {
821 	u32 i;
822 	struct vmx_msr_entry e;
823 	struct msr_data msr;
824 
825 	msr.host_initiated = false;
826 	for (i = 0; i < count; i++) {
827 		if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
828 					&e, sizeof(e))) {
829 			pr_debug_ratelimited(
830 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
831 				__func__, i, gpa + i * sizeof(e));
832 			goto fail;
833 		}
834 		if (nested_vmx_load_msr_check(vcpu, &e)) {
835 			pr_debug_ratelimited(
836 				"%s check failed (%u, 0x%x, 0x%x)\n",
837 				__func__, i, e.index, e.reserved);
838 			goto fail;
839 		}
840 		msr.index = e.index;
841 		msr.data = e.value;
842 		if (kvm_set_msr(vcpu, &msr)) {
843 			pr_debug_ratelimited(
844 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
845 				__func__, i, e.index, e.value);
846 			goto fail;
847 		}
848 	}
849 	return 0;
850 fail:
851 	return i + 1;
852 }
853 
854 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
855 {
856 	u32 i;
857 	struct vmx_msr_entry e;
858 
859 	for (i = 0; i < count; i++) {
860 		struct msr_data msr_info;
861 		if (kvm_vcpu_read_guest(vcpu,
862 					gpa + i * sizeof(e),
863 					&e, 2 * sizeof(u32))) {
864 			pr_debug_ratelimited(
865 				"%s cannot read MSR entry (%u, 0x%08llx)\n",
866 				__func__, i, gpa + i * sizeof(e));
867 			return -EINVAL;
868 		}
869 		if (nested_vmx_store_msr_check(vcpu, &e)) {
870 			pr_debug_ratelimited(
871 				"%s check failed (%u, 0x%x, 0x%x)\n",
872 				__func__, i, e.index, e.reserved);
873 			return -EINVAL;
874 		}
875 		msr_info.host_initiated = false;
876 		msr_info.index = e.index;
877 		if (kvm_get_msr(vcpu, &msr_info)) {
878 			pr_debug_ratelimited(
879 				"%s cannot read MSR (%u, 0x%x)\n",
880 				__func__, i, e.index);
881 			return -EINVAL;
882 		}
883 		if (kvm_vcpu_write_guest(vcpu,
884 					 gpa + i * sizeof(e) +
885 					     offsetof(struct vmx_msr_entry, value),
886 					 &msr_info.data, sizeof(msr_info.data))) {
887 			pr_debug_ratelimited(
888 				"%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
889 				__func__, i, e.index, msr_info.data);
890 			return -EINVAL;
891 		}
892 	}
893 	return 0;
894 }
895 
896 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
897 {
898 	unsigned long invalid_mask;
899 
900 	invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
901 	return (val & invalid_mask) == 0;
902 }
903 
904 /*
905  * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
906  * emulating VM entry into a guest with EPT enabled.
907  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
908  * is assigned to entry_failure_code on failure.
909  */
910 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
911 			       u32 *entry_failure_code)
912 {
913 	if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
914 		if (!nested_cr3_valid(vcpu, cr3)) {
915 			*entry_failure_code = ENTRY_FAIL_DEFAULT;
916 			return 1;
917 		}
918 
919 		/*
920 		 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
921 		 * must not be dereferenced.
922 		 */
923 		if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
924 		    !nested_ept) {
925 			if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
926 				*entry_failure_code = ENTRY_FAIL_PDPTE;
927 				return 1;
928 			}
929 		}
930 	}
931 
932 	if (!nested_ept)
933 		kvm_mmu_new_cr3(vcpu, cr3, false);
934 
935 	vcpu->arch.cr3 = cr3;
936 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
937 
938 	kvm_init_mmu(vcpu, false);
939 
940 	return 0;
941 }
942 
943 /*
944  * Returns if KVM is able to config CPU to tag TLB entries
945  * populated by L2 differently than TLB entries populated
946  * by L1.
947  *
948  * If L1 uses EPT, then TLB entries are tagged with different EPTP.
949  *
950  * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
951  * with different VPID (L1 entries are tagged with vmx->vpid
952  * while L2 entries are tagged with vmx->nested.vpid02).
953  */
954 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
955 {
956 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
957 
958 	return nested_cpu_has_ept(vmcs12) ||
959 	       (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
960 }
961 
962 static u16 nested_get_vpid02(struct kvm_vcpu *vcpu)
963 {
964 	struct vcpu_vmx *vmx = to_vmx(vcpu);
965 
966 	return vmx->nested.vpid02 ? vmx->nested.vpid02 : vmx->vpid;
967 }
968 
969 
970 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
971 {
972 	return fixed_bits_valid(control, low, high);
973 }
974 
975 static inline u64 vmx_control_msr(u32 low, u32 high)
976 {
977 	return low | ((u64)high << 32);
978 }
979 
980 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
981 {
982 	superset &= mask;
983 	subset &= mask;
984 
985 	return (superset | subset) == superset;
986 }
987 
988 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
989 {
990 	const u64 feature_and_reserved =
991 		/* feature (except bit 48; see below) */
992 		BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
993 		/* reserved */
994 		BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
995 	u64 vmx_basic = vmx->nested.msrs.basic;
996 
997 	if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
998 		return -EINVAL;
999 
1000 	/*
1001 	 * KVM does not emulate a version of VMX that constrains physical
1002 	 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1003 	 */
1004 	if (data & BIT_ULL(48))
1005 		return -EINVAL;
1006 
1007 	if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1008 	    vmx_basic_vmcs_revision_id(data))
1009 		return -EINVAL;
1010 
1011 	if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1012 		return -EINVAL;
1013 
1014 	vmx->nested.msrs.basic = data;
1015 	return 0;
1016 }
1017 
1018 static int
1019 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1020 {
1021 	u64 supported;
1022 	u32 *lowp, *highp;
1023 
1024 	switch (msr_index) {
1025 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1026 		lowp = &vmx->nested.msrs.pinbased_ctls_low;
1027 		highp = &vmx->nested.msrs.pinbased_ctls_high;
1028 		break;
1029 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1030 		lowp = &vmx->nested.msrs.procbased_ctls_low;
1031 		highp = &vmx->nested.msrs.procbased_ctls_high;
1032 		break;
1033 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1034 		lowp = &vmx->nested.msrs.exit_ctls_low;
1035 		highp = &vmx->nested.msrs.exit_ctls_high;
1036 		break;
1037 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1038 		lowp = &vmx->nested.msrs.entry_ctls_low;
1039 		highp = &vmx->nested.msrs.entry_ctls_high;
1040 		break;
1041 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1042 		lowp = &vmx->nested.msrs.secondary_ctls_low;
1043 		highp = &vmx->nested.msrs.secondary_ctls_high;
1044 		break;
1045 	default:
1046 		BUG();
1047 	}
1048 
1049 	supported = vmx_control_msr(*lowp, *highp);
1050 
1051 	/* Check must-be-1 bits are still 1. */
1052 	if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1053 		return -EINVAL;
1054 
1055 	/* Check must-be-0 bits are still 0. */
1056 	if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1057 		return -EINVAL;
1058 
1059 	*lowp = data;
1060 	*highp = data >> 32;
1061 	return 0;
1062 }
1063 
1064 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1065 {
1066 	const u64 feature_and_reserved_bits =
1067 		/* feature */
1068 		BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1069 		BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1070 		/* reserved */
1071 		GENMASK_ULL(13, 9) | BIT_ULL(31);
1072 	u64 vmx_misc;
1073 
1074 	vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
1075 				   vmx->nested.msrs.misc_high);
1076 
1077 	if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1078 		return -EINVAL;
1079 
1080 	if ((vmx->nested.msrs.pinbased_ctls_high &
1081 	     PIN_BASED_VMX_PREEMPTION_TIMER) &&
1082 	    vmx_misc_preemption_timer_rate(data) !=
1083 	    vmx_misc_preemption_timer_rate(vmx_misc))
1084 		return -EINVAL;
1085 
1086 	if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1087 		return -EINVAL;
1088 
1089 	if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1090 		return -EINVAL;
1091 
1092 	if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1093 		return -EINVAL;
1094 
1095 	vmx->nested.msrs.misc_low = data;
1096 	vmx->nested.msrs.misc_high = data >> 32;
1097 
1098 	/*
1099 	 * If L1 has read-only VM-exit information fields, use the
1100 	 * less permissive vmx_vmwrite_bitmap to specify write
1101 	 * permissions for the shadow VMCS.
1102 	 */
1103 	if (enable_shadow_vmcs && !nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
1104 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
1105 
1106 	return 0;
1107 }
1108 
1109 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1110 {
1111 	u64 vmx_ept_vpid_cap;
1112 
1113 	vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.msrs.ept_caps,
1114 					   vmx->nested.msrs.vpid_caps);
1115 
1116 	/* Every bit is either reserved or a feature bit. */
1117 	if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1118 		return -EINVAL;
1119 
1120 	vmx->nested.msrs.ept_caps = data;
1121 	vmx->nested.msrs.vpid_caps = data >> 32;
1122 	return 0;
1123 }
1124 
1125 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1126 {
1127 	u64 *msr;
1128 
1129 	switch (msr_index) {
1130 	case MSR_IA32_VMX_CR0_FIXED0:
1131 		msr = &vmx->nested.msrs.cr0_fixed0;
1132 		break;
1133 	case MSR_IA32_VMX_CR4_FIXED0:
1134 		msr = &vmx->nested.msrs.cr4_fixed0;
1135 		break;
1136 	default:
1137 		BUG();
1138 	}
1139 
1140 	/*
1141 	 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1142 	 * must be 1 in the restored value.
1143 	 */
1144 	if (!is_bitwise_subset(data, *msr, -1ULL))
1145 		return -EINVAL;
1146 
1147 	*msr = data;
1148 	return 0;
1149 }
1150 
1151 /*
1152  * Called when userspace is restoring VMX MSRs.
1153  *
1154  * Returns 0 on success, non-0 otherwise.
1155  */
1156 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1157 {
1158 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1159 
1160 	/*
1161 	 * Don't allow changes to the VMX capability MSRs while the vCPU
1162 	 * is in VMX operation.
1163 	 */
1164 	if (vmx->nested.vmxon)
1165 		return -EBUSY;
1166 
1167 	switch (msr_index) {
1168 	case MSR_IA32_VMX_BASIC:
1169 		return vmx_restore_vmx_basic(vmx, data);
1170 	case MSR_IA32_VMX_PINBASED_CTLS:
1171 	case MSR_IA32_VMX_PROCBASED_CTLS:
1172 	case MSR_IA32_VMX_EXIT_CTLS:
1173 	case MSR_IA32_VMX_ENTRY_CTLS:
1174 		/*
1175 		 * The "non-true" VMX capability MSRs are generated from the
1176 		 * "true" MSRs, so we do not support restoring them directly.
1177 		 *
1178 		 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1179 		 * should restore the "true" MSRs with the must-be-1 bits
1180 		 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1181 		 * DEFAULT SETTINGS".
1182 		 */
1183 		return -EINVAL;
1184 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1185 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1186 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1187 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1188 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1189 		return vmx_restore_control_msr(vmx, msr_index, data);
1190 	case MSR_IA32_VMX_MISC:
1191 		return vmx_restore_vmx_misc(vmx, data);
1192 	case MSR_IA32_VMX_CR0_FIXED0:
1193 	case MSR_IA32_VMX_CR4_FIXED0:
1194 		return vmx_restore_fixed0_msr(vmx, msr_index, data);
1195 	case MSR_IA32_VMX_CR0_FIXED1:
1196 	case MSR_IA32_VMX_CR4_FIXED1:
1197 		/*
1198 		 * These MSRs are generated based on the vCPU's CPUID, so we
1199 		 * do not support restoring them directly.
1200 		 */
1201 		return -EINVAL;
1202 	case MSR_IA32_VMX_EPT_VPID_CAP:
1203 		return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1204 	case MSR_IA32_VMX_VMCS_ENUM:
1205 		vmx->nested.msrs.vmcs_enum = data;
1206 		return 0;
1207 	default:
1208 		/*
1209 		 * The rest of the VMX capability MSRs do not support restore.
1210 		 */
1211 		return -EINVAL;
1212 	}
1213 }
1214 
1215 /* Returns 0 on success, non-0 otherwise. */
1216 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1217 {
1218 	switch (msr_index) {
1219 	case MSR_IA32_VMX_BASIC:
1220 		*pdata = msrs->basic;
1221 		break;
1222 	case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1223 	case MSR_IA32_VMX_PINBASED_CTLS:
1224 		*pdata = vmx_control_msr(
1225 			msrs->pinbased_ctls_low,
1226 			msrs->pinbased_ctls_high);
1227 		if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1228 			*pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1229 		break;
1230 	case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1231 	case MSR_IA32_VMX_PROCBASED_CTLS:
1232 		*pdata = vmx_control_msr(
1233 			msrs->procbased_ctls_low,
1234 			msrs->procbased_ctls_high);
1235 		if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1236 			*pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1237 		break;
1238 	case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1239 	case MSR_IA32_VMX_EXIT_CTLS:
1240 		*pdata = vmx_control_msr(
1241 			msrs->exit_ctls_low,
1242 			msrs->exit_ctls_high);
1243 		if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1244 			*pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1245 		break;
1246 	case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1247 	case MSR_IA32_VMX_ENTRY_CTLS:
1248 		*pdata = vmx_control_msr(
1249 			msrs->entry_ctls_low,
1250 			msrs->entry_ctls_high);
1251 		if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1252 			*pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1253 		break;
1254 	case MSR_IA32_VMX_MISC:
1255 		*pdata = vmx_control_msr(
1256 			msrs->misc_low,
1257 			msrs->misc_high);
1258 		break;
1259 	case MSR_IA32_VMX_CR0_FIXED0:
1260 		*pdata = msrs->cr0_fixed0;
1261 		break;
1262 	case MSR_IA32_VMX_CR0_FIXED1:
1263 		*pdata = msrs->cr0_fixed1;
1264 		break;
1265 	case MSR_IA32_VMX_CR4_FIXED0:
1266 		*pdata = msrs->cr4_fixed0;
1267 		break;
1268 	case MSR_IA32_VMX_CR4_FIXED1:
1269 		*pdata = msrs->cr4_fixed1;
1270 		break;
1271 	case MSR_IA32_VMX_VMCS_ENUM:
1272 		*pdata = msrs->vmcs_enum;
1273 		break;
1274 	case MSR_IA32_VMX_PROCBASED_CTLS2:
1275 		*pdata = vmx_control_msr(
1276 			msrs->secondary_ctls_low,
1277 			msrs->secondary_ctls_high);
1278 		break;
1279 	case MSR_IA32_VMX_EPT_VPID_CAP:
1280 		*pdata = msrs->ept_caps |
1281 			((u64)msrs->vpid_caps << 32);
1282 		break;
1283 	case MSR_IA32_VMX_VMFUNC:
1284 		*pdata = msrs->vmfunc_controls;
1285 		break;
1286 	default:
1287 		return 1;
1288 	}
1289 
1290 	return 0;
1291 }
1292 
1293 /*
1294  * Copy the writable VMCS shadow fields back to the VMCS12, in case
1295  * they have been modified by the L1 guest. Note that the "read-only"
1296  * VM-exit information fields are actually writable if the vCPU is
1297  * configured to support "VMWRITE to any supported field in the VMCS."
1298  */
1299 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1300 {
1301 	const u16 *fields[] = {
1302 		shadow_read_write_fields,
1303 		shadow_read_only_fields
1304 	};
1305 	const int max_fields[] = {
1306 		max_shadow_read_write_fields,
1307 		max_shadow_read_only_fields
1308 	};
1309 	int i, q;
1310 	unsigned long field;
1311 	u64 field_value;
1312 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1313 
1314 	preempt_disable();
1315 
1316 	vmcs_load(shadow_vmcs);
1317 
1318 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1319 		for (i = 0; i < max_fields[q]; i++) {
1320 			field = fields[q][i];
1321 			field_value = __vmcs_readl(field);
1322 			vmcs12_write_any(get_vmcs12(&vmx->vcpu), field, field_value);
1323 		}
1324 		/*
1325 		 * Skip the VM-exit information fields if they are read-only.
1326 		 */
1327 		if (!nested_cpu_has_vmwrite_any_field(&vmx->vcpu))
1328 			break;
1329 	}
1330 
1331 	vmcs_clear(shadow_vmcs);
1332 	vmcs_load(vmx->loaded_vmcs->vmcs);
1333 
1334 	preempt_enable();
1335 }
1336 
1337 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1338 {
1339 	const u16 *fields[] = {
1340 		shadow_read_write_fields,
1341 		shadow_read_only_fields
1342 	};
1343 	const int max_fields[] = {
1344 		max_shadow_read_write_fields,
1345 		max_shadow_read_only_fields
1346 	};
1347 	int i, q;
1348 	unsigned long field;
1349 	u64 field_value = 0;
1350 	struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1351 
1352 	vmcs_load(shadow_vmcs);
1353 
1354 	for (q = 0; q < ARRAY_SIZE(fields); q++) {
1355 		for (i = 0; i < max_fields[q]; i++) {
1356 			field = fields[q][i];
1357 			vmcs12_read_any(get_vmcs12(&vmx->vcpu), field, &field_value);
1358 			__vmcs_writel(field, field_value);
1359 		}
1360 	}
1361 
1362 	vmcs_clear(shadow_vmcs);
1363 	vmcs_load(vmx->loaded_vmcs->vmcs);
1364 }
1365 
1366 static int copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx)
1367 {
1368 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1369 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1370 
1371 	/* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1372 	vmcs12->tpr_threshold = evmcs->tpr_threshold;
1373 	vmcs12->guest_rip = evmcs->guest_rip;
1374 
1375 	if (unlikely(!(evmcs->hv_clean_fields &
1376 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1377 		vmcs12->guest_rsp = evmcs->guest_rsp;
1378 		vmcs12->guest_rflags = evmcs->guest_rflags;
1379 		vmcs12->guest_interruptibility_info =
1380 			evmcs->guest_interruptibility_info;
1381 	}
1382 
1383 	if (unlikely(!(evmcs->hv_clean_fields &
1384 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1385 		vmcs12->cpu_based_vm_exec_control =
1386 			evmcs->cpu_based_vm_exec_control;
1387 	}
1388 
1389 	if (unlikely(!(evmcs->hv_clean_fields &
1390 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1391 		vmcs12->exception_bitmap = evmcs->exception_bitmap;
1392 	}
1393 
1394 	if (unlikely(!(evmcs->hv_clean_fields &
1395 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1396 		vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1397 	}
1398 
1399 	if (unlikely(!(evmcs->hv_clean_fields &
1400 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1401 		vmcs12->vm_entry_intr_info_field =
1402 			evmcs->vm_entry_intr_info_field;
1403 		vmcs12->vm_entry_exception_error_code =
1404 			evmcs->vm_entry_exception_error_code;
1405 		vmcs12->vm_entry_instruction_len =
1406 			evmcs->vm_entry_instruction_len;
1407 	}
1408 
1409 	if (unlikely(!(evmcs->hv_clean_fields &
1410 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1411 		vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1412 		vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1413 		vmcs12->host_cr0 = evmcs->host_cr0;
1414 		vmcs12->host_cr3 = evmcs->host_cr3;
1415 		vmcs12->host_cr4 = evmcs->host_cr4;
1416 		vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1417 		vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1418 		vmcs12->host_rip = evmcs->host_rip;
1419 		vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1420 		vmcs12->host_es_selector = evmcs->host_es_selector;
1421 		vmcs12->host_cs_selector = evmcs->host_cs_selector;
1422 		vmcs12->host_ss_selector = evmcs->host_ss_selector;
1423 		vmcs12->host_ds_selector = evmcs->host_ds_selector;
1424 		vmcs12->host_fs_selector = evmcs->host_fs_selector;
1425 		vmcs12->host_gs_selector = evmcs->host_gs_selector;
1426 		vmcs12->host_tr_selector = evmcs->host_tr_selector;
1427 	}
1428 
1429 	if (unlikely(!(evmcs->hv_clean_fields &
1430 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1431 		vmcs12->pin_based_vm_exec_control =
1432 			evmcs->pin_based_vm_exec_control;
1433 		vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1434 		vmcs12->secondary_vm_exec_control =
1435 			evmcs->secondary_vm_exec_control;
1436 	}
1437 
1438 	if (unlikely(!(evmcs->hv_clean_fields &
1439 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1440 		vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1441 		vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1442 	}
1443 
1444 	if (unlikely(!(evmcs->hv_clean_fields &
1445 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1446 		vmcs12->msr_bitmap = evmcs->msr_bitmap;
1447 	}
1448 
1449 	if (unlikely(!(evmcs->hv_clean_fields &
1450 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1451 		vmcs12->guest_es_base = evmcs->guest_es_base;
1452 		vmcs12->guest_cs_base = evmcs->guest_cs_base;
1453 		vmcs12->guest_ss_base = evmcs->guest_ss_base;
1454 		vmcs12->guest_ds_base = evmcs->guest_ds_base;
1455 		vmcs12->guest_fs_base = evmcs->guest_fs_base;
1456 		vmcs12->guest_gs_base = evmcs->guest_gs_base;
1457 		vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1458 		vmcs12->guest_tr_base = evmcs->guest_tr_base;
1459 		vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1460 		vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1461 		vmcs12->guest_es_limit = evmcs->guest_es_limit;
1462 		vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1463 		vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1464 		vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1465 		vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1466 		vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1467 		vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1468 		vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1469 		vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1470 		vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1471 		vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1472 		vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1473 		vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1474 		vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1475 		vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1476 		vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1477 		vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1478 		vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1479 		vmcs12->guest_es_selector = evmcs->guest_es_selector;
1480 		vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1481 		vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1482 		vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1483 		vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1484 		vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1485 		vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1486 		vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1487 	}
1488 
1489 	if (unlikely(!(evmcs->hv_clean_fields &
1490 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1491 		vmcs12->tsc_offset = evmcs->tsc_offset;
1492 		vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1493 		vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1494 	}
1495 
1496 	if (unlikely(!(evmcs->hv_clean_fields &
1497 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1498 		vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1499 		vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1500 		vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1501 		vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1502 		vmcs12->guest_cr0 = evmcs->guest_cr0;
1503 		vmcs12->guest_cr3 = evmcs->guest_cr3;
1504 		vmcs12->guest_cr4 = evmcs->guest_cr4;
1505 		vmcs12->guest_dr7 = evmcs->guest_dr7;
1506 	}
1507 
1508 	if (unlikely(!(evmcs->hv_clean_fields &
1509 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1510 		vmcs12->host_fs_base = evmcs->host_fs_base;
1511 		vmcs12->host_gs_base = evmcs->host_gs_base;
1512 		vmcs12->host_tr_base = evmcs->host_tr_base;
1513 		vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1514 		vmcs12->host_idtr_base = evmcs->host_idtr_base;
1515 		vmcs12->host_rsp = evmcs->host_rsp;
1516 	}
1517 
1518 	if (unlikely(!(evmcs->hv_clean_fields &
1519 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1520 		vmcs12->ept_pointer = evmcs->ept_pointer;
1521 		vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1522 	}
1523 
1524 	if (unlikely(!(evmcs->hv_clean_fields &
1525 		       HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1526 		vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1527 		vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1528 		vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1529 		vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1530 		vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1531 		vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1532 		vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1533 		vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1534 		vmcs12->guest_pending_dbg_exceptions =
1535 			evmcs->guest_pending_dbg_exceptions;
1536 		vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1537 		vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1538 		vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1539 		vmcs12->guest_activity_state = evmcs->guest_activity_state;
1540 		vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1541 	}
1542 
1543 	/*
1544 	 * Not used?
