1 // SPDX-License-Identifier: GPL-2.0
2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3
4 #include <linux/objtool.h>
5 #include <linux/percpu.h>
6
7 #include <asm/debugreg.h>
8 #include <asm/mmu_context.h>
9
10 #include "cpuid.h"
11 #include "hyperv.h"
12 #include "mmu.h"
13 #include "nested.h"
14 #include "pmu.h"
15 #include "posted_intr.h"
16 #include "sgx.h"
17 #include "trace.h"
18 #include "vmx.h"
19 #include "x86.h"
20 #include "smm.h"
21
22 static bool __read_mostly enable_shadow_vmcs = 1;
23 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
24
25 static bool __read_mostly nested_early_check = 0;
26 module_param(nested_early_check, bool, S_IRUGO);
27
28 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
29
30 /*
31 * Hyper-V requires all of these, so mark them as supported even though
32 * they are just treated the same as all-context.
33 */
34 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
35 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
36 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
37 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
38 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
39
40 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
41
42 enum {
43 VMX_VMREAD_BITMAP,
44 VMX_VMWRITE_BITMAP,
45 VMX_BITMAP_NR
46 };
47 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
48
49 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
50 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
51
52 struct shadow_vmcs_field {
53 u16 encoding;
54 u16 offset;
55 };
56 static struct shadow_vmcs_field shadow_read_only_fields[] = {
57 #define SHADOW_FIELD_RO(x, y) { x, offsetof(struct vmcs12, y) },
58 #include "vmcs_shadow_fields.h"
59 };
60 static int max_shadow_read_only_fields =
61 ARRAY_SIZE(shadow_read_only_fields);
62
63 static struct shadow_vmcs_field shadow_read_write_fields[] = {
64 #define SHADOW_FIELD_RW(x, y) { x, offsetof(struct vmcs12, y) },
65 #include "vmcs_shadow_fields.h"
66 };
67 static int max_shadow_read_write_fields =
68 ARRAY_SIZE(shadow_read_write_fields);
69
70 static void init_vmcs_shadow_fields(void)
71 {
72 int i, j;
73
74 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
75 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
76
77 for (i = j = 0; i < max_shadow_read_only_fields; i++) {
78 struct shadow_vmcs_field entry = shadow_read_only_fields[i];
79 u16 field = entry.encoding;
80
81 if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
82 (i + 1 == max_shadow_read_only_fields ||
83 shadow_read_only_fields[i + 1].encoding != field + 1))
84 pr_err("Missing field from shadow_read_only_field %x\n",
85 field + 1);
86
87 clear_bit(field, vmx_vmread_bitmap);
88 if (field & 1)
89 #ifdef CONFIG_X86_64
90 continue;
91 #else
92 entry.offset += sizeof(u32);
93 #endif
94 shadow_read_only_fields[j++] = entry;
95 }
96 max_shadow_read_only_fields = j;
97
98 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
99 struct shadow_vmcs_field entry = shadow_read_write_fields[i];
100 u16 field = entry.encoding;
101
102 if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
103 (i + 1 == max_shadow_read_write_fields ||
104 shadow_read_write_fields[i + 1].encoding != field + 1))
105 pr_err("Missing field from shadow_read_write_field %x\n",
106 field + 1);
107
108 WARN_ONCE(field >= GUEST_ES_AR_BYTES &&
109 field <= GUEST_TR_AR_BYTES,
110 "Update vmcs12_write_any() to drop reserved bits from AR_BYTES");
111
112 /*
113 * PML and the preemption timer can be emulated, but the
114 * processor cannot vmwrite to fields that don't exist
115 * on bare metal.
116 */
117 switch (field) {
118 case GUEST_PML_INDEX:
119 if (!cpu_has_vmx_pml())
120 continue;
121 break;
122 case VMX_PREEMPTION_TIMER_VALUE:
123 if (!cpu_has_vmx_preemption_timer())
124 continue;
125 break;
126 case GUEST_INTR_STATUS:
127 if (!cpu_has_vmx_apicv())
128 continue;
129 break;
130 default:
131 break;
132 }
133
134 clear_bit(field, vmx_vmwrite_bitmap);
135 clear_bit(field, vmx_vmread_bitmap);
136 if (field & 1)
137 #ifdef CONFIG_X86_64
138 continue;
139 #else
140 entry.offset += sizeof(u32);
141 #endif
142 shadow_read_write_fields[j++] = entry;
143 }
144 max_shadow_read_write_fields = j;
145 }
146
147 /*
148 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
149 * set the success or error code of an emulated VMX instruction (as specified
150 * by Vol 2B, VMX Instruction Reference, "Conventions"), and skip the emulated
151 * instruction.
152 */
153 static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
154 {
155 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
156 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
157 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
158 return kvm_skip_emulated_instruction(vcpu);
159 }
160
161 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
162 {
163 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
164 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
165 X86_EFLAGS_SF | X86_EFLAGS_OF))
166 | X86_EFLAGS_CF);
167 return kvm_skip_emulated_instruction(vcpu);
168 }
169
170 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
171 u32 vm_instruction_error)
172 {
173 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
174 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
175 X86_EFLAGS_SF | X86_EFLAGS_OF))
176 | X86_EFLAGS_ZF);
177 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
178 /*
179 * We don't need to force sync to shadow VMCS because
180 * VM_INSTRUCTION_ERROR is not shadowed. Enlightened VMCS 'shadows' all
181 * fields and thus must be synced.
182 */
183 if (nested_vmx_is_evmptr12_set(to_vmx(vcpu)))
184 to_vmx(vcpu)->nested.need_vmcs12_to_shadow_sync = true;
185
186 return kvm_skip_emulated_instruction(vcpu);
187 }
188
189 static int nested_vmx_fail(struct kvm_vcpu *vcpu, u32 vm_instruction_error)
190 {
191 struct vcpu_vmx *vmx = to_vmx(vcpu);
192
193 /*
194 * failValid writes the error number to the current VMCS, which
195 * can't be done if there isn't a current VMCS.
196 */
197 if (vmx->nested.current_vmptr == INVALID_GPA &&
198 !nested_vmx_is_evmptr12_valid(vmx))
199 return nested_vmx_failInvalid(vcpu);
200
201 return nested_vmx_failValid(vcpu, vm_instruction_error);
202 }
203
204 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
205 {
206 /* TODO: not to reset guest simply here. */
207 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
208 pr_debug_ratelimited("nested vmx abort, indicator %d\n", indicator);
209 }
210
211 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
212 {
213 return fixed_bits_valid(control, low, high);
214 }
215
216 static inline u64 vmx_control_msr(u32 low, u32 high)
217 {
218 return low | ((u64)high << 32);
219 }
220
221 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
222 {
223 secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
224 vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
225 vmx->nested.need_vmcs12_to_shadow_sync = false;
226 }
227
228 static inline void nested_release_evmcs(struct kvm_vcpu *vcpu)
229 {
230 #ifdef CONFIG_KVM_HYPERV
231 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
232 struct vcpu_vmx *vmx = to_vmx(vcpu);
233
234 if (nested_vmx_is_evmptr12_valid(vmx)) {
235 kvm_vcpu_unmap(vcpu, &vmx->nested.hv_evmcs_map, true);
236 vmx->nested.hv_evmcs = NULL;
237 }
238
239 vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
240
241 if (hv_vcpu) {
242 hv_vcpu->nested.pa_page_gpa = INVALID_GPA;
243 hv_vcpu->nested.vm_id = 0;
244 hv_vcpu->nested.vp_id = 0;
245 }
246 #endif
247 }
248
249 static bool nested_evmcs_handle_vmclear(struct kvm_vcpu *vcpu, gpa_t vmptr)
250 {
251 #ifdef CONFIG_KVM_HYPERV
252 struct vcpu_vmx *vmx = to_vmx(vcpu);
253 /*
254 * When Enlightened VMEntry is enabled on the calling CPU we treat
255 * memory area pointer by vmptr as Enlightened VMCS (as there's no good
256 * way to distinguish it from VMCS12) and we must not corrupt it by
257 * writing to the non-existent 'launch_state' field. The area doesn't
258 * have to be the currently active EVMCS on the calling CPU and there's
259 * nothing KVM has to do to transition it from 'active' to 'non-active'
260 * state. It is possible that the area will stay mapped as
261 * vmx->nested.hv_evmcs but this shouldn't be a problem.
262 */
263 if (!guest_cpuid_has_evmcs(vcpu) ||
264 !evmptr_is_valid(nested_get_evmptr(vcpu)))
265 return false;
266
267 if (nested_vmx_evmcs(vmx) && vmptr == vmx->nested.hv_evmcs_vmptr)
268 nested_release_evmcs(vcpu);
269
270 return true;
271 #else
272 return false;
273 #endif
274 }
275
276 static void vmx_sync_vmcs_host_state(struct vcpu_vmx *vmx,
277 struct loaded_vmcs *prev)
278 {
279 struct vmcs_host_state *dest, *src;
280
281 if (unlikely(!vmx->guest_state_loaded))
282 return;
283
284 src = &prev->host_state;
285 dest = &vmx->loaded_vmcs->host_state;
286
287 vmx_set_host_fs_gs(dest, src->fs_sel, src->gs_sel, src->fs_base, src->gs_base);
288 dest->ldt_sel = src->ldt_sel;
289 #ifdef CONFIG_X86_64
290 dest->ds_sel = src->ds_sel;
291 dest->es_sel = src->es_sel;
292 #endif
293 }
294
295 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
296 {
297 struct vcpu_vmx *vmx = to_vmx(vcpu);
298 struct loaded_vmcs *prev;
299 int cpu;
300
301 if (WARN_ON_ONCE(vmx->loaded_vmcs == vmcs))
302 return;
303
304 cpu = get_cpu();
305 prev = vmx->loaded_vmcs;
306 vmx->loaded_vmcs = vmcs;
307 vmx_vcpu_load_vmcs(vcpu, cpu, prev);
308 vmx_sync_vmcs_host_state(vmx, prev);
309 put_cpu();
310
311 vcpu->arch.regs_avail = ~VMX_REGS_LAZY_LOAD_SET;
312
313 /*
314 * All lazily updated registers will be reloaded from VMCS12 on both
315 * vmentry and vmexit.
316 */
317 vcpu->arch.regs_dirty = 0;
318 }
319
320 /*
321 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
322 * just stops using VMX.
323 */
324 static void free_nested(struct kvm_vcpu *vcpu)
325 {
326 struct vcpu_vmx *vmx = to_vmx(vcpu);
327
328 if (WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01))
329 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
330
331 if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
332 return;
333
334 kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
335
336 vmx->nested.vmxon = false;
337 vmx->nested.smm.vmxon = false;
338 vmx->nested.vmxon_ptr = INVALID_GPA;
339 free_vpid(vmx->nested.vpid02);
340 vmx->nested.posted_intr_nv = -1;
341 vmx->nested.current_vmptr = INVALID_GPA;
342 if (enable_shadow_vmcs) {
343 vmx_disable_shadow_vmcs(vmx);
344 vmcs_clear(vmx->vmcs01.shadow_vmcs);
345 free_vmcs(vmx->vmcs01.shadow_vmcs);
346 vmx->vmcs01.shadow_vmcs = NULL;
347 }
348 kfree(vmx->nested.cached_vmcs12);
349 vmx->nested.cached_vmcs12 = NULL;
350 kfree(vmx->nested.cached_shadow_vmcs12);
351 vmx->nested.cached_shadow_vmcs12 = NULL;
352 /*
353 * Unpin physical memory we referred to in the vmcs02. The APIC access
354 * page's backing page (yeah, confusing) shouldn't actually be accessed,
355 * and if it is written, the contents are irrelevant.
356 */
357 kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
358 kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
359 kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
360 vmx->nested.pi_desc = NULL;
361
362 kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
363
364 nested_release_evmcs(vcpu);
365
366 free_loaded_vmcs(&vmx->nested.vmcs02);
367 }
368
369 /*
370 * Ensure that the current vmcs of the logical processor is the
371 * vmcs01 of the vcpu before calling free_nested().
372 */
373 void nested_vmx_free_vcpu(struct kvm_vcpu *vcpu)
374 {
375 vcpu_load(vcpu);
376 vmx_leave_nested(vcpu);
377 vcpu_put(vcpu);
378 }
379
380 #define EPTP_PA_MASK GENMASK_ULL(51, 12)
381
382 static bool nested_ept_root_matches(hpa_t root_hpa, u64 root_eptp, u64 eptp)
383 {
384 return VALID_PAGE(root_hpa) &&
385 ((root_eptp & EPTP_PA_MASK) == (eptp & EPTP_PA_MASK));
386 }
387
388 static void nested_ept_invalidate_addr(struct kvm_vcpu *vcpu, gpa_t eptp,
389 gpa_t addr)
390 {
391 unsigned long roots = 0;
392 uint i;
393 struct kvm_mmu_root_info *cached_root;
394
395 WARN_ON_ONCE(!mmu_is_nested(vcpu));
396
397 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
398 cached_root = &vcpu->arch.mmu->prev_roots[i];
399
400 if (nested_ept_root_matches(cached_root->hpa, cached_root->pgd,
401 eptp))
402 roots |= KVM_MMU_ROOT_PREVIOUS(i);
403 }
404 if (roots)
405 kvm_mmu_invalidate_addr(vcpu, vcpu->arch.mmu, addr, roots);
406 }
407
408 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
409 struct x86_exception *fault)
410 {
411 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
412 struct vcpu_vmx *vmx = to_vmx(vcpu);
413 unsigned long exit_qualification;
414 u32 vm_exit_reason;
415
416 if (vmx->nested.pml_full) {
417 vm_exit_reason = EXIT_REASON_PML_FULL;
418 vmx->nested.pml_full = false;
419
420 /*
421 * It should be impossible to trigger a nested PML Full VM-Exit
422 * for anything other than an EPT Violation from L2. KVM *can*
423 * trigger nEPT page fault injection in response to an EPT
424 * Misconfig, e.g. if the MMIO SPTE was stale and L1's EPT
425 * tables also changed, but KVM should not treat EPT Misconfig
426 * VM-Exits as writes.
427 */
428 WARN_ON_ONCE(vmx->exit_reason.basic != EXIT_REASON_EPT_VIOLATION);
429
430 /*
431 * PML Full and EPT Violation VM-Exits both use bit 12 to report
432 * "NMI unblocking due to IRET", i.e. the bit can be propagated
433 * as-is from the original EXIT_QUALIFICATION.
434 */
435 exit_qualification = vmx_get_exit_qual(vcpu) & INTR_INFO_UNBLOCK_NMI;
436 } else {
437 if (fault->error_code & PFERR_RSVD_MASK) {
438 vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
439 exit_qualification = 0;
440 } else {
441 exit_qualification = fault->exit_qualification;
442 exit_qualification |= vmx_get_exit_qual(vcpu) &
443 (EPT_VIOLATION_GVA_IS_VALID |
444 EPT_VIOLATION_GVA_TRANSLATED);
445 vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
446 }
447
448 /*
449 * Although the caller (kvm_inject_emulated_page_fault) would
450 * have already synced the faulting address in the shadow EPT
451 * tables for the current EPTP12, we also need to sync it for
452 * any other cached EPTP02s based on the same EP4TA, since the
453 * TLB associates mappings to the EP4TA rather than the full EPTP.
454 */
455 nested_ept_invalidate_addr(vcpu, vmcs12->ept_pointer,
456 fault->address);
457 }
458
459 nested_vmx_vmexit(vcpu, vm_exit_reason, 0, exit_qualification);
460 vmcs12->guest_physical_address = fault->address;
461 }
462
463 static void nested_ept_new_eptp(struct kvm_vcpu *vcpu)
464 {
465 struct vcpu_vmx *vmx = to_vmx(vcpu);
466 bool execonly = vmx->nested.msrs.ept_caps & VMX_EPT_EXECUTE_ONLY_BIT;
467 int ept_lpage_level = ept_caps_to_lpage_level(vmx->nested.msrs.ept_caps);
468
469 kvm_init_shadow_ept_mmu(vcpu, execonly, ept_lpage_level,
470 nested_ept_ad_enabled(vcpu),
471 nested_ept_get_eptp(vcpu));
472 }
473
474 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
475 {
476 WARN_ON(mmu_is_nested(vcpu));
477
478 vcpu->arch.mmu = &vcpu->arch.guest_mmu;
479 nested_ept_new_eptp(vcpu);
480 vcpu->arch.mmu->get_guest_pgd = nested_ept_get_eptp;
481 vcpu->arch.mmu->inject_page_fault = nested_ept_inject_page_fault;
482 vcpu->arch.mmu->get_pdptr = kvm_pdptr_read;
483
484 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
485 }
486
487 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
488 {
489 vcpu->arch.mmu = &vcpu->arch.root_mmu;
490 vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
491 }
492
493 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
494 u16 error_code)
495 {
496 bool inequality, bit;
497
498 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
499 inequality =
500 (error_code & vmcs12->page_fault_error_code_mask) !=
501 vmcs12->page_fault_error_code_match;
502 return inequality ^ bit;
503 }
504
505 static bool nested_vmx_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
506 u32 error_code)
507 {
508 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
509
510 /*
511 * Drop bits 31:16 of the error code when performing the #PF mask+match
512 * check. All VMCS fields involved are 32 bits, but Intel CPUs never
513 * set bits 31:16 and VMX disallows setting bits 31:16 in the injected
514 * error code. Including the to-be-dropped bits in the check might
515 * result in an "impossible" or missed exit from L1's perspective.
516 */
517 if (vector == PF_VECTOR)
518 return nested_vmx_is_page_fault_vmexit(vmcs12, (u16)error_code);
519
520 return (vmcs12->exception_bitmap & (1u << vector));
521 }
522
523 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
524 struct vmcs12 *vmcs12)
525 {
526 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
527 return 0;
528
529 if (CC(!page_address_valid(vcpu, vmcs12->io_bitmap_a)) ||
530 CC(!page_address_valid(vcpu, vmcs12->io_bitmap_b)))
531 return -EINVAL;
532
533 return 0;
534 }
535
536 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
537 struct vmcs12 *vmcs12)
538 {
539 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
540 return 0;
541
542 if (CC(!page_address_valid(vcpu, vmcs12->msr_bitmap)))
543 return -EINVAL;
544
545 return 0;
546 }
547
548 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
549 struct vmcs12 *vmcs12)
550 {
551 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
552 return 0;
553
554 if (CC(!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr)))
555 return -EINVAL;
556
557 return 0;
558 }
559
560 /*
561 * For x2APIC MSRs, ignore the vmcs01 bitmap. L1 can enable x2APIC without L1
562 * itself utilizing x2APIC. All MSRs were previously set to be intercepted,
563 * only the "disable intercept" case needs to be handled.
564 */
565 static void nested_vmx_disable_intercept_for_x2apic_msr(unsigned long *msr_bitmap_l1,
566 unsigned long *msr_bitmap_l0,
567 u32 msr, int type)
568 {
569 if (type & MSR_TYPE_R && !vmx_test_msr_bitmap_read(msr_bitmap_l1, msr))
570 vmx_clear_msr_bitmap_read(msr_bitmap_l0, msr);
571
572 if (type & MSR_TYPE_W && !vmx_test_msr_bitmap_write(msr_bitmap_l1, msr))
573 vmx_clear_msr_bitmap_write(msr_bitmap_l0, msr);
574 }
575
576 static inline void enable_x2apic_msr_intercepts(unsigned long *msr_bitmap)
577 {
578 int msr;
579
580 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
581 unsigned word = msr / BITS_PER_LONG;
582
583 msr_bitmap[word] = ~0;
584 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
585 }
586 }
587
588 #define BUILD_NVMX_MSR_INTERCEPT_HELPER(rw) \
589 static inline \
590 void nested_vmx_set_msr_##rw##_intercept(struct vcpu_vmx *vmx, \
591 unsigned long *msr_bitmap_l1, \
592 unsigned long *msr_bitmap_l0, u32 msr) \
593 { \
594 if (vmx_test_msr_bitmap_##rw(vmx->vmcs01.msr_bitmap, msr) || \
595 vmx_test_msr_bitmap_##rw(msr_bitmap_l1, msr)) \
596 vmx_set_msr_bitmap_##rw(msr_bitmap_l0, msr); \
597 else \
598 vmx_clear_msr_bitmap_##rw(msr_bitmap_l0, msr); \
599 }
600 BUILD_NVMX_MSR_INTERCEPT_HELPER(read)
601 BUILD_NVMX_MSR_INTERCEPT_HELPER(write)
602
603 static inline void nested_vmx_set_intercept_for_msr(struct vcpu_vmx *vmx,
604 unsigned long *msr_bitmap_l1,
605 unsigned long *msr_bitmap_l0,
606 u32 msr, int types)
607 {
608 if (types & MSR_TYPE_R)
609 nested_vmx_set_msr_read_intercept(vmx, msr_bitmap_l1,
610 msr_bitmap_l0, msr);
611 if (types & MSR_TYPE_W)
612 nested_vmx_set_msr_write_intercept(vmx, msr_bitmap_l1,
613 msr_bitmap_l0, msr);
614 }
615
616 /*
617 * Merge L0's and L1's MSR bitmap, return false to indicate that
618 * we do not use the hardware.
619 */
620 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
621 struct vmcs12 *vmcs12)
622 {
623 struct vcpu_vmx *vmx = to_vmx(vcpu);
624 int msr;
625 unsigned long *msr_bitmap_l1;
626 unsigned long *msr_bitmap_l0 = vmx->nested.vmcs02.msr_bitmap;
627 struct kvm_host_map *map = &vmx->nested.msr_bitmap_map;
628
629 /* Nothing to do if the MSR bitmap is not in use. */
630 if (!cpu_has_vmx_msr_bitmap() ||
631 !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
632 return false;
633
634 /*
635 * MSR bitmap update can be skipped when:
636 * - MSR bitmap for L1 hasn't changed.
637 * - Nested hypervisor (L1) is attempting to launch the same L2 as
638 * before.
639 * - Nested hypervisor (L1) has enabled 'Enlightened MSR Bitmap' feature
640 * and tells KVM (L0) there were no changes in MSR bitmap for L2.
641 */
642 if (!vmx->nested.force_msr_bitmap_recalc) {
643 struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
644
645 if (evmcs && evmcs->hv_enlightenments_control.msr_bitmap &&
646 evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP)
647 return true;
648 }
649
650 if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->msr_bitmap), map))
651 return false;
652
653 msr_bitmap_l1 = (unsigned long *)map->hva;
654
655 /*
656 * To keep the control flow simple, pay eight 8-byte writes (sixteen
657 * 4-byte writes on 32-bit systems) up front to enable intercepts for
658 * the x2APIC MSR range and selectively toggle those relevant to L2.
659 */
660 enable_x2apic_msr_intercepts(msr_bitmap_l0);
661
662 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
663 if (nested_cpu_has_apic_reg_virt(vmcs12)) {
664 /*
665 * L0 need not intercept reads for MSRs between 0x800
666 * and 0x8ff, it just lets the processor take the value
667 * from the virtual-APIC page; take those 256 bits
668 * directly from the L1 bitmap.
669 */
670 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
671 unsigned word = msr / BITS_PER_LONG;
672
673 msr_bitmap_l0[word] = msr_bitmap_l1[word];
674 }
675 }
676
677 nested_vmx_disable_intercept_for_x2apic_msr(
678 msr_bitmap_l1, msr_bitmap_l0,
679 X2APIC_MSR(APIC_TASKPRI),
680 MSR_TYPE_R | MSR_TYPE_W);
681
682 if (nested_cpu_has_vid(vmcs12)) {
683 nested_vmx_disable_intercept_for_x2apic_msr(
684 msr_bitmap_l1, msr_bitmap_l0,
685 X2APIC_MSR(APIC_EOI),
686 MSR_TYPE_W);
687 nested_vmx_disable_intercept_for_x2apic_msr(
688 msr_bitmap_l1, msr_bitmap_l0,
689 X2APIC_MSR(APIC_SELF_IPI),
690 MSR_TYPE_W);
691 }
692 }
693
694 /*
695 * Always check vmcs01's bitmap to honor userspace MSR filters and any
696 * other runtime changes to vmcs01's bitmap, e.g. dynamic pass-through.
697 */
698 #ifdef CONFIG_X86_64
699 nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
700 MSR_FS_BASE, MSR_TYPE_RW);
701
702 nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
703 MSR_GS_BASE, MSR_TYPE_RW);
704
705 nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
706 MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
707 #endif
708 nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
709 MSR_IA32_SPEC_CTRL, MSR_TYPE_RW);
710
711 nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
712 MSR_IA32_PRED_CMD, MSR_TYPE_W);
713
714 nested_vmx_set_intercept_for_msr(vmx, msr_bitmap_l1, msr_bitmap_l0,
715 MSR_IA32_FLUSH_CMD, MSR_TYPE_W);
716
717 kvm_vcpu_unmap(vcpu, &vmx->nested.msr_bitmap_map, false);
718
719 vmx->nested.force_msr_bitmap_recalc = false;
720
721 return true;
722 }
723
724 static void nested_cache_shadow_vmcs12(struct kvm_vcpu *vcpu,
725 struct vmcs12 *vmcs12)
726 {
727 struct vcpu_vmx *vmx = to_vmx(vcpu);
728 struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
729
730 if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
731 vmcs12->vmcs_link_pointer == INVALID_GPA)
732 return;
733
734 if (ghc->gpa != vmcs12->vmcs_link_pointer &&
735 kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
736 vmcs12->vmcs_link_pointer, VMCS12_SIZE))
737 return;
738
739 kvm_read_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
740 VMCS12_SIZE);
741 }
742
743 static void nested_flush_cached_shadow_vmcs12(struct kvm_vcpu *vcpu,
744 struct vmcs12 *vmcs12)
745 {
746 struct vcpu_vmx *vmx = to_vmx(vcpu);
747 struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
748
749 if (!nested_cpu_has_shadow_vmcs(vmcs12) ||
750 vmcs12->vmcs_link_pointer == INVALID_GPA)
751 return;
752
753 if (ghc->gpa != vmcs12->vmcs_link_pointer &&
754 kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
755 vmcs12->vmcs_link_pointer, VMCS12_SIZE))
756 return;
757
758 kvm_write_guest_cached(vmx->vcpu.kvm, ghc, get_shadow_vmcs12(vcpu),
759 VMCS12_SIZE);
760 }
761
762 /*
763 * In nested virtualization, check if L1 has set
764 * VM_EXIT_ACK_INTR_ON_EXIT
765 */
766 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
767 {
768 return get_vmcs12(vcpu)->vm_exit_controls &
769 VM_EXIT_ACK_INTR_ON_EXIT;
770 }
771
772 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
773 struct vmcs12 *vmcs12)
774 {
775 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
776 CC(!page_address_valid(vcpu, vmcs12->apic_access_addr)))
777 return -EINVAL;
778 else
779 return 0;
780 }
781
782 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
783 struct vmcs12 *vmcs12)
784 {
785 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
786 !nested_cpu_has_apic_reg_virt(vmcs12) &&
787 !nested_cpu_has_vid(vmcs12) &&
788 !nested_cpu_has_posted_intr(vmcs12))
789 return 0;
790
791 /*
792 * If virtualize x2apic mode is enabled,
793 * virtualize apic access must be disabled.
794 */
795 if (CC(nested_cpu_has_virt_x2apic_mode(vmcs12) &&
796 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)))
797 return -EINVAL;
798
799 /*
800 * If virtual interrupt delivery is enabled,
801 * we must exit on external interrupts.
802 */
803 if (CC(nested_cpu_has_vid(vmcs12) && !nested_exit_on_intr(vcpu)))
804 return -EINVAL;
805
806 /*
807 * bits 15:8 should be zero in posted_intr_nv,
808 * the descriptor address has been already checked
809 * in nested_get_vmcs12_pages.
810 *
811 * bits 5:0 of posted_intr_desc_addr should be zero.
812 */
813 if (nested_cpu_has_posted_intr(vmcs12) &&
814 (CC(!nested_cpu_has_vid(vmcs12)) ||
815 CC(!nested_exit_intr_ack_set(vcpu)) ||
816 CC((vmcs12->posted_intr_nv & 0xff00)) ||
817 CC(!kvm_vcpu_is_legal_aligned_gpa(vcpu, vmcs12->posted_intr_desc_addr, 64))))
818 return -EINVAL;
819
820 /* tpr shadow is needed by all apicv features. */
821 if (CC(!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)))
822 return -EINVAL;
823
824 return 0;
825 }
826
827 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
828 u32 count, u64 addr)
829 {
830 if (count == 0)
831 return 0;
832
833 if (!kvm_vcpu_is_legal_aligned_gpa(vcpu, addr, 16) ||
834 !kvm_vcpu_is_legal_gpa(vcpu, (addr + count * sizeof(struct vmx_msr_entry) - 1)))
835 return -EINVAL;
836
837 return 0;
838 }
839
840 static int nested_vmx_check_exit_msr_switch_controls(struct kvm_vcpu *vcpu,
841 struct vmcs12 *vmcs12)
842 {
843 if (CC(nested_vmx_check_msr_switch(vcpu,
844 vmcs12->vm_exit_msr_load_count,
845 vmcs12->vm_exit_msr_load_addr)) ||
846 CC(nested_vmx_check_msr_switch(vcpu,
847 vmcs12->vm_exit_msr_store_count,
848 vmcs12->vm_exit_msr_store_addr)))
849 return -EINVAL;
850
851 return 0;
852 }
853
854 static int nested_vmx_check_entry_msr_switch_controls(struct kvm_vcpu *vcpu,
855 struct vmcs12 *vmcs12)
856 {
857 if (CC(nested_vmx_check_msr_switch(vcpu,
858 vmcs12->vm_entry_msr_load_count,
859 vmcs12->vm_entry_msr_load_addr)))
860 return -EINVAL;
861
862 return 0;
863 }
864
865 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
866 struct vmcs12 *vmcs12)
867 {
868 if (!nested_cpu_has_pml(vmcs12))
869 return 0;
870
871 if (CC(!nested_cpu_has_ept(vmcs12)) ||
872 CC(!page_address_valid(vcpu, vmcs12->pml_address)))
873 return -EINVAL;
874
875 return 0;
876 }
877
878 static int nested_vmx_check_unrestricted_guest_controls(struct kvm_vcpu *vcpu,
879 struct vmcs12 *vmcs12)
880 {
881 if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST) &&
882 !nested_cpu_has_ept(vmcs12)))
883 return -EINVAL;
884 return 0;
885 }
886
887 static int nested_vmx_check_mode_based_ept_exec_controls(struct kvm_vcpu *vcpu,
888 struct vmcs12 *vmcs12)
889 {
890 if (CC(nested_cpu_has2(vmcs12, SECONDARY_EXEC_MODE_BASED_EPT_EXEC) &&
891 !nested_cpu_has_ept(vmcs12)))
892 return -EINVAL;
893 return 0;
894 }
895
896 static int nested_vmx_check_shadow_vmcs_controls(struct kvm_vcpu *vcpu,
897 struct vmcs12 *vmcs12)
898 {
899 if (!nested_cpu_has_shadow_vmcs(vmcs12))
900 return 0;
901
902 if (CC(!page_address_valid(vcpu, vmcs12->vmread_bitmap)) ||
903 CC(!page_address_valid(vcpu, vmcs12->vmwrite_bitmap)))
904 return -EINVAL;
905
906 return 0;
907 }
908
909 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
910 struct vmx_msr_entry *e)
911 {
912 /* x2APIC MSR accesses are not allowed */
913 if (CC(vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8))
914 return -EINVAL;
915 if (CC(e->index == MSR_IA32_UCODE_WRITE) || /* SDM Table 35-2 */
916 CC(e->index == MSR_IA32_UCODE_REV))
917 return -EINVAL;
918 if (CC(e->reserved != 0))
919 return -EINVAL;
920 return 0;
921 }
922
923 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
924 struct vmx_msr_entry *e)
925 {
926 if (CC(e->index == MSR_FS_BASE) ||
927 CC(e->index == MSR_GS_BASE) ||
928 CC(e->index == MSR_IA32_SMM_MONITOR_CTL) || /* SMM is not supported */
929 nested_vmx_msr_check_common(vcpu, e))
930 return -EINVAL;
931 return 0;
932 }
933
934 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
935 struct vmx_msr_entry *e)
936 {
937 if (CC(e->index == MSR_IA32_SMBASE) || /* SMM is not supported */
938 nested_vmx_msr_check_common(vcpu, e))
939 return -EINVAL;
940 return 0;
941 }
942
943 static u32 nested_vmx_max_atomic_switch_msrs(struct kvm_vcpu *vcpu)
944 {
945 struct vcpu_vmx *vmx = to_vmx(vcpu);
946 u64 vmx_misc = vmx_control_msr(vmx->nested.msrs.misc_low,
947 vmx->nested.msrs.misc_high);
948
949 return (vmx_misc_max_msr(vmx_misc) + 1) * VMX_MISC_MSR_LIST_MULTIPLIER;
950 }
951
952 /*
953 * Load guest's/host's msr at nested entry/exit.
954 * return 0 for success, entry index for failure.
955 *
956 * One of the failure modes for MSR load/store is when a list exceeds the
957 * virtual hardware's capacity. To maintain compatibility with hardware inasmuch
958 * as possible, process all valid entries before failing rather than precheck
959 * for a capacity violation.
960 */
961 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
962 {
963 u32 i;
964 struct vmx_msr_entry e;
965 u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
966
967 for (i = 0; i < count; i++) {
968 if (unlikely(i >= max_msr_list_size))
969 goto fail;
970
971 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
972 &e, sizeof(e))) {
973 pr_debug_ratelimited(
974 "%s cannot read MSR entry (%u, 0x%08llx)\n",
975 __func__, i, gpa + i * sizeof(e));
976 goto fail;
977 }
978 if (nested_vmx_load_msr_check(vcpu, &e)) {
979 pr_debug_ratelimited(
980 "%s check failed (%u, 0x%x, 0x%x)\n",
981 __func__, i, e.index, e.reserved);
982 goto fail;
983 }
984 if (kvm_set_msr(vcpu, e.index, e.value)) {
985 pr_debug_ratelimited(
986 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
987 __func__, i, e.index, e.value);
988 goto fail;
989 }
990 }
991 return 0;
992 fail:
993 /* Note, max_msr_list_size is at most 4096, i.e. this can't wrap. */
994 return i + 1;
995 }
996
997 static bool nested_vmx_get_vmexit_msr_value(struct kvm_vcpu *vcpu,
998 u32 msr_index,
999 u64 *data)
1000 {
1001 struct vcpu_vmx *vmx = to_vmx(vcpu);
1002
1003 /*
1004 * If the L0 hypervisor stored a more accurate value for the TSC that
1005 * does not include the time taken for emulation of the L2->L1
1006 * VM-exit in L0, use the more accurate value.
