Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * Kernel-based Virtual Machine driver for Linux
0004  *
0005  * AMD SVM support
0006  *
0007  * Copyright (C) 2006 Qumranet, Inc.
0008  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
0009  *
0010  * Authors:
0011  *   Yaniv Kamay  <yaniv@qumranet.com>
0012  *   Avi Kivity   <avi@qumranet.com>
0013  */
0014 
0015 #define pr_fmt(fmt) "SVM: " fmt
0016 
0017 #include <linux/kvm_types.h>
0018 #include <linux/kvm_host.h>
0019 #include <linux/kernel.h>
0020 
0021 #include <asm/msr-index.h>
0022 #include <asm/debugreg.h>
0023 
0024 #include "kvm_emulate.h"
0025 #include "trace.h"
0026 #include "mmu.h"
0027 #include "x86.h"
0028 #include "cpuid.h"
0029 #include "lapic.h"
0030 #include "svm.h"
0031 #include "hyperv.h"
0032 
0033 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
0034 
0035 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
0036                        struct x86_exception *fault)
0037 {
0038     struct vcpu_svm *svm = to_svm(vcpu);
0039     struct vmcb *vmcb = svm->vmcb;
0040 
0041     if (vmcb->control.exit_code != SVM_EXIT_NPF) {
0042         /*
0043          * TODO: track the cause of the nested page fault, and
0044          * correctly fill in the high bits of exit_info_1.
0045          */
0046         vmcb->control.exit_code = SVM_EXIT_NPF;
0047         vmcb->control.exit_code_hi = 0;
0048         vmcb->control.exit_info_1 = (1ULL << 32);
0049         vmcb->control.exit_info_2 = fault->address;
0050     }
0051 
0052     vmcb->control.exit_info_1 &= ~0xffffffffULL;
0053     vmcb->control.exit_info_1 |= fault->error_code;
0054 
0055     nested_svm_vmexit(svm);
0056 }
0057 
0058 static bool nested_svm_handle_page_fault_workaround(struct kvm_vcpu *vcpu,
0059                             struct x86_exception *fault)
0060 {
0061     struct vcpu_svm *svm = to_svm(vcpu);
0062     struct vmcb *vmcb = svm->vmcb;
0063 
0064     WARN_ON(!is_guest_mode(vcpu));
0065 
0066     if (vmcb12_is_intercept(&svm->nested.ctl,
0067                 INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
0068         !WARN_ON_ONCE(svm->nested.nested_run_pending)) {
0069             vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
0070         vmcb->control.exit_code_hi = 0;
0071         vmcb->control.exit_info_1 = fault->error_code;
0072         vmcb->control.exit_info_2 = fault->address;
0073         nested_svm_vmexit(svm);
0074         return true;
0075     }
0076 
0077     return false;
0078 }
0079 
0080 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
0081 {
0082     struct vcpu_svm *svm = to_svm(vcpu);
0083     u64 cr3 = svm->nested.ctl.nested_cr3;
0084     u64 pdpte;
0085     int ret;
0086 
0087     ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
0088                        offset_in_page(cr3) + index * 8, 8);
0089     if (ret)
0090         return 0;
0091     return pdpte;
0092 }
0093 
0094 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
0095 {
0096     struct vcpu_svm *svm = to_svm(vcpu);
0097 
0098     return svm->nested.ctl.nested_cr3;
0099 }
0100 
0101 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
0102 {
0103     struct vcpu_svm *svm = to_svm(vcpu);
0104 
0105     WARN_ON(mmu_is_nested(vcpu));
0106 
0107     vcpu->arch.mmu = &vcpu->arch.guest_mmu;
0108 
0109     /*
0110      * The NPT format depends on L1's CR4 and EFER, which is in vmcb01.  Note,
0111      * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
0112      * vCPU state.  CR0.WP is explicitly ignored, while CR0.PG is required.
0113      */
0114     kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
0115                 svm->vmcb01.ptr->save.efer,
0116                 svm->nested.ctl.nested_cr3);
0117     vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
0118     vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
0119     vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
0120     vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
0121 }
0122 
0123 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
0124 {
0125     vcpu->arch.mmu = &vcpu->arch.root_mmu;
0126     vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
0127 }
0128 
0129 static bool nested_vmcb_needs_vls_intercept(struct vcpu_svm *svm)
0130 {
0131     if (!svm->v_vmload_vmsave_enabled)
0132         return true;
0133 
0134     if (!nested_npt_enabled(svm))
0135         return true;
0136 
0137     if (!(svm->nested.ctl.virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK))
0138         return true;
0139 
0140     return false;
0141 }
0142 
0143 void recalc_intercepts(struct vcpu_svm *svm)
0144 {
0145     struct vmcb_control_area *c, *h;
0146     struct vmcb_ctrl_area_cached *g;
0147     unsigned int i;
0148 
0149     vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
0150 
0151     if (!is_guest_mode(&svm->vcpu))
0152         return;
0153 
0154     c = &svm->vmcb->control;
0155     h = &svm->vmcb01.ptr->control;
0156     g = &svm->nested.ctl;
0157 
0158     for (i = 0; i < MAX_INTERCEPT; i++)
0159         c->intercepts[i] = h->intercepts[i];
0160 
0161     if (g->int_ctl & V_INTR_MASKING_MASK) {
0162         /* We only want the cr8 intercept bits of L1 */
0163         vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
0164         vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
0165 
0166         /*
0167          * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
0168          * affect any interrupt we may want to inject; therefore,
0169          * interrupt window vmexits are irrelevant to L0.
0170          */
0171         vmcb_clr_intercept(c, INTERCEPT_VINTR);
0172     }
0173 
0174     /* We don't want to see VMMCALLs from a nested guest */
0175     vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
0176 
0177     for (i = 0; i < MAX_INTERCEPT; i++)
0178         c->intercepts[i] |= g->intercepts[i];
0179 
0180     /* If SMI is not intercepted, ignore guest SMI intercept as well  */
0181     if (!intercept_smi)
0182         vmcb_clr_intercept(c, INTERCEPT_SMI);
0183 
0184     if (nested_vmcb_needs_vls_intercept(svm)) {
0185         /*
0186          * If the virtual VMLOAD/VMSAVE is not enabled for the L2,
0187          * we must intercept these instructions to correctly
0188          * emulate them in case L1 doesn't intercept them.
0189          */
0190         vmcb_set_intercept(c, INTERCEPT_VMLOAD);
0191         vmcb_set_intercept(c, INTERCEPT_VMSAVE);
0192     } else {
0193         WARN_ON(!(c->virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK));
0194     }
0195 }
0196 
0197 /*
0198  * Merge L0's (KVM) and L1's (Nested VMCB) MSR permission bitmaps. The function
0199  * is optimized in that it only merges the parts where KVM MSR permission bitmap
0200  * may contain zero bits.
0201  */
0202 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
0203 {
0204     struct hv_enlightenments *hve =
0205         (struct hv_enlightenments *)svm->nested.ctl.reserved_sw;
0206     int i;
0207 
0208     /*
0209      * MSR bitmap update can be skipped when:
0210      * - MSR bitmap for L1 hasn't changed.
0211      * - Nested hypervisor (L1) is attempting to launch the same L2 as
0212      *   before.
0213      * - Nested hypervisor (L1) is using Hyper-V emulation interface and
0214      * tells KVM (L0) there were no changes in MSR bitmap for L2.