1545 	 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1546 	 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1547 	 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1548 	 * vmcs12->cr3_target_value0 = evmcs->cr3_target_value0;
1549 	 * vmcs12->cr3_target_value1 = evmcs->cr3_target_value1;
1550 	 * vmcs12->cr3_target_value2 = evmcs->cr3_target_value2;
1551 	 * vmcs12->cr3_target_value3 = evmcs->cr3_target_value3;
1552 	 * vmcs12->page_fault_error_code_mask =
1553 	 *		evmcs->page_fault_error_code_mask;
1554 	 * vmcs12->page_fault_error_code_match =
1555 	 *		evmcs->page_fault_error_code_match;
1556 	 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1557 	 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1558 	 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1559 	 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1560 	 */
1561 
1562 	/*
1563 	 * Read only fields:
1564 	 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1565 	 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1566 	 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1567 	 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1568 	 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1569 	 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1570 	 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1571 	 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1572 	 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1573 	 * vmcs12->exit_qualification = evmcs->exit_qualification;
1574 	 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1575 	 *
1576 	 * Not present in struct vmcs12:
1577 	 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1578 	 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1579 	 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1580 	 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1581 	 */
1582 
1583 	return 0;
1584 }
1585 
1586 static int copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1587 {
1588 	struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1589 	struct hv_enlightened_vmcs *evmcs = vmx->nested.hv_evmcs;
1590 
1591 	/*
1592 	 * Should not be changed by KVM:
1593 	 *
1594 	 * evmcs->host_es_selector = vmcs12->host_es_selector;
1595 	 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1596 	 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1597 	 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1598 	 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1599 	 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1600 	 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1601 	 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1602 	 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1603 	 * evmcs->host_cr0 = vmcs12->host_cr0;
1604 	 * evmcs->host_cr3 = vmcs12->host_cr3;
1605 	 * evmcs->host_cr4 = vmcs12->host_cr4;
1606 	 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1607 	 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1608 	 * evmcs->host_rip = vmcs12->host_rip;
1609 	 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1610 	 * evmcs->host_fs_base = vmcs12->host_fs_base;
1611 	 * evmcs->host_gs_base = vmcs12->host_gs_base;
1612 	 * evmcs->host_tr_base = vmcs12->host_tr_base;
1613 	 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1614 	 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1615 	 * evmcs->host_rsp = vmcs12->host_rsp;
1616 	 * sync_vmcs12() doesn't read these:
1617 	 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1618 	 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1619 	 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1620 	 * evmcs->ept_pointer = vmcs12->ept_pointer;
1621 	 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1622 	 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1623 	 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1624 	 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1625 	 * evmcs->cr3_target_value0 = vmcs12->cr3_target_value0;
1626 	 * evmcs->cr3_target_value1 = vmcs12->cr3_target_value1;
1627 	 * evmcs->cr3_target_value2 = vmcs12->cr3_target_value2;
1628 	 * evmcs->cr3_target_value3 = vmcs12->cr3_target_value3;
1629 	 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1630 	 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1631 	 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1632 	 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1633 	 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1634 	 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1635 	 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1636 	 * evmcs->page_fault_error_code_mask =
1637 	 *		vmcs12->page_fault_error_code_mask;
1638 	 * evmcs->page_fault_error_code_match =
1639 	 *		vmcs12->page_fault_error_code_match;
1640 	 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1641 	 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1642 	 * evmcs->tsc_offset = vmcs12->tsc_offset;
1643 	 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1644 	 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1645 	 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1646 	 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1647 	 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1648 	 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1649 	 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1650 	 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1651 	 *
1652 	 * Not present in struct vmcs12:
1653 	 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1654 	 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1655 	 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1656 	 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1657 	 */
1658 
1659 	evmcs->guest_es_selector = vmcs12->guest_es_selector;
1660 	evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1661 	evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1662 	evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1663 	evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1664 	evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1665 	evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1666 	evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1667 
1668 	evmcs->guest_es_limit = vmcs12->guest_es_limit;
1669 	evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1670 	evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1671 	evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1672 	evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1673 	evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1674 	evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1675 	evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1676 	evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1677 	evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1678 
1679 	evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1680 	evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1681 	evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1682 	evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1683 	evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1684 	evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1685 	evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1686 	evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1687 
1688 	evmcs->guest_es_base = vmcs12->guest_es_base;
1689 	evmcs->guest_cs_base = vmcs12->guest_cs_base;
1690 	evmcs->guest_ss_base = vmcs12->guest_ss_base;
1691 	evmcs->guest_ds_base = vmcs12->guest_ds_base;
1692 	evmcs->guest_fs_base = vmcs12->guest_fs_base;
1693 	evmcs->guest_gs_base = vmcs12->guest_gs_base;
1694 	evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1695 	evmcs->guest_tr_base = vmcs12->guest_tr_base;
1696 	evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1697 	evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1698 
1699 	evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1700 	evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1701 
1702 	evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
1703 	evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
1704 	evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
1705 	evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
1706 
1707 	evmcs->guest_pending_dbg_exceptions =
1708 		vmcs12->guest_pending_dbg_exceptions;
1709 	evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
1710 	evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
1711 
1712 	evmcs->guest_activity_state = vmcs12->guest_activity_state;
1713 	evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
1714 
1715 	evmcs->guest_cr0 = vmcs12->guest_cr0;
1716 	evmcs->guest_cr3 = vmcs12->guest_cr3;
1717 	evmcs->guest_cr4 = vmcs12->guest_cr4;
1718 	evmcs->guest_dr7 = vmcs12->guest_dr7;
1719 
1720 	evmcs->guest_physical_address = vmcs12->guest_physical_address;
1721 
1722 	evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
1723 	evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
1724 	evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
1725 	evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
1726 	evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
1727 	evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
1728 	evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
1729 	evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
1730 
1731 	evmcs->exit_qualification = vmcs12->exit_qualification;
1732 
1733 	evmcs->guest_linear_address = vmcs12->guest_linear_address;
1734 	evmcs->guest_rsp = vmcs12->guest_rsp;
1735 	evmcs->guest_rflags = vmcs12->guest_rflags;
1736 
1737 	evmcs->guest_interruptibility_info =
1738 		vmcs12->guest_interruptibility_info;
1739 	evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
1740 	evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
1741 	evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
1742 	evmcs->vm_entry_exception_error_code =
1743 		vmcs12->vm_entry_exception_error_code;
1744 	evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
1745 
1746 	evmcs->guest_rip = vmcs12->guest_rip;
1747 
1748 	evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
1749 
1750 	return 0;
1751 }
1752 
1753 /*
1754  * This is an equivalent of the nested hypervisor executing the vmptrld
1755  * instruction.
1756  */
1757 static int nested_vmx_handle_enlightened_vmptrld(struct kvm_vcpu *vcpu,
1758 						 bool from_launch)
1759 {
1760 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1761 	struct hv_vp_assist_page assist_page;
1762 
1763 	if (likely(!vmx->nested.enlightened_vmcs_enabled))
1764 		return 1;
1765 
1766 	if (unlikely(!kvm_hv_get_assist_page(vcpu, &assist_page)))
1767 		return 1;
1768 
1769 	if (unlikely(!assist_page.enlighten_vmentry))
1770 		return 1;
1771 
1772 	if (unlikely(assist_page.current_nested_vmcs !=
1773 		     vmx->nested.hv_evmcs_vmptr)) {
1774 
1775 		if (!vmx->nested.hv_evmcs)
1776 			vmx->nested.current_vmptr = -1ull;
1777 
1778 		nested_release_evmcs(vcpu);
1779 
1780 		vmx->nested.hv_evmcs_page = kvm_vcpu_gpa_to_page(
1781 			vcpu, assist_page.current_nested_vmcs);
1782 
1783 		if (unlikely(is_error_page(vmx->nested.hv_evmcs_page)))
1784 			return 0;
1785 
1786 		vmx->nested.hv_evmcs = kmap(vmx->nested.hv_evmcs_page);
1787 
1788 		/*
1789 		 * Currently, KVM only supports eVMCS version 1
1790 		 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
1791 		 * value to first u32 field of eVMCS which should specify eVMCS
1792 		 * VersionNumber.
1793 		 *
1794 		 * Guest should be aware of supported eVMCS versions by host by
1795 		 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
1796 		 * expected to set this CPUID leaf according to the value
1797 		 * returned in vmcs_version from nested_enable_evmcs().
1798 		 *
1799 		 * However, it turns out that Microsoft Hyper-V fails to comply
1800 		 * to their own invented interface: When Hyper-V use eVMCS, it
1801 		 * just sets first u32 field of eVMCS to revision_id specified
1802 		 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
1803 		 * which is one of the supported versions specified in
1804 		 * CPUID.0x4000000A.EAX[0:15].
1805 		 *
1806 		 * To overcome Hyper-V bug, we accept here either a supported
1807 		 * eVMCS version or VMCS12 revision_id as valid values for first
1808 		 * u32 field of eVMCS.
1809 		 */
1810 		if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
1811 		    (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
1812 			nested_release_evmcs(vcpu);
1813 			return 0;
1814 		}
1815 
1816 		vmx->nested.dirty_vmcs12 = true;
1817 		/*
1818 		 * As we keep L2 state for one guest only 'hv_clean_fields' mask
1819 		 * can't be used when we switch between them. Reset it here for
1820 		 * simplicity.
1821 		 */
1822 		vmx->nested.hv_evmcs->hv_clean_fields &=
1823 			~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1824 		vmx->nested.hv_evmcs_vmptr = assist_page.current_nested_vmcs;
1825 
1826 		/*
1827 		 * Unlike normal vmcs12, enlightened vmcs12 is not fully
1828 		 * reloaded from guest's memory (read only fields, fields not
1829 		 * present in struct hv_enlightened_vmcs, ...). Make sure there
1830 		 * are no leftovers.
1831 		 */
1832 		if (from_launch) {
1833 			struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1834 			memset(vmcs12, 0, sizeof(*vmcs12));
1835 			vmcs12->hdr.revision_id = VMCS12_REVISION;
1836 		}
1837 
1838 	}
1839 	return 1;
1840 }
1841 
1842 void nested_sync_from_vmcs12(struct kvm_vcpu *vcpu)
1843 {
1844 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1845 
1846 	/*
1847 	 * hv_evmcs may end up being not mapped after migration (when
1848 	 * L2 was running), map it here to make sure vmcs12 changes are
1849 	 * properly reflected.
1850 	 */
1851 	if (vmx->nested.enlightened_vmcs_enabled && !vmx->nested.hv_evmcs)
1852 		nested_vmx_handle_enlightened_vmptrld(vcpu, false);
1853 
1854 	if (vmx->nested.hv_evmcs) {
1855 		copy_vmcs12_to_enlightened(vmx);
1856 		/* All fields are clean */
1857 		vmx->nested.hv_evmcs->hv_clean_fields |=
1858 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
1859 	} else {
1860 		copy_vmcs12_to_shadow(vmx);
1861 	}
1862 
1863 	vmx->nested.need_vmcs12_sync = false;
1864 }
1865 
1866 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
1867 {
1868 	struct vcpu_vmx *vmx =
1869 		container_of(timer, struct vcpu_vmx, nested.preemption_timer);
1870 
1871 	vmx->nested.preemption_timer_expired = true;
1872 	kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
1873 	kvm_vcpu_kick(&vmx->vcpu);
1874 
1875 	return HRTIMER_NORESTART;
1876 }
1877 
1878 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
1879 {
1880 	u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
1881 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1882 
1883 	/*
1884 	 * A timer value of zero is architecturally guaranteed to cause
1885 	 * a VMExit prior to executing any instructions in the guest.
1886 	 */
1887 	if (preemption_timeout == 0) {
1888 		vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
1889 		return;
1890 	}
1891 
1892 	if (vcpu->arch.virtual_tsc_khz == 0)
1893 		return;
1894 
1895 	preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
1896 	preemption_timeout *= 1000000;
1897 	do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
1898 	hrtimer_start(&vmx->nested.preemption_timer,
1899 		      ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
1900 }
1901 
1902 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1903 {
1904 	if (vmx->nested.nested_run_pending &&
1905 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
1906 		return vmcs12->guest_ia32_efer;
1907 	else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
1908 		return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
1909 	else
1910 		return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
1911 }
1912 
1913 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
1914 {
1915 	/*
1916 	 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
1917 	 * according to L0's settings (vmcs12 is irrelevant here).  Host
1918 	 * fields that come from L0 and are not constant, e.g. HOST_CR3,
1919 	 * will be set as needed prior to VMLAUNCH/VMRESUME.
1920 	 */
1921 	if (vmx->nested.vmcs02_initialized)
1922 		return;
1923 	vmx->nested.vmcs02_initialized = true;
1924 
1925 	/*
1926 	 * We don't care what the EPTP value is we just need to guarantee
1927 	 * it's valid so we don't get a false positive when doing early
1928 	 * consistency checks.
1929 	 */
1930 	if (enable_ept && nested_early_check)
1931 		vmcs_write64(EPT_POINTER, construct_eptp(&vmx->vcpu, 0));
1932 
1933 	/* All VMFUNCs are currently emulated through L0 vmexits.  */
1934 	if (cpu_has_vmx_vmfunc())
1935 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
1936 
1937 	if (cpu_has_vmx_posted_intr())
1938 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
1939 
1940 	if (cpu_has_vmx_msr_bitmap())
1941 		vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
1942 
1943 	if (enable_pml)
1944 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
1945 
1946 	/*
1947 	 * Set the MSR load/store lists to match L0's settings.  Only the
1948 	 * addresses are constant (for vmcs02), the counts can change based
1949 	 * on L2's behavior, e.g. switching to/from long mode.
1950 	 */
1951 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1952 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
1953 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
1954 
1955 	vmx_set_constant_host_state(vmx);
1956 }
1957 
1958 static void prepare_vmcs02_early_full(struct vcpu_vmx *vmx,
1959 				      struct vmcs12 *vmcs12)
1960 {
1961 	prepare_vmcs02_constant_state(vmx);
1962 
1963 	vmcs_write64(VMCS_LINK_POINTER, -1ull);
1964 
1965 	if (enable_vpid) {
1966 		if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
1967 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
1968 		else
1969 			vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
1970 	}
1971 }
1972 
1973 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
1974 {
1975 	u32 exec_control, vmcs12_exec_ctrl;
1976 	u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
1977 
1978 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs)
1979 		prepare_vmcs02_early_full(vmx, vmcs12);
1980 
1981 	/*
1982 	 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
1983 	 * entry, but only if the current (host) sp changed from the value
1984 	 * we wrote last (vmx->host_rsp).  This cache is no longer relevant
1985 	 * if we switch vmcs, and rather than hold a separate cache per vmcs,
1986 	 * here we just force the write to happen on entry.  host_rsp will
1987 	 * also be written unconditionally by nested_vmx_check_vmentry_hw()
1988 	 * if we are doing early consistency checks via hardware.
1989 	 */
1990 	vmx->host_rsp = 0;
1991 
1992 	/*
1993 	 * PIN CONTROLS
1994 	 */
1995 	exec_control = vmcs12->pin_based_vm_exec_control;
1996 
1997 	/* Preemption timer setting is computed directly in vmx_vcpu_run.  */
1998 	exec_control |= vmcs_config.pin_based_exec_ctrl;
1999 	exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2000 	vmx->loaded_vmcs->hv_timer_armed = false;
2001 
2002 	/* Posted interrupts setting is only taken from vmcs12.  */
2003 	if (nested_cpu_has_posted_intr(vmcs12)) {
2004 		vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2005 		vmx->nested.pi_pending = false;
2006 	} else {
2007 		exec_control &= ~PIN_BASED_POSTED_INTR;
2008 	}
2009 	vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
2010 
2011 	/*
2012 	 * EXEC CONTROLS
2013 	 */
2014 	exec_control = vmx_exec_control(vmx); /* L0's desires */
2015 	exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2016 	exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
2017 	exec_control &= ~CPU_BASED_TPR_SHADOW;
2018 	exec_control |= vmcs12->cpu_based_vm_exec_control;
2019 
2020 	/*
2021 	 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
2022 	 * nested_get_vmcs12_pages can't fix it up, the illegal value
2023 	 * will result in a VM entry failure.
2024 	 */
2025 	if (exec_control & CPU_BASED_TPR_SHADOW) {
2026 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
2027 		vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2028 	} else {
2029 #ifdef CONFIG_X86_64
2030 		exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2031 				CPU_BASED_CR8_STORE_EXITING;
2032 #endif
2033 	}
2034 
2035 	/*
2036 	 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2037 	 * for I/O port accesses.
2038 	 */
2039 	exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2040 	exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2041 	vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2042 
2043 	/*
2044 	 * SECONDARY EXEC CONTROLS
2045 	 */
2046 	if (cpu_has_secondary_exec_ctrls()) {
2047 		exec_control = vmx->secondary_exec_control;
2048 
2049 		/* Take the following fields only from vmcs12 */
2050 		exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2051 				  SECONDARY_EXEC_ENABLE_INVPCID |
2052 				  SECONDARY_EXEC_RDTSCP |
2053 				  SECONDARY_EXEC_XSAVES |
2054 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2055 				  SECONDARY_EXEC_APIC_REGISTER_VIRT |
2056 				  SECONDARY_EXEC_ENABLE_VMFUNC);
2057 		if (nested_cpu_has(vmcs12,
2058 				   CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
2059 			vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
2060 				~SECONDARY_EXEC_ENABLE_PML;
2061 			exec_control |= vmcs12_exec_ctrl;
2062 		}
2063 
2064 		/* VMCS shadowing for L2 is emulated for now */
2065 		exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2066 
2067 		if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2068 			vmcs_write16(GUEST_INTR_STATUS,
2069 				vmcs12->guest_intr_status);
2070 
2071 		/*
2072 		 * Write an illegal value to APIC_ACCESS_ADDR. Later,
2073 		 * nested_get_vmcs12_pages will either fix it up or
2074 		 * remove the VM execution control.