1007 */
1008 if (msr_index == MSR_IA32_TSC) {
1009 int i = vmx_find_loadstore_msr_slot(&vmx->msr_autostore.guest,
1010 MSR_IA32_TSC);
1011
1012 if (i >= 0) {
1013 u64 val = vmx->msr_autostore.guest.val[i].value;
1014
1015 *data = kvm_read_l1_tsc(vcpu, val);
1016 return true;
1017 }
1018 }
1019
1020 if (kvm_get_msr(vcpu, msr_index, data)) {
1021 pr_debug_ratelimited("%s cannot read MSR (0x%x)\n", __func__,
1022 msr_index);
1023 return false;
1024 }
1025 return true;
1026 }
1027
1028 static bool read_and_check_msr_entry(struct kvm_vcpu *vcpu, u64 gpa, int i,
1029 struct vmx_msr_entry *e)
1030 {
1031 if (kvm_vcpu_read_guest(vcpu,
1032 gpa + i * sizeof(*e),
1033 e, 2 * sizeof(u32))) {
1034 pr_debug_ratelimited(
1035 "%s cannot read MSR entry (%u, 0x%08llx)\n",
1036 __func__, i, gpa + i * sizeof(*e));
1037 return false;
1038 }
1039 if (nested_vmx_store_msr_check(vcpu, e)) {
1040 pr_debug_ratelimited(
1041 "%s check failed (%u, 0x%x, 0x%x)\n",
1042 __func__, i, e->index, e->reserved);
1043 return false;
1044 }
1045 return true;
1046 }
1047
1048 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
1049 {
1050 u64 data;
1051 u32 i;
1052 struct vmx_msr_entry e;
1053 u32 max_msr_list_size = nested_vmx_max_atomic_switch_msrs(vcpu);
1054
1055 for (i = 0; i < count; i++) {
1056 if (unlikely(i >= max_msr_list_size))
1057 return -EINVAL;
1058
1059 if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1060 return -EINVAL;
1061
1062 if (!nested_vmx_get_vmexit_msr_value(vcpu, e.index, &data))
1063 return -EINVAL;
1064
1065 if (kvm_vcpu_write_guest(vcpu,
1066 gpa + i * sizeof(e) +
1067 offsetof(struct vmx_msr_entry, value),
1068 &data, sizeof(data))) {
1069 pr_debug_ratelimited(
1070 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
1071 __func__, i, e.index, data);
1072 return -EINVAL;
1073 }
1074 }
1075 return 0;
1076 }
1077
1078 static bool nested_msr_store_list_has_msr(struct kvm_vcpu *vcpu, u32 msr_index)
1079 {
1080 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1081 u32 count = vmcs12->vm_exit_msr_store_count;
1082 u64 gpa = vmcs12->vm_exit_msr_store_addr;
1083 struct vmx_msr_entry e;
1084 u32 i;
1085
1086 for (i = 0; i < count; i++) {
1087 if (!read_and_check_msr_entry(vcpu, gpa, i, &e))
1088 return false;
1089
1090 if (e.index == msr_index)
1091 return true;
1092 }
1093 return false;
1094 }
1095
1096 static void prepare_vmx_msr_autostore_list(struct kvm_vcpu *vcpu,
1097 u32 msr_index)
1098 {
1099 struct vcpu_vmx *vmx = to_vmx(vcpu);
1100 struct vmx_msrs *autostore = &vmx->msr_autostore.guest;
1101 bool in_vmcs12_store_list;
1102 int msr_autostore_slot;
1103 bool in_autostore_list;
1104 int last;
1105
1106 msr_autostore_slot = vmx_find_loadstore_msr_slot(autostore, msr_index);
1107 in_autostore_list = msr_autostore_slot >= 0;
1108 in_vmcs12_store_list = nested_msr_store_list_has_msr(vcpu, msr_index);
1109
1110 if (in_vmcs12_store_list && !in_autostore_list) {
1111 if (autostore->nr == MAX_NR_LOADSTORE_MSRS) {
1112 /*
1113 * Emulated VMEntry does not fail here. Instead a less
1114 * accurate value will be returned by
1115 * nested_vmx_get_vmexit_msr_value() using kvm_get_msr()
1116 * instead of reading the value from the vmcs02 VMExit
1117 * MSR-store area.
1118 */
1119 pr_warn_ratelimited(
1120 "Not enough msr entries in msr_autostore. Can't add msr %x\n",
1121 msr_index);
1122 return;
1123 }
1124 last = autostore->nr++;
1125 autostore->val[last].index = msr_index;
1126 } else if (!in_vmcs12_store_list && in_autostore_list) {
1127 last = --autostore->nr;
1128 autostore->val[msr_autostore_slot] = autostore->val[last];
1129 }
1130 }
1131
1132 /*
1133 * Load guest's/host's cr3 at nested entry/exit. @nested_ept is true if we are
1134 * emulating VM-Entry into a guest with EPT enabled. On failure, the expected
1135 * Exit Qualification (for a VM-Entry consistency check VM-Exit) is assigned to
1136 * @entry_failure_code.
1137 */
1138 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
1139 bool nested_ept, bool reload_pdptrs,
1140 enum vm_entry_failure_code *entry_failure_code)
1141 {
1142 if (CC(!kvm_vcpu_is_legal_cr3(vcpu, cr3))) {
1143 *entry_failure_code = ENTRY_FAIL_DEFAULT;
1144 return -EINVAL;
1145 }
1146
1147 /*
1148 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
1149 * must not be dereferenced.
1150 */
1151 if (reload_pdptrs && !nested_ept && is_pae_paging(vcpu) &&
1152 CC(!load_pdptrs(vcpu, cr3))) {
1153 *entry_failure_code = ENTRY_FAIL_PDPTE;
1154 return -EINVAL;
1155 }
1156
1157 vcpu->arch.cr3 = cr3;
1158 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1159
1160 /* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
1161 kvm_init_mmu(vcpu);
1162
1163 if (!nested_ept)
1164 kvm_mmu_new_pgd(vcpu, cr3);
1165
1166 return 0;
1167 }
1168
1169 /*
1170 * Returns if KVM is able to config CPU to tag TLB entries
1171 * populated by L2 differently than TLB entries populated
1172 * by L1.
1173 *
1174 * If L0 uses EPT, L1 and L2 run with different EPTP because
1175 * guest_mode is part of kvm_mmu_page_role. Thus, TLB entries
1176 * are tagged with different EPTP.
1177 *
1178 * If L1 uses VPID and we allocated a vpid02, TLB entries are tagged
1179 * with different VPID (L1 entries are tagged with vmx->vpid
1180 * while L2 entries are tagged with vmx->nested.vpid02).
1181 */
1182 static bool nested_has_guest_tlb_tag(struct kvm_vcpu *vcpu)
1183 {
1184 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1185
1186 return enable_ept ||
1187 (nested_cpu_has_vpid(vmcs12) && to_vmx(vcpu)->nested.vpid02);
1188 }
1189
1190 static void nested_vmx_transition_tlb_flush(struct kvm_vcpu *vcpu,
1191 struct vmcs12 *vmcs12,
1192 bool is_vmenter)
1193 {
1194 struct vcpu_vmx *vmx = to_vmx(vcpu);
1195
1196 /* Handle pending Hyper-V TLB flush requests */
1197 kvm_hv_nested_transtion_tlb_flush(vcpu, enable_ept);
1198
1199 /*
1200 * If vmcs12 doesn't use VPID, L1 expects linear and combined mappings
1201 * for *all* contexts to be flushed on VM-Enter/VM-Exit, i.e. it's a
1202 * full TLB flush from the guest's perspective. This is required even
1203 * if VPID is disabled in the host as KVM may need to synchronize the
1204 * MMU in response to the guest TLB flush.
1205 *
1206 * Note, using TLB_FLUSH_GUEST is correct even if nested EPT is in use.
1207 * EPT is a special snowflake, as guest-physical mappings aren't
1208 * flushed on VPID invalidations, including VM-Enter or VM-Exit with
1209 * VPID disabled. As a result, KVM _never_ needs to sync nEPT
1210 * entries on VM-Enter because L1 can't rely on VM-Enter to flush
1211 * those mappings.
1212 */
1213 if (!nested_cpu_has_vpid(vmcs12)) {
1214 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1215 return;
1216 }
1217
1218 /* L2 should never have a VPID if VPID is disabled. */
1219 WARN_ON(!enable_vpid);
1220
1221 /*
1222 * VPID is enabled and in use by vmcs12. If vpid12 is changing, then
1223 * emulate a guest TLB flush as KVM does not track vpid12 history nor
1224 * is the VPID incorporated into the MMU context. I.e. KVM must assume
1225 * that the new vpid12 has never been used and thus represents a new
1226 * guest ASID that cannot have entries in the TLB.
1227 */
1228 if (is_vmenter && vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
1229 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
1230 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1231 return;
1232 }
1233
1234 /*
1235 * If VPID is enabled, used by vmc12, and vpid12 is not changing but
1236 * does not have a unique TLB tag (ASID), i.e. EPT is disabled and
1237 * KVM was unable to allocate a VPID for L2, flush the current context
1238 * as the effective ASID is common to both L1 and L2.
1239 */
1240 if (!nested_has_guest_tlb_tag(vcpu))
1241 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1242 }
1243
1244 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
1245 {
1246 superset &= mask;
1247 subset &= mask;
1248
1249 return (superset | subset) == superset;
1250 }
1251
1252 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
1253 {
1254 const u64 feature_and_reserved =
1255 /* feature (except bit 48; see below) */
1256 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
1257 /* reserved */
1258 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
1259 u64 vmx_basic = vmcs_config.nested.basic;
1260
1261 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
1262 return -EINVAL;
1263
1264 /*
1265 * KVM does not emulate a version of VMX that constrains physical
1266 * addresses of VMX structures (e.g. VMCS) to 32-bits.
1267 */
1268 if (data & BIT_ULL(48))
1269 return -EINVAL;
1270
1271 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
1272 vmx_basic_vmcs_revision_id(data))
1273 return -EINVAL;
1274
1275 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
1276 return -EINVAL;
1277
1278 vmx->nested.msrs.basic = data;
1279 return 0;
1280 }
1281
1282 static void vmx_get_control_msr(struct nested_vmx_msrs *msrs, u32 msr_index,
1283 u32 **low, u32 **high)
1284 {
1285 switch (msr_index) {
1286 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1287 *low = &msrs->pinbased_ctls_low;
1288 *high = &msrs->pinbased_ctls_high;
1289 break;
1290 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1291 *low = &msrs->procbased_ctls_low;
1292 *high = &msrs->procbased_ctls_high;
1293 break;
1294 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1295 *low = &msrs->exit_ctls_low;
1296 *high = &msrs->exit_ctls_high;
1297 break;
1298 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1299 *low = &msrs->entry_ctls_low;
1300 *high = &msrs->entry_ctls_high;
1301 break;
1302 case MSR_IA32_VMX_PROCBASED_CTLS2:
1303 *low = &msrs->secondary_ctls_low;
1304 *high = &msrs->secondary_ctls_high;
1305 break;
1306 default:
1307 BUG();
1308 }
1309 }
1310
1311 static int
1312 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1313 {
1314 u32 *lowp, *highp;
1315 u64 supported;
1316
1317 vmx_get_control_msr(&vmcs_config.nested, msr_index, &lowp, &highp);
1318
1319 supported = vmx_control_msr(*lowp, *highp);
1320
1321 /* Check must-be-1 bits are still 1. */
1322 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
1323 return -EINVAL;
1324
1325 /* Check must-be-0 bits are still 0. */
1326 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
1327 return -EINVAL;
1328
1329 vmx_get_control_msr(&vmx->nested.msrs, msr_index, &lowp, &highp);
1330 *lowp = data;
1331 *highp = data >> 32;
1332 return 0;
1333 }
1334
1335 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
1336 {
1337 const u64 feature_and_reserved_bits =
1338 /* feature */
1339 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
1340 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
1341 /* reserved */
1342 GENMASK_ULL(13, 9) | BIT_ULL(31);
1343 u64 vmx_misc = vmx_control_msr(vmcs_config.nested.misc_low,
1344 vmcs_config.nested.misc_high);
1345
1346 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
1347 return -EINVAL;
1348
1349 if ((vmx->nested.msrs.pinbased_ctls_high &
1350 PIN_BASED_VMX_PREEMPTION_TIMER) &&
1351 vmx_misc_preemption_timer_rate(data) !=
1352 vmx_misc_preemption_timer_rate(vmx_misc))
1353 return -EINVAL;
1354
1355 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
1356 return -EINVAL;
1357
1358 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
1359 return -EINVAL;
1360
1361 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
1362 return -EINVAL;
1363
1364 vmx->nested.msrs.misc_low = data;
1365 vmx->nested.msrs.misc_high = data >> 32;
1366
1367 return 0;
1368 }
1369
1370 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
1371 {
1372 u64 vmx_ept_vpid_cap = vmx_control_msr(vmcs_config.nested.ept_caps,
1373 vmcs_config.nested.vpid_caps);
1374
1375 /* Every bit is either reserved or a feature bit. */
1376 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
1377 return -EINVAL;
1378
1379 vmx->nested.msrs.ept_caps = data;
1380 vmx->nested.msrs.vpid_caps = data >> 32;
1381 return 0;
1382 }
1383
1384 static u64 *vmx_get_fixed0_msr(struct nested_vmx_msrs *msrs, u32 msr_index)
1385 {
1386 switch (msr_index) {
1387 case MSR_IA32_VMX_CR0_FIXED0:
1388 return &msrs->cr0_fixed0;
1389 case MSR_IA32_VMX_CR4_FIXED0:
1390 return &msrs->cr4_fixed0;
1391 default:
1392 BUG();
1393 }
1394 }
1395
1396 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
1397 {
1398 const u64 *msr = vmx_get_fixed0_msr(&vmcs_config.nested, msr_index);
1399
1400 /*
1401 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
1402 * must be 1 in the restored value.
1403 */
1404 if (!is_bitwise_subset(data, *msr, -1ULL))
1405 return -EINVAL;
1406
1407 *vmx_get_fixed0_msr(&vmx->nested.msrs, msr_index) = data;
1408 return 0;
1409 }
1410
1411 /*
1412 * Called when userspace is restoring VMX MSRs.
1413 *
1414 * Returns 0 on success, non-0 otherwise.
1415 */
1416 int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1417 {
1418 struct vcpu_vmx *vmx = to_vmx(vcpu);
1419
1420 /*
1421 * Don't allow changes to the VMX capability MSRs while the vCPU
1422 * is in VMX operation.
1423 */
1424 if (vmx->nested.vmxon)
1425 return -EBUSY;
1426
1427 switch (msr_index) {
1428 case MSR_IA32_VMX_BASIC:
1429 return vmx_restore_vmx_basic(vmx, data);
1430 case MSR_IA32_VMX_PINBASED_CTLS:
1431 case MSR_IA32_VMX_PROCBASED_CTLS:
1432 case MSR_IA32_VMX_EXIT_CTLS:
1433 case MSR_IA32_VMX_ENTRY_CTLS:
1434 /*
1435 * The "non-true" VMX capability MSRs are generated from the
1436 * "true" MSRs, so we do not support restoring them directly.
1437 *
1438 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
1439 * should restore the "true" MSRs with the must-be-1 bits
1440 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
1441 * DEFAULT SETTINGS".
1442 */
1443 return -EINVAL;
1444 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1445 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1446 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1447 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1448 case MSR_IA32_VMX_PROCBASED_CTLS2:
1449 return vmx_restore_control_msr(vmx, msr_index, data);
1450 case MSR_IA32_VMX_MISC:
1451 return vmx_restore_vmx_misc(vmx, data);
1452 case MSR_IA32_VMX_CR0_FIXED0:
1453 case MSR_IA32_VMX_CR4_FIXED0:
1454 return vmx_restore_fixed0_msr(vmx, msr_index, data);
1455 case MSR_IA32_VMX_CR0_FIXED1:
1456 case MSR_IA32_VMX_CR4_FIXED1:
1457 /*
1458 * These MSRs are generated based on the vCPU's CPUID, so we
1459 * do not support restoring them directly.
1460 */
1461 return -EINVAL;
1462 case MSR_IA32_VMX_EPT_VPID_CAP:
1463 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
1464 case MSR_IA32_VMX_VMCS_ENUM:
1465 vmx->nested.msrs.vmcs_enum = data;
1466 return 0;
1467 case MSR_IA32_VMX_VMFUNC:
1468 if (data & ~vmcs_config.nested.vmfunc_controls)
1469 return -EINVAL;
1470 vmx->nested.msrs.vmfunc_controls = data;
1471 return 0;
1472 default:
1473 /*
1474 * The rest of the VMX capability MSRs do not support restore.
1475 */
1476 return -EINVAL;
1477 }
1478 }
1479
1480 /* Returns 0 on success, non-0 otherwise. */
1481 int vmx_get_vmx_msr(struct nested_vmx_msrs *msrs, u32 msr_index, u64 *pdata)
1482 {
1483 switch (msr_index) {
1484 case MSR_IA32_VMX_BASIC:
1485 *pdata = msrs->basic;
1486 break;
1487 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1488 case MSR_IA32_VMX_PINBASED_CTLS:
1489 *pdata = vmx_control_msr(
1490 msrs->pinbased_ctls_low,
1491 msrs->pinbased_ctls_high);
1492 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
1493 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1494 break;
1495 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1496 case MSR_IA32_VMX_PROCBASED_CTLS:
1497 *pdata = vmx_control_msr(
1498 msrs->procbased_ctls_low,
1499 msrs->procbased_ctls_high);
1500 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
1501 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
1502 break;
1503 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1504 case MSR_IA32_VMX_EXIT_CTLS:
1505 *pdata = vmx_control_msr(
1506 msrs->exit_ctls_low,
1507 msrs->exit_ctls_high);
1508 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
1509 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
1510 break;
1511 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1512 case MSR_IA32_VMX_ENTRY_CTLS:
1513 *pdata = vmx_control_msr(
1514 msrs->entry_ctls_low,
1515 msrs->entry_ctls_high);
1516 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
1517 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
1518 break;
1519 case MSR_IA32_VMX_MISC:
1520 *pdata = vmx_control_msr(
1521 msrs->misc_low,
1522 msrs->misc_high);
1523 break;
1524 case MSR_IA32_VMX_CR0_FIXED0:
1525 *pdata = msrs->cr0_fixed0;
1526 break;
1527 case MSR_IA32_VMX_CR0_FIXED1:
1528 *pdata = msrs->cr0_fixed1;
1529 break;
1530 case MSR_IA32_VMX_CR4_FIXED0:
1531 *pdata = msrs->cr4_fixed0;
1532 break;
1533 case MSR_IA32_VMX_CR4_FIXED1:
1534 *pdata = msrs->cr4_fixed1;
1535 break;
1536 case MSR_IA32_VMX_VMCS_ENUM:
1537 *pdata = msrs->vmcs_enum;
1538 break;
1539 case MSR_IA32_VMX_PROCBASED_CTLS2:
1540 *pdata = vmx_control_msr(
1541 msrs->secondary_ctls_low,
1542 msrs->secondary_ctls_high);
1543 break;
1544 case MSR_IA32_VMX_EPT_VPID_CAP:
1545 *pdata = msrs->ept_caps |
1546 ((u64)msrs->vpid_caps << 32);
1547 break;
1548 case MSR_IA32_VMX_VMFUNC:
1549 *pdata = msrs->vmfunc_controls;
1550 break;
1551 default:
1552 return 1;
1553 }
1554
1555 return 0;
1556 }
1557
1558 /*
1559 * Copy the writable VMCS shadow fields back to the VMCS12, in case they have
1560 * been modified by the L1 guest. Note, "writable" in this context means
1561 * "writable by the guest", i.e. tagged SHADOW_FIELD_RW; the set of
1562 * fields tagged SHADOW_FIELD_RO may or may not align with the "read-only"
1563 * VM-exit information fields (which are actually writable if the vCPU is
1564 * configured to support "VMWRITE to any supported field in the VMCS").
1565 */
1566 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
1567 {
1568 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1569 struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1570 struct shadow_vmcs_field field;
1571 unsigned long val;
1572 int i;
1573
1574 if (WARN_ON(!shadow_vmcs))
1575 return;
1576
1577 preempt_disable();
1578
1579 vmcs_load(shadow_vmcs);
1580
1581 for (i = 0; i < max_shadow_read_write_fields; i++) {
1582 field = shadow_read_write_fields[i];
1583 val = __vmcs_readl(field.encoding);
1584 vmcs12_write_any(vmcs12, field.encoding, field.offset, val);
1585 }
1586
1587 vmcs_clear(shadow_vmcs);
1588 vmcs_load(vmx->loaded_vmcs->vmcs);
1589
1590 preempt_enable();
1591 }
1592
1593 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
1594 {
1595 const struct shadow_vmcs_field *fields[] = {
1596 shadow_read_write_fields,
1597 shadow_read_only_fields
1598 };
1599 const int max_fields[] = {
1600 max_shadow_read_write_fields,
1601 max_shadow_read_only_fields
1602 };
1603 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
1604 struct vmcs12 *vmcs12 = get_vmcs12(&vmx->vcpu);
1605 struct shadow_vmcs_field field;
1606 unsigned long val;
1607 int i, q;
1608
1609 if (WARN_ON(!shadow_vmcs))
1610 return;
1611
1612 vmcs_load(shadow_vmcs);
1613
1614 for (q = 0; q < ARRAY_SIZE(fields); q++) {
1615 for (i = 0; i < max_fields[q]; i++) {
1616 field = fields[q][i];
1617 val = vmcs12_read_any(vmcs12, field.encoding,
1618 field.offset);
1619 __vmcs_writel(field.encoding, val);
1620 }
1621 }
1622
1623 vmcs_clear(shadow_vmcs);
1624 vmcs_load(vmx->loaded_vmcs->vmcs);
1625 }
1626
1627 static void copy_enlightened_to_vmcs12(struct vcpu_vmx *vmx, u32 hv_clean_fields)
1628 {
1629 #ifdef CONFIG_KVM_HYPERV
1630 struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1631 struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
1632 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(&vmx->vcpu);
1633
1634 /* HV_VMX_ENLIGHTENED_CLEAN_FIELD_NONE */
1635 vmcs12->tpr_threshold = evmcs->tpr_threshold;
1636 vmcs12->guest_rip = evmcs->guest_rip;
1637
1638 if (unlikely(!(hv_clean_fields &
1639 HV_VMX_ENLIGHTENED_CLEAN_FIELD_ENLIGHTENMENTSCONTROL))) {
1640 hv_vcpu->nested.pa_page_gpa = evmcs->partition_assist_page;
1641 hv_vcpu->nested.vm_id = evmcs->hv_vm_id;
1642 hv_vcpu->nested.vp_id = evmcs->hv_vp_id;
1643 }
1644
1645 if (unlikely(!(hv_clean_fields &
1646 HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_BASIC))) {
1647 vmcs12->guest_rsp = evmcs->guest_rsp;
1648 vmcs12->guest_rflags = evmcs->guest_rflags;
1649 vmcs12->guest_interruptibility_info =
1650 evmcs->guest_interruptibility_info;
1651 /*
1652 * Not present in struct vmcs12:
1653 * vmcs12->guest_ssp = evmcs->guest_ssp;
1654 */
1655 }
1656
1657 if (unlikely(!(hv_clean_fields &
1658 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_PROC))) {
1659 vmcs12->cpu_based_vm_exec_control =
1660 evmcs->cpu_based_vm_exec_control;
1661 }
1662
1663 if (unlikely(!(hv_clean_fields &
1664 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EXCPN))) {
1665 vmcs12->exception_bitmap = evmcs->exception_bitmap;
1666 }
1667
1668 if (unlikely(!(hv_clean_fields &
1669 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_ENTRY))) {
1670 vmcs12->vm_entry_controls = evmcs->vm_entry_controls;
1671 }
1672
1673 if (unlikely(!(hv_clean_fields &
1674 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_EVENT))) {
1675 vmcs12->vm_entry_intr_info_field =
1676 evmcs->vm_entry_intr_info_field;
1677 vmcs12->vm_entry_exception_error_code =
1678 evmcs->vm_entry_exception_error_code;
1679 vmcs12->vm_entry_instruction_len =
1680 evmcs->vm_entry_instruction_len;
1681 }
1682
1683 if (unlikely(!(hv_clean_fields &
1684 HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_GRP1))) {
1685 vmcs12->host_ia32_pat = evmcs->host_ia32_pat;
1686 vmcs12->host_ia32_efer = evmcs->host_ia32_efer;
1687 vmcs12->host_cr0 = evmcs->host_cr0;
1688 vmcs12->host_cr3 = evmcs->host_cr3;
1689 vmcs12->host_cr4 = evmcs->host_cr4;
1690 vmcs12->host_ia32_sysenter_esp = evmcs->host_ia32_sysenter_esp;
1691 vmcs12->host_ia32_sysenter_eip = evmcs->host_ia32_sysenter_eip;
1692 vmcs12->host_rip = evmcs->host_rip;
1693 vmcs12->host_ia32_sysenter_cs = evmcs->host_ia32_sysenter_cs;
1694 vmcs12->host_es_selector = evmcs->host_es_selector;
1695 vmcs12->host_cs_selector = evmcs->host_cs_selector;
1696 vmcs12->host_ss_selector = evmcs->host_ss_selector;
1697 vmcs12->host_ds_selector = evmcs->host_ds_selector;
1698 vmcs12->host_fs_selector = evmcs->host_fs_selector;
1699 vmcs12->host_gs_selector = evmcs->host_gs_selector;
1700 vmcs12->host_tr_selector = evmcs->host_tr_selector;
1701 vmcs12->host_ia32_perf_global_ctrl = evmcs->host_ia32_perf_global_ctrl;
1702 /*
1703 * Not present in struct vmcs12:
1704 * vmcs12->host_ia32_s_cet = evmcs->host_ia32_s_cet;
1705 * vmcs12->host_ssp = evmcs->host_ssp;
1706 * vmcs12->host_ia32_int_ssp_table_addr = evmcs->host_ia32_int_ssp_table_addr;
1707 */
1708 }
1709
1710 if (unlikely(!(hv_clean_fields &
1711 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP1))) {
1712 vmcs12->pin_based_vm_exec_control =
1713 evmcs->pin_based_vm_exec_control;
1714 vmcs12->vm_exit_controls = evmcs->vm_exit_controls;
1715 vmcs12->secondary_vm_exec_control =
1716 evmcs->secondary_vm_exec_control;
1717 }
1718
1719 if (unlikely(!(hv_clean_fields &
1720 HV_VMX_ENLIGHTENED_CLEAN_FIELD_IO_BITMAP))) {
1721 vmcs12->io_bitmap_a = evmcs->io_bitmap_a;
1722 vmcs12->io_bitmap_b = evmcs->io_bitmap_b;
1723 }
1724
1725 if (unlikely(!(hv_clean_fields &
1726 HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP))) {
1727 vmcs12->msr_bitmap = evmcs->msr_bitmap;
1728 }
1729
1730 if (unlikely(!(hv_clean_fields &
1731 HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2))) {
1732 vmcs12->guest_es_base = evmcs->guest_es_base;
1733 vmcs12->guest_cs_base = evmcs->guest_cs_base;
1734 vmcs12->guest_ss_base = evmcs->guest_ss_base;
1735 vmcs12->guest_ds_base = evmcs->guest_ds_base;
1736 vmcs12->guest_fs_base = evmcs->guest_fs_base;
1737 vmcs12->guest_gs_base = evmcs->guest_gs_base;
1738 vmcs12->guest_ldtr_base = evmcs->guest_ldtr_base;
1739 vmcs12->guest_tr_base = evmcs->guest_tr_base;
1740 vmcs12->guest_gdtr_base = evmcs->guest_gdtr_base;
1741 vmcs12->guest_idtr_base = evmcs->guest_idtr_base;
1742 vmcs12->guest_es_limit = evmcs->guest_es_limit;
1743 vmcs12->guest_cs_limit = evmcs->guest_cs_limit;
1744 vmcs12->guest_ss_limit = evmcs->guest_ss_limit;
1745 vmcs12->guest_ds_limit = evmcs->guest_ds_limit;
1746 vmcs12->guest_fs_limit = evmcs->guest_fs_limit;
1747 vmcs12->guest_gs_limit = evmcs->guest_gs_limit;
1748 vmcs12->guest_ldtr_limit = evmcs->guest_ldtr_limit;
1749 vmcs12->guest_tr_limit = evmcs->guest_tr_limit;
1750 vmcs12->guest_gdtr_limit = evmcs->guest_gdtr_limit;
1751 vmcs12->guest_idtr_limit = evmcs->guest_idtr_limit;
1752 vmcs12->guest_es_ar_bytes = evmcs->guest_es_ar_bytes;
1753 vmcs12->guest_cs_ar_bytes = evmcs->guest_cs_ar_bytes;
1754 vmcs12->guest_ss_ar_bytes = evmcs->guest_ss_ar_bytes;
1755 vmcs12->guest_ds_ar_bytes = evmcs->guest_ds_ar_bytes;
1756 vmcs12->guest_fs_ar_bytes = evmcs->guest_fs_ar_bytes;
1757 vmcs12->guest_gs_ar_bytes = evmcs->guest_gs_ar_bytes;
1758 vmcs12->guest_ldtr_ar_bytes = evmcs->guest_ldtr_ar_bytes;
1759 vmcs12->guest_tr_ar_bytes = evmcs->guest_tr_ar_bytes;
1760 vmcs12->guest_es_selector = evmcs->guest_es_selector;
1761 vmcs12->guest_cs_selector = evmcs->guest_cs_selector;
1762 vmcs12->guest_ss_selector = evmcs->guest_ss_selector;
1763 vmcs12->guest_ds_selector = evmcs->guest_ds_selector;
1764 vmcs12->guest_fs_selector = evmcs->guest_fs_selector;
1765 vmcs12->guest_gs_selector = evmcs->guest_gs_selector;
1766 vmcs12->guest_ldtr_selector = evmcs->guest_ldtr_selector;
1767 vmcs12->guest_tr_selector = evmcs->guest_tr_selector;
1768 }
1769
1770 if (unlikely(!(hv_clean_fields &
1771 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_GRP2))) {
1772 vmcs12->tsc_offset = evmcs->tsc_offset;
1773 vmcs12->virtual_apic_page_addr = evmcs->virtual_apic_page_addr;
1774 vmcs12->xss_exit_bitmap = evmcs->xss_exit_bitmap;
1775 vmcs12->encls_exiting_bitmap = evmcs->encls_exiting_bitmap;
1776 vmcs12->tsc_multiplier = evmcs->tsc_multiplier;
1777 }
1778
1779 if (unlikely(!(hv_clean_fields &
1780 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CRDR))) {
1781 vmcs12->cr0_guest_host_mask = evmcs->cr0_guest_host_mask;
1782 vmcs12->cr4_guest_host_mask = evmcs->cr4_guest_host_mask;
1783 vmcs12->cr0_read_shadow = evmcs->cr0_read_shadow;
1784 vmcs12->cr4_read_shadow = evmcs->cr4_read_shadow;
1785 vmcs12->guest_cr0 = evmcs->guest_cr0;
1786 vmcs12->guest_cr3 = evmcs->guest_cr3;
1787 vmcs12->guest_cr4 = evmcs->guest_cr4;
1788 vmcs12->guest_dr7 = evmcs->guest_dr7;
1789 }
1790
1791 if (unlikely(!(hv_clean_fields &
1792 HV_VMX_ENLIGHTENED_CLEAN_FIELD_HOST_POINTER))) {
1793 vmcs12->host_fs_base = evmcs->host_fs_base;
1794 vmcs12->host_gs_base = evmcs->host_gs_base;
1795 vmcs12->host_tr_base = evmcs->host_tr_base;
1796 vmcs12->host_gdtr_base = evmcs->host_gdtr_base;
1797 vmcs12->host_idtr_base = evmcs->host_idtr_base;
1798 vmcs12->host_rsp = evmcs->host_rsp;
1799 }
1800
1801 if (unlikely(!(hv_clean_fields &
1802 HV_VMX_ENLIGHTENED_CLEAN_FIELD_CONTROL_XLAT))) {
1803 vmcs12->ept_pointer = evmcs->ept_pointer;
1804 vmcs12->virtual_processor_id = evmcs->virtual_processor_id;
1805 }
1806
1807 if (unlikely(!(hv_clean_fields &
1808 HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1))) {
1809 vmcs12->vmcs_link_pointer = evmcs->vmcs_link_pointer;
1810 vmcs12->guest_ia32_debugctl = evmcs->guest_ia32_debugctl;
1811 vmcs12->guest_ia32_pat = evmcs->guest_ia32_pat;
1812 vmcs12->guest_ia32_efer = evmcs->guest_ia32_efer;
1813 vmcs12->guest_pdptr0 = evmcs->guest_pdptr0;
1814 vmcs12->guest_pdptr1 = evmcs->guest_pdptr1;
1815 vmcs12->guest_pdptr2 = evmcs->guest_pdptr2;
1816 vmcs12->guest_pdptr3 = evmcs->guest_pdptr3;
1817 vmcs12->guest_pending_dbg_exceptions =
1818 evmcs->guest_pending_dbg_exceptions;
1819 vmcs12->guest_sysenter_esp = evmcs->guest_sysenter_esp;
1820 vmcs12->guest_sysenter_eip = evmcs->guest_sysenter_eip;
1821 vmcs12->guest_bndcfgs = evmcs->guest_bndcfgs;
1822 vmcs12->guest_activity_state = evmcs->guest_activity_state;
1823 vmcs12->guest_sysenter_cs = evmcs->guest_sysenter_cs;
1824 vmcs12->guest_ia32_perf_global_ctrl = evmcs->guest_ia32_perf_global_ctrl;
1825 /*
1826 * Not present in struct vmcs12:
1827 * vmcs12->guest_ia32_s_cet = evmcs->guest_ia32_s_cet;
1828 * vmcs12->guest_ia32_lbr_ctl = evmcs->guest_ia32_lbr_ctl;
1829 * vmcs12->guest_ia32_int_ssp_table_addr = evmcs->guest_ia32_int_ssp_table_addr;
1830 */
1831 }
1832
1833 /*
1834 * Not used?