0215      */
0216     if (!svm->nested.force_msr_bitmap_recalc &&
0217         kvm_hv_hypercall_enabled(&svm->vcpu) &&
0218         hve->hv_enlightenments_control.msr_bitmap &&
0219         (svm->nested.ctl.clean & BIT(VMCB_HV_NESTED_ENLIGHTENMENTS)))
0220         goto set_msrpm_base_pa;
0221 
0222     if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
0223         return true;
0224 
0225     for (i = 0; i < MSRPM_OFFSETS; i++) {
0226         u32 value, p;
0227         u64 offset;
0228 
0229         if (msrpm_offsets[i] == 0xffffffff)
0230             break;
0231 
0232         p      = msrpm_offsets[i];
0233 
0234         /* x2apic msrs are intercepted always for the nested guest */
0235         if (is_x2apic_msrpm_offset(p))
0236             continue;
0237 
0238         offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
0239 
0240         if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
0241             return false;
0242 
0243         svm->nested.msrpm[p] = svm->msrpm[p] | value;
0244     }
0245 
0246     svm->nested.force_msr_bitmap_recalc = false;
0247 
0248 set_msrpm_base_pa:
0249     svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
0250 
0251     return true;
0252 }
0253 
0254 /*
0255  * Bits 11:0 of bitmap address are ignored by hardware
0256  */
0257 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
0258 {
0259     u64 addr = PAGE_ALIGN(pa);
0260 
0261     return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
0262         kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
0263 }
0264 
0265 static bool nested_svm_check_tlb_ctl(struct kvm_vcpu *vcpu, u8 tlb_ctl)
0266 {
0267     /* Nested FLUSHBYASID is not supported yet.  */
0268     switch(tlb_ctl) {
0269         case TLB_CONTROL_DO_NOTHING:
0270         case TLB_CONTROL_FLUSH_ALL_ASID:
0271             return true;
0272         default:
0273             return false;
0274     }
0275 }
0276 
0277 static bool __nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
0278                      struct vmcb_ctrl_area_cached *control)
0279 {
0280     if (CC(!vmcb12_is_intercept(control, INTERCEPT_VMRUN)))
0281         return false;
0282 
0283     if (CC(control->asid == 0))
0284         return false;
0285 
0286     if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
0287         return false;
0288 
0289     if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
0290                        MSRPM_SIZE)))
0291         return false;
0292     if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
0293                        IOPM_SIZE)))
0294         return false;
0295 
0296     if (CC(!nested_svm_check_tlb_ctl(vcpu, control->tlb_ctl)))
0297         return false;
0298 
0299     return true;
0300 }
0301 
0302 /* Common checks that apply to both L1 and L2 state.  */
0303 static bool __nested_vmcb_check_save(struct kvm_vcpu *vcpu,
0304                      struct vmcb_save_area_cached *save)
0305 {
0306     if (CC(!(save->efer & EFER_SVME)))
0307         return false;
0308 
0309     if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
0310         CC(save->cr0 & ~0xffffffffULL))
0311         return false;
0312 
0313     if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
0314         return false;
0315 
0316     /*
0317      * These checks are also performed by KVM_SET_SREGS,
0318      * except that EFER.LMA is not checked by SVM against
0319      * CR0.PG && EFER.LME.
0320      */
0321     if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
0322         if (CC(!(save->cr4 & X86_CR4_PAE)) ||
0323             CC(!(save->cr0 & X86_CR0_PE)) ||
0324             CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
0325             return false;
0326     }
0327 
0328     /* Note, SVM doesn't have any additional restrictions on CR4. */
0329     if (CC(!__kvm_is_valid_cr4(vcpu, save->cr4)))
0330         return false;
0331 
0332     if (CC(!kvm_valid_efer(vcpu, save->efer)))
0333         return false;
0334 
0335     return true;
0336 }
0337 
0338 static bool nested_vmcb_check_save(struct kvm_vcpu *vcpu)
0339 {
0340     struct vcpu_svm *svm = to_svm(vcpu);
0341     struct vmcb_save_area_cached *save = &svm->nested.save;
0342 
0343     return __nested_vmcb_check_save(vcpu, save);
0344 }
0345 
0346 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu)
0347 {
0348     struct vcpu_svm *svm = to_svm(vcpu);
0349     struct vmcb_ctrl_area_cached *ctl = &svm->nested.ctl;
0350 
0351     return __nested_vmcb_check_controls(vcpu, ctl);
0352 }
0353 
0354 static
0355 void __nested_copy_vmcb_control_to_cache(struct kvm_vcpu *vcpu,
0356                      struct vmcb_ctrl_area_cached *to,
0357                      struct vmcb_control_area *from)
0358 {
0359     unsigned int i;
0360 
0361     for (i = 0; i < MAX_INTERCEPT; i++)
0362         to->intercepts[i] = from->intercepts[i];
0363 
0364     to->iopm_base_pa        = from->iopm_base_pa;
0365     to->msrpm_base_pa       = from->msrpm_base_pa;
0366     to->tsc_offset          = from->tsc_offset;
0367     to->tlb_ctl             = from->tlb_ctl;
0368     to->int_ctl             = from->int_ctl;
0369     to->int_vector          = from->int_vector;
0370     to->int_state           = from->int_state;
0371     to->exit_code           = from->exit_code;
0372     to->exit_code_hi        = from->exit_code_hi;
0373     to->exit_info_1         = from->exit_info_1;
0374     to->exit_info_2         = from->exit_info_2;
0375     to->exit_int_info       = from->exit_int_info;
0376     to->exit_int_info_err   = from->exit_int_info_err;
0377     to->nested_ctl          = from->nested_ctl;
0378     to->event_inj           = from->event_inj;
0379     to->event_inj_err       = from->event_inj_err;
0380     to->next_rip            = from->next_rip;
0381     to->nested_cr3          = from->nested_cr3;
0382     to->virt_ext            = from->virt_ext;
0383     to->pause_filter_count  = from->pause_filter_count;
0384     to->pause_filter_thresh = from->pause_filter_thresh;
0385 
0386     /* Copy asid here because nested_vmcb_check_controls will check it.  */
0387     to->asid           = from->asid;
0388     to->msrpm_base_pa &= ~0x0fffULL;
0389     to->iopm_base_pa  &= ~0x0fffULL;
0390 
0391     /* Hyper-V extensions (Enlightened VMCB) */
0392     if (kvm_hv_hypercall_enabled(vcpu)) {
0393         to->clean = from->clean;
0394         memcpy(to->reserved_sw, from->reserved_sw,
0395                sizeof(struct hv_enlightenments));
0396     }
0397 }
0398 
0399 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
0400                        struct vmcb_control_area *control)
0401 {
0402     __nested_copy_vmcb_control_to_cache(&svm->vcpu, &svm->nested.ctl, control);
0403 }
0404 
0405 static void __nested_copy_vmcb_save_to_cache(struct vmcb_save_area_cached *to,
0406                          struct vmcb_save_area *from)
0407 {
0408     /*
0409      * Copy only fields that are validated, as we need them
0410      * to avoid TOC/TOU races.
0411      */
0412     to->efer = from->efer;
0413     to->cr0 = from->cr0;
0414     to->cr3 = from->cr3;
0415     to->cr4 = from->cr4;
0416 
0417     to->dr6 = from->dr6;
0418     to->dr7 = from->dr7;
0419 }
0420 
0421 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
0422                     struct vmcb_save_area *save)
0423 {
0424     __nested_copy_vmcb_save_to_cache(&svm->nested.save, save);
0425 }
0426 
0427 /*
0428  * Synchronize fields that are written by the processor, so that
0429  * they can be copied back into the vmcb12.