2075 		 */
2076 		if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
2077 			vmcs_write64(APIC_ACCESS_ADDR, -1ull);
2078 
2079 		if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2080 			vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
2081 
2082 		vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2083 	}
2084 
2085 	/*
2086 	 * ENTRY CONTROLS
2087 	 *
2088 	 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2089 	 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2090 	 * on the related bits (if supported by the CPU) in the hope that
2091 	 * we can avoid VMWrites during vmx_set_efer().
2092 	 */
2093 	exec_control = (vmcs12->vm_entry_controls | vmx_vmentry_ctrl()) &
2094 			~VM_ENTRY_IA32E_MODE & ~VM_ENTRY_LOAD_IA32_EFER;
2095 	if (cpu_has_load_ia32_efer()) {
2096 		if (guest_efer & EFER_LMA)
2097 			exec_control |= VM_ENTRY_IA32E_MODE;
2098 		if (guest_efer != host_efer)
2099 			exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2100 	}
2101 	vm_entry_controls_init(vmx, exec_control);
2102 
2103 	/*
2104 	 * EXIT CONTROLS
2105 	 *
2106 	 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2107 	 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2108 	 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2109 	 */
2110 	exec_control = vmx_vmexit_ctrl();
2111 	if (cpu_has_load_ia32_efer() && guest_efer != host_efer)
2112 		exec_control |= VM_EXIT_LOAD_IA32_EFER;
2113 	vm_exit_controls_init(vmx, exec_control);
2114 
2115 	/*
2116 	 * Conceptually we want to copy the PML address and index from
2117 	 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
2118 	 * since we always flush the log on each vmexit and never change
2119 	 * the PML address (once set), this happens to be equivalent to
2120 	 * simply resetting the index in vmcs02.
2121 	 */
2122 	if (enable_pml)
2123 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
2124 
2125 	/*
2126 	 * Interrupt/Exception Fields
2127 	 */
2128 	if (vmx->nested.nested_run_pending) {
2129 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2130 			     vmcs12->vm_entry_intr_info_field);
2131 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2132 			     vmcs12->vm_entry_exception_error_code);
2133 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2134 			     vmcs12->vm_entry_instruction_len);
2135 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2136 			     vmcs12->guest_interruptibility_info);
2137 		vmx->loaded_vmcs->nmi_known_unmasked =
2138 			!(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2139 	} else {
2140 		vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2141 	}
2142 }
2143 
2144 static void prepare_vmcs02_full(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2145 {
2146 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2147 
2148 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2149 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2150 		vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2151 		vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2152 		vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2153 		vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2154 		vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2155 		vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2156 		vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2157 		vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2158 		vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2159 		vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2160 		vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2161 		vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2162 		vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2163 		vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2164 		vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2165 		vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2166 		vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2167 		vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2168 		vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2169 		vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2170 		vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2171 		vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2172 		vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2173 		vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2174 		vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2175 		vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2176 		vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2177 		vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2178 		vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2179 		vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2180 		vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2181 		vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2182 		vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2183 		vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2184 	}
2185 
2186 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2187 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2188 		vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2189 		vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2190 			    vmcs12->guest_pending_dbg_exceptions);
2191 		vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2192 		vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2193 
2194 		/*
2195 		 * L1 may access the L2's PDPTR, so save them to construct
2196 		 * vmcs12
2197 		 */
2198 		if (enable_ept) {
2199 			vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2200 			vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2201 			vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2202 			vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2203 		}
2204 	}
2205 
2206 	if (nested_cpu_has_xsaves(vmcs12))
2207 		vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2208 
2209 	/*
2210 	 * Whether page-faults are trapped is determined by a combination of
2211 	 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
2212 	 * If enable_ept, L0 doesn't care about page faults and we should
2213 	 * set all of these to L1's desires. However, if !enable_ept, L0 does
2214 	 * care about (at least some) page faults, and because it is not easy
2215 	 * (if at all possible?) to merge L0 and L1's desires, we simply ask
2216 	 * to exit on each and every L2 page fault. This is done by setting
2217 	 * MASK=MATCH=0 and (see below) EB.PF=1.
2218 	 * Note that below we don't need special code to set EB.PF beyond the
2219 	 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2220 	 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2221 	 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2222 	 */
2223 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
2224 		enable_ept ? vmcs12->page_fault_error_code_mask : 0);
2225 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
2226 		enable_ept ? vmcs12->page_fault_error_code_match : 0);
2227 
2228 	if (cpu_has_vmx_apicv()) {
2229 		vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2230 		vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2231 		vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2232 		vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2233 	}
2234 
2235 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2236 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2237 
2238 	set_cr4_guest_host_mask(vmx);
2239 
2240 	if (kvm_mpx_supported()) {
2241 		if (vmx->nested.nested_run_pending &&
2242 			(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2243 			vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2244 		else
2245 			vmcs_write64(GUEST_BNDCFGS, vmx->nested.vmcs01_guest_bndcfgs);
2246 	}
2247 }
2248 
2249 /*
2250  * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2251  * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2252  * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2253  * guest in a way that will both be appropriate to L1's requests, and our
2254  * needs. In addition to modifying the active vmcs (which is vmcs02), this
2255  * function also has additional necessary side-effects, like setting various
2256  * vcpu->arch fields.
2257  * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2258  * is assigned to entry_failure_code on failure.
2259  */
2260 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2261 			  u32 *entry_failure_code)
2262 {
2263 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2264 	struct hv_enlightened_vmcs *hv_evmcs = vmx->nested.hv_evmcs;
2265 
2266 	if (vmx->nested.dirty_vmcs12 || vmx->nested.hv_evmcs) {
2267 		prepare_vmcs02_full(vmx, vmcs12);
2268 		vmx->nested.dirty_vmcs12 = false;
2269 	}
2270 
2271 	/*
2272 	 * First, the fields that are shadowed.  This must be kept in sync
2273 	 * with vmcs_shadow_fields.h.
2274 	 */
2275 	if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2276 			   HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2277 		vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2278 		vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2279 	}
2280 
2281 	if (vmx->nested.nested_run_pending &&
2282 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2283 		kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2284 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2285 	} else {
2286 		kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2287 		vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
2288 	}
2289 	vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2290 
2291 	vmx->nested.preemption_timer_expired = false;
2292 	if (nested_cpu_has_preemption_timer(vmcs12))
2293 		vmx_start_preemption_timer(vcpu);
2294 
2295 	/* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2296 	 * bitwise-or of what L1 wants to trap for L2, and what we want to
2297 	 * trap. Note that CR0.TS also needs updating - we do this later.
2298 	 */
2299 	update_exception_bitmap(vcpu);
2300 	vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2301 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2302 
2303 	if (vmx->nested.nested_run_pending &&
2304 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2305 		vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2306 		vcpu->arch.pat = vmcs12->guest_ia32_pat;
2307 	} else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2308 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2309 	}
2310 
2311 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2312 
2313 	if (kvm_has_tsc_control)
2314 		decache_tsc_multiplier(vmx);
2315 
2316 	if (enable_vpid) {
2317 		/*
2318 		 * There is no direct mapping between vpid02 and vpid12, the
2319 		 * vpid02 is per-vCPU for L0 and reused while the value of
2320 		 * vpid12 is changed w/ one invvpid during nested vmentry.
2321 		 * The vpid12 is allocated by L1 for L2, so it will not
2322 		 * influence global bitmap(for vpid01 and vpid02 allocation)
2323 		 * even if spawn a lot of nested vCPUs.
2324 		 */
2325 		if (nested_cpu_has_vpid(vmcs12) && nested_has_guest_tlb_tag(vcpu)) {
2326 			if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
2327 				vmx->nested.last_vpid = vmcs12->virtual_processor_id;
2328 				__vmx_flush_tlb(vcpu, nested_get_vpid02(vcpu), false);
2329 			}
2330 		} else {
2331 			/*
2332 			 * If L1 use EPT, then L0 needs to execute INVEPT on
2333 			 * EPTP02 instead of EPTP01. Therefore, delay TLB
2334 			 * flush until vmcs02->eptp is fully updated by
2335 			 * KVM_REQ_LOAD_CR3. Note that this assumes
2336 			 * KVM_REQ_TLB_FLUSH is evaluated after
2337 			 * KVM_REQ_LOAD_CR3 in vcpu_enter_guest().
2338 			 */
2339 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2340 		}
2341 	}
2342 
2343 	if (nested_cpu_has_ept(vmcs12))
2344 		nested_ept_init_mmu_context(vcpu);
2345 	else if (nested_cpu_has2(vmcs12,
2346 				 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2347 		vmx_flush_tlb(vcpu, true);
2348 
2349 	/*
2350 	 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
2351 	 * bits which we consider mandatory enabled.
2352 	 * The CR0_READ_SHADOW is what L2 should have expected to read given
2353 	 * the specifications by L1; It's not enough to take
2354 	 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
2355 	 * have more bits than L1 expected.
2356 	 */
2357 	vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2358 	vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2359 
2360 	vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2361 	vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2362 
2363 	vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2364 	/* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2365 	vmx_set_efer(vcpu, vcpu->arch.efer);
2366 
2367 	/*
2368 	 * Guest state is invalid and unrestricted guest is disabled,
2369 	 * which means L1 attempted VMEntry to L2 with invalid state.
2370 	 * Fail the VMEntry.
2371 	 */
2372 	if (vmx->emulation_required) {
2373 		*entry_failure_code = ENTRY_FAIL_DEFAULT;
2374 		return 1;
2375 	}
2376 
2377 	/* Shadow page tables on either EPT or shadow page tables. */
2378 	if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2379 				entry_failure_code))
2380 		return 1;
2381 
2382 	if (!enable_ept)
2383 		vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
2384 
2385 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
2386 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
2387 	return 0;
2388 }
2389 
2390 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2391 {
2392 	if (!nested_cpu_has_nmi_exiting(vmcs12) &&
2393 	    nested_cpu_has_virtual_nmis(vmcs12))
2394 		return -EINVAL;
2395 
2396 	if (!nested_cpu_has_virtual_nmis(vmcs12) &&
2397 	    nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING))
2398 		return -EINVAL;
2399 
2400 	return 0;
2401 }
2402 
2403 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
2404 {
2405 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2406 	int maxphyaddr = cpuid_maxphyaddr(vcpu);
2407 
2408 	/* Check for memory type validity */
2409 	switch (address & VMX_EPTP_MT_MASK) {
2410 	case VMX_EPTP_MT_UC:
2411 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT))
2412 			return false;
2413 		break;
2414 	case VMX_EPTP_MT_WB:
2415 		if (!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT))
2416 			return false;
2417 		break;
2418 	default:
2419 		return false;
2420 	}
2421 
2422 	/* only 4 levels page-walk length are valid */
2423 	if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
2424 		return false;
2425 
2426 	/* Reserved bits should not be set */
2427 	if (address >> maxphyaddr || ((address >> 7) & 0x1f))
2428 		return false;
2429 
2430 	/* AD, if set, should be supported */
2431 	if (address & VMX_EPTP_AD_ENABLE_BIT) {
2432 		if (!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT))
2433 			return false;
2434 	}
2435 
2436 	return true;
2437 }
2438 
2439 /*
2440  * Checks related to VM-Execution Control Fields
2441  */
2442 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2443                                               struct vmcs12 *vmcs12)
2444 {
2445 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2446 
2447 	if (!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2448 				vmx->nested.msrs.pinbased_ctls_low,
2449 				vmx->nested.msrs.pinbased_ctls_high) ||
2450 	    !vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2451 				vmx->nested.msrs.procbased_ctls_low,
2452 				vmx->nested.msrs.procbased_ctls_high))
2453 		return -EINVAL;
2454 
2455 	if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2456 	    !vmx_control_verify(vmcs12->secondary_vm_exec_control,
2457 				 vmx->nested.msrs.secondary_ctls_low,
2458 				 vmx->nested.msrs.secondary_ctls_high))
2459 		return -EINVAL;
2460 
2461 	if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu) ||
2462 	    nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2463 	    nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2464 	    nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2465 	    nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2466 	    nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2467 	    nested_vmx_check_nmi_controls(vmcs12) ||
2468 	    nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2469 	    nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2470 	    nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2471 	    nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2472 	    (nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2473 		return -EINVAL;
2474 
2475 	if (nested_cpu_has_ept(vmcs12) &&
2476 	    !valid_ept_address(vcpu, vmcs12->ept_pointer))
2477 		return -EINVAL;
2478 
2479 	if (nested_cpu_has_vmfunc(vmcs12)) {
2480 		if (vmcs12->vm_function_control &
2481 		    ~vmx->nested.msrs.vmfunc_controls)
2482 			return -EINVAL;
2483 
2484 		if (nested_cpu_has_eptp_switching(vmcs12)) {
2485 			if (!nested_cpu_has_ept(vmcs12) ||
2486 			    !page_address_valid(vcpu, vmcs12->eptp_list_address))
2487 				return -EINVAL;
2488 		}
2489 	}
2490 
2491 	return 0;
2492 }
2493 
2494 /*
2495  * Checks related to VM-Exit Control Fields
2496  */
2497 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2498                                          struct vmcs12 *vmcs12)
2499 {
2500 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2501 
2502 	if (!vmx_control_verify(vmcs12->vm_exit_controls,
2503 				vmx->nested.msrs.exit_ctls_low,
2504 				vmx->nested.msrs.exit_ctls_high) ||
2505 	    nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12))
2506 		return -EINVAL;
2507 
2508 	return 0;
2509 }
2510 
2511 /*
2512  * Checks related to VM-Entry Control Fields
2513  */
2514 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2515 					  struct vmcs12 *vmcs12)
2516 {
2517 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2518 
2519 	if (!vmx_control_verify(vmcs12->vm_entry_controls,
2520 				vmx->nested.msrs.entry_ctls_low,
2521 				vmx->nested.msrs.entry_ctls_high))
2522 		return -EINVAL;
2523 
2524 	/*
2525 	 * From the Intel SDM, volume 3:
2526 	 * Fields relevant to VM-entry event injection must be set properly.
2527 	 * These fields are the VM-entry interruption-information field, the
2528 	 * VM-entry exception error code, and the VM-entry instruction length.
2529 	 */
2530 	if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2531 		u32 intr_info = vmcs12->vm_entry_intr_info_field;
2532 		u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2533 		u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2534 		bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2535 		bool should_have_error_code;
2536 		bool urg = nested_cpu_has2(vmcs12,
2537 					   SECONDARY_EXEC_UNRESTRICTED_GUEST);
2538 		bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2539 
2540 		/* VM-entry interruption-info field: interruption type */
2541 		if (intr_type == INTR_TYPE_RESERVED ||
2542 		    (intr_type == INTR_TYPE_OTHER_EVENT &&
2543 		     !nested_cpu_supports_monitor_trap_flag(vcpu)))
2544 			return -EINVAL;
2545 
2546 		/* VM-entry interruption-info field: vector */
2547 		if ((intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2548 		    (intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2549 		    (intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2550 			return -EINVAL;
2551 
2552 		/* VM-entry interruption-info field: deliver error code */
2553 		should_have_error_code =
2554 			intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2555 			x86_exception_has_error_code(vector);
2556 		if (has_error_code != should_have_error_code)
2557 			return -EINVAL;
2558 
2559 		/* VM-entry exception error code */
2560 		if (has_error_code &&
2561 		    vmcs12->vm_entry_exception_error_code & GENMASK(31, 15))
2562 			return -EINVAL;
2563 
2564 		/* VM-entry interruption-info field: reserved bits */
2565 		if (intr_info & INTR_INFO_RESVD_BITS_MASK)
2566 			return -EINVAL;
2567 
2568 		/* VM-entry instruction length */
2569 		switch (intr_type) {
2570 		case INTR_TYPE_SOFT_EXCEPTION:
2571 		case INTR_TYPE_SOFT_INTR:
2572 		case INTR_TYPE_PRIV_SW_EXCEPTION:
2573 			if ((vmcs12->vm_entry_instruction_len > 15) ||
2574 			    (vmcs12->vm_entry_instruction_len == 0 &&
2575 			     !nested_cpu_has_zero_length_injection(vcpu)))
2576 				return -EINVAL;
2577 		}
2578 	}
2579 
2580 	if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2581 		return -EINVAL;
2582 
2583 	return 0;
2584 }
2585 
2586 /*
2587  * Checks related to Host Control Registers and MSRs
2588  */
2589 static int nested_check_host_control_regs(struct kvm_vcpu *vcpu,
2590                                           struct vmcs12 *vmcs12)
2591 {
2592 	bool ia32e;
2593 
2594 	if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
2595 	    !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
2596 	    !nested_cr3_valid(vcpu, vmcs12->host_cr3))
2597 		return -EINVAL;
2598 	/*
2599 	 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
2600 	 * IA32_EFER MSR must be 0 in the field for that register. In addition,
2601 	 * the values of the LMA and LME bits in the field must each be that of
2602 	 * the host address-space size VM-exit control.
2603 	 */
2604 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
2605 		ia32e = (vmcs12->vm_exit_controls &
2606 			 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
2607 		if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
2608 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
2609 		    ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
2610 			return -EINVAL;
2611 	}
2612 
2613 	return 0;
2614 }
2615 
2616 /*
2617  * Checks related to Guest Non-register State
2618  */
2619 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
2620 {
2621 	if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
2622 	    vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
2623 		return -EINVAL;
2624 
2625 	return 0;
2626 }
2627 
2628 static int nested_vmx_check_vmentry_prereqs(struct kvm_vcpu *vcpu,
2629 					    struct vmcs12 *vmcs12)
2630 {
2631 	if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2632 	    nested_check_vm_exit_controls(vcpu, vmcs12) ||
2633 	    nested_check_vm_entry_controls(vcpu, vmcs12))
2634 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2635 
2636 	if (nested_check_host_control_regs(vcpu, vmcs12))
2637 		return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
2638 
2639 	if (nested_check_guest_non_reg_state(vmcs12))
2640 		return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
2641 
2642 	return 0;
2643 }
2644 
2645 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
2646 					  struct vmcs12 *vmcs12)
2647 {
2648 	int r;
2649 	struct page *page;
2650 	struct vmcs12 *shadow;
2651 
2652 	if (vmcs12->vmcs_link_pointer == -1ull)
2653 		return 0;
2654 
2655 	if (!page_address_valid(vcpu, vmcs12->vmcs_link_pointer))
2656 		return -EINVAL;
2657 
2658 	page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->vmcs_link_pointer);
2659 	if (is_error_page(page))
2660 		return -EINVAL;
2661 
2662 	r = 0;
2663 	shadow = kmap(page);
2664 	if (shadow->hdr.revision_id != VMCS12_REVISION ||
2665 	    shadow->hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12))
2666 		r = -EINVAL;
2667 	kunmap(page);
2668 	kvm_release_page_clean(page);
2669 	return r;
2670 }
2671 
2672 static int nested_vmx_check_vmentry_postreqs(struct kvm_vcpu *vcpu,
2673 					     struct vmcs12 *vmcs12,
2674 					     u32 *exit_qual)
2675 {
2676 	bool ia32e;
2677 
2678 	*exit_qual = ENTRY_FAIL_DEFAULT;
2679 
2680 	if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
2681 	    !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
2682 		return 1;
2683 
2684 	if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
2685 		*exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
2686 		return 1;
2687 	}
2688 
2689 	/*
2690 	 * If the load IA32_EFER VM-entry control is 1, the following checks
2691 	 * are performed on the field for the IA32_EFER MSR:
2692 	 * - Bits reserved in the IA32_EFER MSR must be 0.
2693 	 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
2694 	 *   the IA-32e mode guest VM-exit control. It must also be identical
2695 	 *   to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
2696 	 *   CR0.PG) is 1.
2697 	 */
2698 	if (to_vmx(vcpu)->nested.nested_run_pending &&
2699 	    (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
2700 		ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
2701 		if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
2702 		    ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
2703 		    ((vmcs12->guest_cr0 & X86_CR0_PG) &&
2704 		     ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
2705 			return 1;
2706 	}
2707 
2708 	if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
2709 		(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
2710 		(vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
2711 			return 1;
2712 
2713 	return 0;
2714 }
2715 
2716 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
2717 {
2718 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2719 	unsigned long cr3, cr4;
2720 
2721 	if (!nested_early_check)
2722 		return 0;
2723 
2724 	if (vmx->msr_autoload.host.nr)
2725 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2726 	if (vmx->msr_autoload.guest.nr)
2727 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2728 
2729 	preempt_disable();
2730 
2731 	vmx_prepare_switch_to_guest(vcpu);
2732 
2733 	/*
2734 	 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
2735 	 * which is reserved to '1' by hardware.  GUEST_RFLAGS is guaranteed to
2736 	 * be written (by preparve_vmcs02()) before the "real" VMEnter, i.e.
2737 	 * there is no need to preserve other bits or save/restore the field.