1835 * vmcs12->vm_exit_msr_store_addr = evmcs->vm_exit_msr_store_addr;
1836 * vmcs12->vm_exit_msr_load_addr = evmcs->vm_exit_msr_load_addr;
1837 * vmcs12->vm_entry_msr_load_addr = evmcs->vm_entry_msr_load_addr;
1838 * vmcs12->page_fault_error_code_mask =
1839 * evmcs->page_fault_error_code_mask;
1840 * vmcs12->page_fault_error_code_match =
1841 * evmcs->page_fault_error_code_match;
1842 * vmcs12->cr3_target_count = evmcs->cr3_target_count;
1843 * vmcs12->vm_exit_msr_store_count = evmcs->vm_exit_msr_store_count;
1844 * vmcs12->vm_exit_msr_load_count = evmcs->vm_exit_msr_load_count;
1845 * vmcs12->vm_entry_msr_load_count = evmcs->vm_entry_msr_load_count;
1846 */
1847
1848 /*
1849 * Read only fields:
1850 * vmcs12->guest_physical_address = evmcs->guest_physical_address;
1851 * vmcs12->vm_instruction_error = evmcs->vm_instruction_error;
1852 * vmcs12->vm_exit_reason = evmcs->vm_exit_reason;
1853 * vmcs12->vm_exit_intr_info = evmcs->vm_exit_intr_info;
1854 * vmcs12->vm_exit_intr_error_code = evmcs->vm_exit_intr_error_code;
1855 * vmcs12->idt_vectoring_info_field = evmcs->idt_vectoring_info_field;
1856 * vmcs12->idt_vectoring_error_code = evmcs->idt_vectoring_error_code;
1857 * vmcs12->vm_exit_instruction_len = evmcs->vm_exit_instruction_len;
1858 * vmcs12->vmx_instruction_info = evmcs->vmx_instruction_info;
1859 * vmcs12->exit_qualification = evmcs->exit_qualification;
1860 * vmcs12->guest_linear_address = evmcs->guest_linear_address;
1861 *
1862 * Not present in struct vmcs12:
1863 * vmcs12->exit_io_instruction_ecx = evmcs->exit_io_instruction_ecx;
1864 * vmcs12->exit_io_instruction_esi = evmcs->exit_io_instruction_esi;
1865 * vmcs12->exit_io_instruction_edi = evmcs->exit_io_instruction_edi;
1866 * vmcs12->exit_io_instruction_eip = evmcs->exit_io_instruction_eip;
1867 */
1868
1869 return;
1870 #else /* CONFIG_KVM_HYPERV */
1871 KVM_BUG_ON(1, vmx->vcpu.kvm);
1872 #endif /* CONFIG_KVM_HYPERV */
1873 }
1874
1875 static void copy_vmcs12_to_enlightened(struct vcpu_vmx *vmx)
1876 {
1877 #ifdef CONFIG_KVM_HYPERV
1878 struct vmcs12 *vmcs12 = vmx->nested.cached_vmcs12;
1879 struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
1880
1881 /*
1882 * Should not be changed by KVM:
1883 *
1884 * evmcs->host_es_selector = vmcs12->host_es_selector;
1885 * evmcs->host_cs_selector = vmcs12->host_cs_selector;
1886 * evmcs->host_ss_selector = vmcs12->host_ss_selector;
1887 * evmcs->host_ds_selector = vmcs12->host_ds_selector;
1888 * evmcs->host_fs_selector = vmcs12->host_fs_selector;
1889 * evmcs->host_gs_selector = vmcs12->host_gs_selector;
1890 * evmcs->host_tr_selector = vmcs12->host_tr_selector;
1891 * evmcs->host_ia32_pat = vmcs12->host_ia32_pat;
1892 * evmcs->host_ia32_efer = vmcs12->host_ia32_efer;
1893 * evmcs->host_cr0 = vmcs12->host_cr0;
1894 * evmcs->host_cr3 = vmcs12->host_cr3;
1895 * evmcs->host_cr4 = vmcs12->host_cr4;
1896 * evmcs->host_ia32_sysenter_esp = vmcs12->host_ia32_sysenter_esp;
1897 * evmcs->host_ia32_sysenter_eip = vmcs12->host_ia32_sysenter_eip;
1898 * evmcs->host_rip = vmcs12->host_rip;
1899 * evmcs->host_ia32_sysenter_cs = vmcs12->host_ia32_sysenter_cs;
1900 * evmcs->host_fs_base = vmcs12->host_fs_base;
1901 * evmcs->host_gs_base = vmcs12->host_gs_base;
1902 * evmcs->host_tr_base = vmcs12->host_tr_base;
1903 * evmcs->host_gdtr_base = vmcs12->host_gdtr_base;
1904 * evmcs->host_idtr_base = vmcs12->host_idtr_base;
1905 * evmcs->host_rsp = vmcs12->host_rsp;
1906 * sync_vmcs02_to_vmcs12() doesn't read these:
1907 * evmcs->io_bitmap_a = vmcs12->io_bitmap_a;
1908 * evmcs->io_bitmap_b = vmcs12->io_bitmap_b;
1909 * evmcs->msr_bitmap = vmcs12->msr_bitmap;
1910 * evmcs->ept_pointer = vmcs12->ept_pointer;
1911 * evmcs->xss_exit_bitmap = vmcs12->xss_exit_bitmap;
1912 * evmcs->vm_exit_msr_store_addr = vmcs12->vm_exit_msr_store_addr;
1913 * evmcs->vm_exit_msr_load_addr = vmcs12->vm_exit_msr_load_addr;
1914 * evmcs->vm_entry_msr_load_addr = vmcs12->vm_entry_msr_load_addr;
1915 * evmcs->tpr_threshold = vmcs12->tpr_threshold;
1916 * evmcs->virtual_processor_id = vmcs12->virtual_processor_id;
1917 * evmcs->exception_bitmap = vmcs12->exception_bitmap;
1918 * evmcs->vmcs_link_pointer = vmcs12->vmcs_link_pointer;
1919 * evmcs->pin_based_vm_exec_control = vmcs12->pin_based_vm_exec_control;
1920 * evmcs->vm_exit_controls = vmcs12->vm_exit_controls;
1921 * evmcs->secondary_vm_exec_control = vmcs12->secondary_vm_exec_control;
1922 * evmcs->page_fault_error_code_mask =
1923 * vmcs12->page_fault_error_code_mask;
1924 * evmcs->page_fault_error_code_match =
1925 * vmcs12->page_fault_error_code_match;
1926 * evmcs->cr3_target_count = vmcs12->cr3_target_count;
1927 * evmcs->virtual_apic_page_addr = vmcs12->virtual_apic_page_addr;
1928 * evmcs->tsc_offset = vmcs12->tsc_offset;
1929 * evmcs->guest_ia32_debugctl = vmcs12->guest_ia32_debugctl;
1930 * evmcs->cr0_guest_host_mask = vmcs12->cr0_guest_host_mask;
1931 * evmcs->cr4_guest_host_mask = vmcs12->cr4_guest_host_mask;
1932 * evmcs->cr0_read_shadow = vmcs12->cr0_read_shadow;
1933 * evmcs->cr4_read_shadow = vmcs12->cr4_read_shadow;
1934 * evmcs->vm_exit_msr_store_count = vmcs12->vm_exit_msr_store_count;
1935 * evmcs->vm_exit_msr_load_count = vmcs12->vm_exit_msr_load_count;
1936 * evmcs->vm_entry_msr_load_count = vmcs12->vm_entry_msr_load_count;
1937 * evmcs->guest_ia32_perf_global_ctrl = vmcs12->guest_ia32_perf_global_ctrl;
1938 * evmcs->host_ia32_perf_global_ctrl = vmcs12->host_ia32_perf_global_ctrl;
1939 * evmcs->encls_exiting_bitmap = vmcs12->encls_exiting_bitmap;
1940 * evmcs->tsc_multiplier = vmcs12->tsc_multiplier;
1941 *
1942 * Not present in struct vmcs12:
1943 * evmcs->exit_io_instruction_ecx = vmcs12->exit_io_instruction_ecx;
1944 * evmcs->exit_io_instruction_esi = vmcs12->exit_io_instruction_esi;
1945 * evmcs->exit_io_instruction_edi = vmcs12->exit_io_instruction_edi;
1946 * evmcs->exit_io_instruction_eip = vmcs12->exit_io_instruction_eip;
1947 * evmcs->host_ia32_s_cet = vmcs12->host_ia32_s_cet;
1948 * evmcs->host_ssp = vmcs12->host_ssp;
1949 * evmcs->host_ia32_int_ssp_table_addr = vmcs12->host_ia32_int_ssp_table_addr;
1950 * evmcs->guest_ia32_s_cet = vmcs12->guest_ia32_s_cet;
1951 * evmcs->guest_ia32_lbr_ctl = vmcs12->guest_ia32_lbr_ctl;
1952 * evmcs->guest_ia32_int_ssp_table_addr = vmcs12->guest_ia32_int_ssp_table_addr;
1953 * evmcs->guest_ssp = vmcs12->guest_ssp;
1954 */
1955
1956 evmcs->guest_es_selector = vmcs12->guest_es_selector;
1957 evmcs->guest_cs_selector = vmcs12->guest_cs_selector;
1958 evmcs->guest_ss_selector = vmcs12->guest_ss_selector;
1959 evmcs->guest_ds_selector = vmcs12->guest_ds_selector;
1960 evmcs->guest_fs_selector = vmcs12->guest_fs_selector;
1961 evmcs->guest_gs_selector = vmcs12->guest_gs_selector;
1962 evmcs->guest_ldtr_selector = vmcs12->guest_ldtr_selector;
1963 evmcs->guest_tr_selector = vmcs12->guest_tr_selector;
1964
1965 evmcs->guest_es_limit = vmcs12->guest_es_limit;
1966 evmcs->guest_cs_limit = vmcs12->guest_cs_limit;
1967 evmcs->guest_ss_limit = vmcs12->guest_ss_limit;
1968 evmcs->guest_ds_limit = vmcs12->guest_ds_limit;
1969 evmcs->guest_fs_limit = vmcs12->guest_fs_limit;
1970 evmcs->guest_gs_limit = vmcs12->guest_gs_limit;
1971 evmcs->guest_ldtr_limit = vmcs12->guest_ldtr_limit;
1972 evmcs->guest_tr_limit = vmcs12->guest_tr_limit;
1973 evmcs->guest_gdtr_limit = vmcs12->guest_gdtr_limit;
1974 evmcs->guest_idtr_limit = vmcs12->guest_idtr_limit;
1975
1976 evmcs->guest_es_ar_bytes = vmcs12->guest_es_ar_bytes;
1977 evmcs->guest_cs_ar_bytes = vmcs12->guest_cs_ar_bytes;
1978 evmcs->guest_ss_ar_bytes = vmcs12->guest_ss_ar_bytes;
1979 evmcs->guest_ds_ar_bytes = vmcs12->guest_ds_ar_bytes;
1980 evmcs->guest_fs_ar_bytes = vmcs12->guest_fs_ar_bytes;
1981 evmcs->guest_gs_ar_bytes = vmcs12->guest_gs_ar_bytes;
1982 evmcs->guest_ldtr_ar_bytes = vmcs12->guest_ldtr_ar_bytes;
1983 evmcs->guest_tr_ar_bytes = vmcs12->guest_tr_ar_bytes;
1984
1985 evmcs->guest_es_base = vmcs12->guest_es_base;
1986 evmcs->guest_cs_base = vmcs12->guest_cs_base;
1987 evmcs->guest_ss_base = vmcs12->guest_ss_base;
1988 evmcs->guest_ds_base = vmcs12->guest_ds_base;
1989 evmcs->guest_fs_base = vmcs12->guest_fs_base;
1990 evmcs->guest_gs_base = vmcs12->guest_gs_base;
1991 evmcs->guest_ldtr_base = vmcs12->guest_ldtr_base;
1992 evmcs->guest_tr_base = vmcs12->guest_tr_base;
1993 evmcs->guest_gdtr_base = vmcs12->guest_gdtr_base;
1994 evmcs->guest_idtr_base = vmcs12->guest_idtr_base;
1995
1996 evmcs->guest_ia32_pat = vmcs12->guest_ia32_pat;
1997 evmcs->guest_ia32_efer = vmcs12->guest_ia32_efer;
1998
1999 evmcs->guest_pdptr0 = vmcs12->guest_pdptr0;
2000 evmcs->guest_pdptr1 = vmcs12->guest_pdptr1;
2001 evmcs->guest_pdptr2 = vmcs12->guest_pdptr2;
2002 evmcs->guest_pdptr3 = vmcs12->guest_pdptr3;
2003
2004 evmcs->guest_pending_dbg_exceptions =
2005 vmcs12->guest_pending_dbg_exceptions;
2006 evmcs->guest_sysenter_esp = vmcs12->guest_sysenter_esp;
2007 evmcs->guest_sysenter_eip = vmcs12->guest_sysenter_eip;
2008
2009 evmcs->guest_activity_state = vmcs12->guest_activity_state;
2010 evmcs->guest_sysenter_cs = vmcs12->guest_sysenter_cs;
2011
2012 evmcs->guest_cr0 = vmcs12->guest_cr0;
2013 evmcs->guest_cr3 = vmcs12->guest_cr3;
2014 evmcs->guest_cr4 = vmcs12->guest_cr4;
2015 evmcs->guest_dr7 = vmcs12->guest_dr7;
2016
2017 evmcs->guest_physical_address = vmcs12->guest_physical_address;
2018
2019 evmcs->vm_instruction_error = vmcs12->vm_instruction_error;
2020 evmcs->vm_exit_reason = vmcs12->vm_exit_reason;
2021 evmcs->vm_exit_intr_info = vmcs12->vm_exit_intr_info;
2022 evmcs->vm_exit_intr_error_code = vmcs12->vm_exit_intr_error_code;
2023 evmcs->idt_vectoring_info_field = vmcs12->idt_vectoring_info_field;
2024 evmcs->idt_vectoring_error_code = vmcs12->idt_vectoring_error_code;
2025 evmcs->vm_exit_instruction_len = vmcs12->vm_exit_instruction_len;
2026 evmcs->vmx_instruction_info = vmcs12->vmx_instruction_info;
2027
2028 evmcs->exit_qualification = vmcs12->exit_qualification;
2029
2030 evmcs->guest_linear_address = vmcs12->guest_linear_address;
2031 evmcs->guest_rsp = vmcs12->guest_rsp;
2032 evmcs->guest_rflags = vmcs12->guest_rflags;
2033
2034 evmcs->guest_interruptibility_info =
2035 vmcs12->guest_interruptibility_info;
2036 evmcs->cpu_based_vm_exec_control = vmcs12->cpu_based_vm_exec_control;
2037 evmcs->vm_entry_controls = vmcs12->vm_entry_controls;
2038 evmcs->vm_entry_intr_info_field = vmcs12->vm_entry_intr_info_field;
2039 evmcs->vm_entry_exception_error_code =
2040 vmcs12->vm_entry_exception_error_code;
2041 evmcs->vm_entry_instruction_len = vmcs12->vm_entry_instruction_len;
2042
2043 evmcs->guest_rip = vmcs12->guest_rip;
2044
2045 evmcs->guest_bndcfgs = vmcs12->guest_bndcfgs;
2046
2047 return;
2048 #else /* CONFIG_KVM_HYPERV */
2049 KVM_BUG_ON(1, vmx->vcpu.kvm);
2050 #endif /* CONFIG_KVM_HYPERV */
2051 }
2052
2053 /*
2054 * This is an equivalent of the nested hypervisor executing the vmptrld
2055 * instruction.
2056 */
2057 static enum nested_evmptrld_status nested_vmx_handle_enlightened_vmptrld(
2058 struct kvm_vcpu *vcpu, bool from_launch)
2059 {
2060 #ifdef CONFIG_KVM_HYPERV
2061 struct vcpu_vmx *vmx = to_vmx(vcpu);
2062 bool evmcs_gpa_changed = false;
2063 u64 evmcs_gpa;
2064
2065 if (likely(!guest_cpuid_has_evmcs(vcpu)))
2066 return EVMPTRLD_DISABLED;
2067
2068 evmcs_gpa = nested_get_evmptr(vcpu);
2069 if (!evmptr_is_valid(evmcs_gpa)) {
2070 nested_release_evmcs(vcpu);
2071 return EVMPTRLD_DISABLED;
2072 }
2073
2074 if (unlikely(evmcs_gpa != vmx->nested.hv_evmcs_vmptr)) {
2075 vmx->nested.current_vmptr = INVALID_GPA;
2076
2077 nested_release_evmcs(vcpu);
2078
2079 if (kvm_vcpu_map(vcpu, gpa_to_gfn(evmcs_gpa),
2080 &vmx->nested.hv_evmcs_map))
2081 return EVMPTRLD_ERROR;
2082
2083 vmx->nested.hv_evmcs = vmx->nested.hv_evmcs_map.hva;
2084
2085 /*
2086 * Currently, KVM only supports eVMCS version 1
2087 * (== KVM_EVMCS_VERSION) and thus we expect guest to set this
2088 * value to first u32 field of eVMCS which should specify eVMCS
2089 * VersionNumber.
2090 *
2091 * Guest should be aware of supported eVMCS versions by host by
2092 * examining CPUID.0x4000000A.EAX[0:15]. Host userspace VMM is
2093 * expected to set this CPUID leaf according to the value
2094 * returned in vmcs_version from nested_enable_evmcs().
2095 *
2096 * However, it turns out that Microsoft Hyper-V fails to comply
2097 * to their own invented interface: When Hyper-V use eVMCS, it
2098 * just sets first u32 field of eVMCS to revision_id specified
2099 * in MSR_IA32_VMX_BASIC. Instead of used eVMCS version number
2100 * which is one of the supported versions specified in
2101 * CPUID.0x4000000A.EAX[0:15].
2102 *
2103 * To overcome Hyper-V bug, we accept here either a supported
2104 * eVMCS version or VMCS12 revision_id as valid values for first
2105 * u32 field of eVMCS.
2106 */
2107 if ((vmx->nested.hv_evmcs->revision_id != KVM_EVMCS_VERSION) &&
2108 (vmx->nested.hv_evmcs->revision_id != VMCS12_REVISION)) {
2109 nested_release_evmcs(vcpu);
2110 return EVMPTRLD_VMFAIL;
2111 }
2112
2113 vmx->nested.hv_evmcs_vmptr = evmcs_gpa;
2114
2115 evmcs_gpa_changed = true;
2116 /*
2117 * Unlike normal vmcs12, enlightened vmcs12 is not fully
2118 * reloaded from guest's memory (read only fields, fields not
2119 * present in struct hv_enlightened_vmcs, ...). Make sure there
2120 * are no leftovers.
2121 */
2122 if (from_launch) {
2123 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2124 memset(vmcs12, 0, sizeof(*vmcs12));
2125 vmcs12->hdr.revision_id = VMCS12_REVISION;
2126 }
2127
2128 }
2129
2130 /*
2131 * Clean fields data can't be used on VMLAUNCH and when we switch
2132 * between different L2 guests as KVM keeps a single VMCS12 per L1.
2133 */
2134 if (from_launch || evmcs_gpa_changed) {
2135 vmx->nested.hv_evmcs->hv_clean_fields &=
2136 ~HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2137
2138 vmx->nested.force_msr_bitmap_recalc = true;
2139 }
2140
2141 return EVMPTRLD_SUCCEEDED;
2142 #else
2143 return EVMPTRLD_DISABLED;
2144 #endif
2145 }
2146
2147 void nested_sync_vmcs12_to_shadow(struct kvm_vcpu *vcpu)
2148 {
2149 struct vcpu_vmx *vmx = to_vmx(vcpu);
2150
2151 if (nested_vmx_is_evmptr12_valid(vmx))
2152 copy_vmcs12_to_enlightened(vmx);
2153 else
2154 copy_vmcs12_to_shadow(vmx);
2155
2156 vmx->nested.need_vmcs12_to_shadow_sync = false;
2157 }
2158
2159 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
2160 {
2161 struct vcpu_vmx *vmx =
2162 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
2163
2164 vmx->nested.preemption_timer_expired = true;
2165 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
2166 kvm_vcpu_kick(&vmx->vcpu);
2167
2168 return HRTIMER_NORESTART;
2169 }
2170
2171 static u64 vmx_calc_preemption_timer_value(struct kvm_vcpu *vcpu)
2172 {
2173 struct vcpu_vmx *vmx = to_vmx(vcpu);
2174 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2175
2176 u64 l1_scaled_tsc = kvm_read_l1_tsc(vcpu, rdtsc()) >>
2177 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2178
2179 if (!vmx->nested.has_preemption_timer_deadline) {
2180 vmx->nested.preemption_timer_deadline =
2181 vmcs12->vmx_preemption_timer_value + l1_scaled_tsc;
2182 vmx->nested.has_preemption_timer_deadline = true;
2183 }
2184 return vmx->nested.preemption_timer_deadline - l1_scaled_tsc;
2185 }
2186
2187 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu,
2188 u64 preemption_timeout)
2189 {
2190 struct vcpu_vmx *vmx = to_vmx(vcpu);
2191
2192 /*
2193 * A timer value of zero is architecturally guaranteed to cause
2194 * a VMExit prior to executing any instructions in the guest.
2195 */
2196 if (preemption_timeout == 0) {
2197 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
2198 return;
2199 }
2200
2201 if (vcpu->arch.virtual_tsc_khz == 0)
2202 return;
2203
2204 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
2205 preemption_timeout *= 1000000;
2206 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
2207 hrtimer_start(&vmx->nested.preemption_timer,
2208 ktime_add_ns(ktime_get(), preemption_timeout),
2209 HRTIMER_MODE_ABS_PINNED);
2210 }
2211
2212 static u64 nested_vmx_calc_efer(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2213 {
2214 if (vmx->nested.nested_run_pending &&
2215 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
2216 return vmcs12->guest_ia32_efer;
2217 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
2218 return vmx->vcpu.arch.efer | (EFER_LMA | EFER_LME);
2219 else
2220 return vmx->vcpu.arch.efer & ~(EFER_LMA | EFER_LME);
2221 }
2222
2223 static void prepare_vmcs02_constant_state(struct vcpu_vmx *vmx)
2224 {
2225 struct kvm *kvm = vmx->vcpu.kvm;
2226
2227 /*
2228 * If vmcs02 hasn't been initialized, set the constant vmcs02 state
2229 * according to L0's settings (vmcs12 is irrelevant here). Host
2230 * fields that come from L0 and are not constant, e.g. HOST_CR3,
2231 * will be set as needed prior to VMLAUNCH/VMRESUME.
2232 */
2233 if (vmx->nested.vmcs02_initialized)
2234 return;
2235 vmx->nested.vmcs02_initialized = true;
2236
2237 /*
2238 * We don't care what the EPTP value is we just need to guarantee
2239 * it's valid so we don't get a false positive when doing early
2240 * consistency checks.
2241 */
2242 if (enable_ept && nested_early_check)
2243 vmcs_write64(EPT_POINTER,
2244 construct_eptp(&vmx->vcpu, 0, PT64_ROOT_4LEVEL));
2245
2246 if (vmx->ve_info)
2247 vmcs_write64(VE_INFORMATION_ADDRESS, __pa(vmx->ve_info));
2248
2249 /* All VMFUNCs are currently emulated through L0 vmexits. */
2250 if (cpu_has_vmx_vmfunc())
2251 vmcs_write64(VM_FUNCTION_CONTROL, 0);
2252
2253 if (cpu_has_vmx_posted_intr())
2254 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
2255
2256 if (cpu_has_vmx_msr_bitmap())
2257 vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
2258
2259 /*
2260 * PML is emulated for L2, but never enabled in hardware as the MMU
2261 * handles A/D emulation. Disabling PML for L2 also avoids having to
2262 * deal with filtering out L2 GPAs from the buffer.
2263 */
2264 if (enable_pml) {
2265 vmcs_write64(PML_ADDRESS, 0);
2266 vmcs_write16(GUEST_PML_INDEX, -1);
2267 }
2268
2269 if (cpu_has_vmx_encls_vmexit())
2270 vmcs_write64(ENCLS_EXITING_BITMAP, INVALID_GPA);
2271
2272 if (kvm_notify_vmexit_enabled(kvm))
2273 vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
2274
2275 /*
2276 * Set the MSR load/store lists to match L0's settings. Only the
2277 * addresses are constant (for vmcs02), the counts can change based
2278 * on L2's behavior, e.g. switching to/from long mode.
2279 */
2280 vmcs_write64(VM_EXIT_MSR_STORE_ADDR, __pa(vmx->msr_autostore.guest.val));
2281 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
2282 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
2283
2284 vmx_set_constant_host_state(vmx);
2285 }
2286
2287 static void prepare_vmcs02_early_rare(struct vcpu_vmx *vmx,
2288 struct vmcs12 *vmcs12)
2289 {
2290 prepare_vmcs02_constant_state(vmx);
2291
2292 vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA);
2293
2294 if (enable_vpid) {
2295 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
2296 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
2297 else
2298 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2299 }
2300 }
2301
2302 static void prepare_vmcs02_early(struct vcpu_vmx *vmx, struct loaded_vmcs *vmcs01,
2303 struct vmcs12 *vmcs12)
2304 {
2305 u32 exec_control;
2306 u64 guest_efer = nested_vmx_calc_efer(vmx, vmcs12);
2307
2308 if (vmx->nested.dirty_vmcs12 || nested_vmx_is_evmptr12_valid(vmx))
2309 prepare_vmcs02_early_rare(vmx, vmcs12);
2310
2311 /*
2312 * PIN CONTROLS
2313 */
2314 exec_control = __pin_controls_get(vmcs01);
2315 exec_control |= (vmcs12->pin_based_vm_exec_control &
2316 ~PIN_BASED_VMX_PREEMPTION_TIMER);
2317
2318 /* Posted interrupts setting is only taken from vmcs12. */
2319 vmx->nested.pi_pending = false;
2320 if (nested_cpu_has_posted_intr(vmcs12))
2321 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
2322 else
2323 exec_control &= ~PIN_BASED_POSTED_INTR;
2324 pin_controls_set(vmx, exec_control);
2325
2326 /*
2327 * EXEC CONTROLS
2328 */
2329 exec_control = __exec_controls_get(vmcs01); /* L0's desires */
2330 exec_control &= ~CPU_BASED_INTR_WINDOW_EXITING;
2331 exec_control &= ~CPU_BASED_NMI_WINDOW_EXITING;
2332 exec_control &= ~CPU_BASED_TPR_SHADOW;
2333 exec_control |= vmcs12->cpu_based_vm_exec_control;
2334
2335 vmx->nested.l1_tpr_threshold = -1;
2336 if (exec_control & CPU_BASED_TPR_SHADOW)
2337 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
2338 #ifdef CONFIG_X86_64
2339 else
2340 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
2341 CPU_BASED_CR8_STORE_EXITING;
2342 #endif
2343
2344 /*
2345 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
2346 * for I/O port accesses.
2347 */
2348 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
2349 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
2350
2351 /*
2352 * This bit will be computed in nested_get_vmcs12_pages, because
2353 * we do not have access to L1's MSR bitmap yet. For now, keep
2354 * the same bit as before, hoping to avoid multiple VMWRITEs that
2355 * only set/clear this bit.
2356 */
2357 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
2358 exec_control |= exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS;
2359
2360 exec_controls_set(vmx, exec_control);
2361
2362 /*
2363 * SECONDARY EXEC CONTROLS
2364 */
2365 if (cpu_has_secondary_exec_ctrls()) {
2366 exec_control = __secondary_exec_controls_get(vmcs01);
2367
2368 /* Take the following fields only from vmcs12 */
2369 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2370 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2371 SECONDARY_EXEC_ENABLE_INVPCID |
2372 SECONDARY_EXEC_ENABLE_RDTSCP |
2373 SECONDARY_EXEC_ENABLE_XSAVES |
2374 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2375 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2376 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2377 SECONDARY_EXEC_ENABLE_VMFUNC |
2378 SECONDARY_EXEC_DESC);
2379
2380 if (nested_cpu_has(vmcs12,
2381 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
2382 exec_control |= vmcs12->secondary_vm_exec_control;
2383
2384 /* PML is emulated and never enabled in hardware for L2. */
2385 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
2386
2387 /* VMCS shadowing for L2 is emulated for now */
2388 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
2389
2390 /*
2391 * Preset *DT exiting when emulating UMIP, so that vmx_set_cr4()
2392 * will not have to rewrite the controls just for this bit.
2393 */
2394 if (vmx_umip_emulated() && (vmcs12->guest_cr4 & X86_CR4_UMIP))
2395 exec_control |= SECONDARY_EXEC_DESC;
2396
2397 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
2398 vmcs_write16(GUEST_INTR_STATUS,
2399 vmcs12->guest_intr_status);
2400
2401 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
2402 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2403
2404 if (exec_control & SECONDARY_EXEC_ENCLS_EXITING)
2405 vmx_write_encls_bitmap(&vmx->vcpu, vmcs12);
2406
2407 secondary_exec_controls_set(vmx, exec_control);
2408 }
2409
2410 /*
2411 * ENTRY CONTROLS
2412 *
2413 * vmcs12's VM_{ENTRY,EXIT}_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE
2414 * are emulated by vmx_set_efer() in prepare_vmcs02(), but speculate
2415 * on the related bits (if supported by the CPU) in the hope that
2416 * we can avoid VMWrites during vmx_set_efer().
2417 *
2418 * Similarly, take vmcs01's PERF_GLOBAL_CTRL in the hope that if KVM is
2419 * loading PERF_GLOBAL_CTRL via the VMCS for L1, then KVM will want to
2420 * do the same for L2.
2421 */
2422 exec_control = __vm_entry_controls_get(vmcs01);
2423 exec_control |= (vmcs12->vm_entry_controls &
2424 ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
2425 exec_control &= ~(VM_ENTRY_IA32E_MODE | VM_ENTRY_LOAD_IA32_EFER);
2426 if (cpu_has_load_ia32_efer()) {
2427 if (guest_efer & EFER_LMA)
2428 exec_control |= VM_ENTRY_IA32E_MODE;
2429 if (guest_efer != kvm_host.efer)
2430 exec_control |= VM_ENTRY_LOAD_IA32_EFER;
2431 }
2432 vm_entry_controls_set(vmx, exec_control);
2433
2434 /*
2435 * EXIT CONTROLS
2436 *
2437 * L2->L1 exit controls are emulated - the hardware exit is to L0 so
2438 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
2439 * bits may be modified by vmx_set_efer() in prepare_vmcs02().
2440 */
2441 exec_control = __vm_exit_controls_get(vmcs01);
2442 if (cpu_has_load_ia32_efer() && guest_efer != kvm_host.efer)
2443 exec_control |= VM_EXIT_LOAD_IA32_EFER;
2444 else
2445 exec_control &= ~VM_EXIT_LOAD_IA32_EFER;
2446 vm_exit_controls_set(vmx, exec_control);
2447
2448 /*
2449 * Interrupt/Exception Fields
2450 */
2451 if (vmx->nested.nested_run_pending) {
2452 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2453 vmcs12->vm_entry_intr_info_field);
2454 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2455 vmcs12->vm_entry_exception_error_code);
2456 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2457 vmcs12->vm_entry_instruction_len);
2458 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
2459 vmcs12->guest_interruptibility_info);
2460 vmx->loaded_vmcs->nmi_known_unmasked =
2461 !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
2462 } else {
2463 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
2464 }
2465 }
2466
2467 static void prepare_vmcs02_rare(struct vcpu_vmx *vmx, struct vmcs12 *vmcs12)
2468 {
2469 struct hv_enlightened_vmcs *hv_evmcs = nested_vmx_evmcs(vmx);
2470
2471 if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2472 HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP2)) {
2473 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
2474 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
2475 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
2476 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
2477 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
2478 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
2479 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
2480 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
2481 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
2482 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
2483 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
2484 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
2485 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
2486 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
2487 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
2488 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
2489 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
2490 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
2491 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
2492 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
2493 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
2494 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
2495 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
2496 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
2497 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
2498 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
2499 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
2500 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
2501 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
2502 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
2503 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
2504 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
2505 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
2506 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
2507 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
2508 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
2509
2510 vmx->segment_cache.bitmask = 0;
2511 }
2512
2513 if (!hv_evmcs || !(hv_evmcs->hv_clean_fields &
2514 HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1)) {
2515 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
2516 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
2517 vmcs12->guest_pending_dbg_exceptions);
2518 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
2519 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
2520
2521 /*
2522 * L1 may access the L2's PDPTR, so save them to construct
2523 * vmcs12
2524 */
2525 if (enable_ept) {
2526 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2527 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2528 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2529 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2530 }
2531
2532 if (kvm_mpx_supported() && vmx->nested.nested_run_pending &&
2533 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS))
2534 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
2535 }
2536
2537 if (nested_cpu_has_xsaves(vmcs12))
2538 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
2539
2540 /*
2541 * Whether page-faults are trapped is determined by a combination of
2542 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF. If L0
2543 * doesn't care about page faults then we should set all of these to
2544 * L1's desires. However, if L0 does care about (some) page faults, it
2545 * is not easy (if at all possible?) to merge L0 and L1's desires, we
2546 * simply ask to exit on each and every L2 page fault. This is done by
2547 * setting MASK=MATCH=0 and (see below) EB.PF=1.
2548 * Note that below we don't need special code to set EB.PF beyond the
2549 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
2550 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
2551 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
2552 */
2553 if (vmx_need_pf_intercept(&vmx->vcpu)) {
2554 /*
2555 * TODO: if both L0 and L1 need the same MASK and MATCH,
2556 * go ahead and use it?
2557 */
2558 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
2559 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
2560 } else {
2561 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, vmcs12->page_fault_error_code_mask);
2562 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, vmcs12->page_fault_error_code_match);
2563 }
2564
2565 if (cpu_has_vmx_apicv()) {
2566 vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
2567 vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
2568 vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
2569 vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
2570 }
2571
2572 /*
2573 * Make sure the msr_autostore list is up to date before we set the
2574 * count in the vmcs02.
2575 */
2576 prepare_vmx_msr_autostore_list(&vmx->vcpu, MSR_IA32_TSC);
2577
2578 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, vmx->msr_autostore.guest.nr);
2579 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
2580 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
2581
2582 set_cr4_guest_host_mask(vmx);
2583 }
2584
2585 /*
2586 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
2587 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
2588 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
2589 * guest in a way that will both be appropriate to L1's requests, and our
2590 * needs. In addition to modifying the active vmcs (which is vmcs02), this
2591 * function also has additional necessary side-effects, like setting various
2592 * vcpu->arch fields.
2593 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
2594 * is assigned to entry_failure_code on failure.
2595 */
2596 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
2597 bool from_vmentry,
2598 enum vm_entry_failure_code *entry_failure_code)
2599 {
2600 struct vcpu_vmx *vmx = to_vmx(vcpu);
2601 struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
2602 bool load_guest_pdptrs_vmcs12 = false;
2603
2604 if (vmx->nested.dirty_vmcs12 || nested_vmx_is_evmptr12_valid(vmx)) {
2605 prepare_vmcs02_rare(vmx, vmcs12);
2606 vmx->nested.dirty_vmcs12 = false;
2607
2608 load_guest_pdptrs_vmcs12 = !nested_vmx_is_evmptr12_valid(vmx) ||
2609 !(evmcs->hv_clean_fields & HV_VMX_ENLIGHTENED_CLEAN_FIELD_GUEST_GRP1);
2610 }
2611
2612 if (vmx->nested.nested_run_pending &&
2613 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
2614 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
2615 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
2616 } else {
2617 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
2618 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.pre_vmenter_debugctl);
2619 }
2620 if (kvm_mpx_supported() && (!vmx->nested.nested_run_pending ||
2621 !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
2622 vmcs_write64(GUEST_BNDCFGS, vmx->nested.pre_vmenter_bndcfgs);
2623 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
2624
2625 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
2626 * bitwise-or of what L1 wants to trap for L2, and what we want to
2627 * trap. Note that CR0.TS also needs updating - we do this later.