0430  */
0431 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
0432 {
0433     u32 mask;
0434     svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
0435     svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
0436 
0437     /* Only a few fields of int_ctl are written by the processor.  */
0438     mask = V_IRQ_MASK | V_TPR_MASK;
0439     if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
0440         svm_is_intercept(svm, INTERCEPT_VINTR)) {
0441         /*
0442          * In order to request an interrupt window, L0 is usurping
0443          * svm->vmcb->control.int_ctl and possibly setting V_IRQ
0444          * even if it was clear in L1's VMCB.  Restoring it would be
0445          * wrong.  However, in this case V_IRQ will remain true until
0446          * interrupt_window_interception calls svm_clear_vintr and
0447          * restores int_ctl.  We can just leave it aside.
0448          */
0449         mask &= ~V_IRQ_MASK;
0450     }
0451 
0452     if (nested_vgif_enabled(svm))
0453         mask |= V_GIF_MASK;
0454 
0455     svm->nested.ctl.int_ctl        &= ~mask;
0456     svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
0457 }
0458 
0459 /*
0460  * Transfer any event that L0 or L1 wanted to inject into L2 to
0461  * EXIT_INT_INFO.
0462  */
0463 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
0464                         struct vmcb *vmcb12)
0465 {
0466     struct kvm_vcpu *vcpu = &svm->vcpu;
0467     u32 exit_int_info = 0;
0468     unsigned int nr;
0469 
0470     if (vcpu->arch.exception.injected) {
0471         nr = vcpu->arch.exception.nr;
0472         exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
0473 
0474         if (vcpu->arch.exception.has_error_code) {
0475             exit_int_info |= SVM_EVTINJ_VALID_ERR;
0476             vmcb12->control.exit_int_info_err =
0477                 vcpu->arch.exception.error_code;
0478         }
0479 
0480     } else if (vcpu->arch.nmi_injected) {
0481         exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
0482 
0483     } else if (vcpu->arch.interrupt.injected) {
0484         nr = vcpu->arch.interrupt.nr;
0485         exit_int_info = nr | SVM_EVTINJ_VALID;
0486 
0487         if (vcpu->arch.interrupt.soft)
0488             exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
0489         else
0490             exit_int_info |= SVM_EVTINJ_TYPE_INTR;
0491     }
0492 
0493     vmcb12->control.exit_int_info = exit_int_info;
0494 }
0495 
0496 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
0497 {
0498     /*
0499      * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
0500      * things to fix before this can be conditional:
0501      *
0502      *  - Flush TLBs for both L1 and L2 remote TLB flush
0503      *  - Honor L1's request to flush an ASID on nested VMRUN
0504      *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
0505      *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
0506      *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
0507      *
0508      * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
0509      *     NPT guest-physical mappings on VMRUN.
0510      */
0511     kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
0512     kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
0513 }
0514 
0515 /*
0516  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
0517  * if we are emulating VM-Entry into a guest with NPT enabled.
0518  */
0519 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
0520                    bool nested_npt, bool reload_pdptrs)
0521 {
0522     if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
0523         return -EINVAL;
0524 
0525     if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
0526         CC(!load_pdptrs(vcpu, cr3)))
0527         return -EINVAL;
0528 
0529     vcpu->arch.cr3 = cr3;
0530 
0531     /* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
0532     kvm_init_mmu(vcpu);
0533 
0534     if (!nested_npt)
0535         kvm_mmu_new_pgd(vcpu, cr3);
0536 
0537     return 0;
0538 }
0539 
0540 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
0541 {
0542     if (!svm->nested.vmcb02.ptr)
0543         return;
0544 
0545     /* FIXME: merge g_pat from vmcb01 and vmcb12.  */
0546     svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
0547 }
0548 
0549 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
0550 {
0551     bool new_vmcb12 = false;
0552     struct vmcb *vmcb01 = svm->vmcb01.ptr;
0553     struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
0554 
0555     nested_vmcb02_compute_g_pat(svm);
0556 
0557     /* Load the nested guest state */
0558     if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
0559         new_vmcb12 = true;
0560         svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
0561         svm->nested.force_msr_bitmap_recalc = true;
0562     }
0563 
0564     if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
0565         vmcb02->save.es = vmcb12->save.es;
0566         vmcb02->save.cs = vmcb12->save.cs;
0567         vmcb02->save.ss = vmcb12->save.ss;
0568         vmcb02->save.ds = vmcb12->save.ds;
0569         vmcb02->save.cpl = vmcb12->save.cpl;
0570         vmcb_mark_dirty(vmcb02, VMCB_SEG);
0571     }
0572 
0573     if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
0574         vmcb02->save.gdtr = vmcb12->save.gdtr;
0575         vmcb02->save.idtr = vmcb12->save.idtr;
0576         vmcb_mark_dirty(vmcb02, VMCB_DT);
0577     }
0578 
0579     kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
0580 
0581     svm_set_efer(&svm->vcpu, svm->nested.save.efer);
0582 
0583     svm_set_cr0(&svm->vcpu, svm->nested.save.cr0);
0584     svm_set_cr4(&svm->vcpu, svm->nested.save.cr4);
0585 
0586     svm->vcpu.arch.cr2 = vmcb12->save.cr2;
0587 
0588     kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
0589     kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
0590     kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
0591 
0592     /* In case we don't even reach vcpu_run, the fields are not updated */
0593     vmcb02->save.rax = vmcb12->save.rax;
0594     vmcb02->save.rsp = vmcb12->save.rsp;
0595     vmcb02->save.rip = vmcb12->save.rip;
0596 
0597     /* These bits will be set properly on the first execution when new_vmc12 is true */
0598     if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
0599         vmcb02->save.dr7 = svm->nested.save.dr7 | DR7_FIXED_1;
0600         svm->vcpu.arch.dr6  = svm->nested.save.dr6 | DR6_ACTIVE_LOW;
0601         vmcb_mark_dirty(vmcb02, VMCB_DR);
0602     }
0603 
0604     if (unlikely(svm->lbrv_enabled && (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
0605         /*
0606          * Reserved bits of DEBUGCTL are ignored.  Be consistent with
0607          * svm_set_msr's definition of reserved bits.