2738 	 */
2739 	vmcs_writel(GUEST_RFLAGS, 0);
2740 
2741 	cr3 = __get_current_cr3_fast();
2742 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
2743 		vmcs_writel(HOST_CR3, cr3);
2744 		vmx->loaded_vmcs->host_state.cr3 = cr3;
2745 	}
2746 
2747 	cr4 = cr4_read_shadow();
2748 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
2749 		vmcs_writel(HOST_CR4, cr4);
2750 		vmx->loaded_vmcs->host_state.cr4 = cr4;
2751 	}
2752 
2753 	vmx->__launched = vmx->loaded_vmcs->launched;
2754 
2755 	asm(
2756 		/* Set HOST_RSP */
2757 		"sub $%c[wordsize], %%" _ASM_SP "\n\t" /* temporarily adjust RSP for CALL */
2758 		__ex("vmwrite %%" _ASM_SP ", %%" _ASM_DX) "\n\t"
2759 		"mov %%" _ASM_SP ", %c[host_rsp](%1)\n\t"
2760 		"add $%c[wordsize], %%" _ASM_SP "\n\t" /* un-adjust RSP */
2761 
2762 		/* Check if vmlaunch or vmresume is needed */
2763 		"cmpl $0, %c[launched](%% " _ASM_CX")\n\t"
2764 
2765 		"call vmx_vmenter\n\t"
2766 
2767 		/* Set vmx->fail accordingly */
2768 		"setbe %c[fail](%% " _ASM_CX")\n\t"
2769 	      : ASM_CALL_CONSTRAINT
2770 	      : "c"(vmx), "d"((unsigned long)HOST_RSP),
2771 		[launched]"i"(offsetof(struct vcpu_vmx, __launched)),
2772 		[fail]"i"(offsetof(struct vcpu_vmx, fail)),
2773 		[host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
2774 		[wordsize]"i"(sizeof(ulong))
2775 	      : "rax", "cc", "memory"
2776 	);
2777 
2778 	preempt_enable();
2779 
2780 	if (vmx->msr_autoload.host.nr)
2781 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2782 	if (vmx->msr_autoload.guest.nr)
2783 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2784 
2785 	if (vmx->fail) {
2786 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
2787 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
2788 		vmx->fail = 0;
2789 		return 1;
2790 	}
2791 
2792 	/*
2793 	 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
2794 	 */
2795 	local_irq_enable();
2796 	if (hw_breakpoint_active())
2797 		set_debugreg(__this_cpu_read(cpu_dr7), 7);
2798 
2799 	/*
2800 	 * A non-failing VMEntry means we somehow entered guest mode with
2801 	 * an illegal RIP, and that's just the tip of the iceberg.  There
2802 	 * is no telling what memory has been modified or what state has
2803 	 * been exposed to unknown code.  Hitting this all but guarantees
2804 	 * a (very critical) hardware issue.
2805 	 */
2806 	WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
2807 		VMX_EXIT_REASONS_FAILED_VMENTRY));
2808 
2809 	return 0;
2810 }
2811 STACK_FRAME_NON_STANDARD(nested_vmx_check_vmentry_hw);
2812 
2813 
2814 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
2815 						 struct vmcs12 *vmcs12);
2816 
2817 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
2818 {
2819 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2820 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2821 	struct page *page;
2822 	u64 hpa;
2823 
2824 	if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
2825 		/*
2826 		 * Translate L1 physical address to host physical
2827 		 * address for vmcs02. Keep the page pinned, so this
2828 		 * physical address remains valid. We keep a reference
2829 		 * to it so we can release it later.
2830 		 */
2831 		if (vmx->nested.apic_access_page) { /* shouldn't happen */
2832 			kvm_release_page_dirty(vmx->nested.apic_access_page);
2833 			vmx->nested.apic_access_page = NULL;
2834 		}
2835 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
2836 		/*
2837 		 * If translation failed, no matter: This feature asks
2838 		 * to exit when accessing the given address, and if it
2839 		 * can never be accessed, this feature won't do
2840 		 * anything anyway.
2841 		 */
2842 		if (!is_error_page(page)) {
2843 			vmx->nested.apic_access_page = page;
2844 			hpa = page_to_phys(vmx->nested.apic_access_page);
2845 			vmcs_write64(APIC_ACCESS_ADDR, hpa);
2846 		} else {
2847 			vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
2848 					SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
2849 		}
2850 	}
2851 
2852 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
2853 		if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
2854 			kvm_release_page_dirty(vmx->nested.virtual_apic_page);
2855 			vmx->nested.virtual_apic_page = NULL;
2856 		}
2857 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
2858 
2859 		/*
2860 		 * If translation failed, VM entry will fail because
2861 		 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
2862 		 * Failing the vm entry is _not_ what the processor
2863 		 * does but it's basically the only possibility we
2864 		 * have.  We could still enter the guest if CR8 load
2865 		 * exits are enabled, CR8 store exits are enabled, and
2866 		 * virtualize APIC access is disabled; in this case
2867 		 * the processor would never use the TPR shadow and we
2868 		 * could simply clear the bit from the execution
2869 		 * control.  But such a configuration is useless, so
2870 		 * let's keep the code simple.
2871 		 */
2872 		if (!is_error_page(page)) {
2873 			vmx->nested.virtual_apic_page = page;
2874 			hpa = page_to_phys(vmx->nested.virtual_apic_page);
2875 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
2876 		}
2877 	}
2878 
2879 	if (nested_cpu_has_posted_intr(vmcs12)) {
2880 		if (vmx->nested.pi_desc_page) { /* shouldn't happen */
2881 			kunmap(vmx->nested.pi_desc_page);
2882 			kvm_release_page_dirty(vmx->nested.pi_desc_page);
2883 			vmx->nested.pi_desc_page = NULL;
2884 			vmx->nested.pi_desc = NULL;
2885 			vmcs_write64(POSTED_INTR_DESC_ADDR, -1ull);
2886 		}
2887 		page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
2888 		if (is_error_page(page))
2889 			return;
2890 		vmx->nested.pi_desc_page = page;
2891 		vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
2892 		vmx->nested.pi_desc =
2893 			(struct pi_desc *)((void *)vmx->nested.pi_desc +
2894 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2895 			(PAGE_SIZE - 1)));
2896 		vmcs_write64(POSTED_INTR_DESC_ADDR,
2897 			page_to_phys(vmx->nested.pi_desc_page) +
2898 			(unsigned long)(vmcs12->posted_intr_desc_addr &
2899 			(PAGE_SIZE - 1)));
2900 	}
2901 	if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
2902 		vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
2903 			      CPU_BASED_USE_MSR_BITMAPS);
2904 	else
2905 		vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
2906 				CPU_BASED_USE_MSR_BITMAPS);
2907 }
2908 
2909 /*
2910  * Intel's VMX Instruction Reference specifies a common set of prerequisites
2911  * for running VMX instructions (except VMXON, whose prerequisites are
2912  * slightly different). It also specifies what exception to inject otherwise.
2913  * Note that many of these exceptions have priority over VM exits, so they
2914  * don't have to be checked again here.
2915  */
2916 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
2917 {
2918 	if (!to_vmx(vcpu)->nested.vmxon) {
2919 		kvm_queue_exception(vcpu, UD_VECTOR);
2920 		return 0;
2921 	}
2922 
2923 	if (vmx_get_cpl(vcpu)) {
2924 		kvm_inject_gp(vcpu, 0);
2925 		return 0;
2926 	}
2927 
2928 	return 1;
2929 }
2930 
2931 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
2932 {
2933 	u8 rvi = vmx_get_rvi();
2934 	u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
2935 
2936 	return ((rvi & 0xf0) > (vppr & 0xf0));
2937 }
2938 
2939 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
2940 				   struct vmcs12 *vmcs12);
2941 
2942 /*
2943  * If from_vmentry is false, this is being called from state restore (either RSM
2944  * or KVM_SET_NESTED_STATE).  Otherwise it's called from vmlaunch/vmresume.
2945 + *
2946 + * Returns:
2947 + *   0 - success, i.e. proceed with actual VMEnter
2948 + *   1 - consistency check VMExit
2949 + *  -1 - consistency check VMFail
2950  */
2951 int nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
2952 {
2953 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2954 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2955 	bool evaluate_pending_interrupts;
2956 	u32 exit_reason = EXIT_REASON_INVALID_STATE;
2957 	u32 exit_qual;
2958 
2959 	evaluate_pending_interrupts = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2960 		(CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_VIRTUAL_NMI_PENDING);
2961 	if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
2962 		evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
2963 
2964 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
2965 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
2966 	if (kvm_mpx_supported() &&
2967 		!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2968 		vmx->nested.vmcs01_guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
2969 
2970 	vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
2971 
2972 	prepare_vmcs02_early(vmx, vmcs12);
2973 
2974 	if (from_vmentry) {
2975 		nested_get_vmcs12_pages(vcpu);
2976 
2977 		if (nested_vmx_check_vmentry_hw(vcpu)) {
2978 			vmx_switch_vmcs(vcpu, &vmx->vmcs01);
2979 			return -1;
2980 		}
2981 
2982 		if (nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
2983 			goto vmentry_fail_vmexit;
2984 	}
2985 
2986 	enter_guest_mode(vcpu);
2987 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
2988 		vcpu->arch.tsc_offset += vmcs12->tsc_offset;
2989 
2990 	if (prepare_vmcs02(vcpu, vmcs12, &exit_qual))
2991 		goto vmentry_fail_vmexit_guest_mode;
2992 
2993 	if (from_vmentry) {
2994 		exit_reason = EXIT_REASON_MSR_LOAD_FAIL;
2995 		exit_qual = nested_vmx_load_msr(vcpu,
2996 						vmcs12->vm_entry_msr_load_addr,
2997 						vmcs12->vm_entry_msr_load_count);
2998 		if (exit_qual)
2999 			goto vmentry_fail_vmexit_guest_mode;
3000 	} else {
3001 		/*
3002 		 * The MMU is not initialized to point at the right entities yet and
3003 		 * "get pages" would need to read data from the guest (i.e. we will
3004 		 * need to perform gpa to hpa translation). Request a call
3005 		 * to nested_get_vmcs12_pages before the next VM-entry.  The MSRs
3006 		 * have already been set at vmentry time and should not be reset.
3007 		 */
3008 		kvm_make_request(KVM_REQ_GET_VMCS12_PAGES, vcpu);
3009 	}
3010 
3011 	/*
3012 	 * If L1 had a pending IRQ/NMI until it executed
3013 	 * VMLAUNCH/VMRESUME which wasn't delivered because it was
3014 	 * disallowed (e.g. interrupts disabled), L0 needs to
3015 	 * evaluate if this pending event should cause an exit from L2
3016 	 * to L1 or delivered directly to L2 (e.g. In case L1 don't
3017 	 * intercept EXTERNAL_INTERRUPT).
3018 	 *
3019 	 * Usually this would be handled by the processor noticing an
3020 	 * IRQ/NMI window request, or checking RVI during evaluation of
3021 	 * pending virtual interrupts.  However, this setting was done
3022 	 * on VMCS01 and now VMCS02 is active instead. Thus, we force L0
3023 	 * to perform pending event evaluation by requesting a KVM_REQ_EVENT.
3024 	 */
3025 	if (unlikely(evaluate_pending_interrupts))
3026 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3027 
3028 	/*
3029 	 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3030 	 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3031 	 * returned as far as L1 is concerned. It will only return (and set
3032 	 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3033 	 */
3034 	return 0;
3035 
3036 	/*
3037 	 * A failed consistency check that leads to a VMExit during L1's
3038 	 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3039 	 * 26.7 "VM-entry failures during or after loading guest state".
3040 	 */
3041 vmentry_fail_vmexit_guest_mode:
3042 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3043 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3044 	leave_guest_mode(vcpu);
3045 
3046 vmentry_fail_vmexit:
3047 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3048 
3049 	if (!from_vmentry)
3050 		return 1;
3051 
3052 	load_vmcs12_host_state(vcpu, vmcs12);
3053 	vmcs12->vm_exit_reason = exit_reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
3054 	vmcs12->exit_qualification = exit_qual;
3055 	if (enable_shadow_vmcs || vmx->nested.hv_evmcs)
3056 		vmx->nested.need_vmcs12_sync = true;
3057 	return 1;
3058 }
3059 
3060 /*
3061  * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3062  * for running an L2 nested guest.
3063  */
3064 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3065 {
3066 	struct vmcs12 *vmcs12;
3067 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3068 	u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3069 	int ret;
3070 
3071 	if (!nested_vmx_check_permission(vcpu))
3072 		return 1;
3073 
3074 	if (!nested_vmx_handle_enlightened_vmptrld(vcpu, true))
3075 		return 1;
3076 
3077 	if (!vmx->nested.hv_evmcs && vmx->nested.current_vmptr == -1ull)
3078 		return nested_vmx_failInvalid(vcpu);
3079 
3080 	vmcs12 = get_vmcs12(vcpu);
3081 
3082 	/*
3083 	 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3084 	 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3085 	 * rather than RFLAGS.ZF, and no error number is stored to the
3086 	 * VM-instruction error field.
3087 	 */
3088 	if (vmcs12->hdr.shadow_vmcs)
3089 		return nested_vmx_failInvalid(vcpu);
3090 
3091 	if (vmx->nested.hv_evmcs) {
3092 		copy_enlightened_to_vmcs12(vmx);
3093 		/* Enlightened VMCS doesn't have launch state */
3094 		vmcs12->launch_state = !launch;
3095 	} else if (enable_shadow_vmcs) {
3096 		copy_shadow_to_vmcs12(vmx);
3097 	}
3098 
3099 	/*
3100 	 * The nested entry process starts with enforcing various prerequisites
3101 	 * on vmcs12 as required by the Intel SDM, and act appropriately when
3102 	 * they fail: As the SDM explains, some conditions should cause the
3103 	 * instruction to fail, while others will cause the instruction to seem
3104 	 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3105 	 * To speed up the normal (success) code path, we should avoid checking
3106 	 * for misconfigurations which will anyway be caught by the processor
3107 	 * when using the merged vmcs02.
3108 	 */
3109 	if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS)
3110 		return nested_vmx_failValid(vcpu,
3111 			VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3112 
3113 	if (vmcs12->launch_state == launch)
3114 		return nested_vmx_failValid(vcpu,
3115 			launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3116 			       : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3117 
3118 	ret = nested_vmx_check_vmentry_prereqs(vcpu, vmcs12);
3119 	if (ret)
3120 		return nested_vmx_failValid(vcpu, ret);
3121 
3122 	/*
3123 	 * We're finally done with prerequisite checking, and can start with
3124 	 * the nested entry.
3125 	 */
3126 	vmx->nested.nested_run_pending = 1;
3127 	ret = nested_vmx_enter_non_root_mode(vcpu, true);
3128 	vmx->nested.nested_run_pending = !ret;
3129 	if (ret > 0)
3130 		return 1;
3131 	else if (ret)
3132 		return nested_vmx_failValid(vcpu,
3133 			VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3134 
3135 	/* Hide L1D cache contents from the nested guest.  */
3136 	vmx->vcpu.arch.l1tf_flush_l1d = true;
3137 
3138 	/*
3139 	 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3140 	 * also be used as part of restoring nVMX state for
3141 	 * snapshot restore (migration).
3142 	 *
3143 	 * In this flow, it is assumed that vmcs12 cache was
3144 	 * trasferred as part of captured nVMX state and should
3145 	 * therefore not be read from guest memory (which may not
3146 	 * exist on destination host yet).
3147 	 */
3148 	nested_cache_shadow_vmcs12(vcpu, vmcs12);
3149 
3150 	/*
3151 	 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3152 	 * awakened by event injection or by an NMI-window VM-exit or
3153 	 * by an interrupt-window VM-exit, halt the vcpu.
3154 	 */
3155 	if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
3156 	    !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3157 	    !(vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_NMI_PENDING) &&
3158 	    !((vmcs12->cpu_based_vm_exec_control & CPU_BASED_VIRTUAL_INTR_PENDING) &&
3159 	      (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3160 		vmx->nested.nested_run_pending = 0;
3161 		return kvm_vcpu_halt(vcpu);
3162 	}
3163 	return 1;
3164 }
3165 
3166 /*
3167  * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3168  * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
3169  * This function returns the new value we should put in vmcs12.guest_cr0.
3170  * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3171  *  1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3172  *     available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3173  *     didn't trap the bit, because if L1 did, so would L0).
3174  *  2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3175  *     been modified by L2, and L1 knows it. So just leave the old value of
3176  *     the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3177  *     isn't relevant, because if L0 traps this bit it can set it to anything.
3178  *  3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3179  *     changed these bits, and therefore they need to be updated, but L0
3180  *     didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3181  *     put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3182  */
3183 static inline unsigned long
3184 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3185 {
3186 	return
3187 	/*1*/	(vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3188 	/*2*/	(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3189 	/*3*/	(vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3190 			vcpu->arch.cr0_guest_owned_bits));
3191 }
3192 
3193 static inline unsigned long
3194 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3195 {
3196 	return
3197 	/*1*/	(vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3198 	/*2*/	(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3199 	/*3*/	(vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3200 			vcpu->arch.cr4_guest_owned_bits));
3201 }
3202 
3203 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3204 				      struct vmcs12 *vmcs12)
3205 {
3206 	u32 idt_vectoring;
3207 	unsigned int nr;
3208 
3209 	if (vcpu->arch.exception.injected) {
3210 		nr = vcpu->arch.exception.nr;
3211 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3212 
3213 		if (kvm_exception_is_soft(nr)) {
3214 			vmcs12->vm_exit_instruction_len =
3215 				vcpu->arch.event_exit_inst_len;
3216 			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3217 		} else
3218 			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3219 
3220 		if (vcpu->arch.exception.has_error_code) {
3221 			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3222 			vmcs12->idt_vectoring_error_code =
3223 				vcpu->arch.exception.error_code;
3224 		}
3225 
3226 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3227 	} else if (vcpu->arch.nmi_injected) {
3228 		vmcs12->idt_vectoring_info_field =
3229 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3230 	} else if (vcpu->arch.interrupt.injected) {
3231 		nr = vcpu->arch.interrupt.nr;
3232 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3233 
3234 		if (vcpu->arch.interrupt.soft) {
3235 			idt_vectoring |= INTR_TYPE_SOFT_INTR;
3236 			vmcs12->vm_entry_instruction_len =
3237 				vcpu->arch.event_exit_inst_len;
3238 		} else
3239 			idt_vectoring |= INTR_TYPE_EXT_INTR;
3240 
3241 		vmcs12->idt_vectoring_info_field = idt_vectoring;
3242 	}
3243 }
3244 
3245 
3246 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3247 {
3248 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3249 	gfn_t gfn;
3250 
3251 	/*
3252 	 * Don't need to mark the APIC access page dirty; it is never
3253 	 * written to by the CPU during APIC virtualization.
3254 	 */
3255 
3256 	if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3257 		gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3258 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3259 	}
3260 
3261 	if (nested_cpu_has_posted_intr(vmcs12)) {
3262 		gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3263 		kvm_vcpu_mark_page_dirty(vcpu, gfn);
3264 	}
3265 }
3266 
3267 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3268 {
3269 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3270 	int max_irr;
3271 	void *vapic_page;
3272 	u16 status;
3273 
3274 	if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
3275 		return;
3276 
3277 	vmx->nested.pi_pending = false;
3278 	if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3279 		return;
3280 
3281 	max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
3282 	if (max_irr != 256) {
3283 		vapic_page = kmap(vmx->nested.virtual_apic_page);
3284 		__kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3285 			vapic_page, &max_irr);
3286 		kunmap(vmx->nested.virtual_apic_page);
3287 
3288 		status = vmcs_read16(GUEST_INTR_STATUS);
3289 		if ((u8)max_irr > ((u8)status & 0xff)) {
3290 			status &= ~0xff;
3291 			status |= (u8)max_irr;
3292 			vmcs_write16(GUEST_INTR_STATUS, status);
3293 		}
3294 	}
3295 
3296 	nested_mark_vmcs12_pages_dirty(vcpu);
3297 }
3298 
3299 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
3300 					       unsigned long exit_qual)
3301 {
3302 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3303 	unsigned int nr = vcpu->arch.exception.nr;
3304 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
3305 
3306 	if (vcpu->arch.exception.has_error_code) {
3307 		vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
3308 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3309 	}
3310 
3311 	if (kvm_exception_is_soft(nr))
3312 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3313 	else
3314 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
3315 
3316 	if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3317 	    vmx_get_nmi_mask(vcpu))
3318 		intr_info |= INTR_INFO_UNBLOCK_NMI;
3319 
3320 	nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3321 }
3322 
3323 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
3324 {
3325 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3326 	unsigned long exit_qual;
3327 	bool block_nested_events =
3328 	    vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
3329 
3330 	if (vcpu->arch.exception.pending &&
3331 		nested_vmx_check_exception(vcpu, &exit_qual)) {
3332 		if (block_nested_events)
3333 			return -EBUSY;
3334 		nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
3335 		return 0;
3336 	}
3337 
3338 	if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
3339 	    vmx->nested.preemption_timer_expired) {
3340 		if (block_nested_events)
3341 			return -EBUSY;
3342 		nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
3343 		return 0;
3344 	}
3345 
3346 	if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
3347 		if (block_nested_events)
3348 			return -EBUSY;
3349 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
3350 				  NMI_VECTOR | INTR_TYPE_NMI_INTR |
3351 				  INTR_INFO_VALID_MASK, 0);
3352 		/*
3353 		 * The NMI-triggered VM exit counts as injection:
3354 		 * clear this one and block further NMIs.