2628 */
2629 vmx_update_exception_bitmap(vcpu);
2630 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
2631 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2632
2633 if (vmx->nested.nested_run_pending &&
2634 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
2635 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
2636 vcpu->arch.pat = vmcs12->guest_ia32_pat;
2637 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2638 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
2639 }
2640
2641 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2642 vcpu->arch.l1_tsc_offset,
2643 vmx_get_l2_tsc_offset(vcpu),
2644 vmx_get_l2_tsc_multiplier(vcpu));
2645
2646 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2647 vcpu->arch.l1_tsc_scaling_ratio,
2648 vmx_get_l2_tsc_multiplier(vcpu));
2649
2650 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
2651 if (kvm_caps.has_tsc_control)
2652 vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
2653
2654 nested_vmx_transition_tlb_flush(vcpu, vmcs12, true);
2655
2656 if (nested_cpu_has_ept(vmcs12))
2657 nested_ept_init_mmu_context(vcpu);
2658
2659 /*
2660 * Override the CR0/CR4 read shadows after setting the effective guest
2661 * CR0/CR4. The common helpers also set the shadows, but they don't
2662 * account for vmcs12's cr0/4_guest_host_mask.
2663 */
2664 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
2665 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2666
2667 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
2668 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
2669
2670 vcpu->arch.efer = nested_vmx_calc_efer(vmx, vmcs12);
2671 /* Note: may modify VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
2672 vmx_set_efer(vcpu, vcpu->arch.efer);
2673
2674 /*
2675 * Guest state is invalid and unrestricted guest is disabled,
2676 * which means L1 attempted VMEntry to L2 with invalid state.
2677 * Fail the VMEntry.
2678 *
2679 * However when force loading the guest state (SMM exit or
2680 * loading nested state after migration, it is possible to
2681 * have invalid guest state now, which will be later fixed by
2682 * restoring L2 register state
2683 */
2684 if (CC(from_vmentry && !vmx_guest_state_valid(vcpu))) {
2685 *entry_failure_code = ENTRY_FAIL_DEFAULT;
2686 return -EINVAL;
2687 }
2688
2689 /* Shadow page tables on either EPT or shadow page tables. */
2690 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
2691 from_vmentry, entry_failure_code))
2692 return -EINVAL;
2693
2694 /*
2695 * Immediately write vmcs02.GUEST_CR3. It will be propagated to vmcs12
2696 * on nested VM-Exit, which can occur without actually running L2 and
2697 * thus without hitting vmx_load_mmu_pgd(), e.g. if L1 is entering L2 with
2698 * vmcs12.GUEST_ACTIVITYSTATE=HLT, in which case KVM will intercept the
2699 * transition to HLT instead of running L2.
2700 */
2701 if (enable_ept)
2702 vmcs_writel(GUEST_CR3, vmcs12->guest_cr3);
2703
2704 /* Late preparation of GUEST_PDPTRs now that EFER and CRs are set. */
2705 if (load_guest_pdptrs_vmcs12 && nested_cpu_has_ept(vmcs12) &&
2706 is_pae_paging(vcpu)) {
2707 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
2708 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
2709 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
2710 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
2711 }
2712
2713 if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2714 kvm_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)) &&
2715 WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
2716 vmcs12->guest_ia32_perf_global_ctrl))) {
2717 *entry_failure_code = ENTRY_FAIL_DEFAULT;
2718 return -EINVAL;
2719 }
2720
2721 kvm_rsp_write(vcpu, vmcs12->guest_rsp);
2722 kvm_rip_write(vcpu, vmcs12->guest_rip);
2723
2724 /*
2725 * It was observed that genuine Hyper-V running in L1 doesn't reset
2726 * 'hv_clean_fields' by itself, it only sets the corresponding dirty
2727 * bits when it changes a field in eVMCS. Mark all fields as clean
2728 * here.
2729 */
2730 if (nested_vmx_is_evmptr12_valid(vmx))
2731 evmcs->hv_clean_fields |= HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
2732
2733 return 0;
2734 }
2735
2736 static int nested_vmx_check_nmi_controls(struct vmcs12 *vmcs12)
2737 {
2738 if (CC(!nested_cpu_has_nmi_exiting(vmcs12) &&
2739 nested_cpu_has_virtual_nmis(vmcs12)))
2740 return -EINVAL;
2741
2742 if (CC(!nested_cpu_has_virtual_nmis(vmcs12) &&
2743 nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING)))
2744 return -EINVAL;
2745
2746 return 0;
2747 }
2748
2749 static bool nested_vmx_check_eptp(struct kvm_vcpu *vcpu, u64 new_eptp)
2750 {
2751 struct vcpu_vmx *vmx = to_vmx(vcpu);
2752
2753 /* Check for memory type validity */
2754 switch (new_eptp & VMX_EPTP_MT_MASK) {
2755 case VMX_EPTP_MT_UC:
2756 if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_UC_BIT)))
2757 return false;
2758 break;
2759 case VMX_EPTP_MT_WB:
2760 if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPTP_WB_BIT)))
2761 return false;
2762 break;
2763 default:
2764 return false;
2765 }
2766
2767 /* Page-walk levels validity. */
2768 switch (new_eptp & VMX_EPTP_PWL_MASK) {
2769 case VMX_EPTP_PWL_5:
2770 if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_5_BIT)))
2771 return false;
2772 break;
2773 case VMX_EPTP_PWL_4:
2774 if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_PAGE_WALK_4_BIT)))
2775 return false;
2776 break;
2777 default:
2778 return false;
2779 }
2780
2781 /* Reserved bits should not be set */
2782 if (CC(!kvm_vcpu_is_legal_gpa(vcpu, new_eptp) || ((new_eptp >> 7) & 0x1f)))
2783 return false;
2784
2785 /* AD, if set, should be supported */
2786 if (new_eptp & VMX_EPTP_AD_ENABLE_BIT) {
2787 if (CC(!(vmx->nested.msrs.ept_caps & VMX_EPT_AD_BIT)))
2788 return false;
2789 }
2790
2791 return true;
2792 }
2793
2794 /*
2795 * Checks related to VM-Execution Control Fields
2796 */
2797 static int nested_check_vm_execution_controls(struct kvm_vcpu *vcpu,
2798 struct vmcs12 *vmcs12)
2799 {
2800 struct vcpu_vmx *vmx = to_vmx(vcpu);
2801
2802 if (CC(!vmx_control_verify(vmcs12->pin_based_vm_exec_control,
2803 vmx->nested.msrs.pinbased_ctls_low,
2804 vmx->nested.msrs.pinbased_ctls_high)) ||
2805 CC(!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
2806 vmx->nested.msrs.procbased_ctls_low,
2807 vmx->nested.msrs.procbased_ctls_high)))
2808 return -EINVAL;
2809
2810 if (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
2811 CC(!vmx_control_verify(vmcs12->secondary_vm_exec_control,
2812 vmx->nested.msrs.secondary_ctls_low,
2813 vmx->nested.msrs.secondary_ctls_high)))
2814 return -EINVAL;
2815
2816 if (CC(vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu)) ||
2817 nested_vmx_check_io_bitmap_controls(vcpu, vmcs12) ||
2818 nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12) ||
2819 nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12) ||
2820 nested_vmx_check_apic_access_controls(vcpu, vmcs12) ||
2821 nested_vmx_check_apicv_controls(vcpu, vmcs12) ||
2822 nested_vmx_check_nmi_controls(vmcs12) ||
2823 nested_vmx_check_pml_controls(vcpu, vmcs12) ||
2824 nested_vmx_check_unrestricted_guest_controls(vcpu, vmcs12) ||
2825 nested_vmx_check_mode_based_ept_exec_controls(vcpu, vmcs12) ||
2826 nested_vmx_check_shadow_vmcs_controls(vcpu, vmcs12) ||
2827 CC(nested_cpu_has_vpid(vmcs12) && !vmcs12->virtual_processor_id))
2828 return -EINVAL;
2829
2830 if (!nested_cpu_has_preemption_timer(vmcs12) &&
2831 nested_cpu_has_save_preemption_timer(vmcs12))
2832 return -EINVAL;
2833
2834 if (nested_cpu_has_ept(vmcs12) &&
2835 CC(!nested_vmx_check_eptp(vcpu, vmcs12->ept_pointer)))
2836 return -EINVAL;
2837
2838 if (nested_cpu_has_vmfunc(vmcs12)) {
2839 if (CC(vmcs12->vm_function_control &
2840 ~vmx->nested.msrs.vmfunc_controls))
2841 return -EINVAL;
2842
2843 if (nested_cpu_has_eptp_switching(vmcs12)) {
2844 if (CC(!nested_cpu_has_ept(vmcs12)) ||
2845 CC(!page_address_valid(vcpu, vmcs12->eptp_list_address)))
2846 return -EINVAL;
2847 }
2848 }
2849
2850 return 0;
2851 }
2852
2853 /*
2854 * Checks related to VM-Exit Control Fields
2855 */
2856 static int nested_check_vm_exit_controls(struct kvm_vcpu *vcpu,
2857 struct vmcs12 *vmcs12)
2858 {
2859 struct vcpu_vmx *vmx = to_vmx(vcpu);
2860
2861 if (CC(!vmx_control_verify(vmcs12->vm_exit_controls,
2862 vmx->nested.msrs.exit_ctls_low,
2863 vmx->nested.msrs.exit_ctls_high)) ||
2864 CC(nested_vmx_check_exit_msr_switch_controls(vcpu, vmcs12)))
2865 return -EINVAL;
2866
2867 return 0;
2868 }
2869
2870 /*
2871 * Checks related to VM-Entry Control Fields
2872 */
2873 static int nested_check_vm_entry_controls(struct kvm_vcpu *vcpu,
2874 struct vmcs12 *vmcs12)
2875 {
2876 struct vcpu_vmx *vmx = to_vmx(vcpu);
2877
2878 if (CC(!vmx_control_verify(vmcs12->vm_entry_controls,
2879 vmx->nested.msrs.entry_ctls_low,
2880 vmx->nested.msrs.entry_ctls_high)))
2881 return -EINVAL;
2882
2883 /*
2884 * From the Intel SDM, volume 3:
2885 * Fields relevant to VM-entry event injection must be set properly.
2886 * These fields are the VM-entry interruption-information field, the
2887 * VM-entry exception error code, and the VM-entry instruction length.
2888 */
2889 if (vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) {
2890 u32 intr_info = vmcs12->vm_entry_intr_info_field;
2891 u8 vector = intr_info & INTR_INFO_VECTOR_MASK;
2892 u32 intr_type = intr_info & INTR_INFO_INTR_TYPE_MASK;
2893 bool has_error_code = intr_info & INTR_INFO_DELIVER_CODE_MASK;
2894 bool should_have_error_code;
2895 bool urg = nested_cpu_has2(vmcs12,
2896 SECONDARY_EXEC_UNRESTRICTED_GUEST);
2897 bool prot_mode = !urg || vmcs12->guest_cr0 & X86_CR0_PE;
2898
2899 /* VM-entry interruption-info field: interruption type */
2900 if (CC(intr_type == INTR_TYPE_RESERVED) ||
2901 CC(intr_type == INTR_TYPE_OTHER_EVENT &&
2902 !nested_cpu_supports_monitor_trap_flag(vcpu)))
2903 return -EINVAL;
2904
2905 /* VM-entry interruption-info field: vector */
2906 if (CC(intr_type == INTR_TYPE_NMI_INTR && vector != NMI_VECTOR) ||
2907 CC(intr_type == INTR_TYPE_HARD_EXCEPTION && vector > 31) ||
2908 CC(intr_type == INTR_TYPE_OTHER_EVENT && vector != 0))
2909 return -EINVAL;
2910
2911 /* VM-entry interruption-info field: deliver error code */
2912 should_have_error_code =
2913 intr_type == INTR_TYPE_HARD_EXCEPTION && prot_mode &&
2914 x86_exception_has_error_code(vector);
2915 if (CC(has_error_code != should_have_error_code))
2916 return -EINVAL;
2917
2918 /* VM-entry exception error code */
2919 if (CC(has_error_code &&
2920 vmcs12->vm_entry_exception_error_code & GENMASK(31, 16)))
2921 return -EINVAL;
2922
2923 /* VM-entry interruption-info field: reserved bits */
2924 if (CC(intr_info & INTR_INFO_RESVD_BITS_MASK))
2925 return -EINVAL;
2926
2927 /* VM-entry instruction length */
2928 switch (intr_type) {
2929 case INTR_TYPE_SOFT_EXCEPTION:
2930 case INTR_TYPE_SOFT_INTR:
2931 case INTR_TYPE_PRIV_SW_EXCEPTION:
2932 if (CC(vmcs12->vm_entry_instruction_len > 15) ||
2933 CC(vmcs12->vm_entry_instruction_len == 0 &&
2934 CC(!nested_cpu_has_zero_length_injection(vcpu))))
2935 return -EINVAL;
2936 }
2937 }
2938
2939 if (nested_vmx_check_entry_msr_switch_controls(vcpu, vmcs12))
2940 return -EINVAL;
2941
2942 return 0;
2943 }
2944
2945 static int nested_vmx_check_controls(struct kvm_vcpu *vcpu,
2946 struct vmcs12 *vmcs12)
2947 {
2948 if (nested_check_vm_execution_controls(vcpu, vmcs12) ||
2949 nested_check_vm_exit_controls(vcpu, vmcs12) ||
2950 nested_check_vm_entry_controls(vcpu, vmcs12))
2951 return -EINVAL;
2952
2953 #ifdef CONFIG_KVM_HYPERV
2954 if (guest_cpuid_has_evmcs(vcpu))
2955 return nested_evmcs_check_controls(vmcs12);
2956 #endif
2957
2958 return 0;
2959 }
2960
2961 static int nested_vmx_check_address_space_size(struct kvm_vcpu *vcpu,
2962 struct vmcs12 *vmcs12)
2963 {
2964 #ifdef CONFIG_X86_64
2965 if (CC(!!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE) !=
2966 !!(vcpu->arch.efer & EFER_LMA)))
2967 return -EINVAL;
2968 #endif
2969 return 0;
2970 }
2971
2972 static int nested_vmx_check_host_state(struct kvm_vcpu *vcpu,
2973 struct vmcs12 *vmcs12)
2974 {
2975 bool ia32e = !!(vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE);
2976
2977 if (CC(!nested_host_cr0_valid(vcpu, vmcs12->host_cr0)) ||
2978 CC(!nested_host_cr4_valid(vcpu, vmcs12->host_cr4)) ||
2979 CC(!kvm_vcpu_is_legal_cr3(vcpu, vmcs12->host_cr3)))
2980 return -EINVAL;
2981
2982 if (CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_esp, vcpu)) ||
2983 CC(is_noncanonical_address(vmcs12->host_ia32_sysenter_eip, vcpu)))
2984 return -EINVAL;
2985
2986 if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) &&
2987 CC(!kvm_pat_valid(vmcs12->host_ia32_pat)))
2988 return -EINVAL;
2989
2990 if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
2991 CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
2992 vmcs12->host_ia32_perf_global_ctrl)))
2993 return -EINVAL;
2994
2995 if (ia32e) {
2996 if (CC(!(vmcs12->host_cr4 & X86_CR4_PAE)))
2997 return -EINVAL;
2998 } else {
2999 if (CC(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) ||
3000 CC(vmcs12->host_cr4 & X86_CR4_PCIDE) ||
3001 CC((vmcs12->host_rip) >> 32))
3002 return -EINVAL;
3003 }
3004
3005 if (CC(vmcs12->host_cs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3006 CC(vmcs12->host_ss_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3007 CC(vmcs12->host_ds_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3008 CC(vmcs12->host_es_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3009 CC(vmcs12->host_fs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3010 CC(vmcs12->host_gs_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3011 CC(vmcs12->host_tr_selector & (SEGMENT_RPL_MASK | SEGMENT_TI_MASK)) ||
3012 CC(vmcs12->host_cs_selector == 0) ||
3013 CC(vmcs12->host_tr_selector == 0) ||
3014 CC(vmcs12->host_ss_selector == 0 && !ia32e))
3015 return -EINVAL;
3016
3017 if (CC(is_noncanonical_address(vmcs12->host_fs_base, vcpu)) ||
3018 CC(is_noncanonical_address(vmcs12->host_gs_base, vcpu)) ||
3019 CC(is_noncanonical_address(vmcs12->host_gdtr_base, vcpu)) ||
3020 CC(is_noncanonical_address(vmcs12->host_idtr_base, vcpu)) ||
3021 CC(is_noncanonical_address(vmcs12->host_tr_base, vcpu)) ||
3022 CC(is_noncanonical_address(vmcs12->host_rip, vcpu)))
3023 return -EINVAL;
3024
3025 /*
3026 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
3027 * IA32_EFER MSR must be 0 in the field for that register. In addition,
3028 * the values of the LMA and LME bits in the field must each be that of
3029 * the host address-space size VM-exit control.
3030 */
3031 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
3032 if (CC(!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer)) ||
3033 CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA)) ||
3034 CC(ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)))
3035 return -EINVAL;
3036 }
3037
3038 return 0;
3039 }
3040
3041 static int nested_vmx_check_vmcs_link_ptr(struct kvm_vcpu *vcpu,
3042 struct vmcs12 *vmcs12)
3043 {
3044 struct vcpu_vmx *vmx = to_vmx(vcpu);
3045 struct gfn_to_hva_cache *ghc = &vmx->nested.shadow_vmcs12_cache;
3046 struct vmcs_hdr hdr;
3047
3048 if (vmcs12->vmcs_link_pointer == INVALID_GPA)
3049 return 0;
3050
3051 if (CC(!page_address_valid(vcpu, vmcs12->vmcs_link_pointer)))
3052 return -EINVAL;
3053
3054 if (ghc->gpa != vmcs12->vmcs_link_pointer &&
3055 CC(kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc,
3056 vmcs12->vmcs_link_pointer, VMCS12_SIZE)))
3057 return -EINVAL;
3058
3059 if (CC(kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
3060 offsetof(struct vmcs12, hdr),
3061 sizeof(hdr))))
3062 return -EINVAL;
3063
3064 if (CC(hdr.revision_id != VMCS12_REVISION) ||
3065 CC(hdr.shadow_vmcs != nested_cpu_has_shadow_vmcs(vmcs12)))
3066 return -EINVAL;
3067
3068 return 0;
3069 }
3070
3071 /*
3072 * Checks related to Guest Non-register State
3073 */
3074 static int nested_check_guest_non_reg_state(struct vmcs12 *vmcs12)
3075 {
3076 if (CC(vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
3077 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT &&
3078 vmcs12->guest_activity_state != GUEST_ACTIVITY_WAIT_SIPI))
3079 return -EINVAL;
3080
3081 return 0;
3082 }
3083
3084 static int nested_vmx_check_guest_state(struct kvm_vcpu *vcpu,
3085 struct vmcs12 *vmcs12,
3086 enum vm_entry_failure_code *entry_failure_code)
3087 {
3088 bool ia32e = !!(vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE);
3089
3090 *entry_failure_code = ENTRY_FAIL_DEFAULT;
3091
3092 if (CC(!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0)) ||
3093 CC(!nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4)))
3094 return -EINVAL;
3095
3096 if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) &&
3097 CC(!kvm_dr7_valid(vmcs12->guest_dr7)))
3098 return -EINVAL;
3099
3100 if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) &&
3101 CC(!kvm_pat_valid(vmcs12->guest_ia32_pat)))
3102 return -EINVAL;
3103
3104 if (nested_vmx_check_vmcs_link_ptr(vcpu, vmcs12)) {
3105 *entry_failure_code = ENTRY_FAIL_VMCS_LINK_PTR;
3106 return -EINVAL;
3107 }
3108
3109 if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL) &&
3110 CC(!kvm_valid_perf_global_ctrl(vcpu_to_pmu(vcpu),
3111 vmcs12->guest_ia32_perf_global_ctrl)))
3112 return -EINVAL;
3113
3114 if (CC((vmcs12->guest_cr0 & (X86_CR0_PG | X86_CR0_PE)) == X86_CR0_PG))
3115 return -EINVAL;
3116
3117 if (CC(ia32e && !(vmcs12->guest_cr4 & X86_CR4_PAE)) ||
3118 CC(ia32e && !(vmcs12->guest_cr0 & X86_CR0_PG)))
3119 return -EINVAL;
3120
3121 /*
3122 * If the load IA32_EFER VM-entry control is 1, the following checks
3123 * are performed on the field for the IA32_EFER MSR:
3124 * - Bits reserved in the IA32_EFER MSR must be 0.
3125 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
3126 * the IA-32e mode guest VM-exit control. It must also be identical
3127 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
3128 * CR0.PG) is 1.
3129 */
3130 if (to_vmx(vcpu)->nested.nested_run_pending &&
3131 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
3132 if (CC(!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer)) ||
3133 CC(ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA)) ||
3134 CC(((vmcs12->guest_cr0 & X86_CR0_PG) &&
3135 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))))
3136 return -EINVAL;
3137 }
3138
3139 if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
3140 (CC(is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu)) ||
3141 CC((vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD))))
3142 return -EINVAL;
3143
3144 if (nested_check_guest_non_reg_state(vmcs12))
3145 return -EINVAL;
3146
3147 return 0;
3148 }
3149
3150 static int nested_vmx_check_vmentry_hw(struct kvm_vcpu *vcpu)
3151 {
3152 struct vcpu_vmx *vmx = to_vmx(vcpu);
3153 unsigned long cr3, cr4;
3154 bool vm_fail;
3155
3156 if (!nested_early_check)
3157 return 0;
3158
3159 if (vmx->msr_autoload.host.nr)
3160 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3161 if (vmx->msr_autoload.guest.nr)
3162 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3163
3164 preempt_disable();
3165
3166 vmx_prepare_switch_to_guest(vcpu);
3167
3168 /*
3169 * Induce a consistency check VMExit by clearing bit 1 in GUEST_RFLAGS,
3170 * which is reserved to '1' by hardware. GUEST_RFLAGS is guaranteed to
3171 * be written (by prepare_vmcs02()) before the "real" VMEnter, i.e.
3172 * there is no need to preserve other bits or save/restore the field.
3173 */
3174 vmcs_writel(GUEST_RFLAGS, 0);
3175
3176 cr3 = __get_current_cr3_fast();
3177 if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
3178 vmcs_writel(HOST_CR3, cr3);
3179 vmx->loaded_vmcs->host_state.cr3 = cr3;
3180 }
3181
3182 cr4 = cr4_read_shadow();
3183 if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
3184 vmcs_writel(HOST_CR4, cr4);
3185 vmx->loaded_vmcs->host_state.cr4 = cr4;
3186 }
3187
3188 vm_fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
3189 __vmx_vcpu_run_flags(vmx));
3190
3191 if (vmx->msr_autoload.host.nr)
3192 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
3193 if (vmx->msr_autoload.guest.nr)
3194 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
3195
3196 if (vm_fail) {
3197 u32 error = vmcs_read32(VM_INSTRUCTION_ERROR);
3198
3199 preempt_enable();
3200
3201 trace_kvm_nested_vmenter_failed(
3202 "early hardware check VM-instruction error: ", error);
3203 WARN_ON_ONCE(error != VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3204 return 1;
3205 }
3206
3207 /*
3208 * VMExit clears RFLAGS.IF and DR7, even on a consistency check.
3209 */
3210 if (hw_breakpoint_active())
3211 set_debugreg(__this_cpu_read(cpu_dr7), 7);
3212 local_irq_enable();
3213 preempt_enable();
3214
3215 /*
3216 * A non-failing VMEntry means we somehow entered guest mode with
3217 * an illegal RIP, and that's just the tip of the iceberg. There
3218 * is no telling what memory has been modified or what state has
3219 * been exposed to unknown code. Hitting this all but guarantees
3220 * a (very critical) hardware issue.
3221 */
3222 WARN_ON(!(vmcs_read32(VM_EXIT_REASON) &
3223 VMX_EXIT_REASONS_FAILED_VMENTRY));
3224
3225 return 0;
3226 }
3227
3228 #ifdef CONFIG_KVM_HYPERV
3229 static bool nested_get_evmcs_page(struct kvm_vcpu *vcpu)
3230 {
3231 struct vcpu_vmx *vmx = to_vmx(vcpu);
3232
3233 /*
3234 * hv_evmcs may end up being not mapped after migration (when
3235 * L2 was running), map it here to make sure vmcs12 changes are
3236 * properly reflected.
3237 */
3238 if (guest_cpuid_has_evmcs(vcpu) &&
3239 vmx->nested.hv_evmcs_vmptr == EVMPTR_MAP_PENDING) {
3240 enum nested_evmptrld_status evmptrld_status =
3241 nested_vmx_handle_enlightened_vmptrld(vcpu, false);
3242
3243 if (evmptrld_status == EVMPTRLD_VMFAIL ||
3244 evmptrld_status == EVMPTRLD_ERROR)
3245 return false;
3246
3247 /*
3248 * Post migration VMCS12 always provides the most actual
3249 * information, copy it to eVMCS upon entry.
3250 */
3251 vmx->nested.need_vmcs12_to_shadow_sync = true;
3252 }
3253
3254 return true;
3255 }
3256 #endif
3257
3258 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu)
3259 {
3260 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3261 struct vcpu_vmx *vmx = to_vmx(vcpu);
3262 struct kvm_host_map *map;
3263
3264 if (!vcpu->arch.pdptrs_from_userspace &&
3265 !nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
3266 /*
3267 * Reload the guest's PDPTRs since after a migration
3268 * the guest CR3 might be restored prior to setting the nested
3269 * state which can lead to a load of wrong PDPTRs.
3270 */
3271 if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
3272 return false;
3273 }
3274
3275
3276 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3277 map = &vmx->nested.apic_access_page_map;
3278
3279 if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->apic_access_addr), map)) {
3280 vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(map->pfn));
3281 } else {
3282 pr_debug_ratelimited("%s: no backing for APIC-access address in vmcs12\n",
3283 __func__);
3284 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3285 vcpu->run->internal.suberror =
3286 KVM_INTERNAL_ERROR_EMULATION;
3287 vcpu->run->internal.ndata = 0;
3288 return false;
3289 }
3290 }
3291
3292 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3293 map = &vmx->nested.virtual_apic_map;
3294
3295 if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->virtual_apic_page_addr), map)) {
3296 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, pfn_to_hpa(map->pfn));
3297 } else if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING) &&
3298 nested_cpu_has(vmcs12, CPU_BASED_CR8_STORE_EXITING) &&
3299 !nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
3300 /*
3301 * The processor will never use the TPR shadow, simply
3302 * clear the bit from the execution control. Such a
3303 * configuration is useless, but it happens in tests.
3304 * For any other configuration, failing the vm entry is
3305 * _not_ what the processor does but it's basically the
3306 * only possibility we have.
3307 */
3308 exec_controls_clearbit(vmx, CPU_BASED_TPR_SHADOW);
3309 } else {
3310 /*
3311 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR to
3312 * force VM-Entry to fail.
3313 */
3314 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, INVALID_GPA);
3315 }
3316 }
3317
3318 if (nested_cpu_has_posted_intr(vmcs12)) {
3319 map = &vmx->nested.pi_desc_map;
3320
3321 if (!kvm_vcpu_map(vcpu, gpa_to_gfn(vmcs12->posted_intr_desc_addr), map)) {
3322 vmx->nested.pi_desc =
3323 (struct pi_desc *)(((void *)map->hva) +
3324 offset_in_page(vmcs12->posted_intr_desc_addr));
3325 vmcs_write64(POSTED_INTR_DESC_ADDR,
3326 pfn_to_hpa(map->pfn) + offset_in_page(vmcs12->posted_intr_desc_addr));
3327 } else {
3328 /*
3329 * Defer the KVM_INTERNAL_EXIT until KVM tries to
3330 * access the contents of the VMCS12 posted interrupt
3331 * descriptor. (Note that KVM may do this when it
3332 * should not, per the architectural specification.)
3333 */
3334 vmx->nested.pi_desc = NULL;
3335 pin_controls_clearbit(vmx, PIN_BASED_POSTED_INTR);
3336 }
3337 }
3338 if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
3339 exec_controls_setbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3340 else
3341 exec_controls_clearbit(vmx, CPU_BASED_USE_MSR_BITMAPS);
3342
3343 return true;
3344 }
3345
3346 static bool vmx_get_nested_state_pages(struct kvm_vcpu *vcpu)
3347 {
3348 #ifdef CONFIG_KVM_HYPERV
3349 /*
3350 * Note: nested_get_evmcs_page() also updates 'vp_assist_page' copy
3351 * in 'struct kvm_vcpu_hv' in case eVMCS is in use, this is mandatory
3352 * to make nested_evmcs_l2_tlb_flush_enabled() work correctly post
3353 * migration.
3354 */
3355 if (!nested_get_evmcs_page(vcpu)) {
3356 pr_debug_ratelimited("%s: enlightened vmptrld failed\n",
3357 __func__);
3358 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3359 vcpu->run->internal.suberror =
3360 KVM_INTERNAL_ERROR_EMULATION;
3361 vcpu->run->internal.ndata = 0;
3362
3363 return false;
3364 }
3365 #endif
3366
3367 if (is_guest_mode(vcpu) && !nested_get_vmcs12_pages(vcpu))
3368 return false;
3369
3370 return true;
3371 }
3372
3373 static int nested_vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
3374 {
3375 struct vmcs12 *vmcs12;
3376 struct vcpu_vmx *vmx = to_vmx(vcpu);
3377 gpa_t dst;
3378
3379 if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
3380 return 0;
3381
3382 if (WARN_ON_ONCE(vmx->nested.pml_full))
3383 return 1;
3384
3385 /*
3386 * Check if PML is enabled for the nested guest. Whether eptp bit 6 is
3387 * set is already checked as part of A/D emulation.
3388 */
3389 vmcs12 = get_vmcs12(vcpu);
3390 if (!nested_cpu_has_pml(vmcs12))
3391 return 0;
3392
3393 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
3394 vmx->nested.pml_full = true;
3395 return 1;
3396 }
3397
3398 gpa &= ~0xFFFull;
3399 dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
3400
3401 if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
3402 offset_in_page(dst), sizeof(gpa)))
3403 return 0;
3404
3405 vmcs12->guest_pml_index--;
3406
3407 return 0;
3408 }
3409
3410 /*
3411 * Intel's VMX Instruction Reference specifies a common set of prerequisites
3412 * for running VMX instructions (except VMXON, whose prerequisites are
3413 * slightly different). It also specifies what exception to inject otherwise.
3414 * Note that many of these exceptions have priority over VM exits, so they
3415 * don't have to be checked again here.
3416 */
3417 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
3418 {
3419 if (!to_vmx(vcpu)->nested.vmxon) {
3420 kvm_queue_exception(vcpu, UD_VECTOR);
3421 return 0;
3422 }
3423
3424 if (vmx_get_cpl(vcpu)) {
3425 kvm_inject_gp(vcpu, 0);
3426 return 0;
3427 }
3428
3429 return 1;
3430 }
3431
3432 static u8 vmx_has_apicv_interrupt(struct kvm_vcpu *vcpu)
3433 {
3434 u8 rvi = vmx_get_rvi();
3435 u8 vppr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_PROCPRI);
3436
3437 return ((rvi & 0xf0) > (vppr & 0xf0));
3438 }
3439
3440 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
3441 struct vmcs12 *vmcs12);
3442
3443 /*
3444 * If from_vmentry is false, this is being called from state restore (either RSM
3445 * or KVM_SET_NESTED_STATE). Otherwise it's called from vmlaunch/vmresume.
3446 *
3447 * Returns:
3448 * NVMX_VMENTRY_SUCCESS: Entered VMX non-root mode
3449 * NVMX_VMENTRY_VMFAIL: Consistency check VMFail
3450 * NVMX_VMENTRY_VMEXIT: Consistency check VMExit
3451 * NVMX_VMENTRY_KVM_INTERNAL_ERROR: KVM internal error
3452 */
3453 enum nvmx_vmentry_status nested_vmx_enter_non_root_mode(struct kvm_vcpu *vcpu,
3454 bool from_vmentry)
3455 {
3456 struct vcpu_vmx *vmx = to_vmx(vcpu);
3457 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3458 enum vm_entry_failure_code entry_failure_code;
3459 bool evaluate_pending_interrupts;
3460 union vmx_exit_reason exit_reason = {
3461 .basic = EXIT_REASON_INVALID_STATE,
3462 .failed_vmentry = 1,
3463 };
3464 u32 failed_index;
3465
3466 trace_kvm_nested_vmenter(kvm_rip_read(vcpu),
3467 vmx->nested.current_vmptr,
3468 vmcs12->guest_rip,
3469 vmcs12->guest_intr_status,
3470 vmcs12->vm_entry_intr_info_field,
3471 vmcs12->secondary_vm_exec_control & SECONDARY_EXEC_ENABLE_EPT,
3472 vmcs12->ept_pointer,
3473 vmcs12->guest_cr3,
3474 KVM_ISA_VMX);
3475
3476 kvm_service_local_tlb_flush_requests(vcpu);
3477
3478 evaluate_pending_interrupts = exec_controls_get(vmx) &
3479 (CPU_BASED_INTR_WINDOW_EXITING | CPU_BASED_NMI_WINDOW_EXITING);
3480 if (likely(!evaluate_pending_interrupts) && kvm_vcpu_apicv_active(vcpu))
3481 evaluate_pending_interrupts |= vmx_has_apicv_interrupt(vcpu);
3482 if (!evaluate_pending_interrupts)
3483 evaluate_pending_interrupts |= kvm_apic_has_pending_init_or_sipi(vcpu);
3484
3485 if (!vmx->nested.nested_run_pending ||
3486 !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
3487 vmx->nested.pre_vmenter_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
3488 if (kvm_mpx_supported() &&
3489 (!vmx->nested.nested_run_pending ||
3490 !(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)))
3491 vmx->nested.pre_vmenter_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
3492
3493 /*
3494 * Overwrite vmcs01.GUEST_CR3 with L1's CR3 if EPT is disabled *and*
3495 * nested early checks are disabled. In the event of a "late" VM-Fail,
3496 * i.e. a VM-Fail detected by hardware but not KVM, KVM must unwind its
3497 * software model to the pre-VMEntry host state. When EPT is disabled,
3498 * GUEST_CR3 holds KVM's shadow CR3, not L1's "real" CR3, which causes
3499 * nested_vmx_restore_host_state() to corrupt vcpu->arch.cr3. Stuffing
3500 * vmcs01.GUEST_CR3 results in the unwind naturally setting arch.cr3 to
3501 * the correct value. Smashing vmcs01.GUEST_CR3 is safe because nested
3502 * VM-Exits, and the unwind, reset KVM's MMU, i.e. vmcs01.GUEST_CR3 is
3503 * guaranteed to be overwritten with a shadow CR3 prior to re-entering
3504 * L1. Don't stuff vmcs01.GUEST_CR3 when using nested early checks as
3505 * KVM modifies vcpu->arch.cr3 if and only if the early hardware checks
3506 * pass, and early VM-Fails do not reset KVM's MMU, i.e. the VM-Fail
3507 * path would need to manually save/restore vmcs01.GUEST_CR3.