0608          */
0609         svm_copy_lbrs(vmcb02, vmcb12);
0610         vmcb02->save.dbgctl &= ~DEBUGCTL_RESERVED_BITS;
0611         svm_update_lbrv(&svm->vcpu);
0612 
0613     } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
0614         svm_copy_lbrs(vmcb02, vmcb01);
0615     }
0616 }
0617 
0618 static inline bool is_evtinj_soft(u32 evtinj)
0619 {
0620     u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
0621     u8 vector = evtinj & SVM_EVTINJ_VEC_MASK;
0622 
0623     if (!(evtinj & SVM_EVTINJ_VALID))
0624         return false;
0625 
0626     if (type == SVM_EVTINJ_TYPE_SOFT)
0627         return true;
0628 
0629     return type == SVM_EVTINJ_TYPE_EXEPT && kvm_exception_is_soft(vector);
0630 }
0631 
0632 static bool is_evtinj_nmi(u32 evtinj)
0633 {
0634     u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
0635 
0636     if (!(evtinj & SVM_EVTINJ_VALID))
0637         return false;
0638 
0639     return type == SVM_EVTINJ_TYPE_NMI;
0640 }
0641 
0642 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm,
0643                       unsigned long vmcb12_rip,
0644                       unsigned long vmcb12_csbase)
0645 {
0646     u32 int_ctl_vmcb01_bits = V_INTR_MASKING_MASK;
0647     u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
0648 
0649     struct kvm_vcpu *vcpu = &svm->vcpu;
0650     struct vmcb *vmcb01 = svm->vmcb01.ptr;
0651     struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
0652     u32 pause_count12;
0653     u32 pause_thresh12;
0654 
0655     /*
0656      * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
0657      * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
0658      */
0659 
0660     if (svm->vgif_enabled && (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK))
0661         int_ctl_vmcb12_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
0662     else
0663         int_ctl_vmcb01_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
0664 
0665     /* Copied from vmcb01.  msrpm_base can be overwritten later.  */
0666     vmcb02->control.nested_ctl = vmcb01->control.nested_ctl;
0667     vmcb02->control.iopm_base_pa = vmcb01->control.iopm_base_pa;
0668     vmcb02->control.msrpm_base_pa = vmcb01->control.msrpm_base_pa;
0669 
0670     /* Done at vmrun: asid.  */
0671 
0672     /* Also overwritten later if necessary.  */
0673     vmcb02->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
0674 
0675     /* nested_cr3.  */
0676     if (nested_npt_enabled(svm))
0677         nested_svm_init_mmu_context(vcpu);
0678 
0679     vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
0680             vcpu->arch.l1_tsc_offset,
0681             svm->nested.ctl.tsc_offset,
0682             svm->tsc_ratio_msr);
0683 
0684     vmcb02->control.tsc_offset = vcpu->arch.tsc_offset;
0685 
0686     if (svm->tsc_ratio_msr != kvm_caps.default_tsc_scaling_ratio) {
0687         WARN_ON(!svm->tsc_scaling_enabled);
0688         nested_svm_update_tsc_ratio_msr(vcpu);
0689     }
0690 
0691     vmcb02->control.int_ctl             =
0692         (svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
0693         (vmcb01->control.int_ctl & int_ctl_vmcb01_bits);
0694 
0695     vmcb02->control.int_vector          = svm->nested.ctl.int_vector;
0696     vmcb02->control.int_state           = svm->nested.ctl.int_state;
0697     vmcb02->control.event_inj           = svm->nested.ctl.event_inj;
0698     vmcb02->control.event_inj_err       = svm->nested.ctl.event_inj_err;
0699 
0700     /*
0701      * next_rip is consumed on VMRUN as the return address pushed on the
0702      * stack for injected soft exceptions/interrupts.  If nrips is exposed
0703      * to L1, take it verbatim from vmcb12.  If nrips is supported in
0704      * hardware but not exposed to L1, stuff the actual L2 RIP to emulate
0705      * what a nrips=0 CPU would do (L1 is responsible for advancing RIP
0706      * prior to injecting the event).
0707      */
0708     if (svm->nrips_enabled)
0709         vmcb02->control.next_rip    = svm->nested.ctl.next_rip;
0710     else if (boot_cpu_has(X86_FEATURE_NRIPS))
0711         vmcb02->control.next_rip    = vmcb12_rip;
0712 
0713     svm->nmi_l1_to_l2 = is_evtinj_nmi(vmcb02->control.event_inj);
0714     if (is_evtinj_soft(vmcb02->control.event_inj)) {
0715         svm->soft_int_injected = true;
0716         svm->soft_int_csbase = vmcb12_csbase;
0717         svm->soft_int_old_rip = vmcb12_rip;
0718         if (svm->nrips_enabled)
0719             svm->soft_int_next_rip = svm->nested.ctl.next_rip;
0720         else
0721             svm->soft_int_next_rip = vmcb12_rip;
0722     }
0723 
0724     vmcb02->control.virt_ext            = vmcb01->control.virt_ext &
0725                           LBR_CTL_ENABLE_MASK;
0726     if (svm->lbrv_enabled)
0727         vmcb02->control.virt_ext  |=
0728             (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK);
0729 
0730     if (!nested_vmcb_needs_vls_intercept(svm))
0731         vmcb02->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
0732 
0733     pause_count12 = svm->pause_filter_enabled ? svm->nested.ctl.pause_filter_count : 0;
0734     pause_thresh12 = svm->pause_threshold_enabled ? svm->nested.ctl.pause_filter_thresh : 0;
0735     if (kvm_pause_in_guest(svm->vcpu.kvm)) {
0736         /* use guest values since host doesn't intercept PAUSE */
0737         vmcb02->control.pause_filter_count = pause_count12;
0738         vmcb02->control.pause_filter_thresh = pause_thresh12;
0739 
0740     } else {
0741         /* start from host values otherwise */
0742         vmcb02->control.pause_filter_count = vmcb01->control.pause_filter_count;
0743         vmcb02->control.pause_filter_thresh = vmcb01->control.pause_filter_thresh;
0744 
0745         /* ... but ensure filtering is disabled if so requested.  */
0746         if (vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_PAUSE)) {
0747             if (!pause_count12)
0748                 vmcb02->control.pause_filter_count = 0;
0749             if (!pause_thresh12)
0750                 vmcb02->control.pause_filter_thresh = 0;
0751         }
0752     }
0753 
0754     nested_svm_transition_tlb_flush(vcpu);
0755 
0756     /* Enter Guest-Mode */
0757     enter_guest_mode(vcpu);
0758 
0759     /*
0760      * Merge guest and host intercepts - must be called with vcpu in
0761      * guest-mode to take effect.
0762      */
0763     recalc_intercepts(svm);
0764 }
0765 
0766 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
0767 {
0768     /*
0769      * Some VMCB state is shared between L1 and L2 and thus has to be
0770      * moved at the time of nested vmrun and vmexit.
0771      *
0772      * VMLOAD/VMSAVE state would also belong in this category, but KVM
0773      * always performs VMLOAD and VMSAVE from the VMCB01.
0774      */
0775     to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
0776 }
0777 
0778 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
0779              struct vmcb *vmcb12, bool from_vmrun)
0780 {
0781     struct vcpu_svm *svm = to_svm(vcpu);
0782     int ret;
0783 
0784     trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
0785                    vmcb12->save.rip,
0786                    vmcb12->control.int_ctl,
0787                    vmcb12->control.event_inj,
0788                    vmcb12->control.nested_ctl);
0789 
0790     trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
0791                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
0792                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
0793                     vmcb12->control.intercepts[INTERCEPT_WORD3],
0794                     vmcb12->control.intercepts[INTERCEPT_WORD4],
0795                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
0796 
0797 
0798     svm->nested.vmcb12_gpa = vmcb12_gpa;
0799 
0800     WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
0801 
0802     nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
0803 
0804     svm_switch_vmcb(svm, &svm->nested.vmcb02);
0805     nested_vmcb02_prepare_control(svm, vmcb12->save.rip, vmcb12->save.cs.base);
0806     nested_vmcb02_prepare_save(svm, vmcb12);
0807 
0808     ret = nested_svm_load_cr3(&svm->vcpu, svm->nested.save.cr3,
0809                   nested_npt_enabled(svm), from_vmrun);
0810     if (ret)
0811         return ret;
0812 
0813     if (!from_vmrun)
0814         kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
0815 
0816     svm_set_gif(svm, true);
0817 
0818     if (kvm_vcpu_apicv_active(vcpu))
0819         kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
0820 
0821     return 0;
0822 }
0823 
0824 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
0825 {
0826     struct vcpu_svm *svm = to_svm(vcpu);
0827     int ret;
0828     struct vmcb *vmcb12;
0829     struct kvm_host_map map;
0830     u64 vmcb12_gpa;
0831     struct vmcb *vmcb01 = svm->vmcb01.ptr;
0832 
0833     if (!svm->nested.hsave_msr) {
0834         kvm_inject_gp(vcpu, 0);
0835         return 1;
0836     }
0837 
0838     if (is_smm(vcpu)) {
0839         kvm_queue_exception(vcpu, UD_VECTOR);
0840         return 1;
0841     }
0842 
0843     vmcb12_gpa = svm->vmcb->save.rax;
0844     ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
0845     if (ret == -EINVAL) {
0846         kvm_inject_gp(vcpu, 0);
0847         return 1;
0848     } else if (ret) {
0849         return kvm_skip_emulated_instruction(vcpu);
0850     }
0851 
0852     ret = kvm_skip_emulated_instruction(vcpu);
0853 
0854     vmcb12 = map.hva;
0855 
0856     if (WARN_ON_ONCE(!svm->nested.initialized))
0857         return -EINVAL;
0858 
0859     nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
0860     nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
0861 
0862     if (!nested_vmcb_check_save(vcpu) ||
0863         !nested_vmcb_check_controls(vcpu)) {
0864         vmcb12->control.exit_code    = SVM_EXIT_ERR;
0865         vmcb12->control.exit_code_hi = 0;
0866         vmcb12->control.exit_info_1  = 0;
0867         vmcb12->control.exit_info_2  = 0;
0868         goto out;
0869     }
0870 
0871     /*
0872      * Since vmcb01 is not in use, we can use it to store some of the L1
0873      * state.