3355 		 */
3356 		vcpu->arch.nmi_pending = 0;
3357 		vmx_set_nmi_mask(vcpu, true);
3358 		return 0;
3359 	}
3360 
3361 	if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
3362 	    nested_exit_on_intr(vcpu)) {
3363 		if (block_nested_events)
3364 			return -EBUSY;
3365 		nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
3366 		return 0;
3367 	}
3368 
3369 	vmx_complete_nested_posted_interrupt(vcpu);
3370 	return 0;
3371 }
3372 
3373 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
3374 {
3375 	ktime_t remaining =
3376 		hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
3377 	u64 value;
3378 
3379 	if (ktime_to_ns(remaining) <= 0)
3380 		return 0;
3381 
3382 	value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
3383 	do_div(value, 1000000);
3384 	return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
3385 }
3386 
3387 /*
3388  * Update the guest state fields of vmcs12 to reflect changes that
3389  * occurred while L2 was running. (The "IA-32e mode guest" bit of the
3390  * VM-entry controls is also updated, since this is really a guest
3391  * state bit.)
3392  */
3393 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3394 {
3395 	vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
3396 	vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
3397 
3398 	vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3399 	vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
3400 	vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
3401 
3402 	vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
3403 	vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
3404 	vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
3405 	vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
3406 	vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
3407 	vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
3408 	vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
3409 	vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
3410 	vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
3411 	vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
3412 	vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
3413 	vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
3414 	vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
3415 	vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
3416 	vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
3417 	vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
3418 	vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
3419 	vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
3420 	vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
3421 	vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
3422 	vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
3423 	vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
3424 	vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
3425 	vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
3426 	vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
3427 	vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
3428 	vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
3429 	vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
3430 	vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
3431 	vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
3432 	vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
3433 	vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
3434 	vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
3435 	vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
3436 	vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
3437 	vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
3438 
3439 	vmcs12->guest_interruptibility_info =
3440 		vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
3441 	vmcs12->guest_pending_dbg_exceptions =
3442 		vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
3443 	if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
3444 		vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
3445 	else
3446 		vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
3447 
3448 	if (nested_cpu_has_preemption_timer(vmcs12)) {
3449 		if (vmcs12->vm_exit_controls &
3450 		    VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
3451 			vmcs12->vmx_preemption_timer_value =
3452 				vmx_get_preemption_timer_value(vcpu);
3453 		hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
3454 	}
3455 
3456 	/*
3457 	 * In some cases (usually, nested EPT), L2 is allowed to change its
3458 	 * own CR3 without exiting. If it has changed it, we must keep it.
3459 	 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
3460 	 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
3461 	 *
3462 	 * Additionally, restore L2's PDPTR to vmcs12.
3463 	 */
3464 	if (enable_ept) {
3465 		vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
3466 		vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
3467 		vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
3468 		vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
3469 		vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
3470 	}
3471 
3472 	vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
3473 
3474 	if (nested_cpu_has_vid(vmcs12))
3475 		vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
3476 
3477 	vmcs12->vm_entry_controls =
3478 		(vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
3479 		(vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
3480 
3481 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
3482 		kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
3483 		vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3484 	}
3485 
3486 	/* TODO: These cannot have changed unless we have MSR bitmaps and
3487 	 * the relevant bit asks not to trap the change */
3488 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
3489 		vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
3490 	if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
3491 		vmcs12->guest_ia32_efer = vcpu->arch.efer;
3492 	vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
3493 	vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
3494 	vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
3495 	if (kvm_mpx_supported())
3496 		vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3497 }
3498 
3499 /*
3500  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
3501  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
3502  * and this function updates it to reflect the changes to the guest state while
3503  * L2 was running (and perhaps made some exits which were handled directly by L0
3504  * without going back to L1), and to reflect the exit reason.
3505  * Note that we do not have to copy here all VMCS fields, just those that
3506  * could have changed by the L2 guest or the exit - i.e., the guest-state and
3507  * exit-information fields only. Other fields are modified by L1 with VMWRITE,
3508  * which already writes to vmcs12 directly.
3509  */
3510 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
3511 			   u32 exit_reason, u32 exit_intr_info,
3512 			   unsigned long exit_qualification)
3513 {
3514 	/* update guest state fields: */
3515 	sync_vmcs12(vcpu, vmcs12);
3516 
3517 	/* update exit information fields: */
3518 
3519 	vmcs12->vm_exit_reason = exit_reason;
3520 	vmcs12->exit_qualification = exit_qualification;
3521 	vmcs12->vm_exit_intr_info = exit_intr_info;
3522 
3523 	vmcs12->idt_vectoring_info_field = 0;
3524 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3525 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
3526 
3527 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
3528 		vmcs12->launch_state = 1;
3529 
3530 		/* vm_entry_intr_info_field is cleared on exit. Emulate this
3531 		 * instead of reading the real value. */
3532 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
3533 
3534 		/*
3535 		 * Transfer the event that L0 or L1 may wanted to inject into
3536 		 * L2 to IDT_VECTORING_INFO_FIELD.
3537 		 */
3538 		vmcs12_save_pending_event(vcpu, vmcs12);
3539 
3540 		/*
3541 		 * According to spec, there's no need to store the guest's
3542 		 * MSRs if the exit is due to a VM-entry failure that occurs
3543 		 * during or after loading the guest state. Since this exit
3544 		 * does not fall in that category, we need to save the MSRs.
3545 		 */
3546 		if (nested_vmx_store_msr(vcpu,
3547 					 vmcs12->vm_exit_msr_store_addr,
3548 					 vmcs12->vm_exit_msr_store_count))
3549 			nested_vmx_abort(vcpu,
3550 					 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
3551 	}
3552 
3553 	/*
3554 	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
3555 	 * preserved above and would only end up incorrectly in L1.
3556 	 */
3557 	vcpu->arch.nmi_injected = false;
3558 	kvm_clear_exception_queue(vcpu);
3559 	kvm_clear_interrupt_queue(vcpu);
3560 }
3561 
3562 /*
3563  * A part of what we need to when the nested L2 guest exits and we want to
3564  * run its L1 parent, is to reset L1's guest state to the host state specified
3565  * in vmcs12.
3566  * This function is to be called not only on normal nested exit, but also on
3567  * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
3568  * Failures During or After Loading Guest State").
3569  * This function should be called when the active VMCS is L1's (vmcs01).
3570  */
3571 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3572 				   struct vmcs12 *vmcs12)
3573 {
3574 	struct kvm_segment seg;
3575 	u32 entry_failure_code;
3576 
3577 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
3578 		vcpu->arch.efer = vmcs12->host_ia32_efer;
3579 	else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3580 		vcpu->arch.efer |= (EFER_LMA | EFER_LME);
3581 	else
3582 		vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
3583 	vmx_set_efer(vcpu, vcpu->arch.efer);
3584 
3585 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
3586 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
3587 	vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
3588 	vmx_set_interrupt_shadow(vcpu, 0);
3589 
3590 	/*
3591 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
3592 	 * actually changed, because vmx_set_cr0 refers to efer set above.
3593 	 *
3594 	 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
3595 	 * (KVM doesn't change it);
3596 	 */
3597 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3598 	vmx_set_cr0(vcpu, vmcs12->host_cr0);
3599 
3600 	/* Same as above - no reason to call set_cr4_guest_host_mask().  */
3601 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3602 	vmx_set_cr4(vcpu, vmcs12->host_cr4);
3603 
3604 	nested_ept_uninit_mmu_context(vcpu);
3605 
3606 	/*
3607 	 * Only PDPTE load can fail as the value of cr3 was checked on entry and
3608 	 * couldn't have changed.
3609 	 */
3610 	if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
3611 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
3612 
3613 	if (!enable_ept)
3614 		vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
3615 
3616 	/*
3617 	 * If vmcs01 doesn't use VPID, CPU flushes TLB on every
3618 	 * VMEntry/VMExit. Thus, no need to flush TLB.
3619 	 *
3620 	 * If vmcs12 doesn't use VPID, L1 expects TLB to be
3621 	 * flushed on every VMEntry/VMExit.
3622 	 *
3623 	 * Otherwise, we can preserve TLB entries as long as we are
3624 	 * able to tag L1 TLB entries differently than L2 TLB entries.
3625 	 *
3626 	 * If vmcs12 uses EPT, we need to execute this flush on EPTP01
3627 	 * and therefore we request the TLB flush to happen only after VMCS EPTP
3628 	 * has been set by KVM_REQ_LOAD_CR3.
3629 	 */
3630 	if (enable_vpid &&
3631 	    (!nested_cpu_has_vpid(vmcs12) || !nested_has_guest_tlb_tag(vcpu))) {
3632 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3633 	}
3634 
3635 	vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
3636 	vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
3637 	vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
3638 	vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
3639 	vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
3640 	vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
3641 	vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
3642 
3643 	/* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1.  */
3644 	if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
3645 		vmcs_write64(GUEST_BNDCFGS, 0);
3646 
3647 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
3648 		vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
3649 		vcpu->arch.pat = vmcs12->host_ia32_pat;
3650 	}
3651 	if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
3652 		vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
3653 			vmcs12->host_ia32_perf_global_ctrl);
3654 
3655 	/* Set L1 segment info according to Intel SDM
3656 	    27.5.2 Loading Host Segment and Descriptor-Table Registers */
3657 	seg = (struct kvm_segment) {
3658 		.base = 0,
3659 		.limit = 0xFFFFFFFF,
3660 		.selector = vmcs12->host_cs_selector,
3661 		.type = 11,
3662 		.present = 1,
3663 		.s = 1,
3664 		.g = 1
3665 	};
3666 	if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
3667 		seg.l = 1;
3668 	else
3669 		seg.db = 1;
3670 	vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
3671 	seg = (struct kvm_segment) {
3672 		.base = 0,
3673 		.limit = 0xFFFFFFFF,
3674 		.type = 3,
3675 		.present = 1,
3676 		.s = 1,
3677 		.db = 1,
3678 		.g = 1
3679 	};
3680 	seg.selector = vmcs12->host_ds_selector;
3681 	vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
3682 	seg.selector = vmcs12->host_es_selector;
3683 	vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
3684 	seg.selector = vmcs12->host_ss_selector;
3685 	vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
3686 	seg.selector = vmcs12->host_fs_selector;
3687 	seg.base = vmcs12->host_fs_base;
3688 	vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
3689 	seg.selector = vmcs12->host_gs_selector;
3690 	seg.base = vmcs12->host_gs_base;
3691 	vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
3692 	seg = (struct kvm_segment) {
3693 		.base = vmcs12->host_tr_base,
3694 		.limit = 0x67,
3695 		.selector = vmcs12->host_tr_selector,
3696 		.type = 11,
3697 		.present = 1
3698 	};
3699 	vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
3700 
3701 	kvm_set_dr(vcpu, 7, 0x400);
3702 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3703 
3704 	if (cpu_has_vmx_msr_bitmap())
3705 		vmx_update_msr_bitmap(vcpu);
3706 
3707 	if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
3708 				vmcs12->vm_exit_msr_load_count))
3709 		nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3710 }
3711 
3712 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
3713 {
3714 	struct shared_msr_entry *efer_msr;
3715 	unsigned int i;
3716 
3717 	if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
3718 		return vmcs_read64(GUEST_IA32_EFER);
3719 
3720 	if (cpu_has_load_ia32_efer())
3721 		return host_efer;
3722 
3723 	for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
3724 		if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
3725 			return vmx->msr_autoload.guest.val[i].value;
3726 	}
3727 
3728 	efer_msr = find_msr_entry(vmx, MSR_EFER);
3729 	if (efer_msr)
3730 		return efer_msr->data;
3731 
3732 	return host_efer;
3733 }
3734 
3735 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
3736 {
3737 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3738 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3739 	struct vmx_msr_entry g, h;
3740 	struct msr_data msr;
3741 	gpa_t gpa;
3742 	u32 i, j;
3743 
3744 	vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
3745 
3746 	if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
3747 		/*
3748 		 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
3749 		 * as vmcs01.GUEST_DR7 contains a userspace defined value
3750 		 * and vcpu->arch.dr7 is not squirreled away before the
3751 		 * nested VMENTER (not worth adding a variable in nested_vmx).
3752 		 */
3753 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
3754 			kvm_set_dr(vcpu, 7, DR7_FIXED_1);
3755 		else
3756 			WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
3757 	}
3758 
3759 	/*
3760 	 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
3761 	 * handle a variety of side effects to KVM's software model.
3762 	 */
3763 	vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
3764 
3765 	vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
3766 	vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
3767 
3768 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
3769 	vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
3770 
3771 	nested_ept_uninit_mmu_context(vcpu);
3772 	vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3773 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3774 
3775 	/*
3776 	 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
3777 	 * from vmcs01 (if necessary).  The PDPTRs are not loaded on
3778 	 * VMFail, like everything else we just need to ensure our
3779 	 * software model is up-to-date.
3780 	 */
3781 	ept_save_pdptrs(vcpu);
3782 
3783 	kvm_mmu_reset_context(vcpu);
3784 
3785 	if (cpu_has_vmx_msr_bitmap())
3786 		vmx_update_msr_bitmap(vcpu);
3787 
3788 	/*
3789 	 * This nasty bit of open coding is a compromise between blindly
3790 	 * loading L1's MSRs using the exit load lists (incorrect emulation
3791 	 * of VMFail), leaving the nested VM's MSRs in the software model
3792 	 * (incorrect behavior) and snapshotting the modified MSRs (too
3793 	 * expensive since the lists are unbound by hardware).  For each
3794 	 * MSR that was (prematurely) loaded from the nested VMEntry load
3795 	 * list, reload it from the exit load list if it exists and differs
3796 	 * from the guest value.  The intent is to stuff host state as
3797 	 * silently as possible, not to fully process the exit load list.
3798 	 */
3799 	msr.host_initiated = false;
3800 	for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
3801 		gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
3802 		if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
3803 			pr_debug_ratelimited(
3804 				"%s read MSR index failed (%u, 0x%08llx)\n",
3805 				__func__, i, gpa);
3806 			goto vmabort;
3807 		}
3808 
3809 		for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
3810 			gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
3811 			if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
3812 				pr_debug_ratelimited(
3813 					"%s read MSR failed (%u, 0x%08llx)\n",
3814 					__func__, j, gpa);
3815 				goto vmabort;
3816 			}
3817 			if (h.index != g.index)
3818 				continue;
3819 			if (h.value == g.value)
3820 				break;
3821 
3822 			if (nested_vmx_load_msr_check(vcpu, &h)) {
3823 				pr_debug_ratelimited(
3824 					"%s check failed (%u, 0x%x, 0x%x)\n",
3825 					__func__, j, h.index, h.reserved);
3826 				goto vmabort;
3827 			}
3828 
3829 			msr.index = h.index;
3830 			msr.data = h.value;
3831 			if (kvm_set_msr(vcpu, &msr)) {
3832 				pr_debug_ratelimited(
3833 					"%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
3834 					__func__, j, h.index, h.value);
3835 				goto vmabort;
3836 			}
3837 		}
3838 	}
3839 
3840 	return;
3841 
3842 vmabort:
3843 	nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
3844 }
3845 
3846 /*
3847  * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
3848  * and modify vmcs12 to make it see what it would expect to see there if
3849  * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
3850  */
3851 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
3852 		       u32 exit_intr_info, unsigned long exit_qualification)
3853 {
3854 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3855 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3856 
3857 	/* trying to cancel vmlaunch/vmresume is a bug */
3858 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
3859 
3860 	leave_guest_mode(vcpu);
3861 
3862 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
3863 		vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3864 
3865 	if (likely(!vmx->fail)) {
3866 		if (exit_reason == -1)
3867 			sync_vmcs12(vcpu, vmcs12);
3868 		else
3869 			prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
3870 				       exit_qualification);
3871 
3872 		/*
3873 		 * Must happen outside of sync_vmcs12() as it will
3874 		 * also be used to capture vmcs12 cache as part of
3875 		 * capturing nVMX state for snapshot (migration).
3876 		 *
3877 		 * Otherwise, this flush will dirty guest memory at a
3878 		 * point it is already assumed by user-space to be
3879 		 * immutable.
3880 		 */
3881 		nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
3882 	} else {
3883 		/*
3884 		 * The only expected VM-instruction error is "VM entry with
3885 		 * invalid control field(s)." Anything else indicates a
3886 		 * problem with L0.  And we should never get here with a
3887 		 * VMFail of any type if early consistency checks are enabled.
3888 		 */
3889 		WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
3890 			     VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3891 		WARN_ON_ONCE(nested_early_check);
3892 	}
3893 
3894 	vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3895 
3896 	/* Update any VMCS fields that might have changed while L2 ran */
3897 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3898 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3899 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
3900 
3901 	if (kvm_has_tsc_control)
3902 		decache_tsc_multiplier(vmx);
3903 
3904 	if (vmx->nested.change_vmcs01_virtual_apic_mode) {
3905 		vmx->nested.change_vmcs01_virtual_apic_mode = false;
3906 		vmx_set_virtual_apic_mode(vcpu);
3907 	} else if (!nested_cpu_has_ept(vmcs12) &&
3908 		   nested_cpu_has2(vmcs12,
3909 				   SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3910 		vmx_flush_tlb(vcpu, true);
3911 	}
3912 
3913 	/* This is needed for same reason as it was needed in prepare_vmcs02 */
3914 	vmx->host_rsp = 0;
3915 
3916 	/* Unpin physical memory we referred to in vmcs02 */
3917 	if (vmx->nested.apic_access_page) {
3918 		kvm_release_page_dirty(vmx->nested.apic_access_page);
3919 		vmx->nested.apic_access_page = NULL;
3920 	}
3921 	if (vmx->nested.virtual_apic_page) {
3922 		kvm_release_page_dirty(vmx->nested.virtual_apic_page);
3923 		vmx->nested.virtual_apic_page = NULL;
3924 	}
3925 	if (vmx->nested.pi_desc_page) {
3926 		kunmap(vmx->nested.pi_desc_page);
3927 		kvm_release_page_dirty(vmx->nested.pi_desc_page);
3928 		vmx->nested.pi_desc_page = NULL;
3929 		vmx->nested.pi_desc = NULL;
3930 	}
3931 
3932 	/*
3933 	 * We are now running in L2, mmu_notifier will force to reload the
3934 	 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
3935 	 */
3936 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
3937 
3938 	if ((exit_reason != -1) && (enable_shadow_vmcs || vmx->nested.hv_evmcs))
3939 		vmx->nested.need_vmcs12_sync = true;
3940 
3941 	/* in case we halted in L2 */
3942 	vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3943 
3944 	if (likely(!vmx->fail)) {
3945 		/*
3946 		 * TODO: SDM says that with acknowledge interrupt on
3947 		 * exit, bit 31 of the VM-exit interrupt information
3948 		 * (valid interrupt) is always set to 1 on
3949 		 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
3950 		 * need kvm_cpu_has_interrupt().  See the commit
3951 		 * message for details.
3952 		 */
3953 		if (nested_exit_intr_ack_set(vcpu) &&
3954 		    exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
3955 		    kvm_cpu_has_interrupt(vcpu)) {
3956 			int irq = kvm_cpu_get_interrupt(vcpu);
3957 			WARN_ON(irq < 0);
3958 			vmcs12->vm_exit_intr_info = irq |
3959 				INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
3960 		}
3961 
3962 		if (exit_reason != -1)
3963 			trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
3964 						       vmcs12->exit_qualification,
3965 						       vmcs12->idt_vectoring_info_field,
3966 						       vmcs12->vm_exit_intr_info,
3967 						       vmcs12->vm_exit_intr_error_code,
3968 						       KVM_ISA_VMX);
3969 
3970 		load_vmcs12_host_state(vcpu, vmcs12);
3971 
3972 		return;
3973 	}
3974 
3975 	/*
3976 	 * After an early L2 VM-entry failure, we're now back
3977 	 * in L1 which thinks it just finished a VMLAUNCH or
3978 	 * VMRESUME instruction, so we need to set the failure
3979 	 * flag and the VM-instruction error field of the VMCS
3980 	 * accordingly, and skip the emulated instruction.