3508 */
3509 if (!enable_ept && !nested_early_check)
3510 vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3511
3512 vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
3513
3514 prepare_vmcs02_early(vmx, &vmx->vmcs01, vmcs12);
3515
3516 if (from_vmentry) {
3517 if (unlikely(!nested_get_vmcs12_pages(vcpu))) {
3518 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3519 return NVMX_VMENTRY_KVM_INTERNAL_ERROR;
3520 }
3521
3522 if (nested_vmx_check_vmentry_hw(vcpu)) {
3523 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3524 return NVMX_VMENTRY_VMFAIL;
3525 }
3526
3527 if (nested_vmx_check_guest_state(vcpu, vmcs12,
3528 &entry_failure_code)) {
3529 exit_reason.basic = EXIT_REASON_INVALID_STATE;
3530 vmcs12->exit_qualification = entry_failure_code;
3531 goto vmentry_fail_vmexit;
3532 }
3533 }
3534
3535 enter_guest_mode(vcpu);
3536
3537 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &entry_failure_code)) {
3538 exit_reason.basic = EXIT_REASON_INVALID_STATE;
3539 vmcs12->exit_qualification = entry_failure_code;
3540 goto vmentry_fail_vmexit_guest_mode;
3541 }
3542
3543 if (from_vmentry) {
3544 failed_index = nested_vmx_load_msr(vcpu,
3545 vmcs12->vm_entry_msr_load_addr,
3546 vmcs12->vm_entry_msr_load_count);
3547 if (failed_index) {
3548 exit_reason.basic = EXIT_REASON_MSR_LOAD_FAIL;
3549 vmcs12->exit_qualification = failed_index;
3550 goto vmentry_fail_vmexit_guest_mode;
3551 }
3552 } else {
3553 /*
3554 * The MMU is not initialized to point at the right entities yet and
3555 * "get pages" would need to read data from the guest (i.e. we will
3556 * need to perform gpa to hpa translation). Request a call
3557 * to nested_get_vmcs12_pages before the next VM-entry. The MSRs
3558 * have already been set at vmentry time and should not be reset.
3559 */
3560 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
3561 }
3562
3563 /*
3564 * Re-evaluate pending events if L1 had a pending IRQ/NMI/INIT/SIPI
3565 * when it executed VMLAUNCH/VMRESUME, as entering non-root mode can
3566 * effectively unblock various events, e.g. INIT/SIPI cause VM-Exit
3567 * unconditionally.
3568 */
3569 if (unlikely(evaluate_pending_interrupts))
3570 kvm_make_request(KVM_REQ_EVENT, vcpu);
3571
3572 /*
3573 * Do not start the preemption timer hrtimer until after we know
3574 * we are successful, so that only nested_vmx_vmexit needs to cancel
3575 * the timer.
3576 */
3577 vmx->nested.preemption_timer_expired = false;
3578 if (nested_cpu_has_preemption_timer(vmcs12)) {
3579 u64 timer_value = vmx_calc_preemption_timer_value(vcpu);
3580 vmx_start_preemption_timer(vcpu, timer_value);
3581 }
3582
3583 /*
3584 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
3585 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
3586 * returned as far as L1 is concerned. It will only return (and set
3587 * the success flag) when L2 exits (see nested_vmx_vmexit()).
3588 */
3589 return NVMX_VMENTRY_SUCCESS;
3590
3591 /*
3592 * A failed consistency check that leads to a VMExit during L1's
3593 * VMEnter to L2 is a variation of a normal VMexit, as explained in
3594 * 26.7 "VM-entry failures during or after loading guest state".
3595 */
3596 vmentry_fail_vmexit_guest_mode:
3597 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING)
3598 vcpu->arch.tsc_offset -= vmcs12->tsc_offset;
3599 leave_guest_mode(vcpu);
3600
3601 vmentry_fail_vmexit:
3602 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
3603
3604 if (!from_vmentry)
3605 return NVMX_VMENTRY_VMEXIT;
3606
3607 load_vmcs12_host_state(vcpu, vmcs12);
3608 vmcs12->vm_exit_reason = exit_reason.full;
3609 if (enable_shadow_vmcs || nested_vmx_is_evmptr12_valid(vmx))
3610 vmx->nested.need_vmcs12_to_shadow_sync = true;
3611 return NVMX_VMENTRY_VMEXIT;
3612 }
3613
3614 /*
3615 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
3616 * for running an L2 nested guest.
3617 */
3618 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
3619 {
3620 struct vmcs12 *vmcs12;
3621 enum nvmx_vmentry_status status;
3622 struct vcpu_vmx *vmx = to_vmx(vcpu);
3623 u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
3624 enum nested_evmptrld_status evmptrld_status;
3625
3626 if (!nested_vmx_check_permission(vcpu))
3627 return 1;
3628
3629 evmptrld_status = nested_vmx_handle_enlightened_vmptrld(vcpu, launch);
3630 if (evmptrld_status == EVMPTRLD_ERROR) {
3631 kvm_queue_exception(vcpu, UD_VECTOR);
3632 return 1;
3633 }
3634
3635 kvm_pmu_trigger_event(vcpu, kvm_pmu_eventsel.BRANCH_INSTRUCTIONS_RETIRED);
3636
3637 if (CC(evmptrld_status == EVMPTRLD_VMFAIL))
3638 return nested_vmx_failInvalid(vcpu);
3639
3640 if (CC(!nested_vmx_is_evmptr12_valid(vmx) &&
3641 vmx->nested.current_vmptr == INVALID_GPA))
3642 return nested_vmx_failInvalid(vcpu);
3643
3644 vmcs12 = get_vmcs12(vcpu);
3645
3646 /*
3647 * Can't VMLAUNCH or VMRESUME a shadow VMCS. Despite the fact
3648 * that there *is* a valid VMCS pointer, RFLAGS.CF is set
3649 * rather than RFLAGS.ZF, and no error number is stored to the
3650 * VM-instruction error field.
3651 */
3652 if (CC(vmcs12->hdr.shadow_vmcs))
3653 return nested_vmx_failInvalid(vcpu);
3654
3655 if (nested_vmx_is_evmptr12_valid(vmx)) {
3656 struct hv_enlightened_vmcs *evmcs = nested_vmx_evmcs(vmx);
3657
3658 copy_enlightened_to_vmcs12(vmx, evmcs->hv_clean_fields);
3659 /* Enlightened VMCS doesn't have launch state */
3660 vmcs12->launch_state = !launch;
3661 } else if (enable_shadow_vmcs) {
3662 copy_shadow_to_vmcs12(vmx);
3663 }
3664
3665 /*
3666 * The nested entry process starts with enforcing various prerequisites
3667 * on vmcs12 as required by the Intel SDM, and act appropriately when
3668 * they fail: As the SDM explains, some conditions should cause the
3669 * instruction to fail, while others will cause the instruction to seem
3670 * to succeed, but return an EXIT_REASON_INVALID_STATE.
3671 * To speed up the normal (success) code path, we should avoid checking
3672 * for misconfigurations which will anyway be caught by the processor
3673 * when using the merged vmcs02.
3674 */
3675 if (CC(interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS))
3676 return nested_vmx_fail(vcpu, VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
3677
3678 if (CC(vmcs12->launch_state == launch))
3679 return nested_vmx_fail(vcpu,
3680 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
3681 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
3682
3683 if (nested_vmx_check_controls(vcpu, vmcs12))
3684 return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3685
3686 if (nested_vmx_check_address_space_size(vcpu, vmcs12))
3687 return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3688
3689 if (nested_vmx_check_host_state(vcpu, vmcs12))
3690 return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
3691
3692 /*
3693 * We're finally done with prerequisite checking, and can start with
3694 * the nested entry.
3695 */
3696 vmx->nested.nested_run_pending = 1;
3697 vmx->nested.has_preemption_timer_deadline = false;
3698 status = nested_vmx_enter_non_root_mode(vcpu, true);
3699 if (unlikely(status != NVMX_VMENTRY_SUCCESS))
3700 goto vmentry_failed;
3701
3702 /* Emulate processing of posted interrupts on VM-Enter. */
3703 if (nested_cpu_has_posted_intr(vmcs12) &&
3704 kvm_apic_has_interrupt(vcpu) == vmx->nested.posted_intr_nv) {
3705 vmx->nested.pi_pending = true;
3706 kvm_make_request(KVM_REQ_EVENT, vcpu);
3707 kvm_apic_clear_irr(vcpu, vmx->nested.posted_intr_nv);
3708 }
3709
3710 /* Hide L1D cache contents from the nested guest. */
3711 vmx->vcpu.arch.l1tf_flush_l1d = true;
3712
3713 /*
3714 * Must happen outside of nested_vmx_enter_non_root_mode() as it will
3715 * also be used as part of restoring nVMX state for
3716 * snapshot restore (migration).
3717 *
3718 * In this flow, it is assumed that vmcs12 cache was
3719 * transferred as part of captured nVMX state and should
3720 * therefore not be read from guest memory (which may not
3721 * exist on destination host yet).
3722 */
3723 nested_cache_shadow_vmcs12(vcpu, vmcs12);
3724
3725 switch (vmcs12->guest_activity_state) {
3726 case GUEST_ACTIVITY_HLT:
3727 /*
3728 * If we're entering a halted L2 vcpu and the L2 vcpu won't be
3729 * awakened by event injection or by an NMI-window VM-exit or
3730 * by an interrupt-window VM-exit, halt the vcpu.
3731 */
3732 if (!(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK) &&
3733 !nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING) &&
3734 !(nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING) &&
3735 (vmcs12->guest_rflags & X86_EFLAGS_IF))) {
3736 vmx->nested.nested_run_pending = 0;
3737 return kvm_emulate_halt_noskip(vcpu);
3738 }
3739 break;
3740 case GUEST_ACTIVITY_WAIT_SIPI:
3741 vmx->nested.nested_run_pending = 0;
3742 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
3743 break;
3744 default:
3745 break;
3746 }
3747
3748 return 1;
3749
3750 vmentry_failed:
3751 vmx->nested.nested_run_pending = 0;
3752 if (status == NVMX_VMENTRY_KVM_INTERNAL_ERROR)
3753 return 0;
3754 if (status == NVMX_VMENTRY_VMEXIT)
3755 return 1;
3756 WARN_ON_ONCE(status != NVMX_VMENTRY_VMFAIL);
3757 return nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3758 }
3759
3760 /*
3761 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
3762 * because L2 may have changed some cr0 bits directly (CR0_GUEST_HOST_MASK).
3763 * This function returns the new value we should put in vmcs12.guest_cr0.
3764 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
3765 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
3766 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
3767 * didn't trap the bit, because if L1 did, so would L0).
3768 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
3769 * been modified by L2, and L1 knows it. So just leave the old value of
3770 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
3771 * isn't relevant, because if L0 traps this bit it can set it to anything.
3772 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
3773 * changed these bits, and therefore they need to be updated, but L0
3774 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
3775 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
3776 */
3777 static inline unsigned long
3778 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3779 {
3780 return
3781 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
3782 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
3783 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
3784 vcpu->arch.cr0_guest_owned_bits));
3785 }
3786
3787 static inline unsigned long
3788 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
3789 {
3790 return
3791 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
3792 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
3793 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
3794 vcpu->arch.cr4_guest_owned_bits));
3795 }
3796
3797 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
3798 struct vmcs12 *vmcs12,
3799 u32 vm_exit_reason, u32 exit_intr_info)
3800 {
3801 u32 idt_vectoring;
3802 unsigned int nr;
3803
3804 /*
3805 * Per the SDM, VM-Exits due to double and triple faults are never
3806 * considered to occur during event delivery, even if the double/triple
3807 * fault is the result of an escalating vectoring issue.
3808 *
3809 * Note, the SDM qualifies the double fault behavior with "The original
3810 * event results in a double-fault exception". It's unclear why the
3811 * qualification exists since exits due to double fault can occur only
3812 * while vectoring a different exception (injected events are never
3813 * subject to interception), i.e. there's _always_ an original event.
3814 *
3815 * The SDM also uses NMI as a confusing example for the "original event
3816 * causes the VM exit directly" clause. NMI isn't special in any way,
3817 * the same rule applies to all events that cause an exit directly.
3818 * NMI is an odd choice for the example because NMIs can only occur on
3819 * instruction boundaries, i.e. they _can't_ occur during vectoring.
3820 */
3821 if ((u16)vm_exit_reason == EXIT_REASON_TRIPLE_FAULT ||
3822 ((u16)vm_exit_reason == EXIT_REASON_EXCEPTION_NMI &&
3823 is_double_fault(exit_intr_info))) {
3824 vmcs12->idt_vectoring_info_field = 0;
3825 } else if (vcpu->arch.exception.injected) {
3826 nr = vcpu->arch.exception.vector;
3827 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3828
3829 if (kvm_exception_is_soft(nr)) {
3830 vmcs12->vm_exit_instruction_len =
3831 vcpu->arch.event_exit_inst_len;
3832 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
3833 } else
3834 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
3835
3836 if (vcpu->arch.exception.has_error_code) {
3837 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
3838 vmcs12->idt_vectoring_error_code =
3839 vcpu->arch.exception.error_code;
3840 }
3841
3842 vmcs12->idt_vectoring_info_field = idt_vectoring;
3843 } else if (vcpu->arch.nmi_injected) {
3844 vmcs12->idt_vectoring_info_field =
3845 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
3846 } else if (vcpu->arch.interrupt.injected) {
3847 nr = vcpu->arch.interrupt.nr;
3848 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
3849
3850 if (vcpu->arch.interrupt.soft) {
3851 idt_vectoring |= INTR_TYPE_SOFT_INTR;
3852 vmcs12->vm_entry_instruction_len =
3853 vcpu->arch.event_exit_inst_len;
3854 } else
3855 idt_vectoring |= INTR_TYPE_EXT_INTR;
3856
3857 vmcs12->idt_vectoring_info_field = idt_vectoring;
3858 } else {
3859 vmcs12->idt_vectoring_info_field = 0;
3860 }
3861 }
3862
3863
3864 void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
3865 {
3866 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3867 gfn_t gfn;
3868
3869 /*
3870 * Don't need to mark the APIC access page dirty; it is never
3871 * written to by the CPU during APIC virtualization.
3872 */
3873
3874 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
3875 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
3876 kvm_vcpu_mark_page_dirty(vcpu, gfn);
3877 }
3878
3879 if (nested_cpu_has_posted_intr(vmcs12)) {
3880 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
3881 kvm_vcpu_mark_page_dirty(vcpu, gfn);
3882 }
3883 }
3884
3885 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
3886 {
3887 struct vcpu_vmx *vmx = to_vmx(vcpu);
3888 int max_irr;
3889 void *vapic_page;
3890 u16 status;
3891
3892 if (!vmx->nested.pi_pending)
3893 return 0;
3894
3895 if (!vmx->nested.pi_desc)
3896 goto mmio_needed;
3897
3898 vmx->nested.pi_pending = false;
3899
3900 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
3901 return 0;
3902
3903 max_irr = pi_find_highest_vector(vmx->nested.pi_desc);
3904 if (max_irr > 0) {
3905 vapic_page = vmx->nested.virtual_apic_map.hva;
3906 if (!vapic_page)
3907 goto mmio_needed;
3908
3909 __kvm_apic_update_irr(vmx->nested.pi_desc->pir,
3910 vapic_page, &max_irr);
3911 status = vmcs_read16(GUEST_INTR_STATUS);
3912 if ((u8)max_irr > ((u8)status & 0xff)) {
3913 status &= ~0xff;
3914 status |= (u8)max_irr;
3915 vmcs_write16(GUEST_INTR_STATUS, status);
3916 }
3917 }
3918
3919 nested_mark_vmcs12_pages_dirty(vcpu);
3920 return 0;
3921
3922 mmio_needed:
3923 kvm_handle_memory_failure(vcpu, X86EMUL_IO_NEEDED, NULL);
3924 return -ENXIO;
3925 }
3926
3927 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu)
3928 {
3929 struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
3930 u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
3931 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
3932 unsigned long exit_qual;
3933
3934 if (ex->has_payload) {
3935 exit_qual = ex->payload;
3936 } else if (ex->vector == PF_VECTOR) {
3937 exit_qual = vcpu->arch.cr2;
3938 } else if (ex->vector == DB_VECTOR) {
3939 exit_qual = vcpu->arch.dr6;
3940 exit_qual &= ~DR6_BT;
3941 exit_qual ^= DR6_ACTIVE_LOW;
3942 } else {
3943 exit_qual = 0;
3944 }
3945
3946 /*
3947 * Unlike AMD's Paged Real Mode, which reports an error code on #PF
3948 * VM-Exits even if the CPU is in Real Mode, Intel VMX never sets the
3949 * "has error code" flags on VM-Exit if the CPU is in Real Mode.
3950 */
3951 if (ex->has_error_code && is_protmode(vcpu)) {
3952 /*
3953 * Intel CPUs do not generate error codes with bits 31:16 set,
3954 * and more importantly VMX disallows setting bits 31:16 in the
3955 * injected error code for VM-Entry. Drop the bits to mimic
3956 * hardware and avoid inducing failure on nested VM-Entry if L1
3957 * chooses to inject the exception back to L2. AMD CPUs _do_
3958 * generate "full" 32-bit error codes, so KVM allows userspace
3959 * to inject exception error codes with bits 31:16 set.
3960 */
3961 vmcs12->vm_exit_intr_error_code = (u16)ex->error_code;
3962 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
3963 }
3964
3965 if (kvm_exception_is_soft(ex->vector))
3966 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
3967 else
3968 intr_info |= INTR_TYPE_HARD_EXCEPTION;
3969
3970 if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
3971 vmx_get_nmi_mask(vcpu))
3972 intr_info |= INTR_INFO_UNBLOCK_NMI;
3973
3974 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
3975 }
3976
3977 /*
3978 * Returns true if a debug trap is (likely) pending delivery. Infer the class
3979 * of a #DB (trap-like vs. fault-like) from the exception payload (to-be-DR6).
3980 * Using the payload is flawed because code breakpoints (fault-like) and data
3981 * breakpoints (trap-like) set the same bits in DR6 (breakpoint detected), i.e.
3982 * this will return false positives if a to-be-injected code breakpoint #DB is
3983 * pending (from KVM's perspective, but not "pending" across an instruction
3984 * boundary). ICEBP, a.k.a. INT1, is also not reflected here even though it
3985 * too is trap-like.
3986 *
3987 * KVM "works" despite these flaws as ICEBP isn't currently supported by the
3988 * emulator, Monitor Trap Flag is not marked pending on intercepted #DBs (the
3989 * #DB has already happened), and MTF isn't marked pending on code breakpoints
3990 * from the emulator (because such #DBs are fault-like and thus don't trigger
3991 * actions that fire on instruction retire).
3992 */
3993 static unsigned long vmx_get_pending_dbg_trap(struct kvm_queued_exception *ex)
3994 {
3995 if (!ex->pending || ex->vector != DB_VECTOR)
3996 return 0;
3997
3998 /* General Detect #DBs are always fault-like. */
3999 return ex->payload & ~DR6_BD;
4000 }
4001
4002 /*
4003 * Returns true if there's a pending #DB exception that is lower priority than
4004 * a pending Monitor Trap Flag VM-Exit. TSS T-flag #DBs are not emulated by
4005 * KVM, but could theoretically be injected by userspace. Note, this code is
4006 * imperfect, see above.
4007 */
4008 static bool vmx_is_low_priority_db_trap(struct kvm_queued_exception *ex)
4009 {
4010 return vmx_get_pending_dbg_trap(ex) & ~DR6_BT;
4011 }
4012
4013 /*
4014 * Certain VM-exits set the 'pending debug exceptions' field to indicate a
4015 * recognized #DB (data or single-step) that has yet to be delivered. Since KVM
4016 * represents these debug traps with a payload that is said to be compatible
4017 * with the 'pending debug exceptions' field, write the payload to the VMCS
4018 * field if a VM-exit is delivered before the debug trap.
4019 */
4020 static void nested_vmx_update_pending_dbg(struct kvm_vcpu *vcpu)
4021 {
4022 unsigned long pending_dbg;
4023
4024 pending_dbg = vmx_get_pending_dbg_trap(&vcpu->arch.exception);
4025 if (pending_dbg)
4026 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, pending_dbg);
4027 }
4028
4029 static bool nested_vmx_preemption_timer_pending(struct kvm_vcpu *vcpu)
4030 {
4031 return nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
4032 to_vmx(vcpu)->nested.preemption_timer_expired;
4033 }
4034
4035 static bool vmx_has_nested_events(struct kvm_vcpu *vcpu, bool for_injection)
4036 {
4037 struct vcpu_vmx *vmx = to_vmx(vcpu);
4038 void *vapic = vmx->nested.virtual_apic_map.hva;
4039 int max_irr, vppr;
4040
4041 if (nested_vmx_preemption_timer_pending(vcpu) ||
4042 vmx->nested.mtf_pending)
4043 return true;
4044
4045 /*
4046 * Virtual Interrupt Delivery doesn't require manual injection. Either
4047 * the interrupt is already in GUEST_RVI and will be recognized by CPU
4048 * at VM-Entry, or there is a KVM_REQ_EVENT pending and KVM will move
4049 * the interrupt from the PIR to RVI prior to entering the guest.
4050 */
4051 if (for_injection)
4052 return false;
4053
4054 if (!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
4055 __vmx_interrupt_blocked(vcpu))
4056 return false;
4057
4058 if (!vapic)
4059 return false;
4060
4061 vppr = *((u32 *)(vapic + APIC_PROCPRI));
4062
4063 max_irr = vmx_get_rvi();
4064 if ((max_irr & 0xf0) > (vppr & 0xf0))
4065 return true;
4066
4067 if (vmx->nested.pi_pending && vmx->nested.pi_desc &&
4068 pi_test_on(vmx->nested.pi_desc)) {
4069 max_irr = pi_find_highest_vector(vmx->nested.pi_desc);
4070 if (max_irr > 0 && (max_irr & 0xf0) > (vppr & 0xf0))
4071 return true;
4072 }
4073
4074 return false;
4075 }
4076
4077 /*
4078 * Per the Intel SDM's table "Priority Among Concurrent Events", with minor
4079 * edits to fill in missing examples, e.g. #DB due to split-lock accesses,
4080 * and less minor edits to splice in the priority of VMX Non-Root specific
4081 * events, e.g. MTF and NMI/INTR-window exiting.
4082 *
4083 * 1 Hardware Reset and Machine Checks
4084 * - RESET
4085 * - Machine Check
4086 *
4087 * 2 Trap on Task Switch
4088 * - T flag in TSS is set (on task switch)
4089 *
4090 * 3 External Hardware Interventions
4091 * - FLUSH
4092 * - STOPCLK
4093 * - SMI
4094 * - INIT
4095 *
4096 * 3.5 Monitor Trap Flag (MTF) VM-exit[1]
4097 *
4098 * 4 Traps on Previous Instruction
4099 * - Breakpoints
4100 * - Trap-class Debug Exceptions (#DB due to TF flag set, data/I-O
4101 * breakpoint, or #DB due to a split-lock access)
4102 *
4103 * 4.3 VMX-preemption timer expired VM-exit
4104 *
4105 * 4.6 NMI-window exiting VM-exit[2]
4106 *
4107 * 5 Nonmaskable Interrupts (NMI)
4108 *
4109 * 5.5 Interrupt-window exiting VM-exit and Virtual-interrupt delivery
4110 *
4111 * 6 Maskable Hardware Interrupts
4112 *
4113 * 7 Code Breakpoint Fault
4114 *
4115 * 8 Faults from Fetching Next Instruction
4116 * - Code-Segment Limit Violation
4117 * - Code Page Fault
4118 * - Control protection exception (missing ENDBRANCH at target of indirect
4119 * call or jump)
4120 *
4121 * 9 Faults from Decoding Next Instruction
4122 * - Instruction length > 15 bytes
4123 * - Invalid Opcode
4124 * - Coprocessor Not Available
4125 *
4126 *10 Faults on Executing Instruction
4127 * - Overflow
4128 * - Bound error
4129 * - Invalid TSS
4130 * - Segment Not Present
4131 * - Stack fault
4132 * - General Protection
4133 * - Data Page Fault
4134 * - Alignment Check
4135 * - x86 FPU Floating-point exception
4136 * - SIMD floating-point exception
4137 * - Virtualization exception
4138 * - Control protection exception
4139 *
4140 * [1] Per the "Monitor Trap Flag" section: System-management interrupts (SMIs),
4141 * INIT signals, and higher priority events take priority over MTF VM exits.
4142 * MTF VM exits take priority over debug-trap exceptions and lower priority
4143 * events.
4144 *
4145 * [2] Debug-trap exceptions and higher priority events take priority over VM exits
4146 * caused by the VMX-preemption timer. VM exits caused by the VMX-preemption
4147 * timer take priority over VM exits caused by the "NMI-window exiting"
4148 * VM-execution control and lower priority events.
4149 *
4150 * [3] Debug-trap exceptions and higher priority events take priority over VM exits
4151 * caused by "NMI-window exiting". VM exits caused by this control take
4152 * priority over non-maskable interrupts (NMIs) and lower priority events.
4153 *
4154 * [4] Virtual-interrupt delivery has the same priority as that of VM exits due to
4155 * the 1-setting of the "interrupt-window exiting" VM-execution control. Thus,
4156 * non-maskable interrupts (NMIs) and higher priority events take priority over
4157 * delivery of a virtual interrupt; delivery of a virtual interrupt takes
4158 * priority over external interrupts and lower priority events.
4159 */
4160 static int vmx_check_nested_events(struct kvm_vcpu *vcpu)
4161 {
4162 struct kvm_lapic *apic = vcpu->arch.apic;
4163 struct vcpu_vmx *vmx = to_vmx(vcpu);
4164 /*
4165 * Only a pending nested run blocks a pending exception. If there is a
4166 * previously injected event, the pending exception occurred while said
4167 * event was being delivered and thus needs to be handled.
4168 */
4169 bool block_nested_exceptions = vmx->nested.nested_run_pending;
4170 /*
4171 * New events (not exceptions) are only recognized at instruction
4172 * boundaries. If an event needs reinjection, then KVM is handling a
4173 * VM-Exit that occurred _during_ instruction execution; new events are
4174 * blocked until the instruction completes.
4175 */
4176 bool block_nested_events = block_nested_exceptions ||
4177 kvm_event_needs_reinjection(vcpu);
4178
4179 if (lapic_in_kernel(vcpu) &&
4180 test_bit(KVM_APIC_INIT, &apic->pending_events)) {
4181 if (block_nested_events)
4182 return -EBUSY;
4183 nested_vmx_update_pending_dbg(vcpu);
4184 clear_bit(KVM_APIC_INIT, &apic->pending_events);
4185 if (vcpu->arch.mp_state != KVM_MP_STATE_INIT_RECEIVED)
4186 nested_vmx_vmexit(vcpu, EXIT_REASON_INIT_SIGNAL, 0, 0);
4187
4188 /* MTF is discarded if the vCPU is in WFS. */
4189 vmx->nested.mtf_pending = false;
4190 return 0;
4191 }
4192
4193 if (lapic_in_kernel(vcpu) &&
4194 test_bit(KVM_APIC_SIPI, &apic->pending_events)) {
4195 if (block_nested_events)
4196 return -EBUSY;
4197
4198 clear_bit(KVM_APIC_SIPI, &apic->pending_events);
4199 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
4200 nested_vmx_vmexit(vcpu, EXIT_REASON_SIPI_SIGNAL, 0,
4201 apic->sipi_vector & 0xFFUL);
4202 return 0;
4203 }
4204 /* Fallthrough, the SIPI is completely ignored. */
4205 }
4206
4207 /*
4208 * Process exceptions that are higher priority than Monitor Trap Flag:
4209 * fault-like exceptions, TSS T flag #DB (not emulated by KVM, but
4210 * could theoretically come in from userspace), and ICEBP (INT1).
4211 *
4212 * TODO: SMIs have higher priority than MTF and trap-like #DBs (except
4213 * for TSS T flag #DBs). KVM also doesn't save/restore pending MTF
4214 * across SMI/RSM as it should; that needs to be addressed in order to
4215 * prioritize SMI over MTF and trap-like #DBs.
4216 */
4217 if (vcpu->arch.exception_vmexit.pending &&
4218 !vmx_is_low_priority_db_trap(&vcpu->arch.exception_vmexit)) {
4219 if (block_nested_exceptions)
4220 return -EBUSY;
4221
4222 nested_vmx_inject_exception_vmexit(vcpu);
4223 return 0;
4224 }
4225
4226 if (vcpu->arch.exception.pending &&
4227 !vmx_is_low_priority_db_trap(&vcpu->arch.exception)) {
4228 if (block_nested_exceptions)
4229 return -EBUSY;
4230 goto no_vmexit;
4231 }
4232
4233 if (vmx->nested.mtf_pending) {
4234 if (block_nested_events)
4235 return -EBUSY;
4236 nested_vmx_update_pending_dbg(vcpu);
4237 nested_vmx_vmexit(vcpu, EXIT_REASON_MONITOR_TRAP_FLAG, 0, 0);
4238 return 0;
4239 }
4240
4241 if (vcpu->arch.exception_vmexit.pending) {
4242 if (block_nested_exceptions)
4243 return -EBUSY;
4244
4245 nested_vmx_inject_exception_vmexit(vcpu);
4246 return 0;
4247 }
4248
4249 if (vcpu->arch.exception.pending) {
4250 if (block_nested_exceptions)
4251 return -EBUSY;
4252 goto no_vmexit;
4253 }
4254
4255 if (nested_vmx_preemption_timer_pending(vcpu)) {
4256 if (block_nested_events)
4257 return -EBUSY;
4258 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
4259 return 0;
4260 }
4261
4262 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
4263 if (block_nested_events)
4264 return -EBUSY;
4265 goto no_vmexit;
4266 }
4267
4268 if (vcpu->arch.nmi_pending && !vmx_nmi_blocked(vcpu)) {
4269 if (block_nested_events)
4270 return -EBUSY;
4271 if (!nested_exit_on_nmi(vcpu))
4272 goto no_vmexit;
4273
4274 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
4275 NMI_VECTOR | INTR_TYPE_NMI_INTR |
4276 INTR_INFO_VALID_MASK, 0);
4277 /*
4278 * The NMI-triggered VM exit counts as injection:
4279 * clear this one and block further NMIs.
4280 */
4281 vcpu->arch.nmi_pending = 0;
4282 vmx_set_nmi_mask(vcpu, true);
4283 return 0;
4284 }
4285
4286 if (kvm_cpu_has_interrupt(vcpu) && !vmx_interrupt_blocked(vcpu)) {
4287 if (block_nested_events)
4288 return -EBUSY;
4289 if (!nested_exit_on_intr(vcpu))
4290 goto no_vmexit;
4291 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
4292 return 0;
4293 }
4294
4295 no_vmexit:
4296 return vmx_complete_nested_posted_interrupt(vcpu);
4297 }
4298
4299 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
4300 {
4301 ktime_t remaining =
4302 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
4303 u64 value;
4304
4305 if (ktime_to_ns(remaining) <= 0)
4306 return 0;
4307
4308 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
4309 do_div(value, 1000000);
4310 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
4311 }
4312
4313 static bool is_vmcs12_ext_field(unsigned long field)
4314 {
4315 switch (field) {
4316 case GUEST_ES_SELECTOR:
4317 case GUEST_CS_SELECTOR:
4318 case GUEST_SS_SELECTOR:
4319 case GUEST_DS_SELECTOR:
4320 case GUEST_FS_SELECTOR:
4321 case GUEST_GS_SELECTOR:
4322 case GUEST_LDTR_SELECTOR:
4323 case GUEST_TR_SELECTOR:
4324 case GUEST_ES_LIMIT:
4325 case GUEST_CS_LIMIT:
4326 case GUEST_SS_LIMIT:
4327 case GUEST_DS_LIMIT:
4328 case GUEST_FS_LIMIT:
4329 case GUEST_GS_LIMIT:
4330 case GUEST_LDTR_LIMIT:
4331 case GUEST_TR_LIMIT:
4332 case GUEST_GDTR_LIMIT:
4333 case GUEST_IDTR_LIMIT:
4334 case GUEST_ES_AR_BYTES:
4335 case GUEST_DS_AR_BYTES:
4336 case GUEST_FS_AR_BYTES:
4337 case GUEST_GS_AR_BYTES:
4338 case GUEST_LDTR_AR_BYTES:
4339 case GUEST_TR_AR_BYTES:
4340 case GUEST_ES_BASE:
4341 case GUEST_CS_BASE:
4342 case GUEST_SS_BASE:
4343 case GUEST_DS_BASE:
4344 case GUEST_FS_BASE:
4345 case GUEST_GS_BASE:
4346 case GUEST_LDTR_BASE:
4347 case GUEST_TR_BASE:
4348 case GUEST_GDTR_BASE:
4349 case GUEST_IDTR_BASE:
4350 case GUEST_PENDING_DBG_EXCEPTIONS:
4351 case GUEST_BNDCFGS:
4352 return true;
4353 default:
4354 break;
4355 }
4356
4357 return false;
4358 }
4359
4360 static void sync_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4361 struct vmcs12 *vmcs12)
4362 {
4363 struct vcpu_vmx *vmx = to_vmx(vcpu);
4364
4365 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
4366 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
4367 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
4368 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
4369 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
4370 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
4371 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
4372 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
4373 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
4374 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
4375 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
4376 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
4377 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
4378 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
4379 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
4380 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
4381 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
4382 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
4383 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
4384 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
4385 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
4386 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
4387 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
4388 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
4389 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
4390 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
4391 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
4392 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
4393 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
4394 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
4395 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
4396 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
4397 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
4398 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
4399 vmcs12->guest_pending_dbg_exceptions =
4400 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
4401
4402 vmx->nested.need_sync_vmcs02_to_vmcs12_rare = false;
4403 }
4404
4405 static void copy_vmcs02_to_vmcs12_rare(struct kvm_vcpu *vcpu,
4406 struct vmcs12 *vmcs12)
4407 {
4408 struct vcpu_vmx *vmx = to_vmx(vcpu);
4409 int cpu;
4410
4411 if (!vmx->nested.need_sync_vmcs02_to_vmcs12_rare)
4412 return;
4413
4414
4415 WARN_ON_ONCE(vmx->loaded_vmcs != &vmx->vmcs01);
4416
4417 cpu = get_cpu();
4418 vmx->loaded_vmcs = &vmx->nested.vmcs02;
4419 vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->vmcs01);
4420
4421 sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4422
4423 vmx->loaded_vmcs = &vmx->vmcs01;
4424 vmx_vcpu_load_vmcs(vcpu, cpu, &vmx->nested.vmcs02);
4425 put_cpu();
4426 }
4427
4428 /*
4429 * Update the guest state fields of vmcs12 to reflect changes that
4430 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
4431 * VM-entry controls is also updated, since this is really a guest
4432 * state bit.)