0874      */
0875     vmcb01->save.efer   = vcpu->arch.efer;
0876     vmcb01->save.cr0    = kvm_read_cr0(vcpu);
0877     vmcb01->save.cr4    = vcpu->arch.cr4;
0878     vmcb01->save.rflags = kvm_get_rflags(vcpu);
0879     vmcb01->save.rip    = kvm_rip_read(vcpu);
0880 
0881     if (!npt_enabled)
0882         vmcb01->save.cr3 = kvm_read_cr3(vcpu);
0883 
0884     svm->nested.nested_run_pending = 1;
0885 
0886     if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
0887         goto out_exit_err;
0888 
0889     if (nested_svm_vmrun_msrpm(svm))
0890         goto out;
0891 
0892 out_exit_err:
0893     svm->nested.nested_run_pending = 0;
0894     svm->nmi_l1_to_l2 = false;
0895     svm->soft_int_injected = false;
0896 
0897     svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
0898     svm->vmcb->control.exit_code_hi = 0;
0899     svm->vmcb->control.exit_info_1  = 0;
0900     svm->vmcb->control.exit_info_2  = 0;
0901 
0902     nested_svm_vmexit(svm);
0903 
0904 out:
0905     kvm_vcpu_unmap(vcpu, &map, true);
0906 
0907     return ret;
0908 }
0909 
0910 /* Copy state save area fields which are handled by VMRUN */
0911 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
0912               struct vmcb_save_area *from_save)
0913 {
0914     to_save->es = from_save->es;
0915     to_save->cs = from_save->cs;
0916     to_save->ss = from_save->ss;
0917     to_save->ds = from_save->ds;
0918     to_save->gdtr = from_save->gdtr;
0919     to_save->idtr = from_save->idtr;
0920     to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
0921     to_save->efer = from_save->efer;
0922     to_save->cr0 = from_save->cr0;
0923     to_save->cr3 = from_save->cr3;
0924     to_save->cr4 = from_save->cr4;
0925     to_save->rax = from_save->rax;
0926     to_save->rsp = from_save->rsp;
0927     to_save->rip = from_save->rip;
0928     to_save->cpl = 0;
0929 }
0930 
0931 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
0932 {
0933     to_vmcb->save.fs = from_vmcb->save.fs;
0934     to_vmcb->save.gs = from_vmcb->save.gs;
0935     to_vmcb->save.tr = from_vmcb->save.tr;
0936     to_vmcb->save.ldtr = from_vmcb->save.ldtr;
0937     to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
0938     to_vmcb->save.star = from_vmcb->save.star;
0939     to_vmcb->save.lstar = from_vmcb->save.lstar;
0940     to_vmcb->save.cstar = from_vmcb->save.cstar;
0941     to_vmcb->save.sfmask = from_vmcb->save.sfmask;
0942     to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
0943     to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
0944     to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
0945 }
0946 
0947 int nested_svm_vmexit(struct vcpu_svm *svm)
0948 {
0949     struct kvm_vcpu *vcpu = &svm->vcpu;
0950     struct vmcb *vmcb01 = svm->vmcb01.ptr;
0951     struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
0952     struct vmcb *vmcb12;
0953     struct kvm_host_map map;
0954     int rc;
0955 
0956     rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
0957     if (rc) {
0958         if (rc == -EINVAL)
0959             kvm_inject_gp(vcpu, 0);
0960         return 1;
0961     }
0962 
0963     vmcb12 = map.hva;
0964 
0965     /* Exit Guest-Mode */
0966     leave_guest_mode(vcpu);
0967     svm->nested.vmcb12_gpa = 0;
0968     WARN_ON_ONCE(svm->nested.nested_run_pending);
0969 
0970     kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
0971 
0972     /* in case we halted in L2 */
0973     svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
0974 
0975     /* Give the current vmcb to the guest */
0976 
0977     vmcb12->save.es     = vmcb02->save.es;
0978     vmcb12->save.cs     = vmcb02->save.cs;
0979     vmcb12->save.ss     = vmcb02->save.ss;
0980     vmcb12->save.ds     = vmcb02->save.ds;
0981     vmcb12->save.gdtr   = vmcb02->save.gdtr;
0982     vmcb12->save.idtr   = vmcb02->save.idtr;
0983     vmcb12->save.efer   = svm->vcpu.arch.efer;
0984     vmcb12->save.cr0    = kvm_read_cr0(vcpu);
0985     vmcb12->save.cr3    = kvm_read_cr3(vcpu);
0986     vmcb12->save.cr2    = vmcb02->save.cr2;
0987     vmcb12->save.cr4    = svm->vcpu.arch.cr4;
0988     vmcb12->save.rflags = kvm_get_rflags(vcpu);
0989     vmcb12->save.rip    = kvm_rip_read(vcpu);
0990     vmcb12->save.rsp    = kvm_rsp_read(vcpu);
0991     vmcb12->save.rax    = kvm_rax_read(vcpu);
0992     vmcb12->save.dr7    = vmcb02->save.dr7;
0993     vmcb12->save.dr6    = svm->vcpu.arch.dr6;
0994     vmcb12->save.cpl    = vmcb02->save.cpl;
0995 
0996     vmcb12->control.int_state         = vmcb02->control.int_state;
0997     vmcb12->control.exit_code         = vmcb02->control.exit_code;
0998     vmcb12->control.exit_code_hi      = vmcb02->control.exit_code_hi;
0999     vmcb12->control.exit_info_1       = vmcb02->control.exit_info_1;
1000     vmcb12->control.exit_info_2       = vmcb02->control.exit_info_2;
1001 
1002     if (vmcb12->control.exit_code != SVM_EXIT_ERR)
1003         nested_save_pending_event_to_vmcb12(svm, vmcb12);
1004 
1005     if (svm->nrips_enabled)
1006         vmcb12->control.next_rip  = vmcb02->control.next_rip;
1007 
1008     vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
1009     vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
1010     vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
1011     vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
1012 
1013     if (!kvm_pause_in_guest(vcpu->kvm)) {
1014         vmcb01->control.pause_filter_count = vmcb02->control.pause_filter_count;
1015         vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1016 
1017     }
1018 
1019     nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
1020 
1021     svm_switch_vmcb(svm, &svm->vmcb01);
1022 
1023     if (unlikely(svm->lbrv_enabled && (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
1024         svm_copy_lbrs(vmcb12, vmcb02);
1025         svm_update_lbrv(vcpu);
1026     } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
1027         svm_copy_lbrs(vmcb01, vmcb02);
1028         svm_update_lbrv(vcpu);
1029     }
1030 
1031     /*
1032      * On vmexit the  GIF is set to false and
1033      * no event can be injected in L1.