3981 	 */
3982 	(void)nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3983 
3984 	/*
3985 	 * Restore L1's host state to KVM's software model.  We're here
3986 	 * because a consistency check was caught by hardware, which
3987 	 * means some amount of guest state has been propagated to KVM's
3988 	 * model and needs to be unwound to the host's state.
3989 	 */
3990 	nested_vmx_restore_host_state(vcpu);
3991 
3992 	vmx->fail = 0;
3993 }
3994 
3995 /*
3996  * Decode the memory-address operand of a vmx instruction, as recorded on an
3997  * exit caused by such an instruction (run by a guest hypervisor).
3998  * On success, returns 0. When the operand is invalid, returns 1 and throws
3999  * #UD or #GP.
4000  */
4001 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
4002 			u32 vmx_instruction_info, bool wr, gva_t *ret)
4003 {
4004 	gva_t off;
4005 	bool exn;
4006 	struct kvm_segment s;
4007 
4008 	/*
4009 	 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4010 	 * Execution", on an exit, vmx_instruction_info holds most of the
4011 	 * addressing components of the operand. Only the displacement part
4012 	 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4013 	 * For how an actual address is calculated from all these components,
4014 	 * refer to Vol. 1, "Operand Addressing".
4015 	 */
4016 	int  scaling = vmx_instruction_info & 3;
4017 	int  addr_size = (vmx_instruction_info >> 7) & 7;
4018 	bool is_reg = vmx_instruction_info & (1u << 10);
4019 	int  seg_reg = (vmx_instruction_info >> 15) & 7;
4020 	int  index_reg = (vmx_instruction_info >> 18) & 0xf;
4021 	bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4022 	int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
4023 	bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
4024 
4025 	if (is_reg) {
4026 		kvm_queue_exception(vcpu, UD_VECTOR);
4027 		return 1;
4028 	}
4029 
4030 	/* Addr = segment_base + offset */
4031 	/* offset = base + [index * scale] + displacement */
4032 	off = exit_qualification; /* holds the displacement */
4033 	if (base_is_valid)
4034 		off += kvm_register_read(vcpu, base_reg);
4035 	if (index_is_valid)
4036 		off += kvm_register_read(vcpu, index_reg)<<scaling;
4037 	vmx_get_segment(vcpu, &s, seg_reg);
4038 	*ret = s.base + off;
4039 
4040 	if (addr_size == 1) /* 32 bit */
4041 		*ret &= 0xffffffff;
4042 
4043 	/* Checks for #GP/#SS exceptions. */
4044 	exn = false;
4045 	if (is_long_mode(vcpu)) {
4046 		/* Long mode: #GP(0)/#SS(0) if the memory address is in a
4047 		 * non-canonical form. This is the only check on the memory
4048 		 * destination for long mode!
4049 		 */
4050 		exn = is_noncanonical_address(*ret, vcpu);
4051 	} else if (is_protmode(vcpu)) {
4052 		/* Protected mode: apply checks for segment validity in the
4053 		 * following order:
4054 		 * - segment type check (#GP(0) may be thrown)
4055 		 * - usability check (#GP(0)/#SS(0))
4056 		 * - limit check (#GP(0)/#SS(0))
4057 		 */
4058 		if (wr)
4059 			/* #GP(0) if the destination operand is located in a
4060 			 * read-only data segment or any code segment.
4061 			 */
4062 			exn = ((s.type & 0xa) == 0 || (s.type & 8));
4063 		else
4064 			/* #GP(0) if the source operand is located in an
4065 			 * execute-only code segment
4066 			 */
4067 			exn = ((s.type & 0xa) == 8);
4068 		if (exn) {
4069 			kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4070 			return 1;
4071 		}
4072 		/* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
4073 		 */
4074 		exn = (s.unusable != 0);
4075 		/* Protected mode: #GP(0)/#SS(0) if the memory
4076 		 * operand is outside the segment limit.
4077 		 */
4078 		exn = exn || (off + sizeof(u64) > s.limit);
4079 	}
4080 	if (exn) {
4081 		kvm_queue_exception_e(vcpu,
4082 				      seg_reg == VCPU_SREG_SS ?
4083 						SS_VECTOR : GP_VECTOR,
4084 				      0);
4085 		return 1;
4086 	}
4087 
4088 	return 0;
4089 }
4090 
4091 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
4092 {
4093 	gva_t gva;
4094 	struct x86_exception e;
4095 
4096 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4097 			vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
4098 		return 1;
4099 
4100 	if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
4101 		kvm_inject_page_fault(vcpu, &e);
4102 		return 1;
4103 	}
4104 
4105 	return 0;
4106 }
4107 
4108 /*
4109  * Allocate a shadow VMCS and associate it with the currently loaded
4110  * VMCS, unless such a shadow VMCS already exists. The newly allocated
4111  * VMCS is also VMCLEARed, so that it is ready for use.
4112  */
4113 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
4114 {
4115 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4116 	struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
4117 
4118 	/*
4119 	 * We should allocate a shadow vmcs for vmcs01 only when L1
4120 	 * executes VMXON and free it when L1 executes VMXOFF.
4121 	 * As it is invalid to execute VMXON twice, we shouldn't reach
4122 	 * here when vmcs01 already have an allocated shadow vmcs.
4123 	 */
4124 	WARN_ON(loaded_vmcs == &vmx->vmcs01 && loaded_vmcs->shadow_vmcs);
4125 
4126 	if (!loaded_vmcs->shadow_vmcs) {
4127 		loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
4128 		if (loaded_vmcs->shadow_vmcs)
4129 			vmcs_clear(loaded_vmcs->shadow_vmcs);
4130 	}
4131 	return loaded_vmcs->shadow_vmcs;
4132 }
4133 
4134 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
4135 {
4136 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4137 	int r;
4138 
4139 	r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
4140 	if (r < 0)
4141 		goto out_vmcs02;
4142 
4143 	vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
4144 	if (!vmx->nested.cached_vmcs12)
4145 		goto out_cached_vmcs12;
4146 
4147 	vmx->nested.cached_shadow_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
4148 	if (!vmx->nested.cached_shadow_vmcs12)
4149 		goto out_cached_shadow_vmcs12;
4150 
4151 	if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
4152 		goto out_shadow_vmcs;
4153 
4154 	hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
4155 		     HRTIMER_MODE_REL_PINNED);
4156 	vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
4157 
4158 	vmx->nested.vpid02 = allocate_vpid();
4159 
4160 	vmx->nested.vmcs02_initialized = false;
4161 	vmx->nested.vmxon = true;
4162 
4163 	if (pt_mode == PT_MODE_HOST_GUEST) {
4164 		vmx->pt_desc.guest.ctl = 0;
4165 		pt_update_intercept_for_msr(vmx);
4166 	}
4167 
4168 	return 0;
4169 
4170 out_shadow_vmcs:
4171 	kfree(vmx->nested.cached_shadow_vmcs12);
4172 
4173 out_cached_shadow_vmcs12:
4174 	kfree(vmx->nested.cached_vmcs12);
4175 
4176 out_cached_vmcs12:
4177 	free_loaded_vmcs(&vmx->nested.vmcs02);
4178 
4179 out_vmcs02:
4180 	return -ENOMEM;
4181 }
4182 
4183 /*
4184  * Emulate the VMXON instruction.
4185  * Currently, we just remember that VMX is active, and do not save or even
4186  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4187  * do not currently need to store anything in that guest-allocated memory
4188  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4189  * argument is different from the VMXON pointer (which the spec says they do).
4190  */
4191 static int handle_vmon(struct kvm_vcpu *vcpu)
4192 {
4193 	int ret;
4194 	gpa_t vmptr;
4195 	struct page *page;
4196 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4197 	const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
4198 		| FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
4199 
4200 	/*
4201 	 * The Intel VMX Instruction Reference lists a bunch of bits that are
4202 	 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
4203 	 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
4204 	 * Otherwise, we should fail with #UD.  But most faulting conditions
4205 	 * have already been checked by hardware, prior to the VM-exit for
4206 	 * VMXON.  We do test guest cr4.VMXE because processor CR4 always has
4207 	 * that bit set to 1 in non-root mode.
4208 	 */
4209 	if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
4210 		kvm_queue_exception(vcpu, UD_VECTOR);
4211 		return 1;
4212 	}
4213 
4214 	/* CPL=0 must be checked manually. */
4215 	if (vmx_get_cpl(vcpu)) {
4216 		kvm_inject_gp(vcpu, 0);
4217 		return 1;
4218 	}
4219 
4220 	if (vmx->nested.vmxon)
4221 		return nested_vmx_failValid(vcpu,
4222 			VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
4223 
4224 	if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
4225 			!= VMXON_NEEDED_FEATURES) {
4226 		kvm_inject_gp(vcpu, 0);
4227 		return 1;
4228 	}
4229 
4230 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4231 		return 1;
4232 
4233 	/*
4234 	 * SDM 3: 24.11.5
4235 	 * The first 4 bytes of VMXON region contain the supported
4236 	 * VMCS revision identifier
4237 	 *
4238 	 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
4239 	 * which replaces physical address width with 32
4240 	 */
4241 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4242 		return nested_vmx_failInvalid(vcpu);
4243 
4244 	page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4245 	if (is_error_page(page))
4246 		return nested_vmx_failInvalid(vcpu);
4247 
4248 	if (*(u32 *)kmap(page) != VMCS12_REVISION) {
4249 		kunmap(page);
4250 		kvm_release_page_clean(page);
4251 		return nested_vmx_failInvalid(vcpu);
4252 	}
4253 	kunmap(page);
4254 	kvm_release_page_clean(page);
4255 
4256 	vmx->nested.vmxon_ptr = vmptr;
4257 	ret = enter_vmx_operation(vcpu);
4258 	if (ret)
4259 		return ret;
4260 
4261 	return nested_vmx_succeed(vcpu);
4262 }
4263 
4264 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
4265 {
4266 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4267 
4268 	if (vmx->nested.current_vmptr == -1ull)
4269 		return;
4270 
4271 	if (enable_shadow_vmcs) {
4272 		/* copy to memory all shadowed fields in case
4273 		   they were modified */
4274 		copy_shadow_to_vmcs12(vmx);
4275 		vmx->nested.need_vmcs12_sync = false;
4276 		vmx_disable_shadow_vmcs(vmx);
4277 	}
4278 	vmx->nested.posted_intr_nv = -1;
4279 
4280 	/* Flush VMCS12 to guest memory */
4281 	kvm_vcpu_write_guest_page(vcpu,
4282 				  vmx->nested.current_vmptr >> PAGE_SHIFT,
4283 				  vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
4284 
4285 	kvm_mmu_free_roots(vcpu, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
4286 
4287 	vmx->nested.current_vmptr = -1ull;
4288 }
4289 
4290 /* Emulate the VMXOFF instruction */
4291 static int handle_vmoff(struct kvm_vcpu *vcpu)
4292 {
4293 	if (!nested_vmx_check_permission(vcpu))
4294 		return 1;
4295 	free_nested(vcpu);
4296 	return nested_vmx_succeed(vcpu);
4297 }
4298 
4299 /* Emulate the VMCLEAR instruction */
4300 static int handle_vmclear(struct kvm_vcpu *vcpu)
4301 {
4302 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4303 	u32 zero = 0;
4304 	gpa_t vmptr;
4305 
4306 	if (!nested_vmx_check_permission(vcpu))
4307 		return 1;
4308 
4309 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4310 		return 1;
4311 
4312 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4313 		return nested_vmx_failValid(vcpu,
4314 			VMXERR_VMCLEAR_INVALID_ADDRESS);
4315 
4316 	if (vmptr == vmx->nested.vmxon_ptr)
4317 		return nested_vmx_failValid(vcpu,
4318 			VMXERR_VMCLEAR_VMXON_POINTER);
4319 
4320 	if (vmx->nested.hv_evmcs_page) {
4321 		if (vmptr == vmx->nested.hv_evmcs_vmptr)
4322 			nested_release_evmcs(vcpu);
4323 	} else {
4324 		if (vmptr == vmx->nested.current_vmptr)
4325 			nested_release_vmcs12(vcpu);
4326 
4327 		kvm_vcpu_write_guest(vcpu,
4328 				     vmptr + offsetof(struct vmcs12,
4329 						      launch_state),
4330 				     &zero, sizeof(zero));
4331 	}
4332 
4333 	return nested_vmx_succeed(vcpu);
4334 }
4335 
4336 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4337 
4338 /* Emulate the VMLAUNCH instruction */
4339 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4340 {
4341 	return nested_vmx_run(vcpu, true);
4342 }
4343 
4344 /* Emulate the VMRESUME instruction */
4345 static int handle_vmresume(struct kvm_vcpu *vcpu)
4346 {
4347 
4348 	return nested_vmx_run(vcpu, false);
4349 }
4350 
4351 static int handle_vmread(struct kvm_vcpu *vcpu)
4352 {
4353 	unsigned long field;
4354 	u64 field_value;
4355 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4356 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4357 	gva_t gva = 0;
4358 	struct vmcs12 *vmcs12;
4359 
4360 	if (!nested_vmx_check_permission(vcpu))
4361 		return 1;
4362 
4363 	if (to_vmx(vcpu)->nested.current_vmptr == -1ull)
4364 		return nested_vmx_failInvalid(vcpu);
4365 
4366 	if (!is_guest_mode(vcpu))
4367 		vmcs12 = get_vmcs12(vcpu);
4368 	else {
4369 		/*
4370 		 * When vmcs->vmcs_link_pointer is -1ull, any VMREAD
4371 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4372 		 */
4373 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4374 			return nested_vmx_failInvalid(vcpu);
4375 		vmcs12 = get_shadow_vmcs12(vcpu);
4376 	}
4377 
4378 	/* Decode instruction info and find the field to read */
4379 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4380 	/* Read the field, zero-extended to a u64 field_value */
4381 	if (vmcs12_read_any(vmcs12, field, &field_value) < 0)
4382 		return nested_vmx_failValid(vcpu,
4383 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4384 
4385 	/*
4386 	 * Now copy part of this value to register or memory, as requested.
4387 	 * Note that the number of bits actually copied is 32 or 64 depending
4388 	 * on the guest's mode (32 or 64 bit), not on the given field's length.
4389 	 */
4390 	if (vmx_instruction_info & (1u << 10)) {
4391 		kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
4392 			field_value);
4393 	} else {
4394 		if (get_vmx_mem_address(vcpu, exit_qualification,
4395 				vmx_instruction_info, true, &gva))
4396 			return 1;
4397 		/* _system ok, nested_vmx_check_permission has verified cpl=0 */
4398 		kvm_write_guest_virt_system(vcpu, gva, &field_value,
4399 					    (is_long_mode(vcpu) ? 8 : 4), NULL);
4400 	}
4401 
4402 	return nested_vmx_succeed(vcpu);
4403 }
4404 
4405 
4406 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4407 {
4408 	unsigned long field;
4409 	gva_t gva;
4410 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4411 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4412 	u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4413 
4414 	/* The value to write might be 32 or 64 bits, depending on L1's long
4415 	 * mode, and eventually we need to write that into a field of several
4416 	 * possible lengths. The code below first zero-extends the value to 64
4417 	 * bit (field_value), and then copies only the appropriate number of
4418 	 * bits into the vmcs12 field.
4419 	 */
4420 	u64 field_value = 0;
4421 	struct x86_exception e;
4422 	struct vmcs12 *vmcs12;
4423 
4424 	if (!nested_vmx_check_permission(vcpu))
4425 		return 1;
4426 
4427 	if (vmx->nested.current_vmptr == -1ull)
4428 		return nested_vmx_failInvalid(vcpu);
4429 
4430 	if (vmx_instruction_info & (1u << 10))
4431 		field_value = kvm_register_readl(vcpu,
4432 			(((vmx_instruction_info) >> 3) & 0xf));
4433 	else {
4434 		if (get_vmx_mem_address(vcpu, exit_qualification,
4435 				vmx_instruction_info, false, &gva))
4436 			return 1;
4437 		if (kvm_read_guest_virt(vcpu, gva, &field_value,
4438 					(is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
4439 			kvm_inject_page_fault(vcpu, &e);
4440 			return 1;
4441 		}
4442 	}
4443 
4444 
4445 	field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4446 	/*
4447 	 * If the vCPU supports "VMWRITE to any supported field in the
4448 	 * VMCS," then the "read-only" fields are actually read/write.
4449 	 */
4450 	if (vmcs_field_readonly(field) &&
4451 	    !nested_cpu_has_vmwrite_any_field(vcpu))
4452 		return nested_vmx_failValid(vcpu,
4453 			VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
4454 
4455 	if (!is_guest_mode(vcpu))
4456 		vmcs12 = get_vmcs12(vcpu);
4457 	else {
4458 		/*
4459 		 * When vmcs->vmcs_link_pointer is -1ull, any VMWRITE
4460 		 * to shadowed-field sets the ALU flags for VMfailInvalid.
4461 		 */
4462 		if (get_vmcs12(vcpu)->vmcs_link_pointer == -1ull)
4463 			return nested_vmx_failInvalid(vcpu);
4464 		vmcs12 = get_shadow_vmcs12(vcpu);
4465 	}
4466 
4467 	if (vmcs12_write_any(vmcs12, field, field_value) < 0)
4468 		return nested_vmx_failValid(vcpu,
4469 			VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4470 
4471 	/*
4472 	 * Do not track vmcs12 dirty-state if in guest-mode
4473 	 * as we actually dirty shadow vmcs12 instead of vmcs12.
4474 	 */
4475 	if (!is_guest_mode(vcpu)) {
4476 		switch (field) {
4477 #define SHADOW_FIELD_RW(x) case x:
4478 #include "vmcs_shadow_fields.h"
4479 			/*
4480 			 * The fields that can be updated by L1 without a vmexit are
4481 			 * always updated in the vmcs02, the others go down the slow
4482 			 * path of prepare_vmcs02.
4483 			 */
4484 			break;
4485 		default:
4486 			vmx->nested.dirty_vmcs12 = true;
4487 			break;
4488 		}
4489 	}
4490 
4491 	return nested_vmx_succeed(vcpu);
4492 }
4493 
4494 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
4495 {
4496 	vmx->nested.current_vmptr = vmptr;
4497 	if (enable_shadow_vmcs) {
4498 		vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4499 			      SECONDARY_EXEC_SHADOW_VMCS);
4500 		vmcs_write64(VMCS_LINK_POINTER,
4501 			     __pa(vmx->vmcs01.shadow_vmcs));
4502 		vmx->nested.need_vmcs12_sync = true;
4503 	}
4504 	vmx->nested.dirty_vmcs12 = true;
4505 }
4506 
4507 /* Emulate the VMPTRLD instruction */
4508 static int handle_vmptrld(struct kvm_vcpu *vcpu)
4509 {
4510 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4511 	gpa_t vmptr;
4512 
4513 	if (!nested_vmx_check_permission(vcpu))
4514 		return 1;
4515 
4516 	if (nested_vmx_get_vmptr(vcpu, &vmptr))
4517 		return 1;
4518 
4519 	if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu)))
4520 		return nested_vmx_failValid(vcpu,
4521 			VMXERR_VMPTRLD_INVALID_ADDRESS);
4522 
4523 	if (vmptr == vmx->nested.vmxon_ptr)
4524 		return nested_vmx_failValid(vcpu,
4525 			VMXERR_VMPTRLD_VMXON_POINTER);
4526 
4527 	/* Forbid normal VMPTRLD if Enlightened version was used */
4528 	if (vmx->nested.hv_evmcs)
4529 		return 1;
4530 
4531 	if (vmx->nested.current_vmptr != vmptr) {
4532 		struct vmcs12 *new_vmcs12;
4533 		struct page *page;
4534 
4535 		page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
4536 		if (is_error_page(page)) {
4537 			/*
4538 			 * Reads from an unbacked page return all 1s,
4539 			 * which means that the 32 bits located at the
4540 			 * given physical address won't match the required
4541 			 * VMCS12_REVISION identifier.
4542 			 */
4543 			return nested_vmx_failValid(vcpu,
4544 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4545 		}
4546 		new_vmcs12 = kmap(page);
4547 		if (new_vmcs12->hdr.revision_id != VMCS12_REVISION ||
4548 		    (new_vmcs12->hdr.shadow_vmcs &&
4549 		     !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
4550 			kunmap(page);
4551 			kvm_release_page_clean(page);
4552 			return nested_vmx_failValid(vcpu,
4553 				VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
4554 		}
4555 
4556 		nested_release_vmcs12(vcpu);
4557 
4558 		/*
4559 		 * Load VMCS12 from guest memory since it is not already
4560 		 * cached.