4433 */
4434 static void sync_vmcs02_to_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
4435 {
4436 struct vcpu_vmx *vmx = to_vmx(vcpu);
4437
4438 if (nested_vmx_is_evmptr12_valid(vmx))
4439 sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
4440
4441 vmx->nested.need_sync_vmcs02_to_vmcs12_rare =
4442 !nested_vmx_is_evmptr12_valid(vmx);
4443
4444 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
4445 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
4446
4447 vmcs12->guest_rsp = kvm_rsp_read(vcpu);
4448 vmcs12->guest_rip = kvm_rip_read(vcpu);
4449 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
4450
4451 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
4452 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
4453
4454 vmcs12->guest_interruptibility_info =
4455 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
4456
4457 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
4458 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
4459 else if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4460 vmcs12->guest_activity_state = GUEST_ACTIVITY_WAIT_SIPI;
4461 else
4462 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
4463
4464 if (nested_cpu_has_preemption_timer(vmcs12) &&
4465 vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER &&
4466 !vmx->nested.nested_run_pending)
4467 vmcs12->vmx_preemption_timer_value =
4468 vmx_get_preemption_timer_value(vcpu);
4469
4470 /*
4471 * In some cases (usually, nested EPT), L2 is allowed to change its
4472 * own CR3 without exiting. If it has changed it, we must keep it.
4473 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
4474 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
4475 *
4476 * Additionally, restore L2's PDPTR to vmcs12.
4477 */
4478 if (enable_ept) {
4479 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
4480 if (nested_cpu_has_ept(vmcs12) && is_pae_paging(vcpu)) {
4481 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
4482 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
4483 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
4484 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
4485 }
4486 }
4487
4488 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
4489
4490 if (nested_cpu_has_vid(vmcs12))
4491 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
4492
4493 vmcs12->vm_entry_controls =
4494 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
4495 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
4496
4497 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS)
4498 vmcs12->guest_dr7 = vcpu->arch.dr7;
4499
4500 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
4501 vmcs12->guest_ia32_efer = vcpu->arch.efer;
4502 }
4503
4504 /*
4505 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
4506 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
4507 * and this function updates it to reflect the changes to the guest state while
4508 * L2 was running (and perhaps made some exits which were handled directly by L0
4509 * without going back to L1), and to reflect the exit reason.
4510 * Note that we do not have to copy here all VMCS fields, just those that
4511 * could have changed by the L2 guest or the exit - i.e., the guest-state and
4512 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
4513 * which already writes to vmcs12 directly.
4514 */
4515 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
4516 u32 vm_exit_reason, u32 exit_intr_info,
4517 unsigned long exit_qualification)
4518 {
4519 /* update exit information fields: */
4520 vmcs12->vm_exit_reason = vm_exit_reason;
4521 if (to_vmx(vcpu)->exit_reason.enclave_mode)
4522 vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
4523 vmcs12->exit_qualification = exit_qualification;
4524
4525 /*
4526 * On VM-Exit due to a failed VM-Entry, the VMCS isn't marked launched
4527 * and only EXIT_REASON and EXIT_QUALIFICATION are updated, all other
4528 * exit info fields are unmodified.
4529 */
4530 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
4531 vmcs12->launch_state = 1;
4532
4533 /* vm_entry_intr_info_field is cleared on exit. Emulate this
4534 * instead of reading the real value. */
4535 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
4536
4537 /*
4538 * Transfer the event that L0 or L1 may wanted to inject into
4539 * L2 to IDT_VECTORING_INFO_FIELD.
4540 */
4541 vmcs12_save_pending_event(vcpu, vmcs12,
4542 vm_exit_reason, exit_intr_info);
4543
4544 vmcs12->vm_exit_intr_info = exit_intr_info;
4545 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4546 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4547
4548 /*
4549 * According to spec, there's no need to store the guest's
4550 * MSRs if the exit is due to a VM-entry failure that occurs
4551 * during or after loading the guest state. Since this exit
4552 * does not fall in that category, we need to save the MSRs.
4553 */
4554 if (nested_vmx_store_msr(vcpu,
4555 vmcs12->vm_exit_msr_store_addr,
4556 vmcs12->vm_exit_msr_store_count))
4557 nested_vmx_abort(vcpu,
4558 VMX_ABORT_SAVE_GUEST_MSR_FAIL);
4559 }
4560 }
4561
4562 /*
4563 * A part of what we need to when the nested L2 guest exits and we want to
4564 * run its L1 parent, is to reset L1's guest state to the host state specified
4565 * in vmcs12.
4566 * This function is to be called not only on normal nested exit, but also on
4567 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
4568 * Failures During or After Loading Guest State").
4569 * This function should be called when the active VMCS is L1's (vmcs01).
4570 */
4571 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
4572 struct vmcs12 *vmcs12)
4573 {
4574 enum vm_entry_failure_code ignored;
4575 struct kvm_segment seg;
4576
4577 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
4578 vcpu->arch.efer = vmcs12->host_ia32_efer;
4579 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4580 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
4581 else
4582 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
4583 vmx_set_efer(vcpu, vcpu->arch.efer);
4584
4585 kvm_rsp_write(vcpu, vmcs12->host_rsp);
4586 kvm_rip_write(vcpu, vmcs12->host_rip);
4587 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
4588 vmx_set_interrupt_shadow(vcpu, 0);
4589
4590 /*
4591 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
4592 * actually changed, because vmx_set_cr0 refers to efer set above.
4593 *
4594 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
4595 * (KVM doesn't change it);
4596 */
4597 vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4598 vmx_set_cr0(vcpu, vmcs12->host_cr0);
4599
4600 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
4601 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4602 vmx_set_cr4(vcpu, vmcs12->host_cr4);
4603
4604 nested_ept_uninit_mmu_context(vcpu);
4605
4606 /*
4607 * Only PDPTE load can fail as the value of cr3 was checked on entry and
4608 * couldn't have changed.
4609 */
4610 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, true, &ignored))
4611 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
4612
4613 nested_vmx_transition_tlb_flush(vcpu, vmcs12, false);
4614
4615 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
4616 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
4617 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
4618 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
4619 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
4620 vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
4621 vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
4622
4623 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
4624 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
4625 vmcs_write64(GUEST_BNDCFGS, 0);
4626
4627 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
4628 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
4629 vcpu->arch.pat = vmcs12->host_ia32_pat;
4630 }
4631 if ((vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL) &&
4632 kvm_pmu_has_perf_global_ctrl(vcpu_to_pmu(vcpu)))
4633 WARN_ON_ONCE(kvm_set_msr(vcpu, MSR_CORE_PERF_GLOBAL_CTRL,
4634 vmcs12->host_ia32_perf_global_ctrl));
4635
4636 /* Set L1 segment info according to Intel SDM
4637 27.5.2 Loading Host Segment and Descriptor-Table Registers */
4638 seg = (struct kvm_segment) {
4639 .base = 0,
4640 .limit = 0xFFFFFFFF,
4641 .selector = vmcs12->host_cs_selector,
4642 .type = 11,
4643 .present = 1,
4644 .s = 1,
4645 .g = 1
4646 };
4647 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
4648 seg.l = 1;
4649 else
4650 seg.db = 1;
4651 __vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
4652 seg = (struct kvm_segment) {
4653 .base = 0,
4654 .limit = 0xFFFFFFFF,
4655 .type = 3,
4656 .present = 1,
4657 .s = 1,
4658 .db = 1,
4659 .g = 1
4660 };
4661 seg.selector = vmcs12->host_ds_selector;
4662 __vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
4663 seg.selector = vmcs12->host_es_selector;
4664 __vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
4665 seg.selector = vmcs12->host_ss_selector;
4666 __vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
4667 seg.selector = vmcs12->host_fs_selector;
4668 seg.base = vmcs12->host_fs_base;
4669 __vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
4670 seg.selector = vmcs12->host_gs_selector;
4671 seg.base = vmcs12->host_gs_base;
4672 __vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
4673 seg = (struct kvm_segment) {
4674 .base = vmcs12->host_tr_base,
4675 .limit = 0x67,
4676 .selector = vmcs12->host_tr_selector,
4677 .type = 11,
4678 .present = 1
4679 };
4680 __vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
4681
4682 memset(&seg, 0, sizeof(seg));
4683 seg.unusable = 1;
4684 __vmx_set_segment(vcpu, &seg, VCPU_SREG_LDTR);
4685
4686 kvm_set_dr(vcpu, 7, 0x400);
4687 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4688
4689 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
4690 vmcs12->vm_exit_msr_load_count))
4691 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4692
4693 to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
4694 }
4695
4696 static inline u64 nested_vmx_get_vmcs01_guest_efer(struct vcpu_vmx *vmx)
4697 {
4698 struct vmx_uret_msr *efer_msr;
4699 unsigned int i;
4700
4701 if (vm_entry_controls_get(vmx) & VM_ENTRY_LOAD_IA32_EFER)
4702 return vmcs_read64(GUEST_IA32_EFER);
4703
4704 if (cpu_has_load_ia32_efer())
4705 return kvm_host.efer;
4706
4707 for (i = 0; i < vmx->msr_autoload.guest.nr; ++i) {
4708 if (vmx->msr_autoload.guest.val[i].index == MSR_EFER)
4709 return vmx->msr_autoload.guest.val[i].value;
4710 }
4711
4712 efer_msr = vmx_find_uret_msr(vmx, MSR_EFER);
4713 if (efer_msr)
4714 return efer_msr->data;
4715
4716 return kvm_host.efer;
4717 }
4718
4719 static void nested_vmx_restore_host_state(struct kvm_vcpu *vcpu)
4720 {
4721 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4722 struct vcpu_vmx *vmx = to_vmx(vcpu);
4723 struct vmx_msr_entry g, h;
4724 gpa_t gpa;
4725 u32 i, j;
4726
4727 vcpu->arch.pat = vmcs_read64(GUEST_IA32_PAT);
4728
4729 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
4730 /*
4731 * L1's host DR7 is lost if KVM_GUESTDBG_USE_HW_BP is set
4732 * as vmcs01.GUEST_DR7 contains a userspace defined value
4733 * and vcpu->arch.dr7 is not squirreled away before the
4734 * nested VMENTER (not worth adding a variable in nested_vmx).
4735 */
4736 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
4737 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
4738 else
4739 WARN_ON(kvm_set_dr(vcpu, 7, vmcs_readl(GUEST_DR7)));
4740 }
4741
4742 /*
4743 * Note that calling vmx_set_{efer,cr0,cr4} is important as they
4744 * handle a variety of side effects to KVM's software model.
4745 */
4746 vmx_set_efer(vcpu, nested_vmx_get_vmcs01_guest_efer(vmx));
4747
4748 vcpu->arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4749 vmx_set_cr0(vcpu, vmcs_readl(CR0_READ_SHADOW));
4750
4751 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
4752 vmx_set_cr4(vcpu, vmcs_readl(CR4_READ_SHADOW));
4753
4754 nested_ept_uninit_mmu_context(vcpu);
4755 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4756 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
4757
4758 /*
4759 * Use ept_save_pdptrs(vcpu) to load the MMU's cached PDPTRs
4760 * from vmcs01 (if necessary). The PDPTRs are not loaded on
4761 * VMFail, like everything else we just need to ensure our
4762 * software model is up-to-date.
4763 */
4764 if (enable_ept && is_pae_paging(vcpu))
4765 ept_save_pdptrs(vcpu);
4766
4767 kvm_mmu_reset_context(vcpu);
4768
4769 /*
4770 * This nasty bit of open coding is a compromise between blindly
4771 * loading L1's MSRs using the exit load lists (incorrect emulation
4772 * of VMFail), leaving the nested VM's MSRs in the software model
4773 * (incorrect behavior) and snapshotting the modified MSRs (too
4774 * expensive since the lists are unbound by hardware). For each
4775 * MSR that was (prematurely) loaded from the nested VMEntry load
4776 * list, reload it from the exit load list if it exists and differs
4777 * from the guest value. The intent is to stuff host state as
4778 * silently as possible, not to fully process the exit load list.
4779 */
4780 for (i = 0; i < vmcs12->vm_entry_msr_load_count; i++) {
4781 gpa = vmcs12->vm_entry_msr_load_addr + (i * sizeof(g));
4782 if (kvm_vcpu_read_guest(vcpu, gpa, &g, sizeof(g))) {
4783 pr_debug_ratelimited(
4784 "%s read MSR index failed (%u, 0x%08llx)\n",
4785 __func__, i, gpa);
4786 goto vmabort;
4787 }
4788
4789 for (j = 0; j < vmcs12->vm_exit_msr_load_count; j++) {
4790 gpa = vmcs12->vm_exit_msr_load_addr + (j * sizeof(h));
4791 if (kvm_vcpu_read_guest(vcpu, gpa, &h, sizeof(h))) {
4792 pr_debug_ratelimited(
4793 "%s read MSR failed (%u, 0x%08llx)\n",
4794 __func__, j, gpa);
4795 goto vmabort;
4796 }
4797 if (h.index != g.index)
4798 continue;
4799 if (h.value == g.value)
4800 break;
4801
4802 if (nested_vmx_load_msr_check(vcpu, &h)) {
4803 pr_debug_ratelimited(
4804 "%s check failed (%u, 0x%x, 0x%x)\n",
4805 __func__, j, h.index, h.reserved);
4806 goto vmabort;
4807 }
4808
4809 if (kvm_set_msr(vcpu, h.index, h.value)) {
4810 pr_debug_ratelimited(
4811 "%s WRMSR failed (%u, 0x%x, 0x%llx)\n",
4812 __func__, j, h.index, h.value);
4813 goto vmabort;
4814 }
4815 }
4816 }
4817
4818 return;
4819
4820 vmabort:
4821 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
4822 }
4823
4824 /*
4825 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
4826 * and modify vmcs12 to make it see what it would expect to see there if
4827 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
4828 */
4829 void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
4830 u32 exit_intr_info, unsigned long exit_qualification)
4831 {
4832 struct vcpu_vmx *vmx = to_vmx(vcpu);
4833 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4834
4835 /* Pending MTF traps are discarded on VM-Exit. */
4836 vmx->nested.mtf_pending = false;
4837
4838 /* trying to cancel vmlaunch/vmresume is a bug */
4839 WARN_ON_ONCE(vmx->nested.nested_run_pending);
4840
4841 #ifdef CONFIG_KVM_HYPERV
4842 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
4843 /*
4844 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
4845 * Enlightened VMCS after migration and we still need to
4846 * do that when something is forcing L2->L1 exit prior to
4847 * the first L2 run.
4848 */
4849 (void)nested_get_evmcs_page(vcpu);
4850 }
4851 #endif
4852
4853 /* Service pending TLB flush requests for L2 before switching to L1. */
4854 kvm_service_local_tlb_flush_requests(vcpu);
4855
4856 /*
4857 * VCPU_EXREG_PDPTR will be clobbered in arch/x86/kvm/vmx/vmx.h between
4858 * now and the new vmentry. Ensure that the VMCS02 PDPTR fields are
4859 * up-to-date before switching to L1.
4860 */
4861 if (enable_ept && is_pae_paging(vcpu))
4862 vmx_ept_load_pdptrs(vcpu);
4863
4864 leave_guest_mode(vcpu);
4865
4866 if (nested_cpu_has_preemption_timer(vmcs12))
4867 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
4868
4869 if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING)) {
4870 vcpu->arch.tsc_offset = vcpu->arch.l1_tsc_offset;
4871 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
4872 vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
4873 }
4874
4875 if (likely(!vmx->fail)) {
4876 sync_vmcs02_to_vmcs12(vcpu, vmcs12);
4877
4878 if (vm_exit_reason != -1)
4879 prepare_vmcs12(vcpu, vmcs12, vm_exit_reason,
4880 exit_intr_info, exit_qualification);
4881
4882 /*
4883 * Must happen outside of sync_vmcs02_to_vmcs12() as it will
4884 * also be used to capture vmcs12 cache as part of
4885 * capturing nVMX state for snapshot (migration).
4886 *
4887 * Otherwise, this flush will dirty guest memory at a
4888 * point it is already assumed by user-space to be
4889 * immutable.
4890 */
4891 nested_flush_cached_shadow_vmcs12(vcpu, vmcs12);
4892 } else {
4893 /*
4894 * The only expected VM-instruction error is "VM entry with
4895 * invalid control field(s)." Anything else indicates a
4896 * problem with L0. And we should never get here with a
4897 * VMFail of any type if early consistency checks are enabled.
4898 */
4899 WARN_ON_ONCE(vmcs_read32(VM_INSTRUCTION_ERROR) !=
4900 VMXERR_ENTRY_INVALID_CONTROL_FIELD);
4901 WARN_ON_ONCE(nested_early_check);
4902 }
4903
4904 /*
4905 * Drop events/exceptions that were queued for re-injection to L2
4906 * (picked up via vmx_complete_interrupts()), as well as exceptions
4907 * that were pending for L2. Note, this must NOT be hoisted above
4908 * prepare_vmcs12(), events/exceptions queued for re-injection need to
4909 * be captured in vmcs12 (see vmcs12_save_pending_event()).
4910 */
4911 vcpu->arch.nmi_injected = false;
4912 kvm_clear_exception_queue(vcpu);
4913 kvm_clear_interrupt_queue(vcpu);
4914
4915 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
4916
4917 /*
4918 * If IBRS is advertised to the vCPU, KVM must flush the indirect
4919 * branch predictors when transitioning from L2 to L1, as L1 expects
4920 * hardware (KVM in this case) to provide separate predictor modes.
4921 * Bare metal isolates VMX root (host) from VMX non-root (guest), but
4922 * doesn't isolate different VMCSs, i.e. in this case, doesn't provide
4923 * separate modes for L2 vs L1.
4924 */
4925 if (guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
4926 indirect_branch_prediction_barrier();
4927
4928 /* Update any VMCS fields that might have changed while L2 ran */
4929 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.host.nr);
4930 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.guest.nr);
4931 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
4932 if (kvm_caps.has_tsc_control)
4933 vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
4934
4935 if (vmx->nested.l1_tpr_threshold != -1)
4936 vmcs_write32(TPR_THRESHOLD, vmx->nested.l1_tpr_threshold);
4937
4938 if (vmx->nested.change_vmcs01_virtual_apic_mode) {
4939 vmx->nested.change_vmcs01_virtual_apic_mode = false;
4940 vmx_set_virtual_apic_mode(vcpu);
4941 }
4942
4943 if (vmx->nested.update_vmcs01_cpu_dirty_logging) {
4944 vmx->nested.update_vmcs01_cpu_dirty_logging = false;
4945 vmx_update_cpu_dirty_logging(vcpu);
4946 }
4947
4948 /* Unpin physical memory we referred to in vmcs02 */
4949 kvm_vcpu_unmap(vcpu, &vmx->nested.apic_access_page_map, false);
4950 kvm_vcpu_unmap(vcpu, &vmx->nested.virtual_apic_map, true);
4951 kvm_vcpu_unmap(vcpu, &vmx->nested.pi_desc_map, true);
4952 vmx->nested.pi_desc = NULL;
4953
4954 if (vmx->nested.reload_vmcs01_apic_access_page) {
4955 vmx->nested.reload_vmcs01_apic_access_page = false;
4956 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4957 }
4958
4959 if (vmx->nested.update_vmcs01_apicv_status) {
4960 vmx->nested.update_vmcs01_apicv_status = false;
4961 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
4962 }
4963
4964 if ((vm_exit_reason != -1) &&
4965 (enable_shadow_vmcs || nested_vmx_is_evmptr12_valid(vmx)))
4966 vmx->nested.need_vmcs12_to_shadow_sync = true;
4967
4968 /* in case we halted in L2 */
4969 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4970
4971 if (likely(!vmx->fail)) {
4972 if ((u16)vm_exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
4973 nested_exit_intr_ack_set(vcpu)) {
4974 int irq = kvm_cpu_get_interrupt(vcpu);
4975 WARN_ON(irq < 0);
4976 vmcs12->vm_exit_intr_info = irq |
4977 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
4978 }
4979
4980 if (vm_exit_reason != -1)
4981 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
4982 vmcs12->exit_qualification,
4983 vmcs12->idt_vectoring_info_field,
4984 vmcs12->vm_exit_intr_info,
4985 vmcs12->vm_exit_intr_error_code,
4986 KVM_ISA_VMX);
4987
4988 load_vmcs12_host_state(vcpu, vmcs12);
4989
4990 return;
4991 }
4992
4993 /*
4994 * After an early L2 VM-entry failure, we're now back
4995 * in L1 which thinks it just finished a VMLAUNCH or
4996 * VMRESUME instruction, so we need to set the failure
4997 * flag and the VM-instruction error field of the VMCS
4998 * accordingly, and skip the emulated instruction.
4999 */
5000 (void)nested_vmx_fail(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
5001
5002 /*
5003 * Restore L1's host state to KVM's software model. We're here
5004 * because a consistency check was caught by hardware, which
5005 * means some amount of guest state has been propagated to KVM's
5006 * model and needs to be unwound to the host's state.
5007 */
5008 nested_vmx_restore_host_state(vcpu);
5009
5010 vmx->fail = 0;
5011 }
5012
5013 static void nested_vmx_triple_fault(struct kvm_vcpu *vcpu)
5014 {
5015 kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5016 nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
5017 }
5018
5019 /*
5020 * Decode the memory-address operand of a vmx instruction, as recorded on an
5021 * exit caused by such an instruction (run by a guest hypervisor).
5022 * On success, returns 0. When the operand is invalid, returns 1 and throws
5023 * #UD, #GP, or #SS.
5024 */
5025 int get_vmx_mem_address(struct kvm_vcpu *vcpu, unsigned long exit_qualification,
5026 u32 vmx_instruction_info, bool wr, int len, gva_t *ret)
5027 {
5028 gva_t off;
5029 bool exn;
5030 struct kvm_segment s;
5031
5032 /*
5033 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5034 * Execution", on an exit, vmx_instruction_info holds most of the
5035 * addressing components of the operand. Only the displacement part
5036 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5037 * For how an actual address is calculated from all these components,
5038 * refer to Vol. 1, "Operand Addressing".
5039 */
5040 int scaling = vmx_instruction_info & 3;
5041 int addr_size = (vmx_instruction_info >> 7) & 7;
5042 bool is_reg = vmx_instruction_info & (1u << 10);
5043 int seg_reg = (vmx_instruction_info >> 15) & 7;
5044 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5045 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5046 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5047 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5048
5049 if (is_reg) {
5050 kvm_queue_exception(vcpu, UD_VECTOR);
5051 return 1;
5052 }
5053
5054 /* Addr = segment_base + offset */
5055 /* offset = base + [index * scale] + displacement */
5056 off = exit_qualification; /* holds the displacement */
5057 if (addr_size == 1)
5058 off = (gva_t)sign_extend64(off, 31);
5059 else if (addr_size == 0)
5060 off = (gva_t)sign_extend64(off, 15);
5061 if (base_is_valid)
5062 off += kvm_register_read(vcpu, base_reg);
5063 if (index_is_valid)
5064 off += kvm_register_read(vcpu, index_reg) << scaling;
5065 vmx_get_segment(vcpu, &s, seg_reg);
5066
5067 /*
5068 * The effective address, i.e. @off, of a memory operand is truncated
5069 * based on the address size of the instruction. Note that this is
5070 * the *effective address*, i.e. the address prior to accounting for
5071 * the segment's base.
5072 */
5073 if (addr_size == 1) /* 32 bit */
5074 off &= 0xffffffff;
5075 else if (addr_size == 0) /* 16 bit */
5076 off &= 0xffff;
5077
5078 /* Checks for #GP/#SS exceptions. */
5079 exn = false;
5080 if (is_long_mode(vcpu)) {
5081 /*
5082 * The virtual/linear address is never truncated in 64-bit
5083 * mode, e.g. a 32-bit address size can yield a 64-bit virtual
5084 * address when using FS/GS with a non-zero base.
5085 */
5086 if (seg_reg == VCPU_SREG_FS || seg_reg == VCPU_SREG_GS)
5087 *ret = s.base + off;
5088 else
5089 *ret = off;
5090
5091 *ret = vmx_get_untagged_addr(vcpu, *ret, 0);
5092 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
5093 * non-canonical form. This is the only check on the memory
5094 * destination for long mode!
5095 */
5096 exn = is_noncanonical_address(*ret, vcpu);
5097 } else {
5098 /*
5099 * When not in long mode, the virtual/linear address is
5100 * unconditionally truncated to 32 bits regardless of the
5101 * address size.
5102 */
5103 *ret = (s.base + off) & 0xffffffff;
5104
5105 /* Protected mode: apply checks for segment validity in the
5106 * following order:
5107 * - segment type check (#GP(0) may be thrown)
5108 * - usability check (#GP(0)/#SS(0))
5109 * - limit check (#GP(0)/#SS(0))
5110 */
5111 if (wr)
5112 /* #GP(0) if the destination operand is located in a
5113 * read-only data segment or any code segment.
5114 */
5115 exn = ((s.type & 0xa) == 0 || (s.type & 8));
5116 else
5117 /* #GP(0) if the source operand is located in an
5118 * execute-only code segment
5119 */
5120 exn = ((s.type & 0xa) == 8);
5121 if (exn) {
5122 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5123 return 1;
5124 }
5125 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
5126 */
5127 exn = (s.unusable != 0);
5128
5129 /*
5130 * Protected mode: #GP(0)/#SS(0) if the memory operand is
5131 * outside the segment limit. All CPUs that support VMX ignore
5132 * limit checks for flat segments, i.e. segments with base==0,
5133 * limit==0xffffffff and of type expand-up data or code.
5134 */
5135 if (!(s.base == 0 && s.limit == 0xffffffff &&
5136 ((s.type & 8) || !(s.type & 4))))
5137 exn = exn || ((u64)off + len - 1 > s.limit);
5138 }
5139 if (exn) {
5140 kvm_queue_exception_e(vcpu,
5141 seg_reg == VCPU_SREG_SS ?
5142 SS_VECTOR : GP_VECTOR,
5143 0);
5144 return 1;
5145 }
5146
5147 return 0;
5148 }
5149
5150 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer,
5151 int *ret)
5152 {
5153 gva_t gva;
5154 struct x86_exception e;
5155 int r;
5156
5157 if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5158 vmcs_read32(VMX_INSTRUCTION_INFO), false,
5159 sizeof(*vmpointer), &gva)) {
5160 *ret = 1;
5161 return -EINVAL;
5162 }
5163
5164 r = kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e);
5165 if (r != X86EMUL_CONTINUE) {
5166 *ret = kvm_handle_memory_failure(vcpu, r, &e);
5167 return -EINVAL;
5168 }
5169
5170 return 0;
5171 }
5172
5173 /*
5174 * Allocate a shadow VMCS and associate it with the currently loaded
5175 * VMCS, unless such a shadow VMCS already exists. The newly allocated
5176 * VMCS is also VMCLEARed, so that it is ready for use.
5177 */
5178 static struct vmcs *alloc_shadow_vmcs(struct kvm_vcpu *vcpu)
5179 {
5180 struct vcpu_vmx *vmx = to_vmx(vcpu);
5181 struct loaded_vmcs *loaded_vmcs = vmx->loaded_vmcs;
5182
5183 /*
5184 * KVM allocates a shadow VMCS only when L1 executes VMXON and frees it
5185 * when L1 executes VMXOFF or the vCPU is forced out of nested
5186 * operation. VMXON faults if the CPU is already post-VMXON, so it
5187 * should be impossible to already have an allocated shadow VMCS. KVM
5188 * doesn't support virtualization of VMCS shadowing, so vmcs01 should
5189 * always be the loaded VMCS.
5190 */
5191 if (WARN_ON(loaded_vmcs != &vmx->vmcs01 || loaded_vmcs->shadow_vmcs))
5192 return loaded_vmcs->shadow_vmcs;
5193
5194 loaded_vmcs->shadow_vmcs = alloc_vmcs(true);
5195 if (loaded_vmcs->shadow_vmcs)
5196 vmcs_clear(loaded_vmcs->shadow_vmcs);
5197
5198 return loaded_vmcs->shadow_vmcs;
5199 }
5200
5201 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
5202 {
5203 struct vcpu_vmx *vmx = to_vmx(vcpu);
5204 int r;
5205
5206 r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
5207 if (r < 0)
5208 goto out_vmcs02;
5209
5210 vmx->nested.cached_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5211 if (!vmx->nested.cached_vmcs12)
5212 goto out_cached_vmcs12;
5213
5214 vmx->nested.shadow_vmcs12_cache.gpa = INVALID_GPA;
5215 vmx->nested.cached_shadow_vmcs12 = kzalloc(VMCS12_SIZE, GFP_KERNEL_ACCOUNT);
5216 if (!vmx->nested.cached_shadow_vmcs12)
5217 goto out_cached_shadow_vmcs12;
5218
5219 if (enable_shadow_vmcs && !alloc_shadow_vmcs(vcpu))
5220 goto out_shadow_vmcs;
5221
5222 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
5223 HRTIMER_MODE_ABS_PINNED);
5224 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
5225
5226 vmx->nested.vpid02 = allocate_vpid();
5227
5228 vmx->nested.vmcs02_initialized = false;
5229 vmx->nested.vmxon = true;
5230
5231 if (vmx_pt_mode_is_host_guest()) {
5232 vmx->pt_desc.guest.ctl = 0;
5233 pt_update_intercept_for_msr(vcpu);
5234 }
5235
5236 return 0;
5237
5238 out_shadow_vmcs:
5239 kfree(vmx->nested.cached_shadow_vmcs12);
5240
5241 out_cached_shadow_vmcs12:
5242 kfree(vmx->nested.cached_vmcs12);
5243
5244 out_cached_vmcs12:
5245 free_loaded_vmcs(&vmx->nested.vmcs02);
5246
5247 out_vmcs02:
5248 return -ENOMEM;
5249 }
5250
5251 /* Emulate the VMXON instruction. */
5252 static int handle_vmxon(struct kvm_vcpu *vcpu)
5253 {
5254 int ret;
5255 gpa_t vmptr;
5256 uint32_t revision;
5257 struct vcpu_vmx *vmx = to_vmx(vcpu);
5258 const u64 VMXON_NEEDED_FEATURES = FEAT_CTL_LOCKED
5259 | FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
5260
5261 /*
5262 * Manually check CR4.VMXE checks, KVM must force CR4.VMXE=1 to enter
5263 * the guest and so cannot rely on hardware to perform the check,
5264 * which has higher priority than VM-Exit (see Intel SDM's pseudocode
5265 * for VMXON).
5266 *
5267 * Rely on hardware for the other pre-VM-Exit checks, CR0.PE=1, !VM86
5268 * and !COMPATIBILITY modes. For an unrestricted guest, KVM doesn't
5269 * force any of the relevant guest state. For a restricted guest, KVM
5270 * does force CR0.PE=1, but only to also force VM86 in order to emulate
5271 * Real Mode, and so there's no need to check CR0.PE manually.
5272 */
5273 if (!kvm_is_cr4_bit_set(vcpu, X86_CR4_VMXE)) {
5274 kvm_queue_exception(vcpu, UD_VECTOR);
5275 return 1;
5276 }
5277
5278 /*
5279 * The CPL is checked for "not in VMX operation" and for "in VMX root",
5280 * and has higher priority than the VM-Fail due to being post-VMXON,
5281 * i.e. VMXON #GPs outside of VMX non-root if CPL!=0. In VMX non-root,
5282 * VMXON causes VM-Exit and KVM unconditionally forwards VMXON VM-Exits
5283 * from L2 to L1, i.e. there's no need to check for the vCPU being in
5284 * VMX non-root.
5285 *
5286 * Forwarding the VM-Exit unconditionally, i.e. without performing the
5287 * #UD checks (see above), is functionally ok because KVM doesn't allow
5288 * L1 to run L2 without CR4.VMXE=0, and because KVM never modifies L2's
5289 * CR0 or CR4, i.e. it's L2's responsibility to emulate #UDs that are
5290 * missed by hardware due to shadowing CR0 and/or CR4.
5291 */
5292 if (vmx_get_cpl(vcpu)) {
5293 kvm_inject_gp(vcpu, 0);
5294 return 1;
5295 }
5296
5297 if (vmx->nested.vmxon)
5298 return nested_vmx_fail(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5299
5300 /*
5301 * Invalid CR0/CR4 generates #GP. These checks are performed if and
5302 * only if the vCPU isn't already in VMX operation, i.e. effectively
5303 * have lower priority than the VM-Fail above.
5304 */
5305 if (!nested_host_cr0_valid(vcpu, kvm_read_cr0(vcpu)) ||
5306 !nested_host_cr4_valid(vcpu, kvm_read_cr4(vcpu))) {
5307 kvm_inject_gp(vcpu, 0);
5308 return 1;
5309 }
5310
5311 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5312 != VMXON_NEEDED_FEATURES) {
5313 kvm_inject_gp(vcpu, 0);
5314 return 1;
5315 }
5316
5317 if (nested_vmx_get_vmptr(vcpu, &vmptr, &ret))
5318 return ret;
5319
5320 /*
5321 * SDM 3: 24.11.5
5322 * The first 4 bytes of VMXON region contain the supported
5323 * VMCS revision identifier
5324 *
5325 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
5326 * which replaces physical address width with 32
5327 */
5328 if (!page_address_valid(vcpu, vmptr))
5329 return nested_vmx_failInvalid(vcpu);
5330
5331 if (kvm_read_guest(vcpu->kvm, vmptr, &revision, sizeof(revision)) ||
5332 revision != VMCS12_REVISION)
5333 return nested_vmx_failInvalid(vcpu);
5334
5335 vmx->nested.vmxon_ptr = vmptr;
5336 ret = enter_vmx_operation(vcpu);
5337 if (ret)
5338 return ret;
5339
5340 return nested_vmx_succeed(vcpu);
5341 }
5342
5343 static inline void nested_release_vmcs12(struct kvm_vcpu *vcpu)
5344 {
5345 struct vcpu_vmx *vmx = to_vmx(vcpu);
5346
5347 if (vmx->nested.current_vmptr == INVALID_GPA)
5348 return;
5349
5350 copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
5351
5352 if (enable_shadow_vmcs) {
5353 /* copy to memory all shadowed fields in case
5354 they were modified */
5355 copy_shadow_to_vmcs12(vmx);
5356 vmx_disable_shadow_vmcs(vmx);
5357 }
5358 vmx->nested.posted_intr_nv = -1;
5359
5360 /* Flush VMCS12 to guest memory */
5361 kvm_vcpu_write_guest_page(vcpu,
5362 vmx->nested.current_vmptr >> PAGE_SHIFT,
5363 vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
5364
5365 kvm_mmu_free_roots(vcpu->kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL);
5366
5367 vmx->nested.current_vmptr = INVALID_GPA;
5368 }
5369
5370 /* Emulate the VMXOFF instruction */
5371 static int handle_vmxoff(struct kvm_vcpu *vcpu)
5372 {
5373 if (!nested_vmx_check_permission(vcpu))
5374 return 1;
5375
5376 free_nested(vcpu);
5377
5378 if (kvm_apic_has_pending_init_or_sipi(vcpu))
5379 kvm_make_request(KVM_REQ_EVENT, vcpu);
5380
5381 return nested_vmx_succeed(vcpu);
5382 }
5383
5384 /* Emulate the VMCLEAR instruction */
5385 static int handle_vmclear(struct kvm_vcpu *vcpu)
5386 {
5387 struct vcpu_vmx *vmx = to_vmx(vcpu);
5388 u32 zero = 0;
5389 gpa_t vmptr;
5390 int r;
5391
5392 if (!nested_vmx_check_permission(vcpu))
5393 return 1;
5394
5395 if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5396 return r;
5397
5398 if (!page_address_valid(vcpu, vmptr))
5399 return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5400
5401 if (vmptr == vmx->nested.vmxon_ptr)
5402 return nested_vmx_fail(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
5403
5404 if (likely(!nested_evmcs_handle_vmclear(vcpu, vmptr))) {
5405 if (vmptr == vmx->nested.current_vmptr)
5406 nested_release_vmcs12(vcpu);
5407
5408 /*
5409 * Silently ignore memory errors on VMCLEAR, Intel's pseudocode
5410 * for VMCLEAR includes a "ensure that data for VMCS referenced
5411 * by the operand is in memory" clause that guards writes to
5412 * memory, i.e. doing nothing for I/O is architecturally valid.