1034      */
1035     svm_set_gif(svm, false);
1036     vmcb01->control.exit_int_info = 0;
1037 
1038     svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
1039     if (vmcb01->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
1040         vmcb01->control.tsc_offset = svm->vcpu.arch.tsc_offset;
1041         vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1042     }
1043 
1044     if (svm->tsc_ratio_msr != kvm_caps.default_tsc_scaling_ratio) {
1045         WARN_ON(!svm->tsc_scaling_enabled);
1046         vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
1047         __svm_write_tsc_multiplier(vcpu->arch.tsc_scaling_ratio);
1048     }
1049 
1050     svm->nested.ctl.nested_cr3 = 0;
1051 
1052     /*
1053      * Restore processor state that had been saved in vmcb01
1054      */
1055     kvm_set_rflags(vcpu, vmcb01->save.rflags);
1056     svm_set_efer(vcpu, vmcb01->save.efer);
1057     svm_set_cr0(vcpu, vmcb01->save.cr0 | X86_CR0_PE);
1058     svm_set_cr4(vcpu, vmcb01->save.cr4);
1059     kvm_rax_write(vcpu, vmcb01->save.rax);
1060     kvm_rsp_write(vcpu, vmcb01->save.rsp);
1061     kvm_rip_write(vcpu, vmcb01->save.rip);
1062 
1063     svm->vcpu.arch.dr7 = DR7_FIXED_1;
1064     kvm_update_dr7(&svm->vcpu);
1065 
1066     trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
1067                        vmcb12->control.exit_info_1,
1068                        vmcb12->control.exit_info_2,
1069                        vmcb12->control.exit_int_info,
1070                        vmcb12->control.exit_int_info_err,
1071                        KVM_ISA_SVM);
1072 
1073     kvm_vcpu_unmap(vcpu, &map, true);
1074 
1075     nested_svm_transition_tlb_flush(vcpu);
1076 
1077     nested_svm_uninit_mmu_context(vcpu);
1078 
1079     rc = nested_svm_load_cr3(vcpu, vmcb01->save.cr3, false, true);
1080     if (rc)
1081         return 1;
1082 
1083     /*
1084      * Drop what we picked up for L2 via svm_complete_interrupts() so it
1085      * doesn't end up in L1.
1086      */
1087     svm->vcpu.arch.nmi_injected = false;
1088     kvm_clear_exception_queue(vcpu);
1089     kvm_clear_interrupt_queue(vcpu);
1090 
1091     /*
1092      * If we are here following the completion of a VMRUN that
1093      * is being single-stepped, queue the pending #DB intercept
1094      * right now so that it an be accounted for before we execute
1095      * L1's next instruction.
1096      */
1097     if (unlikely(vmcb01->save.rflags & X86_EFLAGS_TF))
1098         kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
1099 
1100     /*
1101      * Un-inhibit the AVIC right away, so that other vCPUs can start
1102      * to benefit from it right away.
1103      */
1104     if (kvm_apicv_activated(vcpu->kvm))
1105         kvm_vcpu_update_apicv(vcpu);
1106 
1107     return 0;
1108 }
1109 
1110 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
1111 {
1112     nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
1113 }
1114 
1115 int svm_allocate_nested(struct vcpu_svm *svm)
1116 {
1117     struct page *vmcb02_page;
1118 
1119     if (svm->nested.initialized)
1120         return 0;
1121 
1122     vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1123     if (!vmcb02_page)
1124         return -ENOMEM;
1125     svm->nested.vmcb02.ptr = page_address(vmcb02_page);
1126     svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
1127 
1128     svm->nested.msrpm = svm_vcpu_alloc_msrpm();
1129     if (!svm->nested.msrpm)
1130         goto err_free_vmcb02;
1131     svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
1132 
1133     svm->nested.initialized = true;
1134     return 0;
1135 
1136 err_free_vmcb02:
1137     __free_page(vmcb02_page);
1138     return -ENOMEM;
1139 }
1140 
1141 void svm_free_nested(struct vcpu_svm *svm)
1142 {
1143     if (!svm->nested.initialized)
1144         return;
1145 
1146     svm_vcpu_free_msrpm(svm->nested.msrpm);
1147     svm->nested.msrpm = NULL;
1148 
1149     __free_page(virt_to_page(svm->nested.vmcb02.ptr));
1150     svm->nested.vmcb02.ptr = NULL;
1151 
1152     /*
1153      * When last_vmcb12_gpa matches the current vmcb12 gpa,
1154      * some vmcb12 fields are not loaded if they are marked clean
1155      * in the vmcb12, since in this case they are up to date already.
1156      *
1157      * When the vmcb02 is freed, this optimization becomes invalid.
1158      */
1159     svm->nested.last_vmcb12_gpa = INVALID_GPA;
1160 
1161     svm->nested.initialized = false;
1162 }
1163 
1164 /*
1165  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
1166  */
1167 void svm_leave_nested(struct kvm_vcpu *vcpu)
1168 {
1169     struct vcpu_svm *svm = to_svm(vcpu);
1170 
1171     if (is_guest_mode(vcpu)) {
1172         svm->nested.nested_run_pending = 0;
1173         svm->nested.vmcb12_gpa = INVALID_GPA;
1174 
1175         leave_guest_mode(vcpu);
1176 
1177         svm_switch_vmcb(svm, &svm->vmcb01);
1178 
1179         nested_svm_uninit_mmu_context(vcpu);
1180         vmcb_mark_all_dirty(svm->vmcb);
1181     }
1182 
1183     kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1184 }
1185 
1186 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1187 {
1188     u32 offset, msr, value;
1189     int write, mask;
1190 
1191     if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
1192         return NESTED_EXIT_HOST;
1193 
1194     msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1195     offset = svm_msrpm_offset(msr);
1196     write  = svm->vmcb->control.exit_info_1 & 1;
1197     mask   = 1 << ((2 * (msr & 0xf)) + write);
1198 
1199     if (offset == MSR_INVALID)
1200         return NESTED_EXIT_DONE;
1201 
1202     /* Offset is in 32 bit units but need in 8 bit units */
1203     offset *= 4;
1204 
1205     if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
1206         return NESTED_EXIT_DONE;
1207 
1208     return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1209 }
1210 
1211 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1212 {
1213     unsigned port, size, iopm_len;
1214     u16 val, mask;
1215     u8 start_bit;
1216     u64 gpa;
1217 
1218     if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
1219         return NESTED_EXIT_HOST;
1220 
1221     port = svm->vmcb->control.exit_info_1 >> 16;
1222     size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1223         SVM_IOIO_SIZE_SHIFT;
1224     gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1225     start_bit = port % 8;
1226     iopm_len = (start_bit + size > 8) ? 2 : 1;
1227     mask = (0xf >> (4 - size)) << start_bit;
1228     val = 0;
1229 
1230     if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1231         return NESTED_EXIT_DONE;
1232 
1233     return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1234 }
1235 
1236 static int nested_svm_intercept(struct vcpu_svm *svm)
1237 {
1238     u32 exit_code = svm->vmcb->control.exit_code;
1239     int vmexit = NESTED_EXIT_HOST;
1240 
1241     switch (exit_code) {
1242     case SVM_EXIT_MSR:
1243         vmexit = nested_svm_exit_handled_msr(svm);
1244         break;
1245     case SVM_EXIT_IOIO:
1246         vmexit = nested_svm_intercept_ioio(svm);
1247         break;
1248     case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1249         if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1250             vmexit = NESTED_EXIT_DONE;
1251         break;
1252     }
1253     case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1254         if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1255             vmexit = NESTED_EXIT_DONE;
1256         break;
1257     }
1258     case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1259         /*
1260          * Host-intercepted exceptions have been checked already in
1261          * nested_svm_exit_special.  There is nothing to do here,
1262          * the vmexit is injected by svm_check_nested_events.