4561 		 */
4562 		memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
4563 		kunmap(page);
4564 		kvm_release_page_clean(page);
4565 
4566 		set_current_vmptr(vmx, vmptr);
4567 	}
4568 
4569 	return nested_vmx_succeed(vcpu);
4570 }
4571 
4572 /* Emulate the VMPTRST instruction */
4573 static int handle_vmptrst(struct kvm_vcpu *vcpu)
4574 {
4575 	unsigned long exit_qual = vmcs_readl(EXIT_QUALIFICATION);
4576 	u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4577 	gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
4578 	struct x86_exception e;
4579 	gva_t gva;
4580 
4581 	if (!nested_vmx_check_permission(vcpu))
4582 		return 1;
4583 
4584 	if (unlikely(to_vmx(vcpu)->nested.hv_evmcs))
4585 		return 1;
4586 
4587 	if (get_vmx_mem_address(vcpu, exit_qual, instr_info, true, &gva))
4588 		return 1;
4589 	/* *_system ok, nested_vmx_check_permission has verified cpl=0 */
4590 	if (kvm_write_guest_virt_system(vcpu, gva, (void *)&current_vmptr,
4591 					sizeof(gpa_t), &e)) {
4592 		kvm_inject_page_fault(vcpu, &e);
4593 		return 1;
4594 	}
4595 	return nested_vmx_succeed(vcpu);
4596 }
4597 
4598 /* Emulate the INVEPT instruction */
4599 static int handle_invept(struct kvm_vcpu *vcpu)
4600 {
4601 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4602 	u32 vmx_instruction_info, types;
4603 	unsigned long type;
4604 	gva_t gva;
4605 	struct x86_exception e;
4606 	struct {
4607 		u64 eptp, gpa;
4608 	} operand;
4609 
4610 	if (!(vmx->nested.msrs.secondary_ctls_high &
4611 	      SECONDARY_EXEC_ENABLE_EPT) ||
4612 	    !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
4613 		kvm_queue_exception(vcpu, UD_VECTOR);
4614 		return 1;
4615 	}
4616 
4617 	if (!nested_vmx_check_permission(vcpu))
4618 		return 1;
4619 
4620 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4621 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4622 
4623 	types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
4624 
4625 	if (type >= 32 || !(types & (1 << type)))
4626 		return nested_vmx_failValid(vcpu,
4627 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4628 
4629 	/* According to the Intel VMX instruction reference, the memory
4630 	 * operand is read even if it isn't needed (e.g., for type==global)
4631 	 */
4632 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4633 			vmx_instruction_info, false, &gva))
4634 		return 1;
4635 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4636 		kvm_inject_page_fault(vcpu, &e);
4637 		return 1;
4638 	}
4639 
4640 	switch (type) {
4641 	case VMX_EPT_EXTENT_GLOBAL:
4642 	/*
4643 	 * TODO: track mappings and invalidate
4644 	 * single context requests appropriately
4645 	 */
4646 	case VMX_EPT_EXTENT_CONTEXT:
4647 		kvm_mmu_sync_roots(vcpu);
4648 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4649 		break;
4650 	default:
4651 		BUG_ON(1);
4652 		break;
4653 	}
4654 
4655 	return nested_vmx_succeed(vcpu);
4656 }
4657 
4658 static int handle_invvpid(struct kvm_vcpu *vcpu)
4659 {
4660 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4661 	u32 vmx_instruction_info;
4662 	unsigned long type, types;
4663 	gva_t gva;
4664 	struct x86_exception e;
4665 	struct {
4666 		u64 vpid;
4667 		u64 gla;
4668 	} operand;
4669 	u16 vpid02;
4670 
4671 	if (!(vmx->nested.msrs.secondary_ctls_high &
4672 	      SECONDARY_EXEC_ENABLE_VPID) ||
4673 			!(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
4674 		kvm_queue_exception(vcpu, UD_VECTOR);
4675 		return 1;
4676 	}
4677 
4678 	if (!nested_vmx_check_permission(vcpu))
4679 		return 1;
4680 
4681 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4682 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
4683 
4684 	types = (vmx->nested.msrs.vpid_caps &
4685 			VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
4686 
4687 	if (type >= 32 || !(types & (1 << type)))
4688 		return nested_vmx_failValid(vcpu,
4689 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4690 
4691 	/* according to the intel vmx instruction reference, the memory
4692 	 * operand is read even if it isn't needed (e.g., for type==global)
4693 	 */
4694 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4695 			vmx_instruction_info, false, &gva))
4696 		return 1;
4697 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
4698 		kvm_inject_page_fault(vcpu, &e);
4699 		return 1;
4700 	}
4701 	if (operand.vpid >> 16)
4702 		return nested_vmx_failValid(vcpu,
4703 			VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4704 
4705 	vpid02 = nested_get_vpid02(vcpu);
4706 	switch (type) {
4707 	case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
4708 		if (!operand.vpid ||
4709 		    is_noncanonical_address(operand.gla, vcpu))
4710 			return nested_vmx_failValid(vcpu,
4711 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4712 		if (cpu_has_vmx_invvpid_individual_addr()) {
4713 			__invvpid(VMX_VPID_EXTENT_INDIVIDUAL_ADDR,
4714 				vpid02, operand.gla);
4715 		} else
4716 			__vmx_flush_tlb(vcpu, vpid02, false);
4717 		break;
4718 	case VMX_VPID_EXTENT_SINGLE_CONTEXT:
4719 	case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
4720 		if (!operand.vpid)
4721 			return nested_vmx_failValid(vcpu,
4722 				VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
4723 		__vmx_flush_tlb(vcpu, vpid02, false);
4724 		break;
4725 	case VMX_VPID_EXTENT_ALL_CONTEXT:
4726 		__vmx_flush_tlb(vcpu, vpid02, false);
4727 		break;
4728 	default:
4729 		WARN_ON_ONCE(1);
4730 		return kvm_skip_emulated_instruction(vcpu);
4731 	}
4732 
4733 	return nested_vmx_succeed(vcpu);
4734 }
4735 
4736 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
4737 				     struct vmcs12 *vmcs12)
4738 {
4739 	u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
4740 	u64 address;
4741 	bool accessed_dirty;
4742 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4743 
4744 	if (!nested_cpu_has_eptp_switching(vmcs12) ||
4745 	    !nested_cpu_has_ept(vmcs12))
4746 		return 1;
4747 
4748 	if (index >= VMFUNC_EPTP_ENTRIES)
4749 		return 1;
4750 
4751 
4752 	if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
4753 				     &address, index * 8, 8))
4754 		return 1;
4755 
4756 	accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
4757 
4758 	/*
4759 	 * If the (L2) guest does a vmfunc to the currently
4760 	 * active ept pointer, we don't have to do anything else
4761 	 */
4762 	if (vmcs12->ept_pointer != address) {
4763 		if (!valid_ept_address(vcpu, address))
4764 			return 1;
4765 
4766 		kvm_mmu_unload(vcpu);
4767 		mmu->ept_ad = accessed_dirty;
4768 		mmu->mmu_role.base.ad_disabled = !accessed_dirty;
4769 		vmcs12->ept_pointer = address;
4770 		/*
4771 		 * TODO: Check what's the correct approach in case
4772 		 * mmu reload fails. Currently, we just let the next
4773 		 * reload potentially fail
4774 		 */
4775 		kvm_mmu_reload(vcpu);
4776 	}
4777 
4778 	return 0;
4779 }
4780 
4781 static int handle_vmfunc(struct kvm_vcpu *vcpu)
4782 {
4783 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4784 	struct vmcs12 *vmcs12;
4785 	u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
4786 
4787 	/*
4788 	 * VMFUNC is only supported for nested guests, but we always enable the
4789 	 * secondary control for simplicity; for non-nested mode, fake that we
4790 	 * didn't by injecting #UD.
4791 	 */
4792 	if (!is_guest_mode(vcpu)) {
4793 		kvm_queue_exception(vcpu, UD_VECTOR);
4794 		return 1;
4795 	}
4796 
4797 	vmcs12 = get_vmcs12(vcpu);
4798 	if ((vmcs12->vm_function_control & (1 << function)) == 0)
4799 		goto fail;
4800 
4801 	switch (function) {
4802 	case 0:
4803 		if (nested_vmx_eptp_switching(vcpu, vmcs12))
4804 			goto fail;
4805 		break;
4806 	default:
4807 		goto fail;
4808 	}
4809 	return kvm_skip_emulated_instruction(vcpu);
4810 
4811 fail:
4812 	nested_vmx_vmexit(vcpu, vmx->exit_reason,
4813 			  vmcs_read32(VM_EXIT_INTR_INFO),
4814 			  vmcs_readl(EXIT_QUALIFICATION));
4815 	return 1;
4816 }
4817 
4818 
4819 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
4820 				       struct vmcs12 *vmcs12)
4821 {
4822 	unsigned long exit_qualification;
4823 	gpa_t bitmap, last_bitmap;
4824 	unsigned int port;
4825 	int size;
4826 	u8 b;
4827 
4828 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
4829 		return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
4830 
4831 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4832 
4833 	port = exit_qualification >> 16;
4834 	size = (exit_qualification & 7) + 1;
4835 
4836 	last_bitmap = (gpa_t)-1;
4837 	b = -1;
4838 
4839 	while (size > 0) {
4840 		if (port < 0x8000)
4841 			bitmap = vmcs12->io_bitmap_a;
4842 		else if (port < 0x10000)
4843 			bitmap = vmcs12->io_bitmap_b;
4844 		else
4845 			return true;
4846 		bitmap += (port & 0x7fff) / 8;
4847 
4848 		if (last_bitmap != bitmap)
4849 			if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
4850 				return true;
4851 		if (b & (1 << (port & 7)))
4852 			return true;
4853 
4854 		port++;
4855 		size--;
4856 		last_bitmap = bitmap;
4857 	}
4858 
4859 	return false;
4860 }
4861 
4862 /*
4863  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
4864  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
4865  * disinterest in the current event (read or write a specific MSR) by using an
4866  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
4867  */
4868 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
4869 	struct vmcs12 *vmcs12, u32 exit_reason)
4870 {
4871 	u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
4872 	gpa_t bitmap;
4873 
4874 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
4875 		return true;
4876 
4877 	/*
4878 	 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
4879 	 * for the four combinations of read/write and low/high MSR numbers.
4880 	 * First we need to figure out which of the four to use:
4881 	 */
4882 	bitmap = vmcs12->msr_bitmap;
4883 	if (exit_reason == EXIT_REASON_MSR_WRITE)
4884 		bitmap += 2048;
4885 	if (msr_index >= 0xc0000000) {
4886 		msr_index -= 0xc0000000;
4887 		bitmap += 1024;
4888 	}
4889 
4890 	/* Then read the msr_index'th bit from this bitmap: */
4891 	if (msr_index < 1024*8) {
4892 		unsigned char b;
4893 		if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
4894 			return true;
4895 		return 1 & (b >> (msr_index & 7));
4896 	} else
4897 		return true; /* let L1 handle the wrong parameter */
4898 }
4899 
4900 /*
4901  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
4902  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
4903  * intercept (via guest_host_mask etc.) the current event.
4904  */
4905 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
4906 	struct vmcs12 *vmcs12)
4907 {
4908 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4909 	int cr = exit_qualification & 15;
4910 	int reg;
4911 	unsigned long val;
4912 
4913 	switch ((exit_qualification >> 4) & 3) {
4914 	case 0: /* mov to cr */
4915 		reg = (exit_qualification >> 8) & 15;
4916 		val = kvm_register_readl(vcpu, reg);
4917 		switch (cr) {
4918 		case 0:
4919 			if (vmcs12->cr0_guest_host_mask &
4920 			    (val ^ vmcs12->cr0_read_shadow))
4921 				return true;
4922 			break;
4923 		case 3:
4924 			if ((vmcs12->cr3_target_count >= 1 &&
4925 					vmcs12->cr3_target_value0 == val) ||
4926 				(vmcs12->cr3_target_count >= 2 &&
4927 					vmcs12->cr3_target_value1 == val) ||
4928 				(vmcs12->cr3_target_count >= 3 &&
4929 					vmcs12->cr3_target_value2 == val) ||
4930 				(vmcs12->cr3_target_count >= 4 &&
4931 					vmcs12->cr3_target_value3 == val))
4932 				return false;
4933 			if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
4934 				return true;
4935 			break;
4936 		case 4:
4937 			if (vmcs12->cr4_guest_host_mask &
4938 			    (vmcs12->cr4_read_shadow ^ val))
4939 				return true;
4940 			break;
4941 		case 8:
4942 			if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
4943 				return true;
4944 			break;
4945 		}
4946 		break;
4947 	case 2: /* clts */
4948 		if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
4949 		    (vmcs12->cr0_read_shadow & X86_CR0_TS))
4950 			return true;
4951 		break;
4952 	case 1: /* mov from cr */
4953 		switch (cr) {
4954 		case 3:
4955 			if (vmcs12->cpu_based_vm_exec_control &
4956 			    CPU_BASED_CR3_STORE_EXITING)
4957 				return true;
4958 			break;
4959 		case 8:
4960 			if (vmcs12->cpu_based_vm_exec_control &
4961 			    CPU_BASED_CR8_STORE_EXITING)
4962 				return true;
4963 			break;
4964 		}
4965 		break;
4966 	case 3: /* lmsw */
4967 		/*
4968 		 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
4969 		 * cr0. Other attempted changes are ignored, with no exit.
4970 		 */
4971 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4972 		if (vmcs12->cr0_guest_host_mask & 0xe &
4973 		    (val ^ vmcs12->cr0_read_shadow))
4974 			return true;
4975 		if ((vmcs12->cr0_guest_host_mask & 0x1) &&
4976 		    !(vmcs12->cr0_read_shadow & 0x1) &&
4977 		    (val & 0x1))
4978 			return true;
4979 		break;
4980 	}
4981 	return false;
4982 }
4983 
4984 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
4985 	struct vmcs12 *vmcs12, gpa_t bitmap)
4986 {
4987 	u32 vmx_instruction_info;
4988 	unsigned long field;
4989 	u8 b;
4990 
4991 	if (!nested_cpu_has_shadow_vmcs(vmcs12))
4992 		return true;
4993 
4994 	/* Decode instruction info and find the field to access */
4995 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4996 	field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4997 
4998 	/* Out-of-range fields always cause a VM exit from L2 to L1 */
4999 	if (field >> 15)
5000 		return true;
5001 
5002 	if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
5003 		return true;
5004 
5005 	return 1 & (b >> (field & 7));
5006 }
5007 
5008 /*
5009  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5010  * should handle it ourselves in L0 (and then continue L2). Only call this
5011  * when in is_guest_mode (L2).
5012  */
5013 bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
5014 {
5015 	u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5016 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5017 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5018 
5019 	if (vmx->nested.nested_run_pending)
5020 		return false;
5021 
5022 	if (unlikely(vmx->fail)) {
5023 		pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5024 				    vmcs_read32(VM_INSTRUCTION_ERROR));
5025 		return true;
5026 	}
5027 
5028 	/*
5029 	 * The host physical addresses of some pages of guest memory
5030 	 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5031 	 * Page). The CPU may write to these pages via their host
5032 	 * physical address while L2 is running, bypassing any
5033 	 * address-translation-based dirty tracking (e.g. EPT write
5034 	 * protection).
5035 	 *
5036 	 * Mark them dirty on every exit from L2 to prevent them from
5037 	 * getting out of sync with dirty tracking.
5038 	 */
5039 	nested_mark_vmcs12_pages_dirty(vcpu);
5040 
5041 	trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
5042 				vmcs_readl(EXIT_QUALIFICATION),
5043 				vmx->idt_vectoring_info,
5044 				intr_info,
5045 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5046 				KVM_ISA_VMX);
5047 
5048 	switch (exit_reason) {
5049 	case EXIT_REASON_EXCEPTION_NMI:
5050 		if (is_nmi(intr_info))
5051 			return false;
5052 		else if (is_page_fault(intr_info))
5053 			return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
5054 		else if (is_debug(intr_info) &&
5055 			 vcpu->guest_debug &
5056 			 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5057 			return false;
5058 		else if (is_breakpoint(intr_info) &&
5059 			 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5060 			return false;
5061 		return vmcs12->exception_bitmap &
5062 				(1u << (intr_info & INTR_INFO_VECTOR_MASK));
5063 	case EXIT_REASON_EXTERNAL_INTERRUPT:
5064 		return false;
5065 	case EXIT_REASON_TRIPLE_FAULT:
5066 		return true;
5067 	case EXIT_REASON_PENDING_INTERRUPT:
5068 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
5069 	case EXIT_REASON_NMI_WINDOW:
5070 		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
5071 	case EXIT_REASON_TASK_SWITCH:
5072 		return true;
5073 	case EXIT_REASON_CPUID:
5074 		return true;
5075 	case EXIT_REASON_HLT:
5076 		return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5077 	case EXIT_REASON_INVD:
5078 		return true;
5079 	case EXIT_REASON_INVLPG:
5080 		return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5081 	case EXIT_REASON_RDPMC:
5082 		return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5083 	case EXIT_REASON_RDRAND:
5084 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
5085 	case EXIT_REASON_RDSEED:
5086 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
5087 	case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
5088 		return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5089 	case EXIT_REASON_VMREAD:
5090 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5091 			vmcs12->vmread_bitmap);
5092 	case EXIT_REASON_VMWRITE:
5093 		return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
5094 			vmcs12->vmwrite_bitmap);
5095 	case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5096 	case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5097 	case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
5098 	case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5099 	case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
5100 		/*
5101 		 * VMX instructions trap unconditionally. This allows L1 to
5102 		 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5103 		 */
5104 		return true;
5105 	case EXIT_REASON_CR_ACCESS:
5106 		return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5107 	case EXIT_REASON_DR_ACCESS:
5108 		return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5109 	case EXIT_REASON_IO_INSTRUCTION:
5110 		return nested_vmx_exit_handled_io(vcpu, vmcs12);
5111 	case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
5112 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
5113 	case EXIT_REASON_MSR_READ:
5114 	case EXIT_REASON_MSR_WRITE:
5115 		return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5116 	case EXIT_REASON_INVALID_STATE:
5117 		return true;
5118 	case EXIT_REASON_MWAIT_INSTRUCTION:
5119 		return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5120 	case EXIT_REASON_MONITOR_TRAP_FLAG:
5121 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
5122 	case EXIT_REASON_MONITOR_INSTRUCTION:
5123 		return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5124 	case EXIT_REASON_PAUSE_INSTRUCTION:
5125 		return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5126 			nested_cpu_has2(vmcs12,
5127 				SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5128 	case EXIT_REASON_MCE_DURING_VMENTRY:
5129 		return false;
5130 	case EXIT_REASON_TPR_BELOW_THRESHOLD:
5131 		return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
5132 	case EXIT_REASON_APIC_ACCESS:
5133 	case EXIT_REASON_APIC_WRITE:
5134 	case EXIT_REASON_EOI_INDUCED:
5135 		/*
5136 		 * The controls for "virtualize APIC accesses," "APIC-
5137 		 * register virtualization," and "virtual-interrupt
5138 		 * delivery" only come from vmcs12.
5139 		 */
5140 		return true;
5141 	case EXIT_REASON_EPT_VIOLATION:
5142 		/*
5143 		 * L0 always deals with the EPT violation. If nested EPT is
5144 		 * used, and the nested mmu code discovers that the address is
5145 		 * missing in the guest EPT table (EPT12), the EPT violation
5146 		 * will be injected with nested_ept_inject_page_fault()
5147 		 */
5148 		return false;
5149 	case EXIT_REASON_EPT_MISCONFIG:
5150 		/*
5151 		 * L2 never uses directly L1's EPT, but rather L0's own EPT
5152 		 * table (shadow on EPT) or a merged EPT table that L0 built
5153 		 * (EPT on EPT). So any problems with the structure of the
5154 		 * table is L0's fault.
5155 		 */
5156 		return false;
5157 	case EXIT_REASON_INVPCID:
5158 		return
5159 			nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
5160 			nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5161 	case EXIT_REASON_WBINVD:
5162 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5163 	case EXIT_REASON_XSETBV:
5164 		return true;
5165 	case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
5166 		/*
5167 		 * This should never happen, since it is not possible to
5168 		 * set XSS to a non-zero value---neither in L1 nor in L2.
5169 		 * If if it were, XSS would have to be checked against
5170 		 * the XSS exit bitmap in vmcs12.