5413 *
5414 * FIXME: Suppress failures if and only if no memslot is found,
5415 * i.e. exit to userspace if __copy_to_user() fails.
5416 */
5417 (void)kvm_vcpu_write_guest(vcpu,
5418 vmptr + offsetof(struct vmcs12,
5419 launch_state),
5420 &zero, sizeof(zero));
5421 }
5422
5423 return nested_vmx_succeed(vcpu);
5424 }
5425
5426 /* Emulate the VMLAUNCH instruction */
5427 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5428 {
5429 return nested_vmx_run(vcpu, true);
5430 }
5431
5432 /* Emulate the VMRESUME instruction */
5433 static int handle_vmresume(struct kvm_vcpu *vcpu)
5434 {
5435
5436 return nested_vmx_run(vcpu, false);
5437 }
5438
5439 static int handle_vmread(struct kvm_vcpu *vcpu)
5440 {
5441 struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5442 : get_vmcs12(vcpu);
5443 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5444 u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5445 struct vcpu_vmx *vmx = to_vmx(vcpu);
5446 struct x86_exception e;
5447 unsigned long field;
5448 u64 value;
5449 gva_t gva = 0;
5450 short offset;
5451 int len, r;
5452
5453 if (!nested_vmx_check_permission(vcpu))
5454 return 1;
5455
5456 /* Decode instruction info and find the field to read */
5457 field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5458
5459 if (!nested_vmx_is_evmptr12_valid(vmx)) {
5460 /*
5461 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5462 * any VMREAD sets the ALU flags for VMfailInvalid.
5463 */
5464 if (vmx->nested.current_vmptr == INVALID_GPA ||
5465 (is_guest_mode(vcpu) &&
5466 get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5467 return nested_vmx_failInvalid(vcpu);
5468
5469 offset = get_vmcs12_field_offset(field);
5470 if (offset < 0)
5471 return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5472
5473 if (!is_guest_mode(vcpu) && is_vmcs12_ext_field(field))
5474 copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5475
5476 /* Read the field, zero-extended to a u64 value */
5477 value = vmcs12_read_any(vmcs12, field, offset);
5478 } else {
5479 /*
5480 * Hyper-V TLFS (as of 6.0b) explicitly states, that while an
5481 * enlightened VMCS is active VMREAD/VMWRITE instructions are
5482 * unsupported. Unfortunately, certain versions of Windows 11
5483 * don't comply with this requirement which is not enforced in
5484 * genuine Hyper-V. Allow VMREAD from an enlightened VMCS as a
5485 * workaround, as misbehaving guests will panic on VM-Fail.
5486 * Note, enlightened VMCS is incompatible with shadow VMCS so
5487 * all VMREADs from L2 should go to L1.
5488 */
5489 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5490 return nested_vmx_failInvalid(vcpu);
5491
5492 offset = evmcs_field_offset(field, NULL);
5493 if (offset < 0)
5494 return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5495
5496 /* Read the field, zero-extended to a u64 value */
5497 value = evmcs_read_any(nested_vmx_evmcs(vmx), field, offset);
5498 }
5499
5500 /*
5501 * Now copy part of this value to register or memory, as requested.
5502 * Note that the number of bits actually copied is 32 or 64 depending
5503 * on the guest's mode (32 or 64 bit), not on the given field's length.
5504 */
5505 if (instr_info & BIT(10)) {
5506 kvm_register_write(vcpu, (((instr_info) >> 3) & 0xf), value);
5507 } else {
5508 len = is_64_bit_mode(vcpu) ? 8 : 4;
5509 if (get_vmx_mem_address(vcpu, exit_qualification,
5510 instr_info, true, len, &gva))
5511 return 1;
5512 /* _system ok, nested_vmx_check_permission has verified cpl=0 */
5513 r = kvm_write_guest_virt_system(vcpu, gva, &value, len, &e);
5514 if (r != X86EMUL_CONTINUE)
5515 return kvm_handle_memory_failure(vcpu, r, &e);
5516 }
5517
5518 return nested_vmx_succeed(vcpu);
5519 }
5520
5521 static bool is_shadow_field_rw(unsigned long field)
5522 {
5523 switch (field) {
5524 #define SHADOW_FIELD_RW(x, y) case x:
5525 #include "vmcs_shadow_fields.h"
5526 return true;
5527 default:
5528 break;
5529 }
5530 return false;
5531 }
5532
5533 static bool is_shadow_field_ro(unsigned long field)
5534 {
5535 switch (field) {
5536 #define SHADOW_FIELD_RO(x, y) case x:
5537 #include "vmcs_shadow_fields.h"
5538 return true;
5539 default:
5540 break;
5541 }
5542 return false;
5543 }
5544
5545 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5546 {
5547 struct vmcs12 *vmcs12 = is_guest_mode(vcpu) ? get_shadow_vmcs12(vcpu)
5548 : get_vmcs12(vcpu);
5549 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5550 u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5551 struct vcpu_vmx *vmx = to_vmx(vcpu);
5552 struct x86_exception e;
5553 unsigned long field;
5554 short offset;
5555 gva_t gva;
5556 int len, r;
5557
5558 /*
5559 * The value to write might be 32 or 64 bits, depending on L1's long
5560 * mode, and eventually we need to write that into a field of several
5561 * possible lengths. The code below first zero-extends the value to 64
5562 * bit (value), and then copies only the appropriate number of
5563 * bits into the vmcs12 field.
5564 */
5565 u64 value = 0;
5566
5567 if (!nested_vmx_check_permission(vcpu))
5568 return 1;
5569
5570 /*
5571 * In VMX non-root operation, when the VMCS-link pointer is INVALID_GPA,
5572 * any VMWRITE sets the ALU flags for VMfailInvalid.
5573 */
5574 if (vmx->nested.current_vmptr == INVALID_GPA ||
5575 (is_guest_mode(vcpu) &&
5576 get_vmcs12(vcpu)->vmcs_link_pointer == INVALID_GPA))
5577 return nested_vmx_failInvalid(vcpu);
5578
5579 if (instr_info & BIT(10))
5580 value = kvm_register_read(vcpu, (((instr_info) >> 3) & 0xf));
5581 else {
5582 len = is_64_bit_mode(vcpu) ? 8 : 4;
5583 if (get_vmx_mem_address(vcpu, exit_qualification,
5584 instr_info, false, len, &gva))
5585 return 1;
5586 r = kvm_read_guest_virt(vcpu, gva, &value, len, &e);
5587 if (r != X86EMUL_CONTINUE)
5588 return kvm_handle_memory_failure(vcpu, r, &e);
5589 }
5590
5591 field = kvm_register_read(vcpu, (((instr_info) >> 28) & 0xf));
5592
5593 offset = get_vmcs12_field_offset(field);
5594 if (offset < 0)
5595 return nested_vmx_fail(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5596
5597 /*
5598 * If the vCPU supports "VMWRITE to any supported field in the
5599 * VMCS," then the "read-only" fields are actually read/write.
5600 */
5601 if (vmcs_field_readonly(field) &&
5602 !nested_cpu_has_vmwrite_any_field(vcpu))
5603 return nested_vmx_fail(vcpu, VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5604
5605 /*
5606 * Ensure vmcs12 is up-to-date before any VMWRITE that dirties
5607 * vmcs12, else we may crush a field or consume a stale value.
5608 */
5609 if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field))
5610 copy_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
5611
5612 /*
5613 * Some Intel CPUs intentionally drop the reserved bits of the AR byte
5614 * fields on VMWRITE. Emulate this behavior to ensure consistent KVM
5615 * behavior regardless of the underlying hardware, e.g. if an AR_BYTE
5616 * field is intercepted for VMWRITE but not VMREAD (in L1), then VMREAD
5617 * from L1 will return a different value than VMREAD from L2 (L1 sees
5618 * the stripped down value, L2 sees the full value as stored by KVM).
5619 */
5620 if (field >= GUEST_ES_AR_BYTES && field <= GUEST_TR_AR_BYTES)
5621 value &= 0x1f0ff;
5622
5623 vmcs12_write_any(vmcs12, field, offset, value);
5624
5625 /*
5626 * Do not track vmcs12 dirty-state if in guest-mode as we actually
5627 * dirty shadow vmcs12 instead of vmcs12. Fields that can be updated
5628 * by L1 without a vmexit are always updated in the vmcs02, i.e. don't
5629 * "dirty" vmcs12, all others go down the prepare_vmcs02() slow path.
5630 */
5631 if (!is_guest_mode(vcpu) && !is_shadow_field_rw(field)) {
5632 /*
5633 * L1 can read these fields without exiting, ensure the
5634 * shadow VMCS is up-to-date.
5635 */
5636 if (enable_shadow_vmcs && is_shadow_field_ro(field)) {
5637 preempt_disable();
5638 vmcs_load(vmx->vmcs01.shadow_vmcs);
5639
5640 __vmcs_writel(field, value);
5641
5642 vmcs_clear(vmx->vmcs01.shadow_vmcs);
5643 vmcs_load(vmx->loaded_vmcs->vmcs);
5644 preempt_enable();
5645 }
5646 vmx->nested.dirty_vmcs12 = true;
5647 }
5648
5649 return nested_vmx_succeed(vcpu);
5650 }
5651
5652 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
5653 {
5654 vmx->nested.current_vmptr = vmptr;
5655 if (enable_shadow_vmcs) {
5656 secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_SHADOW_VMCS);
5657 vmcs_write64(VMCS_LINK_POINTER,
5658 __pa(vmx->vmcs01.shadow_vmcs));
5659 vmx->nested.need_vmcs12_to_shadow_sync = true;
5660 }
5661 vmx->nested.dirty_vmcs12 = true;
5662 vmx->nested.force_msr_bitmap_recalc = true;
5663 }
5664
5665 /* Emulate the VMPTRLD instruction */
5666 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5667 {
5668 struct vcpu_vmx *vmx = to_vmx(vcpu);
5669 gpa_t vmptr;
5670 int r;
5671
5672 if (!nested_vmx_check_permission(vcpu))
5673 return 1;
5674
5675 if (nested_vmx_get_vmptr(vcpu, &vmptr, &r))
5676 return r;
5677
5678 if (!page_address_valid(vcpu, vmptr))
5679 return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5680
5681 if (vmptr == vmx->nested.vmxon_ptr)
5682 return nested_vmx_fail(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
5683
5684 /* Forbid normal VMPTRLD if Enlightened version was used */
5685 if (nested_vmx_is_evmptr12_valid(vmx))
5686 return 1;
5687
5688 if (vmx->nested.current_vmptr != vmptr) {
5689 struct gfn_to_hva_cache *ghc = &vmx->nested.vmcs12_cache;
5690 struct vmcs_hdr hdr;
5691
5692 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, vmptr, VMCS12_SIZE)) {
5693 /*
5694 * Reads from an unbacked page return all 1s,
5695 * which means that the 32 bits located at the
5696 * given physical address won't match the required
5697 * VMCS12_REVISION identifier.
5698 */
5699 return nested_vmx_fail(vcpu,
5700 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5701 }
5702
5703 if (kvm_read_guest_offset_cached(vcpu->kvm, ghc, &hdr,
5704 offsetof(struct vmcs12, hdr),
5705 sizeof(hdr))) {
5706 return nested_vmx_fail(vcpu,
5707 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5708 }
5709
5710 if (hdr.revision_id != VMCS12_REVISION ||
5711 (hdr.shadow_vmcs &&
5712 !nested_cpu_has_vmx_shadow_vmcs(vcpu))) {
5713 return nested_vmx_fail(vcpu,
5714 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5715 }
5716
5717 nested_release_vmcs12(vcpu);
5718
5719 /*
5720 * Load VMCS12 from guest memory since it is not already
5721 * cached.
5722 */
5723 if (kvm_read_guest_cached(vcpu->kvm, ghc, vmx->nested.cached_vmcs12,
5724 VMCS12_SIZE)) {
5725 return nested_vmx_fail(vcpu,
5726 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5727 }
5728
5729 set_current_vmptr(vmx, vmptr);
5730 }
5731
5732 return nested_vmx_succeed(vcpu);
5733 }
5734
5735 /* Emulate the VMPTRST instruction */
5736 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5737 {
5738 unsigned long exit_qual = vmx_get_exit_qual(vcpu);
5739 u32 instr_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5740 gpa_t current_vmptr = to_vmx(vcpu)->nested.current_vmptr;
5741 struct x86_exception e;
5742 gva_t gva;
5743 int r;
5744
5745 if (!nested_vmx_check_permission(vcpu))
5746 return 1;
5747
5748 if (unlikely(nested_vmx_is_evmptr12_valid(to_vmx(vcpu))))
5749 return 1;
5750
5751 if (get_vmx_mem_address(vcpu, exit_qual, instr_info,
5752 true, sizeof(gpa_t), &gva))
5753 return 1;
5754 /* *_system ok, nested_vmx_check_permission has verified cpl=0 */
5755 r = kvm_write_guest_virt_system(vcpu, gva, (void *)¤t_vmptr,
5756 sizeof(gpa_t), &e);
5757 if (r != X86EMUL_CONTINUE)
5758 return kvm_handle_memory_failure(vcpu, r, &e);
5759
5760 return nested_vmx_succeed(vcpu);
5761 }
5762
5763 /* Emulate the INVEPT instruction */
5764 static int handle_invept(struct kvm_vcpu *vcpu)
5765 {
5766 struct vcpu_vmx *vmx = to_vmx(vcpu);
5767 u32 vmx_instruction_info, types;
5768 unsigned long type, roots_to_free;
5769 struct kvm_mmu *mmu;
5770 gva_t gva;
5771 struct x86_exception e;
5772 struct {
5773 u64 eptp, gpa;
5774 } operand;
5775 int i, r, gpr_index;
5776
5777 if (!(vmx->nested.msrs.secondary_ctls_high &
5778 SECONDARY_EXEC_ENABLE_EPT) ||
5779 !(vmx->nested.msrs.ept_caps & VMX_EPT_INVEPT_BIT)) {
5780 kvm_queue_exception(vcpu, UD_VECTOR);
5781 return 1;
5782 }
5783
5784 if (!nested_vmx_check_permission(vcpu))
5785 return 1;
5786
5787 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5788 gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5789 type = kvm_register_read(vcpu, gpr_index);
5790
5791 types = (vmx->nested.msrs.ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
5792
5793 if (type >= 32 || !(types & (1 << type)))
5794 return nested_vmx_fail(vcpu, VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5795
5796 /* According to the Intel VMX instruction reference, the memory
5797 * operand is read even if it isn't needed (e.g., for type==global)
5798 */
5799 if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5800 vmx_instruction_info, false, sizeof(operand), &gva))
5801 return 1;
5802 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5803 if (r != X86EMUL_CONTINUE)
5804 return kvm_handle_memory_failure(vcpu, r, &e);
5805
5806 /*
5807 * Nested EPT roots are always held through guest_mmu,
5808 * not root_mmu.
5809 */
5810 mmu = &vcpu->arch.guest_mmu;
5811
5812 switch (type) {
5813 case VMX_EPT_EXTENT_CONTEXT:
5814 if (!nested_vmx_check_eptp(vcpu, operand.eptp))
5815 return nested_vmx_fail(vcpu,
5816 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5817
5818 roots_to_free = 0;
5819 if (nested_ept_root_matches(mmu->root.hpa, mmu->root.pgd,
5820 operand.eptp))
5821 roots_to_free |= KVM_MMU_ROOT_CURRENT;
5822
5823 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) {
5824 if (nested_ept_root_matches(mmu->prev_roots[i].hpa,
5825 mmu->prev_roots[i].pgd,
5826 operand.eptp))
5827 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5828 }
5829 break;
5830 case VMX_EPT_EXTENT_GLOBAL:
5831 roots_to_free = KVM_MMU_ROOTS_ALL;
5832 break;
5833 default:
5834 BUG();
5835 break;
5836 }
5837
5838 if (roots_to_free)
5839 kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
5840
5841 return nested_vmx_succeed(vcpu);
5842 }
5843
5844 static int handle_invvpid(struct kvm_vcpu *vcpu)
5845 {
5846 struct vcpu_vmx *vmx = to_vmx(vcpu);
5847 u32 vmx_instruction_info;
5848 unsigned long type, types;
5849 gva_t gva;
5850 struct x86_exception e;
5851 struct {
5852 u64 vpid;
5853 u64 gla;
5854 } operand;
5855 u16 vpid02;
5856 int r, gpr_index;
5857
5858 if (!(vmx->nested.msrs.secondary_ctls_high &
5859 SECONDARY_EXEC_ENABLE_VPID) ||
5860 !(vmx->nested.msrs.vpid_caps & VMX_VPID_INVVPID_BIT)) {
5861 kvm_queue_exception(vcpu, UD_VECTOR);
5862 return 1;
5863 }
5864
5865 if (!nested_vmx_check_permission(vcpu))
5866 return 1;
5867
5868 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5869 gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5870 type = kvm_register_read(vcpu, gpr_index);
5871
5872 types = (vmx->nested.msrs.vpid_caps &
5873 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
5874
5875 if (type >= 32 || !(types & (1 << type)))
5876 return nested_vmx_fail(vcpu,
5877 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5878
5879 /* according to the intel vmx instruction reference, the memory
5880 * operand is read even if it isn't needed (e.g., for type==global)
5881 */
5882 if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5883 vmx_instruction_info, false, sizeof(operand), &gva))
5884 return 1;
5885 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5886 if (r != X86EMUL_CONTINUE)
5887 return kvm_handle_memory_failure(vcpu, r, &e);
5888
5889 if (operand.vpid >> 16)
5890 return nested_vmx_fail(vcpu,
5891 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5892
5893 vpid02 = nested_get_vpid02(vcpu);
5894 switch (type) {
5895 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
5896 /*
5897 * LAM doesn't apply to addresses that are inputs to TLB
5898 * invalidation.
5899 */
5900 if (!operand.vpid ||
5901 is_noncanonical_address(operand.gla, vcpu))
5902 return nested_vmx_fail(vcpu,
5903 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5904 vpid_sync_vcpu_addr(vpid02, operand.gla);
5905 break;
5906 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
5907 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
5908 if (!operand.vpid)
5909 return nested_vmx_fail(vcpu,
5910 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
5911 vpid_sync_context(vpid02);
5912 break;
5913 case VMX_VPID_EXTENT_ALL_CONTEXT:
5914 vpid_sync_context(vpid02);
5915 break;
5916 default:
5917 WARN_ON_ONCE(1);
5918 return kvm_skip_emulated_instruction(vcpu);
5919 }
5920
5921 /*
5922 * Sync the shadow page tables if EPT is disabled, L1 is invalidating
5923 * linear mappings for L2 (tagged with L2's VPID). Free all guest
5924 * roots as VPIDs are not tracked in the MMU role.
5925 *
5926 * Note, this operates on root_mmu, not guest_mmu, as L1 and L2 share
5927 * an MMU when EPT is disabled.
5928 *
5929 * TODO: sync only the affected SPTEs for INVDIVIDUAL_ADDR.
5930 */
5931 if (!enable_ept)
5932 kvm_mmu_free_guest_mode_roots(vcpu->kvm, &vcpu->arch.root_mmu);
5933
5934 return nested_vmx_succeed(vcpu);
5935 }
5936
5937 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
5938 struct vmcs12 *vmcs12)
5939 {
5940 u32 index = kvm_rcx_read(vcpu);
5941 u64 new_eptp;
5942
5943 if (WARN_ON_ONCE(!nested_cpu_has_ept(vmcs12)))
5944 return 1;
5945 if (index >= VMFUNC_EPTP_ENTRIES)
5946 return 1;
5947
5948 if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
5949 &new_eptp, index * 8, 8))
5950 return 1;
5951
5952 /*
5953 * If the (L2) guest does a vmfunc to the currently
5954 * active ept pointer, we don't have to do anything else
5955 */
5956 if (vmcs12->ept_pointer != new_eptp) {
5957 if (!nested_vmx_check_eptp(vcpu, new_eptp))
5958 return 1;
5959
5960 vmcs12->ept_pointer = new_eptp;
5961 nested_ept_new_eptp(vcpu);
5962
5963 if (!nested_cpu_has_vpid(vmcs12))
5964 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
5965 }
5966
5967 return 0;
5968 }
5969
5970 static int handle_vmfunc(struct kvm_vcpu *vcpu)
5971 {
5972 struct vcpu_vmx *vmx = to_vmx(vcpu);
5973 struct vmcs12 *vmcs12;
5974 u32 function = kvm_rax_read(vcpu);
5975
5976 /*
5977 * VMFUNC should never execute cleanly while L1 is active; KVM supports
5978 * VMFUNC for nested VMs, but not for L1.
5979 */
5980 if (WARN_ON_ONCE(!is_guest_mode(vcpu))) {
5981 kvm_queue_exception(vcpu, UD_VECTOR);
5982 return 1;
5983 }
5984
5985 vmcs12 = get_vmcs12(vcpu);
5986
5987 /*
5988 * #UD on out-of-bounds function has priority over VM-Exit, and VMFUNC
5989 * is enabled in vmcs02 if and only if it's enabled in vmcs12.
5990 */
5991 if (WARN_ON_ONCE((function > 63) || !nested_cpu_has_vmfunc(vmcs12))) {
5992 kvm_queue_exception(vcpu, UD_VECTOR);
5993 return 1;
5994 }
5995
5996 if (!(vmcs12->vm_function_control & BIT_ULL(function)))
5997 goto fail;
5998
5999 switch (function) {
6000 case 0:
6001 if (nested_vmx_eptp_switching(vcpu, vmcs12))
6002 goto fail;
6003 break;
6004 default:
6005 goto fail;
6006 }
6007 return kvm_skip_emulated_instruction(vcpu);
6008
6009 fail:
6010 /*
6011 * This is effectively a reflected VM-Exit, as opposed to a synthesized
6012 * nested VM-Exit. Pass the original exit reason, i.e. don't hardcode
6013 * EXIT_REASON_VMFUNC as the exit reason.
6014 */
6015 nested_vmx_vmexit(vcpu, vmx->exit_reason.full,
6016 vmx_get_intr_info(vcpu),
6017 vmx_get_exit_qual(vcpu));
6018 return 1;
6019 }
6020
6021 /*
6022 * Return true if an IO instruction with the specified port and size should cause
6023 * a VM-exit into L1.
6024 */
6025 bool nested_vmx_check_io_bitmaps(struct kvm_vcpu *vcpu, unsigned int port,
6026 int size)
6027 {
6028 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6029 gpa_t bitmap, last_bitmap;
6030 u8 b;
6031
6032 last_bitmap = INVALID_GPA;
6033 b = -1;
6034
6035 while (size > 0) {
6036 if (port < 0x8000)
6037 bitmap = vmcs12->io_bitmap_a;
6038 else if (port < 0x10000)
6039 bitmap = vmcs12->io_bitmap_b;
6040 else
6041 return true;
6042 bitmap += (port & 0x7fff) / 8;
6043
6044 if (last_bitmap != bitmap)
6045 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
6046 return true;
6047 if (b & (1 << (port & 7)))
6048 return true;
6049
6050 port++;
6051 size--;
6052 last_bitmap = bitmap;
6053 }
6054
6055 return false;
6056 }
6057
6058 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6059 struct vmcs12 *vmcs12)
6060 {
6061 unsigned long exit_qualification;
6062 unsigned short port;
6063 int size;
6064
6065 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6066 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
6067
6068 exit_qualification = vmx_get_exit_qual(vcpu);
6069
6070 port = exit_qualification >> 16;
6071 size = (exit_qualification & 7) + 1;
6072
6073 return nested_vmx_check_io_bitmaps(vcpu, port, size);
6074 }
6075
6076 /*
6077 * Return 1 if we should exit from L2 to L1 to handle an MSR access,
6078 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6079 * disinterest in the current event (read or write a specific MSR) by using an
6080 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6081 */
6082 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6083 struct vmcs12 *vmcs12,
6084 union vmx_exit_reason exit_reason)
6085 {
6086 u32 msr_index = kvm_rcx_read(vcpu);
6087 gpa_t bitmap;
6088
6089 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6090 return true;
6091
6092 /*
6093 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6094 * for the four combinations of read/write and low/high MSR numbers.
6095 * First we need to figure out which of the four to use:
6096 */
6097 bitmap = vmcs12->msr_bitmap;
6098 if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
6099 bitmap += 2048;
6100 if (msr_index >= 0xc0000000) {
6101 msr_index -= 0xc0000000;
6102 bitmap += 1024;
6103 }
6104
6105 /* Then read the msr_index'th bit from this bitmap: */
6106 if (msr_index < 1024*8) {
6107 unsigned char b;
6108 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
6109 return true;
6110 return 1 & (b >> (msr_index & 7));
6111 } else
6112 return true; /* let L1 handle the wrong parameter */
6113 }
6114
6115 /*
6116 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6117 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6118 * intercept (via guest_host_mask etc.) the current event.
6119 */
6120 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6121 struct vmcs12 *vmcs12)
6122 {
6123 unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
6124 int cr = exit_qualification & 15;
6125 int reg;
6126 unsigned long val;
6127
6128 switch ((exit_qualification >> 4) & 3) {
6129 case 0: /* mov to cr */
6130 reg = (exit_qualification >> 8) & 15;
6131 val = kvm_register_read(vcpu, reg);
6132 switch (cr) {
6133 case 0:
6134 if (vmcs12->cr0_guest_host_mask &
6135 (val ^ vmcs12->cr0_read_shadow))
6136 return true;
6137 break;
6138 case 3:
6139 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6140 return true;
6141 break;
6142 case 4:
6143 if (vmcs12->cr4_guest_host_mask &
6144 (vmcs12->cr4_read_shadow ^ val))
6145 return true;
6146 break;
6147 case 8:
6148 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6149 return true;
6150 break;
6151 }
6152 break;
6153 case 2: /* clts */
6154 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6155 (vmcs12->cr0_read_shadow & X86_CR0_TS))
6156 return true;
6157 break;
6158 case 1: /* mov from cr */
6159 switch (cr) {
6160 case 3:
6161 if (vmcs12->cpu_based_vm_exec_control &
6162 CPU_BASED_CR3_STORE_EXITING)
6163 return true;
6164 break;
6165 case 8:
6166 if (vmcs12->cpu_based_vm_exec_control &
6167 CPU_BASED_CR8_STORE_EXITING)
6168 return true;
6169 break;
6170 }
6171 break;
6172 case 3: /* lmsw */
6173 /*
6174 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6175 * cr0. Other attempted changes are ignored, with no exit.
6176 */
6177 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6178 if (vmcs12->cr0_guest_host_mask & 0xe &
6179 (val ^ vmcs12->cr0_read_shadow))
6180 return true;
6181 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6182 !(vmcs12->cr0_read_shadow & 0x1) &&
6183 (val & 0x1))
6184 return true;
6185 break;
6186 }
6187 return false;
6188 }
6189
6190 static bool nested_vmx_exit_handled_encls(struct kvm_vcpu *vcpu,
6191 struct vmcs12 *vmcs12)
6192 {
6193 u32 encls_leaf;
6194
6195 if (!guest_cpuid_has(vcpu, X86_FEATURE_SGX) ||
6196 !nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENCLS_EXITING))
6197 return false;
6198
6199 encls_leaf = kvm_rax_read(vcpu);
6200 if (encls_leaf > 62)
6201 encls_leaf = 63;
6202 return vmcs12->encls_exiting_bitmap & BIT_ULL(encls_leaf);
6203 }
6204
6205 static bool nested_vmx_exit_handled_vmcs_access(struct kvm_vcpu *vcpu,
6206 struct vmcs12 *vmcs12, gpa_t bitmap)
6207 {
6208 u32 vmx_instruction_info;
6209 unsigned long field;
6210 u8 b;
6211
6212 if (!nested_cpu_has_shadow_vmcs(vmcs12))
6213 return true;
6214
6215 /* Decode instruction info and find the field to access */
6216 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6217 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6218
6219 /* Out-of-range fields always cause a VM exit from L2 to L1 */
6220 if (field >> 15)
6221 return true;
6222
6223 if (kvm_vcpu_read_guest(vcpu, bitmap + field/8, &b, 1))
6224 return true;
6225
6226 return 1 & (b >> (field & 7));
6227 }
6228
6229 static bool nested_vmx_exit_handled_mtf(struct vmcs12 *vmcs12)
6230 {
6231 u32 entry_intr_info = vmcs12->vm_entry_intr_info_field;
6232
6233 if (nested_cpu_has_mtf(vmcs12))
6234 return true;
6235
6236 /*
6237 * An MTF VM-exit may be injected into the guest by setting the
6238 * interruption-type to 7 (other event) and the vector field to 0. Such
6239 * is the case regardless of the 'monitor trap flag' VM-execution
6240 * control.
6241 */
6242 return entry_intr_info == (INTR_INFO_VALID_MASK
6243 | INTR_TYPE_OTHER_EVENT);
6244 }
6245
6246 /*
6247 * Return true if L0 wants to handle an exit from L2 regardless of whether or not
6248 * L1 wants the exit. Only call this when in is_guest_mode (L2).
6249 */
6250 static bool nested_vmx_l0_wants_exit(struct kvm_vcpu *vcpu,
6251 union vmx_exit_reason exit_reason)
6252 {
6253 u32 intr_info;
6254
6255 switch ((u16)exit_reason.basic) {
6256 case EXIT_REASON_EXCEPTION_NMI:
6257 intr_info = vmx_get_intr_info(vcpu);
6258 if (is_nmi(intr_info))
6259 return true;
6260 else if (is_page_fault(intr_info))
6261 return vcpu->arch.apf.host_apf_flags ||
6262 vmx_need_pf_intercept(vcpu);
6263 else if (is_debug(intr_info) &&
6264 vcpu->guest_debug &
6265 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6266 return true;
6267 else if (is_breakpoint(intr_info) &&
6268 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6269 return true;
6270 else if (is_alignment_check(intr_info) &&
6271 !vmx_guest_inject_ac(vcpu))
6272 return true;
6273 else if (is_ve_fault(intr_info))
6274 return true;
6275 return false;
6276 case EXIT_REASON_EXTERNAL_INTERRUPT:
6277 return true;
6278 case EXIT_REASON_MCE_DURING_VMENTRY:
6279 return true;
6280 case EXIT_REASON_EPT_VIOLATION:
6281 /*
6282 * L0 always deals with the EPT violation. If nested EPT is
6283 * used, and the nested mmu code discovers that the address is
6284 * missing in the guest EPT table (EPT12), the EPT violation
6285 * will be injected with nested_ept_inject_page_fault()
6286 */
6287 return true;
6288 case EXIT_REASON_EPT_MISCONFIG:
6289 /*
6290 * L2 never uses directly L1's EPT, but rather L0's own EPT
6291 * table (shadow on EPT) or a merged EPT table that L0 built
6292 * (EPT on EPT). So any problems with the structure of the
6293 * table is L0's fault.
6294 */
6295 return true;
6296 case EXIT_REASON_PREEMPTION_TIMER:
6297 return true;
6298 case EXIT_REASON_PML_FULL:
6299 /*
6300 * PML is emulated for an L1 VMM and should never be enabled in
6301 * vmcs02, always "handle" PML_FULL by exiting to userspace.
6302 */
6303 return true;
6304 case EXIT_REASON_VMFUNC:
6305 /* VM functions are emulated through L2->L0 vmexits. */
6306 return true;
6307 case EXIT_REASON_BUS_LOCK:
6308 /*
6309 * At present, bus lock VM exit is never exposed to L1.
6310 * Handle L2's bus locks in L0 directly.
6311 */
6312 return true;
6313 #ifdef CONFIG_KVM_HYPERV
6314 case EXIT_REASON_VMCALL:
6315 /* Hyper-V L2 TLB flush hypercall is handled by L0 */
6316 return guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
6317 nested_evmcs_l2_tlb_flush_enabled(vcpu) &&
6318 kvm_hv_is_tlb_flush_hcall(vcpu);
6319 #endif
6320 default:
6321 break;
6322 }
6323 return false;
6324 }
6325
6326 /*
6327 * Return 1 if L1 wants to intercept an exit from L2. Only call this when in
6328 * is_guest_mode (L2).