1263          */
1264         vmexit = NESTED_EXIT_DONE;
1265         break;
1266     }
1267     case SVM_EXIT_ERR: {
1268         vmexit = NESTED_EXIT_DONE;
1269         break;
1270     }
1271     default: {
1272         if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1273             vmexit = NESTED_EXIT_DONE;
1274     }
1275     }
1276 
1277     return vmexit;
1278 }
1279 
1280 int nested_svm_exit_handled(struct vcpu_svm *svm)
1281 {
1282     int vmexit;
1283 
1284     vmexit = nested_svm_intercept(svm);
1285 
1286     if (vmexit == NESTED_EXIT_DONE)
1287         nested_svm_vmexit(svm);
1288 
1289     return vmexit;
1290 }
1291 
1292 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1293 {
1294     if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1295         kvm_queue_exception(vcpu, UD_VECTOR);
1296         return 1;
1297     }
1298 
1299     if (to_svm(vcpu)->vmcb->save.cpl) {
1300         kvm_inject_gp(vcpu, 0);
1301         return 1;
1302     }
1303 
1304     return 0;
1305 }
1306 
1307 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1308 {
1309     unsigned int nr = svm->vcpu.arch.exception.nr;
1310 
1311     return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1312 }
1313 
1314 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1315 {
1316     unsigned int nr = svm->vcpu.arch.exception.nr;
1317     struct vmcb *vmcb = svm->vmcb;
1318 
1319     vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1320     vmcb->control.exit_code_hi = 0;
1321 
1322     if (svm->vcpu.arch.exception.has_error_code)
1323         vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1324 
1325     /*
1326      * EXITINFO2 is undefined for all exception intercepts other
1327      * than #PF.
1328      */
1329     if (nr == PF_VECTOR) {
1330         if (svm->vcpu.arch.exception.nested_apf)
1331             vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1332         else if (svm->vcpu.arch.exception.has_payload)
1333             vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1334         else
1335             vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1336     } else if (nr == DB_VECTOR) {
1337         /* See inject_pending_event.  */
1338         kvm_deliver_exception_payload(&svm->vcpu);
1339         if (svm->vcpu.arch.dr7 & DR7_GD) {
1340             svm->vcpu.arch.dr7 &= ~DR7_GD;
1341             kvm_update_dr7(&svm->vcpu);
1342         }
1343     } else
1344         WARN_ON(svm->vcpu.arch.exception.has_payload);
1345 
1346     nested_svm_vmexit(svm);
1347 }
1348 
1349 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1350 {
1351     return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1352 }
1353 
1354 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1355 {
1356     struct vcpu_svm *svm = to_svm(vcpu);
1357     bool block_nested_events =
1358         kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1359     struct kvm_lapic *apic = vcpu->arch.apic;
1360 
1361     if (lapic_in_kernel(vcpu) &&
1362         test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1363         if (block_nested_events)
1364             return -EBUSY;
1365         if (!nested_exit_on_init(svm))
1366             return 0;
1367         nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1368         return 0;
1369     }
1370 
1371     if (vcpu->arch.exception.pending) {
1372         /*
1373          * Only a pending nested run can block a pending exception.
1374          * Otherwise an injected NMI/interrupt should either be
1375          * lost or delivered to the nested hypervisor in the EXITINTINFO
1376          * vmcb field, while delivering the pending exception.
1377          */
1378         if (svm->nested.nested_run_pending)
1379                         return -EBUSY;
1380         if (!nested_exit_on_exception(svm))
1381             return 0;
1382         nested_svm_inject_exception_vmexit(svm);
1383         return 0;
1384     }
1385 
1386     if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1387         if (block_nested_events)
1388             return -EBUSY;
1389         if (!nested_exit_on_smi(svm))
1390             return 0;
1391         nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1392         return 0;
1393     }
1394 
1395     if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1396         if (block_nested_events)
1397             return -EBUSY;
1398         if (!nested_exit_on_nmi(svm))
1399             return 0;
1400         nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1401         return 0;
1402     }
1403 
1404     if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1405         if (block_nested_events)
1406             return -EBUSY;
1407         if (!nested_exit_on_intr(svm))
1408             return 0;
1409         trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1410         nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1411         return 0;
1412     }
1413 
1414     return 0;
1415 }
1416 
1417 int nested_svm_exit_special(struct vcpu_svm *svm)
1418 {
1419     u32 exit_code = svm->vmcb->control.exit_code;
1420 
1421     switch (exit_code) {
1422     case SVM_EXIT_INTR:
1423     case SVM_EXIT_NMI:
1424     case SVM_EXIT_NPF:
1425         return NESTED_EXIT_HOST;
1426     case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1427         u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1428 
1429         if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1430             excp_bits)
1431             return NESTED_EXIT_HOST;
1432         else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1433              svm->vcpu.arch.apf.host_apf_flags)
1434             /* Trap async PF even if not shadowing */
1435             return NESTED_EXIT_HOST;
1436         break;
1437     }
1438     default:
1439         break;
1440     }
1441 
1442     return NESTED_EXIT_CONTINUE;
1443 }
1444 
1445 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
1446 {
1447     struct vcpu_svm *svm = to_svm(vcpu);
1448 
1449     vcpu->arch.tsc_scaling_ratio =
1450         kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
1451                            svm->tsc_ratio_msr);
1452     __svm_write_tsc_multiplier(vcpu->arch.tsc_scaling_ratio);
1453 }
1454 
1455 /* Inverse operation of nested_copy_vmcb_control_to_cache(). asid is copied too. */
1456 static void nested_copy_vmcb_cache_to_control(struct vmcb_control_area *dst,
1457                           struct vmcb_ctrl_area_cached *from)
1458 {
1459     unsigned int i;
1460 
1461     memset(dst, 0, sizeof(struct vmcb_control_area));
1462 
1463     for (i = 0; i < MAX_INTERCEPT; i++)
1464         dst->intercepts[i] = from->intercepts[i];
1465 
1466     dst->iopm_base_pa         = from->iopm_base_pa;
1467     dst->msrpm_base_pa        = from->msrpm_base_pa;
1468     dst->tsc_offset           = from->tsc_offset;
1469     dst->asid                 = from->asid;
1470     dst->tlb_ctl              = from->tlb_ctl;
1471     dst->int_ctl              = from->int_ctl;
1472     dst->int_vector           = from->int_vector;
1473     dst->int_state            = from->int_state;
1474     dst->exit_code            = from->exit_code;
1475     dst->exit_code_hi         = from->exit_code_hi;
1476     dst->exit_info_1          = from->exit_info_1;
1477     dst->exit_info_2          = from->exit_info_2;
1478     dst->exit_int_info        = from->exit_int_info;
1479     dst->exit_int_info_err    = from->exit_int_info_err;
1480     dst->nested_ctl           = from->nested_ctl;
1481     dst->event_inj            = from->event_inj;
1482     dst->event_inj_err        = from->event_inj_err;
1483     dst->next_rip             = from->next_rip;
1484     dst->nested_cr3           = from->nested_cr3;
1485     dst->virt_ext              = from->virt_ext;
1486     dst->pause_filter_count   = from->pause_filter_count;
1487     dst->pause_filter_thresh  = from->pause_filter_thresh;
1488     /* 'clean' and 'reserved_sw' are not changed by KVM */
1489 }
1490 
1491 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1492                 struct kvm_nested_state __user *user_kvm_nested_state,
1493                 u32 user_data_size)
1494 {
1495     struct vcpu_svm *svm;
1496     struct vmcb_control_area *ctl;
1497     unsigned long r;
1498     struct kvm_nested_state kvm_state = {
1499         .flags = 0,
1500         .format = KVM_STATE_NESTED_FORMAT_SVM,
1501         .size = sizeof(kvm_state),
1502     };
1503     struct vmcb __user *user_vmcb = (struct vmcb __user *)
1504         &user_kvm_nested_state->data.svm[0];
1505 
1506     if (!vcpu)
1507         return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1508 
1509     svm = to_svm(vcpu);
1510 
1511     if (user_data_size < kvm_state.size)
1512         goto out;
1513 
1514     /* First fill in the header and copy it out.  */
1515     if (is_guest_mode(vcpu)) {
1516         kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1517         kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1518         kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1519 
1520         if (svm->nested.nested_run_pending)
1521             kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1522     }
1523 
1524     if (gif_set(svm))
1525         kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1526 
1527     if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1528         return -EFAULT;
1529 
1530     if (!is_guest_mode(vcpu))
1531         goto out;
1532 
1533     /*
1534      * Copy over the full size of the VMCB rather than just the size
1535      * of the structs.