5171 		 */
5172 		return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
5173 	case EXIT_REASON_PREEMPTION_TIMER:
5174 		return false;
5175 	case EXIT_REASON_PML_FULL:
5176 		/* We emulate PML support to L1. */
5177 		return false;
5178 	case EXIT_REASON_VMFUNC:
5179 		/* VM functions are emulated through L2->L0 vmexits. */
5180 		return false;
5181 	case EXIT_REASON_ENCLS:
5182 		/* SGX is never exposed to L1 */
5183 		return false;
5184 	default:
5185 		return true;
5186 	}
5187 }
5188 
5189 
5190 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
5191 				struct kvm_nested_state __user *user_kvm_nested_state,
5192 				u32 user_data_size)
5193 {
5194 	struct vcpu_vmx *vmx;
5195 	struct vmcs12 *vmcs12;
5196 	struct kvm_nested_state kvm_state = {
5197 		.flags = 0,
5198 		.format = 0,
5199 		.size = sizeof(kvm_state),
5200 		.vmx.vmxon_pa = -1ull,
5201 		.vmx.vmcs_pa = -1ull,
5202 	};
5203 
5204 	if (!vcpu)
5205 		return kvm_state.size + 2 * VMCS12_SIZE;
5206 
5207 	vmx = to_vmx(vcpu);
5208 	vmcs12 = get_vmcs12(vcpu);
5209 
5210 	if (nested_vmx_allowed(vcpu) && vmx->nested.enlightened_vmcs_enabled)
5211 		kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
5212 
5213 	if (nested_vmx_allowed(vcpu) &&
5214 	    (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
5215 		kvm_state.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
5216 		kvm_state.vmx.vmcs_pa = vmx->nested.current_vmptr;
5217 
5218 		if (vmx_has_valid_vmcs12(vcpu)) {
5219 			kvm_state.size += VMCS12_SIZE;
5220 
5221 			if (is_guest_mode(vcpu) &&
5222 			    nested_cpu_has_shadow_vmcs(vmcs12) &&
5223 			    vmcs12->vmcs_link_pointer != -1ull)
5224 				kvm_state.size += VMCS12_SIZE;
5225 		}
5226 
5227 		if (vmx->nested.smm.vmxon)
5228 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
5229 
5230 		if (vmx->nested.smm.guest_mode)
5231 			kvm_state.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
5232 
5233 		if (is_guest_mode(vcpu)) {
5234 			kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
5235 
5236 			if (vmx->nested.nested_run_pending)
5237 				kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
5238 		}
5239 	}
5240 
5241 	if (user_data_size < kvm_state.size)
5242 		goto out;
5243 
5244 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
5245 		return -EFAULT;
5246 
5247 	if (!vmx_has_valid_vmcs12(vcpu))
5248 		goto out;
5249 
5250 	/*
5251 	 * When running L2, the authoritative vmcs12 state is in the
5252 	 * vmcs02. When running L1, the authoritative vmcs12 state is
5253 	 * in the shadow or enlightened vmcs linked to vmcs01, unless
5254 	 * need_vmcs12_sync is set, in which case, the authoritative
5255 	 * vmcs12 state is in the vmcs12 already.
5256 	 */
5257 	if (is_guest_mode(vcpu)) {
5258 		sync_vmcs12(vcpu, vmcs12);
5259 	} else if (!vmx->nested.need_vmcs12_sync) {
5260 		if (vmx->nested.hv_evmcs)
5261 			copy_enlightened_to_vmcs12(vmx);
5262 		else if (enable_shadow_vmcs)
5263 			copy_shadow_to_vmcs12(vmx);
5264 	}
5265 
5266 	if (copy_to_user(user_kvm_nested_state->data, vmcs12, sizeof(*vmcs12)))
5267 		return -EFAULT;
5268 
5269 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5270 	    vmcs12->vmcs_link_pointer != -1ull) {
5271 		if (copy_to_user(user_kvm_nested_state->data + VMCS12_SIZE,
5272 				 get_shadow_vmcs12(vcpu), sizeof(*vmcs12)))
5273 			return -EFAULT;
5274 	}
5275 
5276 out:
5277 	return kvm_state.size;
5278 }
5279 
5280 /*
5281  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
5282  */
5283 void vmx_leave_nested(struct kvm_vcpu *vcpu)
5284 {
5285 	if (is_guest_mode(vcpu)) {
5286 		to_vmx(vcpu)->nested.nested_run_pending = 0;
5287 		nested_vmx_vmexit(vcpu, -1, 0, 0);
5288 	}
5289 	free_nested(vcpu);
5290 }
5291 
5292 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
5293 				struct kvm_nested_state __user *user_kvm_nested_state,
5294 				struct kvm_nested_state *kvm_state)
5295 {
5296 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5297 	struct vmcs12 *vmcs12;
5298 	u32 exit_qual;
5299 	int ret;
5300 
5301 	if (kvm_state->format != 0)
5302 		return -EINVAL;
5303 
5304 	if (kvm_state->flags & KVM_STATE_NESTED_EVMCS)
5305 		nested_enable_evmcs(vcpu, NULL);
5306 
5307 	if (!nested_vmx_allowed(vcpu))
5308 		return kvm_state->vmx.vmxon_pa == -1ull ? 0 : -EINVAL;
5309 
5310 	if (kvm_state->vmx.vmxon_pa == -1ull) {
5311 		if (kvm_state->vmx.smm.flags)
5312 			return -EINVAL;
5313 
5314 		if (kvm_state->vmx.vmcs_pa != -1ull)
5315 			return -EINVAL;
5316 
5317 		vmx_leave_nested(vcpu);
5318 		return 0;
5319 	}
5320 
5321 	if (!page_address_valid(vcpu, kvm_state->vmx.vmxon_pa))
5322 		return -EINVAL;
5323 
5324 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5325 	    (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5326 		return -EINVAL;
5327 
5328 	if (kvm_state->vmx.smm.flags &
5329 	    ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
5330 		return -EINVAL;
5331 
5332 	/*
5333 	 * SMM temporarily disables VMX, so we cannot be in guest mode,
5334 	 * nor can VMLAUNCH/VMRESUME be pending.  Outside SMM, SMM flags
5335 	 * must be zero.
5336 	 */
5337 	if (is_smm(vcpu) ? kvm_state->flags : kvm_state->vmx.smm.flags)
5338 		return -EINVAL;
5339 
5340 	if ((kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
5341 	    !(kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
5342 		return -EINVAL;
5343 
5344 	vmx_leave_nested(vcpu);
5345 	if (kvm_state->vmx.vmxon_pa == -1ull)
5346 		return 0;
5347 
5348 	vmx->nested.vmxon_ptr = kvm_state->vmx.vmxon_pa;
5349 	ret = enter_vmx_operation(vcpu);
5350 	if (ret)
5351 		return ret;
5352 
5353 	/* Empty 'VMXON' state is permitted */
5354 	if (kvm_state->size < sizeof(kvm_state) + sizeof(*vmcs12))
5355 		return 0;
5356 
5357 	if (kvm_state->vmx.vmcs_pa != -1ull) {
5358 		if (kvm_state->vmx.vmcs_pa == kvm_state->vmx.vmxon_pa ||
5359 		    !page_address_valid(vcpu, kvm_state->vmx.vmcs_pa))
5360 			return -EINVAL;
5361 
5362 		set_current_vmptr(vmx, kvm_state->vmx.vmcs_pa);
5363 	} else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
5364 		/*
5365 		 * Sync eVMCS upon entry as we may not have
5366 		 * HV_X64_MSR_VP_ASSIST_PAGE set up yet.
5367 		 */
5368 		vmx->nested.need_vmcs12_sync = true;
5369 	} else {
5370 		return -EINVAL;
5371 	}
5372 
5373 	if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
5374 		vmx->nested.smm.vmxon = true;
5375 		vmx->nested.vmxon = false;
5376 
5377 		if (kvm_state->vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
5378 			vmx->nested.smm.guest_mode = true;
5379 	}
5380 
5381 	vmcs12 = get_vmcs12(vcpu);
5382 	if (copy_from_user(vmcs12, user_kvm_nested_state->data, sizeof(*vmcs12)))
5383 		return -EFAULT;
5384 
5385 	if (vmcs12->hdr.revision_id != VMCS12_REVISION)
5386 		return -EINVAL;
5387 
5388 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
5389 		return 0;
5390 
5391 	vmx->nested.nested_run_pending =
5392 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
5393 
5394 	if (nested_cpu_has_shadow_vmcs(vmcs12) &&
5395 	    vmcs12->vmcs_link_pointer != -1ull) {
5396 		struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
5397 
5398 		if (kvm_state->size < sizeof(kvm_state) + 2 * sizeof(*vmcs12))
5399 			return -EINVAL;
5400 
5401 		if (copy_from_user(shadow_vmcs12,
5402 				   user_kvm_nested_state->data + VMCS12_SIZE,
5403 				   sizeof(*vmcs12)))
5404 			return -EFAULT;
5405 
5406 		if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
5407 		    !shadow_vmcs12->hdr.shadow_vmcs)
5408 			return -EINVAL;
5409 	}
5410 
5411 	if (nested_vmx_check_vmentry_prereqs(vcpu, vmcs12) ||
5412 	    nested_vmx_check_vmentry_postreqs(vcpu, vmcs12, &exit_qual))
5413 		return -EINVAL;
5414 
5415 	vmx->nested.dirty_vmcs12 = true;
5416 	ret = nested_vmx_enter_non_root_mode(vcpu, false);
5417 	if (ret)
5418 		return -EINVAL;
5419 
5420 	return 0;
5421 }
5422 
5423 void nested_vmx_vcpu_setup(void)
5424 {
5425 	if (enable_shadow_vmcs) {
5426 		/*
5427 		 * At vCPU creation, "VMWRITE to any supported field
5428 		 * in the VMCS" is supported, so use the more
5429 		 * permissive vmx_vmread_bitmap to specify both read
5430 		 * and write permissions for the shadow VMCS.
5431 		 */
5432 		vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5433 		vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmread_bitmap));
5434 	}
5435 }
5436 
5437 /*
5438  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
5439  * returned for the various VMX controls MSRs when nested VMX is enabled.
5440  * The same values should also be used to verify that vmcs12 control fields are
5441  * valid during nested entry from L1 to L2.
5442  * Each of these control msrs has a low and high 32-bit half: A low bit is on
5443  * if the corresponding bit in the (32-bit) control field *must* be on, and a
5444  * bit in the high half is on if the corresponding bit in the control field
5445  * may be on. See also vmx_control_verify().
5446  */
5447 void nested_vmx_setup_ctls_msrs(struct nested_vmx_msrs *msrs, u32 ept_caps,
5448 				bool apicv)
5449 {
5450 	/*
5451 	 * Note that as a general rule, the high half of the MSRs (bits in
5452 	 * the control fields which may be 1) should be initialized by the
5453 	 * intersection of the underlying hardware's MSR (i.e., features which
5454 	 * can be supported) and the list of features we want to expose -
5455 	 * because they are known to be properly supported in our code.
5456 	 * Also, usually, the low half of the MSRs (bits which must be 1) can
5457 	 * be set to 0, meaning that L1 may turn off any of these bits. The
5458 	 * reason is that if one of these bits is necessary, it will appear
5459 	 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
5460 	 * fields of vmcs01 and vmcs02, will turn these bits off - and
5461 	 * nested_vmx_exit_reflected() will not pass related exits to L1.
5462 	 * These rules have exceptions below.
5463 	 */
5464 
5465 	/* pin-based controls */
5466 	rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
5467 		msrs->pinbased_ctls_low,
5468 		msrs->pinbased_ctls_high);
5469 	msrs->pinbased_ctls_low |=
5470 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5471 	msrs->pinbased_ctls_high &=
5472 		PIN_BASED_EXT_INTR_MASK |
5473 		PIN_BASED_NMI_EXITING |
5474 		PIN_BASED_VIRTUAL_NMIS |
5475 		(apicv ? PIN_BASED_POSTED_INTR : 0);
5476 	msrs->pinbased_ctls_high |=
5477 		PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5478 		PIN_BASED_VMX_PREEMPTION_TIMER;
5479 
5480 	/* exit controls */
5481 	rdmsr(MSR_IA32_VMX_EXIT_CTLS,
5482 		msrs->exit_ctls_low,
5483 		msrs->exit_ctls_high);
5484 	msrs->exit_ctls_low =
5485 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
5486 
5487 	msrs->exit_ctls_high &=
5488 #ifdef CONFIG_X86_64
5489 		VM_EXIT_HOST_ADDR_SPACE_SIZE |
5490 #endif
5491 		VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
5492 	msrs->exit_ctls_high |=
5493 		VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
5494 		VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
5495 		VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
5496 
5497 	/* We support free control of debug control saving. */
5498 	msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
5499 
5500 	/* entry controls */
5501 	rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
5502 		msrs->entry_ctls_low,
5503 		msrs->entry_ctls_high);
5504 	msrs->entry_ctls_low =
5505 		VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
5506 	msrs->entry_ctls_high &=
5507 #ifdef CONFIG_X86_64
5508 		VM_ENTRY_IA32E_MODE |
5509 #endif
5510 		VM_ENTRY_LOAD_IA32_PAT;
5511 	msrs->entry_ctls_high |=
5512 		(VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
5513 
5514 	/* We support free control of debug control loading. */
5515 	msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
5516 
5517 	/* cpu-based controls */
5518 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
5519 		msrs->procbased_ctls_low,
5520 		msrs->procbased_ctls_high);
5521 	msrs->procbased_ctls_low =
5522 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
5523 	msrs->procbased_ctls_high &=
5524 		CPU_BASED_VIRTUAL_INTR_PENDING |
5525 		CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
5526 		CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
5527 		CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
5528 		CPU_BASED_CR3_STORE_EXITING |
5529 #ifdef CONFIG_X86_64
5530 		CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
5531 #endif
5532 		CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
5533 		CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
5534 		CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
5535 		CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
5536 		CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
5537 	/*
5538 	 * We can allow some features even when not supported by the
5539 	 * hardware. For example, L1 can specify an MSR bitmap - and we
5540 	 * can use it to avoid exits to L1 - even when L0 runs L2
5541 	 * without MSR bitmaps.
5542 	 */
5543 	msrs->procbased_ctls_high |=
5544 		CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
5545 		CPU_BASED_USE_MSR_BITMAPS;
5546 
5547 	/* We support free control of CR3 access interception. */
5548 	msrs->procbased_ctls_low &=
5549 		~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
5550 
5551 	/*
5552 	 * secondary cpu-based controls.  Do not include those that
5553 	 * depend on CPUID bits, they are added later by vmx_cpuid_update.
5554 	 */
5555 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
5556 		msrs->secondary_ctls_low,
5557 		msrs->secondary_ctls_high);
5558 	msrs->secondary_ctls_low = 0;
5559 	msrs->secondary_ctls_high &=
5560 		SECONDARY_EXEC_DESC |
5561 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
5562 		SECONDARY_EXEC_APIC_REGISTER_VIRT |
5563 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
5564 		SECONDARY_EXEC_WBINVD_EXITING;
5565 
5566 	/*
5567 	 * We can emulate "VMCS shadowing," even if the hardware
5568 	 * doesn't support it.
5569 	 */
5570 	msrs->secondary_ctls_high |=
5571 		SECONDARY_EXEC_SHADOW_VMCS;
5572 
5573 	if (enable_ept) {
5574 		/* nested EPT: emulate EPT also to L1 */
5575 		msrs->secondary_ctls_high |=
5576 			SECONDARY_EXEC_ENABLE_EPT;
5577 		msrs->ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
5578 			 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
5579 		if (cpu_has_vmx_ept_execute_only())
5580 			msrs->ept_caps |=
5581 				VMX_EPT_EXECUTE_ONLY_BIT;
5582 		msrs->ept_caps &= ept_caps;
5583 		msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
5584 			VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
5585 			VMX_EPT_1GB_PAGE_BIT;
5586 		if (enable_ept_ad_bits) {
5587 			msrs->secondary_ctls_high |=
5588 				SECONDARY_EXEC_ENABLE_PML;
5589 			msrs->ept_caps |= VMX_EPT_AD_BIT;
5590 		}
5591 	}
5592 
5593 	if (cpu_has_vmx_vmfunc()) {
5594 		msrs->secondary_ctls_high |=
5595 			SECONDARY_EXEC_ENABLE_VMFUNC;
5596 		/*
5597 		 * Advertise EPTP switching unconditionally
5598 		 * since we emulate it
5599 		 */
5600 		if (enable_ept)
5601 			msrs->vmfunc_controls =
5602 				VMX_VMFUNC_EPTP_SWITCHING;
5603 	}
5604 
5605 	/*
5606 	 * Old versions of KVM use the single-context version without
5607 	 * checking for support, so declare that it is supported even
5608 	 * though it is treated as global context.  The alternative is
5609 	 * not failing the single-context invvpid, and it is worse.
5610 	 */
5611 	if (enable_vpid) {
5612 		msrs->secondary_ctls_high |=
5613 			SECONDARY_EXEC_ENABLE_VPID;
5614 		msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
5615 			VMX_VPID_EXTENT_SUPPORTED_MASK;
5616 	}
5617 
5618 	if (enable_unrestricted_guest)
5619 		msrs->secondary_ctls_high |=
5620 			SECONDARY_EXEC_UNRESTRICTED_GUEST;
5621 
5622 	if (flexpriority_enabled)
5623 		msrs->secondary_ctls_high |=
5624 			SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5625 
5626 	/* miscellaneous data */
5627 	rdmsr(MSR_IA32_VMX_MISC,
5628 		msrs->misc_low,
5629 		msrs->misc_high);
5630 	msrs->misc_low &= VMX_MISC_SAVE_EFER_LMA;
5631 	msrs->misc_low |=
5632 		MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
5633 		VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
5634 		VMX_MISC_ACTIVITY_HLT;
5635 	msrs->misc_high = 0;
5636 
5637 	/*
5638 	 * This MSR reports some information about VMX support. We
5639 	 * should return information about the VMX we emulate for the
5640 	 * guest, and the VMCS structure we give it - not about the
5641 	 * VMX support of the underlying hardware.
5642 	 */
5643 	msrs->basic =
5644 		VMCS12_REVISION |
5645 		VMX_BASIC_TRUE_CTLS |
5646 		((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
5647 		(VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
5648 
5649 	if (cpu_has_vmx_basic_inout())
5650 		msrs->basic |= VMX_BASIC_INOUT;
5651 
5652 	/*
5653 	 * These MSRs specify bits which the guest must keep fixed on
5654 	 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
5655 	 * We picked the standard core2 setting.
5656 	 */
5657 #define VMXON_CR0_ALWAYSON     (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
5658 #define VMXON_CR4_ALWAYSON     X86_CR4_VMXE
5659 	msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
5660 	msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
5661 
5662 	/* These MSRs specify bits which the guest must keep fixed off. */
5663 	rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
5664 	rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
5665 
5666 	/* highest index: VMX_PREEMPTION_TIMER_VALUE */
5667 	msrs->vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
5668 }
5669 
5670 void nested_vmx_hardware_unsetup(void)
5671 {
5672 	int i;
5673 
5674 	if (enable_shadow_vmcs) {
5675 		for (i = 0; i < VMX_BITMAP_NR; i++)
5676 			free_page((unsigned long)vmx_bitmap[i]);
5677 	}
5678 }
5679 
5680 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
5681 {
5682 	int i;
5683 
5684 	if (!cpu_has_vmx_shadow_vmcs())
5685 		enable_shadow_vmcs = 0;
5686 	if (enable_shadow_vmcs) {
5687 		for (i = 0; i < VMX_BITMAP_NR; i++) {
5688 			vmx_bitmap[i] = (unsigned long *)
5689 				__get_free_page(GFP_KERNEL);
5690 			if (!vmx_bitmap[i]) {
5691 				nested_vmx_hardware_unsetup();
5692 				return -ENOMEM;
5693 			}
5694 		}
5695 
5696 		init_vmcs_shadow_fields();
5697 	}
5698 
5699 	exit_handlers[EXIT_REASON_VMCLEAR]	= handle_vmclear,
5700 	exit_handlers[EXIT_REASON_VMLAUNCH]	= handle_vmlaunch,
5701 	exit_handlers[EXIT_REASON_VMPTRLD]	= handle_vmptrld,
5702 	exit_handlers[EXIT_REASON_VMPTRST]	= handle_vmptrst,
5703 	exit_handlers[EXIT_REASON_VMREAD]	= handle_vmread,
5704 	exit_handlers[EXIT_REASON_VMRESUME]	= handle_vmresume,
5705 	exit_handlers[EXIT_REASON_VMWRITE]	= handle_vmwrite,
5706 	exit_handlers[EXIT_REASON_VMOFF]	= handle_vmoff,
5707 	exit_handlers[EXIT_REASON_VMON]		= handle_vmon,
5708 	exit_handlers[EXIT_REASON_INVEPT]	= handle_invept,
5709 	exit_handlers[EXIT_REASON_INVVPID]	= handle_invvpid,
5710 	exit_handlers[EXIT_REASON_VMFUNC]	= handle_vmfunc,
5711 
5712 	kvm_x86_ops->check_nested_events = vmx_check_nested_events;
5713 	kvm_x86_ops->get_nested_state = vmx_get_nested_state;
5714 	kvm_x86_ops->set_nested_state = vmx_set_nested_state;
5715 	kvm_x86_ops->get_vmcs12_pages = nested_get_vmcs12_pages,
5716 	kvm_x86_ops->nested_enable_evmcs = nested_enable_evmcs;
5717 	kvm_x86_ops->nested_get_evmcs_version = nested_get_evmcs_version;
5718 
5719 	return 0;
5720 }
5721