6329 */
6330 static bool nested_vmx_l1_wants_exit(struct kvm_vcpu *vcpu,
6331 union vmx_exit_reason exit_reason)
6332 {
6333 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6334 u32 intr_info;
6335
6336 switch ((u16)exit_reason.basic) {
6337 case EXIT_REASON_EXCEPTION_NMI:
6338 intr_info = vmx_get_intr_info(vcpu);
6339 if (is_nmi(intr_info))
6340 return true;
6341 else if (is_page_fault(intr_info))
6342 return true;
6343 return vmcs12->exception_bitmap &
6344 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
6345 case EXIT_REASON_EXTERNAL_INTERRUPT:
6346 return nested_exit_on_intr(vcpu);
6347 case EXIT_REASON_TRIPLE_FAULT:
6348 return true;
6349 case EXIT_REASON_INTERRUPT_WINDOW:
6350 return nested_cpu_has(vmcs12, CPU_BASED_INTR_WINDOW_EXITING);
6351 case EXIT_REASON_NMI_WINDOW:
6352 return nested_cpu_has(vmcs12, CPU_BASED_NMI_WINDOW_EXITING);
6353 case EXIT_REASON_TASK_SWITCH:
6354 return true;
6355 case EXIT_REASON_CPUID:
6356 return true;
6357 case EXIT_REASON_HLT:
6358 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6359 case EXIT_REASON_INVD:
6360 return true;
6361 case EXIT_REASON_INVLPG:
6362 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6363 case EXIT_REASON_RDPMC:
6364 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6365 case EXIT_REASON_RDRAND:
6366 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
6367 case EXIT_REASON_RDSEED:
6368 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
6369 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
6370 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6371 case EXIT_REASON_VMREAD:
6372 return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6373 vmcs12->vmread_bitmap);
6374 case EXIT_REASON_VMWRITE:
6375 return nested_vmx_exit_handled_vmcs_access(vcpu, vmcs12,
6376 vmcs12->vmwrite_bitmap);
6377 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6378 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6379 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMRESUME:
6380 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6381 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
6382 /*
6383 * VMX instructions trap unconditionally. This allows L1 to
6384 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6385 */
6386 return true;
6387 case EXIT_REASON_CR_ACCESS:
6388 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6389 case EXIT_REASON_DR_ACCESS:
6390 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6391 case EXIT_REASON_IO_INSTRUCTION:
6392 return nested_vmx_exit_handled_io(vcpu, vmcs12);
6393 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
6394 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
6395 case EXIT_REASON_MSR_READ:
6396 case EXIT_REASON_MSR_WRITE:
6397 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6398 case EXIT_REASON_INVALID_STATE:
6399 return true;
6400 case EXIT_REASON_MWAIT_INSTRUCTION:
6401 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6402 case EXIT_REASON_MONITOR_TRAP_FLAG:
6403 return nested_vmx_exit_handled_mtf(vmcs12);
6404 case EXIT_REASON_MONITOR_INSTRUCTION:
6405 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6406 case EXIT_REASON_PAUSE_INSTRUCTION:
6407 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6408 nested_cpu_has2(vmcs12,
6409 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6410 case EXIT_REASON_MCE_DURING_VMENTRY:
6411 return true;
6412 case EXIT_REASON_TPR_BELOW_THRESHOLD:
6413 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
6414 case EXIT_REASON_APIC_ACCESS:
6415 case EXIT_REASON_APIC_WRITE:
6416 case EXIT_REASON_EOI_INDUCED:
6417 /*
6418 * The controls for "virtualize APIC accesses," "APIC-
6419 * register virtualization," and "virtual-interrupt
6420 * delivery" only come from vmcs12.
6421 */
6422 return true;
6423 case EXIT_REASON_INVPCID:
6424 return
6425 nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
6426 nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6427 case EXIT_REASON_WBINVD:
6428 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6429 case EXIT_REASON_XSETBV:
6430 return true;
6431 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
6432 /*
6433 * This should never happen, since it is not possible to
6434 * set XSS to a non-zero value---neither in L1 nor in L2.
6435 * If if it were, XSS would have to be checked against
6436 * the XSS exit bitmap in vmcs12.
6437 */
6438 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_XSAVES);
6439 case EXIT_REASON_UMWAIT:
6440 case EXIT_REASON_TPAUSE:
6441 return nested_cpu_has2(vmcs12,
6442 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE);
6443 case EXIT_REASON_ENCLS:
6444 return nested_vmx_exit_handled_encls(vcpu, vmcs12);
6445 case EXIT_REASON_NOTIFY:
6446 /* Notify VM exit is not exposed to L1 */
6447 return false;
6448 default:
6449 return true;
6450 }
6451 }
6452
6453 /*
6454 * Conditionally reflect a VM-Exit into L1. Returns %true if the VM-Exit was
6455 * reflected into L1.
6456 */
6457 bool nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu)
6458 {
6459 struct vcpu_vmx *vmx = to_vmx(vcpu);
6460 union vmx_exit_reason exit_reason = vmx->exit_reason;
6461 unsigned long exit_qual;
6462 u32 exit_intr_info;
6463
6464 WARN_ON_ONCE(vmx->nested.nested_run_pending);
6465
6466 /*
6467 * Late nested VM-Fail shares the same flow as nested VM-Exit since KVM
6468 * has already loaded L2's state.
6469 */
6470 if (unlikely(vmx->fail)) {
6471 trace_kvm_nested_vmenter_failed(
6472 "hardware VM-instruction error: ",
6473 vmcs_read32(VM_INSTRUCTION_ERROR));
6474 exit_intr_info = 0;
6475 exit_qual = 0;
6476 goto reflect_vmexit;
6477 }
6478
6479 trace_kvm_nested_vmexit(vcpu, KVM_ISA_VMX);
6480
6481 /* If L0 (KVM) wants the exit, it trumps L1's desires. */
6482 if (nested_vmx_l0_wants_exit(vcpu, exit_reason))
6483 return false;
6484
6485 /* If L1 doesn't want the exit, handle it in L0. */
6486 if (!nested_vmx_l1_wants_exit(vcpu, exit_reason))
6487 return false;
6488
6489 /*
6490 * vmcs.VM_EXIT_INTR_INFO is only valid for EXCEPTION_NMI exits. For
6491 * EXTERNAL_INTERRUPT, the value for vmcs12->vm_exit_intr_info would
6492 * need to be synthesized by querying the in-kernel LAPIC, but external
6493 * interrupts are never reflected to L1 so it's a non-issue.
6494 */
6495 exit_intr_info = vmx_get_intr_info(vcpu);
6496 if (is_exception_with_error_code(exit_intr_info)) {
6497 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6498
6499 vmcs12->vm_exit_intr_error_code =
6500 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6501 }
6502 exit_qual = vmx_get_exit_qual(vcpu);
6503
6504 reflect_vmexit:
6505 nested_vmx_vmexit(vcpu, exit_reason.full, exit_intr_info, exit_qual);
6506 return true;
6507 }
6508
6509 static int vmx_get_nested_state(struct kvm_vcpu *vcpu,
6510 struct kvm_nested_state __user *user_kvm_nested_state,
6511 u32 user_data_size)
6512 {
6513 struct vcpu_vmx *vmx;
6514 struct vmcs12 *vmcs12;
6515 struct kvm_nested_state kvm_state = {
6516 .flags = 0,
6517 .format = KVM_STATE_NESTED_FORMAT_VMX,
6518 .size = sizeof(kvm_state),
6519 .hdr.vmx.flags = 0,
6520 .hdr.vmx.vmxon_pa = INVALID_GPA,
6521 .hdr.vmx.vmcs12_pa = INVALID_GPA,
6522 .hdr.vmx.preemption_timer_deadline = 0,
6523 };
6524 struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6525 &user_kvm_nested_state->data.vmx[0];
6526
6527 if (!vcpu)
6528 return kvm_state.size + sizeof(*user_vmx_nested_state);
6529
6530 vmx = to_vmx(vcpu);
6531 vmcs12 = get_vmcs12(vcpu);
6532
6533 if (guest_can_use(vcpu, X86_FEATURE_VMX) &&
6534 (vmx->nested.vmxon || vmx->nested.smm.vmxon)) {
6535 kvm_state.hdr.vmx.vmxon_pa = vmx->nested.vmxon_ptr;
6536 kvm_state.hdr.vmx.vmcs12_pa = vmx->nested.current_vmptr;
6537
6538 if (vmx_has_valid_vmcs12(vcpu)) {
6539 kvm_state.size += sizeof(user_vmx_nested_state->vmcs12);
6540
6541 /* 'hv_evmcs_vmptr' can also be EVMPTR_MAP_PENDING here */
6542 if (nested_vmx_is_evmptr12_set(vmx))
6543 kvm_state.flags |= KVM_STATE_NESTED_EVMCS;
6544
6545 if (is_guest_mode(vcpu) &&
6546 nested_cpu_has_shadow_vmcs(vmcs12) &&
6547 vmcs12->vmcs_link_pointer != INVALID_GPA)
6548 kvm_state.size += sizeof(user_vmx_nested_state->shadow_vmcs12);
6549 }
6550
6551 if (vmx->nested.smm.vmxon)
6552 kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_VMXON;
6553
6554 if (vmx->nested.smm.guest_mode)
6555 kvm_state.hdr.vmx.smm.flags |= KVM_STATE_NESTED_SMM_GUEST_MODE;
6556
6557 if (is_guest_mode(vcpu)) {
6558 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
6559
6560 if (vmx->nested.nested_run_pending)
6561 kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
6562
6563 if (vmx->nested.mtf_pending)
6564 kvm_state.flags |= KVM_STATE_NESTED_MTF_PENDING;
6565
6566 if (nested_cpu_has_preemption_timer(vmcs12) &&
6567 vmx->nested.has_preemption_timer_deadline) {
6568 kvm_state.hdr.vmx.flags |=
6569 KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE;
6570 kvm_state.hdr.vmx.preemption_timer_deadline =
6571 vmx->nested.preemption_timer_deadline;
6572 }
6573 }
6574 }
6575
6576 if (user_data_size < kvm_state.size)
6577 goto out;
6578
6579 if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
6580 return -EFAULT;
6581
6582 if (!vmx_has_valid_vmcs12(vcpu))
6583 goto out;
6584
6585 /*
6586 * When running L2, the authoritative vmcs12 state is in the
6587 * vmcs02. When running L1, the authoritative vmcs12 state is
6588 * in the shadow or enlightened vmcs linked to vmcs01, unless
6589 * need_vmcs12_to_shadow_sync is set, in which case, the authoritative
6590 * vmcs12 state is in the vmcs12 already.
6591 */
6592 if (is_guest_mode(vcpu)) {
6593 sync_vmcs02_to_vmcs12(vcpu, vmcs12);
6594 sync_vmcs02_to_vmcs12_rare(vcpu, vmcs12);
6595 } else {
6596 copy_vmcs02_to_vmcs12_rare(vcpu, get_vmcs12(vcpu));
6597 if (!vmx->nested.need_vmcs12_to_shadow_sync) {
6598 if (nested_vmx_is_evmptr12_valid(vmx))
6599 /*
6600 * L1 hypervisor is not obliged to keep eVMCS
6601 * clean fields data always up-to-date while
6602 * not in guest mode, 'hv_clean_fields' is only
6603 * supposed to be actual upon vmentry so we need
6604 * to ignore it here and do full copy.
6605 */
6606 copy_enlightened_to_vmcs12(vmx, 0);
6607 else if (enable_shadow_vmcs)
6608 copy_shadow_to_vmcs12(vmx);
6609 }
6610 }
6611
6612 BUILD_BUG_ON(sizeof(user_vmx_nested_state->vmcs12) < VMCS12_SIZE);
6613 BUILD_BUG_ON(sizeof(user_vmx_nested_state->shadow_vmcs12) < VMCS12_SIZE);
6614
6615 /*
6616 * Copy over the full allocated size of vmcs12 rather than just the size
6617 * of the struct.
6618 */
6619 if (copy_to_user(user_vmx_nested_state->vmcs12, vmcs12, VMCS12_SIZE))
6620 return -EFAULT;
6621
6622 if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6623 vmcs12->vmcs_link_pointer != INVALID_GPA) {
6624 if (copy_to_user(user_vmx_nested_state->shadow_vmcs12,
6625 get_shadow_vmcs12(vcpu), VMCS12_SIZE))
6626 return -EFAULT;
6627 }
6628 out:
6629 return kvm_state.size;
6630 }
6631
6632 void vmx_leave_nested(struct kvm_vcpu *vcpu)
6633 {
6634 if (is_guest_mode(vcpu)) {
6635 to_vmx(vcpu)->nested.nested_run_pending = 0;
6636 nested_vmx_vmexit(vcpu, -1, 0, 0);
6637 }
6638 free_nested(vcpu);
6639 }
6640
6641 static int vmx_set_nested_state(struct kvm_vcpu *vcpu,
6642 struct kvm_nested_state __user *user_kvm_nested_state,
6643 struct kvm_nested_state *kvm_state)
6644 {
6645 struct vcpu_vmx *vmx = to_vmx(vcpu);
6646 struct vmcs12 *vmcs12;
6647 enum vm_entry_failure_code ignored;
6648 struct kvm_vmx_nested_state_data __user *user_vmx_nested_state =
6649 &user_kvm_nested_state->data.vmx[0];
6650 int ret;
6651
6652 if (kvm_state->format != KVM_STATE_NESTED_FORMAT_VMX)
6653 return -EINVAL;
6654
6655 if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA) {
6656 if (kvm_state->hdr.vmx.smm.flags)
6657 return -EINVAL;
6658
6659 if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA)
6660 return -EINVAL;
6661
6662 /*
6663 * KVM_STATE_NESTED_EVMCS used to signal that KVM should
6664 * enable eVMCS capability on vCPU. However, since then
6665 * code was changed such that flag signals vmcs12 should
6666 * be copied into eVMCS in guest memory.
6667 *
6668 * To preserve backwards compatibility, allow user
6669 * to set this flag even when there is no VMXON region.
6670 */
6671 if (kvm_state->flags & ~KVM_STATE_NESTED_EVMCS)
6672 return -EINVAL;
6673 } else {
6674 if (!guest_can_use(vcpu, X86_FEATURE_VMX))
6675 return -EINVAL;
6676
6677 if (!page_address_valid(vcpu, kvm_state->hdr.vmx.vmxon_pa))
6678 return -EINVAL;
6679 }
6680
6681 if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6682 (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6683 return -EINVAL;
6684
6685 if (kvm_state->hdr.vmx.smm.flags &
6686 ~(KVM_STATE_NESTED_SMM_GUEST_MODE | KVM_STATE_NESTED_SMM_VMXON))
6687 return -EINVAL;
6688
6689 if (kvm_state->hdr.vmx.flags & ~KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE)
6690 return -EINVAL;
6691
6692 /*
6693 * SMM temporarily disables VMX, so we cannot be in guest mode,
6694 * nor can VMLAUNCH/VMRESUME be pending. Outside SMM, SMM flags
6695 * must be zero.
6696 */
6697 if (is_smm(vcpu) ?
6698 (kvm_state->flags &
6699 (KVM_STATE_NESTED_GUEST_MODE | KVM_STATE_NESTED_RUN_PENDING))
6700 : kvm_state->hdr.vmx.smm.flags)
6701 return -EINVAL;
6702
6703 if ((kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE) &&
6704 !(kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON))
6705 return -EINVAL;
6706
6707 if ((kvm_state->flags & KVM_STATE_NESTED_EVMCS) &&
6708 (!guest_can_use(vcpu, X86_FEATURE_VMX) ||
6709 !vmx->nested.enlightened_vmcs_enabled))
6710 return -EINVAL;
6711
6712 vmx_leave_nested(vcpu);
6713
6714 if (kvm_state->hdr.vmx.vmxon_pa == INVALID_GPA)
6715 return 0;
6716
6717 vmx->nested.vmxon_ptr = kvm_state->hdr.vmx.vmxon_pa;
6718 ret = enter_vmx_operation(vcpu);
6719 if (ret)
6720 return ret;
6721
6722 /* Empty 'VMXON' state is permitted if no VMCS loaded */
6723 if (kvm_state->size < sizeof(*kvm_state) + sizeof(*vmcs12)) {
6724 /* See vmx_has_valid_vmcs12. */
6725 if ((kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE) ||
6726 (kvm_state->flags & KVM_STATE_NESTED_EVMCS) ||
6727 (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA))
6728 return -EINVAL;
6729 else
6730 return 0;
6731 }
6732
6733 if (kvm_state->hdr.vmx.vmcs12_pa != INVALID_GPA) {
6734 if (kvm_state->hdr.vmx.vmcs12_pa == kvm_state->hdr.vmx.vmxon_pa ||
6735 !page_address_valid(vcpu, kvm_state->hdr.vmx.vmcs12_pa))
6736 return -EINVAL;
6737
6738 set_current_vmptr(vmx, kvm_state->hdr.vmx.vmcs12_pa);
6739 #ifdef CONFIG_KVM_HYPERV
6740 } else if (kvm_state->flags & KVM_STATE_NESTED_EVMCS) {
6741 /*
6742 * nested_vmx_handle_enlightened_vmptrld() cannot be called
6743 * directly from here as HV_X64_MSR_VP_ASSIST_PAGE may not be
6744 * restored yet. EVMCS will be mapped from
6745 * nested_get_vmcs12_pages().
6746 */
6747 vmx->nested.hv_evmcs_vmptr = EVMPTR_MAP_PENDING;
6748 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
6749 #endif
6750 } else {
6751 return -EINVAL;
6752 }
6753
6754 if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON) {
6755 vmx->nested.smm.vmxon = true;
6756 vmx->nested.vmxon = false;
6757
6758 if (kvm_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_GUEST_MODE)
6759 vmx->nested.smm.guest_mode = true;
6760 }
6761
6762 vmcs12 = get_vmcs12(vcpu);
6763 if (copy_from_user(vmcs12, user_vmx_nested_state->vmcs12, sizeof(*vmcs12)))
6764 return -EFAULT;
6765
6766 if (vmcs12->hdr.revision_id != VMCS12_REVISION)
6767 return -EINVAL;
6768
6769 if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
6770 return 0;
6771
6772 vmx->nested.nested_run_pending =
6773 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
6774
6775 vmx->nested.mtf_pending =
6776 !!(kvm_state->flags & KVM_STATE_NESTED_MTF_PENDING);
6777
6778 ret = -EINVAL;
6779 if (nested_cpu_has_shadow_vmcs(vmcs12) &&
6780 vmcs12->vmcs_link_pointer != INVALID_GPA) {
6781 struct vmcs12 *shadow_vmcs12 = get_shadow_vmcs12(vcpu);
6782
6783 if (kvm_state->size <
6784 sizeof(*kvm_state) +
6785 sizeof(user_vmx_nested_state->vmcs12) + sizeof(*shadow_vmcs12))
6786 goto error_guest_mode;
6787
6788 if (copy_from_user(shadow_vmcs12,
6789 user_vmx_nested_state->shadow_vmcs12,
6790 sizeof(*shadow_vmcs12))) {
6791 ret = -EFAULT;
6792 goto error_guest_mode;
6793 }
6794
6795 if (shadow_vmcs12->hdr.revision_id != VMCS12_REVISION ||
6796 !shadow_vmcs12->hdr.shadow_vmcs)
6797 goto error_guest_mode;
6798 }
6799
6800 vmx->nested.has_preemption_timer_deadline = false;
6801 if (kvm_state->hdr.vmx.flags & KVM_STATE_VMX_PREEMPTION_TIMER_DEADLINE) {
6802 vmx->nested.has_preemption_timer_deadline = true;
6803 vmx->nested.preemption_timer_deadline =
6804 kvm_state->hdr.vmx.preemption_timer_deadline;
6805 }
6806
6807 if (nested_vmx_check_controls(vcpu, vmcs12) ||
6808 nested_vmx_check_host_state(vcpu, vmcs12) ||
6809 nested_vmx_check_guest_state(vcpu, vmcs12, &ignored))
6810 goto error_guest_mode;
6811
6812 vmx->nested.dirty_vmcs12 = true;
6813 vmx->nested.force_msr_bitmap_recalc = true;
6814 ret = nested_vmx_enter_non_root_mode(vcpu, false);
6815 if (ret)
6816 goto error_guest_mode;
6817
6818 if (vmx->nested.mtf_pending)
6819 kvm_make_request(KVM_REQ_EVENT, vcpu);
6820
6821 return 0;
6822
6823 error_guest_mode:
6824 vmx->nested.nested_run_pending = 0;
6825 return ret;
6826 }
6827
6828 void nested_vmx_set_vmcs_shadowing_bitmap(void)
6829 {
6830 if (enable_shadow_vmcs) {
6831 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
6832 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
6833 }
6834 }
6835
6836 /*
6837 * Indexing into the vmcs12 uses the VMCS encoding rotated left by 6. Undo
6838 * that madness to get the encoding for comparison.
6839 */
6840 #define VMCS12_IDX_TO_ENC(idx) ((u16)(((u16)(idx) >> 6) | ((u16)(idx) << 10)))
6841
6842 static u64 nested_vmx_calc_vmcs_enum_msr(void)
6843 {
6844 /*
6845 * Note these are the so called "index" of the VMCS field encoding, not
6846 * the index into vmcs12.
6847 */
6848 unsigned int max_idx, idx;
6849 int i;
6850
6851 /*
6852 * For better or worse, KVM allows VMREAD/VMWRITE to all fields in
6853 * vmcs12, regardless of whether or not the associated feature is
6854 * exposed to L1. Simply find the field with the highest index.
6855 */
6856 max_idx = 0;
6857 for (i = 0; i < nr_vmcs12_fields; i++) {
6858 /* The vmcs12 table is very, very sparsely populated. */
6859 if (!vmcs12_field_offsets[i])
6860 continue;
6861
6862 idx = vmcs_field_index(VMCS12_IDX_TO_ENC(i));
6863 if (idx > max_idx)
6864 max_idx = idx;
6865 }
6866
6867 return (u64)max_idx << VMCS_FIELD_INDEX_SHIFT;
6868 }
6869
6870 static void nested_vmx_setup_pinbased_ctls(struct vmcs_config *vmcs_conf,
6871 struct nested_vmx_msrs *msrs)
6872 {
6873 msrs->pinbased_ctls_low =
6874 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6875
6876 msrs->pinbased_ctls_high = vmcs_conf->pin_based_exec_ctrl;
6877 msrs->pinbased_ctls_high &=
6878 PIN_BASED_EXT_INTR_MASK |
6879 PIN_BASED_NMI_EXITING |
6880 PIN_BASED_VIRTUAL_NMIS |
6881 (enable_apicv ? PIN_BASED_POSTED_INTR : 0);
6882 msrs->pinbased_ctls_high |=
6883 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6884 PIN_BASED_VMX_PREEMPTION_TIMER;
6885 }
6886
6887 static void nested_vmx_setup_exit_ctls(struct vmcs_config *vmcs_conf,
6888 struct nested_vmx_msrs *msrs)
6889 {
6890 msrs->exit_ctls_low =
6891 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
6892
6893 msrs->exit_ctls_high = vmcs_conf->vmexit_ctrl;
6894 msrs->exit_ctls_high &=
6895 #ifdef CONFIG_X86_64
6896 VM_EXIT_HOST_ADDR_SPACE_SIZE |
6897 #endif
6898 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
6899 VM_EXIT_CLEAR_BNDCFGS;
6900 msrs->exit_ctls_high |=
6901 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
6902 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
6903 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT |
6904 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
6905
6906 /* We support free control of debug control saving. */
6907 msrs->exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
6908 }
6909
6910 static void nested_vmx_setup_entry_ctls(struct vmcs_config *vmcs_conf,
6911 struct nested_vmx_msrs *msrs)
6912 {
6913 msrs->entry_ctls_low =
6914 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
6915
6916 msrs->entry_ctls_high = vmcs_conf->vmentry_ctrl;
6917 msrs->entry_ctls_high &=
6918 #ifdef CONFIG_X86_64
6919 VM_ENTRY_IA32E_MODE |
6920 #endif
6921 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
6922 msrs->entry_ctls_high |=
6923 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER |
6924 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL);
6925
6926 /* We support free control of debug control loading. */
6927 msrs->entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
6928 }
6929
6930 static void nested_vmx_setup_cpubased_ctls(struct vmcs_config *vmcs_conf,
6931 struct nested_vmx_msrs *msrs)
6932 {
6933 msrs->procbased_ctls_low =
6934 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
6935
6936 msrs->procbased_ctls_high = vmcs_conf->cpu_based_exec_ctrl;
6937 msrs->procbased_ctls_high &=
6938 CPU_BASED_INTR_WINDOW_EXITING |
6939 CPU_BASED_NMI_WINDOW_EXITING | CPU_BASED_USE_TSC_OFFSETTING |
6940 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
6941 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
6942 CPU_BASED_CR3_STORE_EXITING |
6943 #ifdef CONFIG_X86_64
6944 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
6945 #endif
6946 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
6947 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
6948 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
6949 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
6950 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
6951 /*
6952 * We can allow some features even when not supported by the
6953 * hardware. For example, L1 can specify an MSR bitmap - and we
6954 * can use it to avoid exits to L1 - even when L0 runs L2
6955 * without MSR bitmaps.
6956 */
6957 msrs->procbased_ctls_high |=
6958 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
6959 CPU_BASED_USE_MSR_BITMAPS;
6960
6961 /* We support free control of CR3 access interception. */
6962 msrs->procbased_ctls_low &=
6963 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
6964 }
6965
6966 static void nested_vmx_setup_secondary_ctls(u32 ept_caps,
6967 struct vmcs_config *vmcs_conf,
6968 struct nested_vmx_msrs *msrs)
6969 {
6970 msrs->secondary_ctls_low = 0;
6971
6972 msrs->secondary_ctls_high = vmcs_conf->cpu_based_2nd_exec_ctrl;
6973 msrs->secondary_ctls_high &=
6974 SECONDARY_EXEC_DESC |
6975 SECONDARY_EXEC_ENABLE_RDTSCP |
6976 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6977 SECONDARY_EXEC_WBINVD_EXITING |
6978 SECONDARY_EXEC_APIC_REGISTER_VIRT |
6979 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
6980 SECONDARY_EXEC_RDRAND_EXITING |
6981 SECONDARY_EXEC_ENABLE_INVPCID |
6982 SECONDARY_EXEC_ENABLE_VMFUNC |
6983 SECONDARY_EXEC_RDSEED_EXITING |
6984 SECONDARY_EXEC_ENABLE_XSAVES |
6985 SECONDARY_EXEC_TSC_SCALING |
6986 SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
6987
6988 /*
6989 * We can emulate "VMCS shadowing," even if the hardware
6990 * doesn't support it.
6991 */
6992 msrs->secondary_ctls_high |=
6993 SECONDARY_EXEC_SHADOW_VMCS;
6994
6995 if (enable_ept) {
6996 /* nested EPT: emulate EPT also to L1 */
6997 msrs->secondary_ctls_high |=
6998 SECONDARY_EXEC_ENABLE_EPT;
6999 msrs->ept_caps =
7000 VMX_EPT_PAGE_WALK_4_BIT |
7001 VMX_EPT_PAGE_WALK_5_BIT |
7002 VMX_EPTP_WB_BIT |
7003 VMX_EPT_INVEPT_BIT |
7004 VMX_EPT_EXECUTE_ONLY_BIT;
7005
7006 msrs->ept_caps &= ept_caps;
7007 msrs->ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
7008 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
7009 VMX_EPT_1GB_PAGE_BIT;
7010 if (enable_ept_ad_bits) {
7011 msrs->secondary_ctls_high |=
7012 SECONDARY_EXEC_ENABLE_PML;
7013 msrs->ept_caps |= VMX_EPT_AD_BIT;
7014 }
7015
7016 /*
7017 * Advertise EPTP switching irrespective of hardware support,
7018 * KVM emulates it in software so long as VMFUNC is supported.
7019 */
7020 if (cpu_has_vmx_vmfunc())
7021 msrs->vmfunc_controls = VMX_VMFUNC_EPTP_SWITCHING;
7022 }
7023
7024 /*
7025 * Old versions of KVM use the single-context version without
7026 * checking for support, so declare that it is supported even
7027 * though it is treated as global context. The alternative is
7028 * not failing the single-context invvpid, and it is worse.
7029 */
7030 if (enable_vpid) {
7031 msrs->secondary_ctls_high |=
7032 SECONDARY_EXEC_ENABLE_VPID;
7033 msrs->vpid_caps = VMX_VPID_INVVPID_BIT |
7034 VMX_VPID_EXTENT_SUPPORTED_MASK;
7035 }
7036
7037 if (enable_unrestricted_guest)
7038 msrs->secondary_ctls_high |=
7039 SECONDARY_EXEC_UNRESTRICTED_GUEST;
7040
7041 if (flexpriority_enabled)
7042 msrs->secondary_ctls_high |=
7043 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7044
7045 if (enable_sgx)
7046 msrs->secondary_ctls_high |= SECONDARY_EXEC_ENCLS_EXITING;
7047 }
7048
7049 static void nested_vmx_setup_misc_data(struct vmcs_config *vmcs_conf,
7050 struct nested_vmx_msrs *msrs)
7051 {
7052 msrs->misc_low = (u32)vmcs_conf->misc & VMX_MISC_SAVE_EFER_LMA;
7053 msrs->misc_low |=
7054 MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS |
7055 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
7056 VMX_MISC_ACTIVITY_HLT |
7057 VMX_MISC_ACTIVITY_WAIT_SIPI;
7058 msrs->misc_high = 0;
7059 }
7060
7061 static void nested_vmx_setup_basic(struct nested_vmx_msrs *msrs)
7062 {
7063 /*
7064 * This MSR reports some information about VMX support. We
7065 * should return information about the VMX we emulate for the
7066 * guest, and the VMCS structure we give it - not about the
7067 * VMX support of the underlying hardware.
7068 */
7069 msrs->basic =
7070 VMCS12_REVISION |
7071 VMX_BASIC_TRUE_CTLS |
7072 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
7073 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
7074
7075 if (cpu_has_vmx_basic_inout())
7076 msrs->basic |= VMX_BASIC_INOUT;
7077 }
7078
7079 static void nested_vmx_setup_cr_fixed(struct nested_vmx_msrs *msrs)
7080 {
7081 /*
7082 * These MSRs specify bits which the guest must keep fixed on
7083 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
7084 * We picked the standard core2 setting.
7085 */
7086 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
7087 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
7088 msrs->cr0_fixed0 = VMXON_CR0_ALWAYSON;
7089 msrs->cr4_fixed0 = VMXON_CR4_ALWAYSON;
7090
7091 /* These MSRs specify bits which the guest must keep fixed off. */
7092 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, msrs->cr0_fixed1);
7093 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, msrs->cr4_fixed1);
7094
7095 if (vmx_umip_emulated())
7096 msrs->cr4_fixed1 |= X86_CR4_UMIP;
7097 }
7098
7099 /*
7100 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
7101 * returned for the various VMX controls MSRs when nested VMX is enabled.
7102 * The same values should also be used to verify that vmcs12 control fields are
7103 * valid during nested entry from L1 to L2.
7104 * Each of these control msrs has a low and high 32-bit half: A low bit is on
7105 * if the corresponding bit in the (32-bit) control field *must* be on, and a
7106 * bit in the high half is on if the corresponding bit in the control field
7107 * may be on. See also vmx_control_verify().
7108 */
7109 void nested_vmx_setup_ctls_msrs(struct vmcs_config *vmcs_conf, u32 ept_caps)
7110 {
7111 struct nested_vmx_msrs *msrs = &vmcs_conf->nested;
7112
7113 /*
7114 * Note that as a general rule, the high half of the MSRs (bits in
7115 * the control fields which may be 1) should be initialized by the
7116 * intersection of the underlying hardware's MSR (i.e., features which
7117 * can be supported) and the list of features we want to expose -
7118 * because they are known to be properly supported in our code.
7119 * Also, usually, the low half of the MSRs (bits which must be 1) can
7120 * be set to 0, meaning that L1 may turn off any of these bits. The
7121 * reason is that if one of these bits is necessary, it will appear
7122 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
7123 * fields of vmcs01 and vmcs02, will turn these bits off - and
7124 * nested_vmx_l1_wants_exit() will not pass related exits to L1.
7125 * These rules have exceptions below.
7126 */
7127 nested_vmx_setup_pinbased_ctls(vmcs_conf, msrs);
7128
7129 nested_vmx_setup_exit_ctls(vmcs_conf, msrs);
7130
7131 nested_vmx_setup_entry_ctls(vmcs_conf, msrs);
7132
7133 nested_vmx_setup_cpubased_ctls(vmcs_conf, msrs);
7134
7135 nested_vmx_setup_secondary_ctls(ept_caps, vmcs_conf, msrs);
7136
7137 nested_vmx_setup_misc_data(vmcs_conf, msrs);
7138
7139 nested_vmx_setup_basic(msrs);
7140
7141 nested_vmx_setup_cr_fixed(msrs);
7142
7143 msrs->vmcs_enum = nested_vmx_calc_vmcs_enum_msr();
7144 }
7145
7146 void nested_vmx_hardware_unsetup(void)
7147 {
7148 int i;
7149
7150 if (enable_shadow_vmcs) {
7151 for (i = 0; i < VMX_BITMAP_NR; i++)
7152 free_page((unsigned long)vmx_bitmap[i]);
7153 }
7154 }
7155
7156 __init int nested_vmx_hardware_setup(int (*exit_handlers[])(struct kvm_vcpu *))
7157 {
7158 int i;
7159
7160 if (!cpu_has_vmx_shadow_vmcs())
7161 enable_shadow_vmcs = 0;
7162 if (enable_shadow_vmcs) {
7163 for (i = 0; i < VMX_BITMAP_NR; i++) {
7164 /*
7165 * The vmx_bitmap is not tied to a VM and so should
7166 * not be charged to a memcg.
7167 */
7168 vmx_bitmap[i] = (unsigned long *)
7169 __get_free_page(GFP_KERNEL);
7170 if (!vmx_bitmap[i]) {
7171 nested_vmx_hardware_unsetup();
7172 return -ENOMEM;
7173 }
7174 }
7175
7176 init_vmcs_shadow_fields();
7177 }
7178
7179 exit_handlers[EXIT_REASON_VMCLEAR] = handle_vmclear;
7180 exit_handlers[EXIT_REASON_VMLAUNCH] = handle_vmlaunch;
7181 exit_handlers[EXIT_REASON_VMPTRLD] = handle_vmptrld;
7182 exit_handlers[EXIT_REASON_VMPTRST] = handle_vmptrst;
7183 exit_handlers[EXIT_REASON_VMREAD] = handle_vmread;
7184 exit_handlers[EXIT_REASON_VMRESUME] = handle_vmresume;
7185 exit_handlers[EXIT_REASON_VMWRITE] = handle_vmwrite;
7186 exit_handlers[EXIT_REASON_VMOFF] = handle_vmxoff;
7187 exit_handlers[EXIT_REASON_VMON] = handle_vmxon;
7188 exit_handlers[EXIT_REASON_INVEPT] = handle_invept;
7189 exit_handlers[EXIT_REASON_INVVPID] = handle_invvpid;
7190 exit_handlers[EXIT_REASON_VMFUNC] = handle_vmfunc;
7191
7192 return 0;
7193 }
7194
7195 struct kvm_x86_nested_ops vmx_nested_ops = {
7196 .leave_nested = vmx_leave_nested,
7197 .is_exception_vmexit = nested_vmx_is_exception_vmexit,
7198 .check_events = vmx_check_nested_events,
7199 .has_events = vmx_has_nested_events,
7200 .triple_fault = nested_vmx_triple_fault,
7201 .get_state = vmx_get_nested_state,
7202 .set_state = vmx_set_nested_state,
7203 .get_nested_state_pages = vmx_get_nested_state_pages,
7204 .write_log_dirty = nested_vmx_write_pml_buffer,
7205 #ifdef CONFIG_KVM_HYPERV
7206 .enable_evmcs = nested_enable_evmcs,
7207 .get_evmcs_version = nested_get_evmcs_version,
7208 .hv_inject_synthetic_vmexit_post_tlb_flush = vmx_hv_inject_synthetic_vmexit_post_tlb_flush,
7209 #endif
7210 };
7211
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.