1536      */
1537     if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1538         return -EFAULT;
1539 
1540     ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
1541     if (!ctl)
1542         return -ENOMEM;
1543 
1544     nested_copy_vmcb_cache_to_control(ctl, &svm->nested.ctl);
1545     r = copy_to_user(&user_vmcb->control, ctl,
1546              sizeof(user_vmcb->control));
1547     kfree(ctl);
1548     if (r)
1549         return -EFAULT;
1550 
1551     if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1552              sizeof(user_vmcb->save)))
1553         return -EFAULT;
1554 out:
1555     return kvm_state.size;
1556 }
1557 
1558 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1559                 struct kvm_nested_state __user *user_kvm_nested_state,
1560                 struct kvm_nested_state *kvm_state)
1561 {
1562     struct vcpu_svm *svm = to_svm(vcpu);
1563     struct vmcb __user *user_vmcb = (struct vmcb __user *)
1564         &user_kvm_nested_state->data.svm[0];
1565     struct vmcb_control_area *ctl;
1566     struct vmcb_save_area *save;
1567     struct vmcb_save_area_cached save_cached;
1568     struct vmcb_ctrl_area_cached ctl_cached;
1569     unsigned long cr0;
1570     int ret;
1571 
1572     BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1573              KVM_STATE_NESTED_SVM_VMCB_SIZE);
1574 
1575     if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1576         return -EINVAL;
1577 
1578     if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1579                  KVM_STATE_NESTED_RUN_PENDING |
1580                  KVM_STATE_NESTED_GIF_SET))
1581         return -EINVAL;
1582 
1583     /*
1584      * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1585      * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1586      */
1587     if (!(vcpu->arch.efer & EFER_SVME)) {
1588         /* GIF=1 and no guest mode are required if SVME=0.  */
1589         if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1590             return -EINVAL;
1591     }
1592 
1593     /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1594     if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1595         return -EINVAL;
1596 
1597     if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1598         svm_leave_nested(vcpu);
1599         svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1600         return 0;
1601     }
1602 
1603     if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1604         return -EINVAL;
1605     if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1606         return -EINVAL;
1607 
1608     ret  = -ENOMEM;
1609     ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1610     save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1611     if (!ctl || !save)
1612         goto out_free;
1613 
1614     ret = -EFAULT;
1615     if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1616         goto out_free;
1617     if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1618         goto out_free;
1619 
1620     ret = -EINVAL;
1621     __nested_copy_vmcb_control_to_cache(vcpu, &ctl_cached, ctl);
1622     if (!__nested_vmcb_check_controls(vcpu, &ctl_cached))
1623         goto out_free;
1624 
1625     /*
1626      * Processor state contains L2 state.  Check that it is
1627      * valid for guest mode (see nested_vmcb_check_save).
1628      */
1629     cr0 = kvm_read_cr0(vcpu);
1630         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1631         goto out_free;
1632 
1633     /*
1634      * Validate host state saved from before VMRUN (see
1635      * nested_svm_check_permissions).
1636      */
1637     __nested_copy_vmcb_save_to_cache(&save_cached, save);
1638     if (!(save->cr0 & X86_CR0_PG) ||
1639         !(save->cr0 & X86_CR0_PE) ||
1640         (save->rflags & X86_EFLAGS_VM) ||
1641         !__nested_vmcb_check_save(vcpu, &save_cached))
1642         goto out_free;
1643 
1644 
1645     /*
1646      * All checks done, we can enter guest mode. Userspace provides
1647      * vmcb12.control, which will be combined with L1 and stored into
1648      * vmcb02, and the L1 save state which we store in vmcb01.
1649      * L2 registers if needed are moved from the current VMCB to VMCB02.
1650      */
1651 
1652     if (is_guest_mode(vcpu))
1653         svm_leave_nested(vcpu);
1654     else
1655         svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1656 
1657     svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1658 
1659     svm->nested.nested_run_pending =
1660         !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1661 
1662     svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1663 
1664     svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1665     nested_copy_vmcb_control_to_cache(svm, ctl);
1666 
1667     svm_switch_vmcb(svm, &svm->nested.vmcb02);
1668     nested_vmcb02_prepare_control(svm, svm->vmcb->save.rip, svm->vmcb->save.cs.base);
1669 
1670     /*
1671      * While the nested guest CR3 is already checked and set by
1672      * KVM_SET_SREGS, it was set when nested state was yet loaded,
1673      * thus MMU might not be initialized correctly.
1674      * Set it again to fix this.
1675      */
1676 
1677     ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1678                   nested_npt_enabled(svm), false);
1679     if (WARN_ON_ONCE(ret))
1680         goto out_free;
1681 
1682     svm->nested.force_msr_bitmap_recalc = true;
1683 
1684     kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1685     ret = 0;
1686 out_free:
1687     kfree(save);
1688     kfree(ctl);
1689 
1690     return ret;
1691 }
1692 
1693 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1694 {
1695     struct vcpu_svm *svm = to_svm(vcpu);
1696 
1697     if (WARN_ON(!is_guest_mode(vcpu)))
1698         return true;
1699 
1700     if (!vcpu->arch.pdptrs_from_userspace &&
1701         !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1702         /*
1703          * Reload the guest's PDPTRs since after a migration
1704          * the guest CR3 might be restored prior to setting the nested
1705          * state which can lead to a load of wrong PDPTRs.
1706          */
1707         if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
1708             return false;
1709 
1710     if (!nested_svm_vmrun_msrpm(svm)) {
1711         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1712         vcpu->run->internal.suberror =
1713             KVM_INTERNAL_ERROR_EMULATION;
1714         vcpu->run->internal.ndata = 0;
1715         return false;
1716     }
1717 
1718     return true;
1719 }
1720 
1721 struct kvm_x86_nested_ops svm_nested_ops = {
1722     .leave_nested = svm_leave_nested,
1723     .check_events = svm_check_nested_events,
1724     .handle_page_fault_workaround = nested_svm_handle_page_fault_workaround,
1725     .triple_fault = nested_svm_triple_fault,
1726     .get_nested_state_pages = svm_get_nested_state_pages,
1727     .get_state = svm_get_nested_state,
1728     .set_state = svm_set_nested_state,
1729 };
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!