arch/x86/kvm/svm/nested.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Kernel-based Virtual Machine driver for Linux
   4  *
   5  * AMD SVM support
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *   Avi Kivity   <avi@qumranet.com>
  13  */
  14
  15 #define pr_fmt(fmt) "SVM: " fmt
  16
  17 #include <linux/kvm_types.h>
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20
  21 #include <asm/msr-index.h>
  22 #include <asm/debugreg.h>
  23
  24 #include "kvm_emulate.h"
  25 #include "trace.h"
  26 #include "mmu.h"
  27 #include "x86.h"
  28 #include "cpuid.h"
  29 #include "lapic.h"
  30 #include "svm.h"
  31 #include "hyperv.h"
  32
  33 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
  34
  35 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  36                                        struct x86_exception *fault)
  37 {
  38         struct vcpu_svm *svm = to_svm(vcpu);
  39         struct vmcb *vmcb = svm->vmcb;
  40
  41         if (vmcb->control.exit_code != SVM_EXIT_NPF) {
  42                 /*
  43                  * TODO: track the cause of the nested page fault, and
  44                  * correctly fill in the high bits of exit_info_1.
  45                  */
  46                 vmcb->control.exit_code = SVM_EXIT_NPF;
  47                 vmcb->control.exit_code_hi = 0;
  48                 vmcb->control.exit_info_1 = (1ULL << 32);
  49                 vmcb->control.exit_info_2 = fault->address;
  50         }
  51
  52         vmcb->control.exit_info_1 &= ~0xffffffffULL;
  53         vmcb->control.exit_info_1 |= fault->error_code;
  54
  55         nested_svm_vmexit(svm);
  56 }
  57
  58 static bool nested_svm_handle_page_fault_workaround(struct kvm_vcpu *vcpu,
  59                                                     struct x86_exception *fault)
  60 {
  61         struct vcpu_svm *svm = to_svm(vcpu);
  62         struct vmcb *vmcb = svm->vmcb;
  63
  64         WARN_ON(!is_guest_mode(vcpu));
  65
  66         if (vmcb12_is_intercept(&svm->nested.ctl,
  67                                 INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
  68             !WARN_ON_ONCE(svm->nested.nested_run_pending)) {
  69                 vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
  70                 vmcb->control.exit_code_hi = 0;
  71                 vmcb->control.exit_info_1 = fault->error_code;
  72                 vmcb->control.exit_info_2 = fault->address;
  73                 nested_svm_vmexit(svm);
  74                 return true;
  75         }
  76
  77         return false;
  78 }
  79
  80 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  81 {
  82         struct vcpu_svm *svm = to_svm(vcpu);
  83         u64 cr3 = svm->nested.ctl.nested_cr3;
  84         u64 pdpte;
  85         int ret;
  86
  87         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
  88                                        offset_in_page(cr3) + index * 8, 8);
  89         if (ret)
  90                 return 0;
  91         return pdpte;
  92 }
  93
  94 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  95 {
  96         struct vcpu_svm *svm = to_svm(vcpu);
  97
  98         return svm->nested.ctl.nested_cr3;
  99 }
 100
 101 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
 102 {
 103         struct vcpu_svm *svm = to_svm(vcpu);
 104
 105         WARN_ON(mmu_is_nested(vcpu));
 106
 107         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
 108
 109         /*
 110          * The NPT format depends on L1's CR4 and EFER, which is in vmcb01.  Note,
 111          * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
 112          * vCPU state.  CR0.WP is explicitly ignored, while CR0.PG is required.
 113          */
 114         kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
 115                                 svm->vmcb01.ptr->save.efer,
 116                                 svm->nested.ctl.nested_cr3);
 117         vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
 118         vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
 119         vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
 120         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 121 }
 122
 123 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 124 {
 125         vcpu->arch.mmu = &vcpu->arch.root_mmu;
 126         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 127 }
 128
 129 static bool nested_vmcb_needs_vls_intercept(struct vcpu_svm *svm)
 130 {
 131         if (!svm->v_vmload_vmsave_enabled)
 132                 return true;
 133
 134         if (!nested_npt_enabled(svm))
 135                 return true;
 136
 137         if (!(svm->nested.ctl.virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK))
 138                 return true;
 139
 140         return false;
 141 }
 142
 143 void recalc_intercepts(struct vcpu_svm *svm)
 144 {
 145         struct vmcb_control_area *c, *h;
 146         struct vmcb_ctrl_area_cached *g;
 147         unsigned int i;
 148
 149         vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 150
 151         if (!is_guest_mode(&svm->vcpu))
 152                 return;
 153
 154         c = &svm->vmcb->control;
 155         h = &svm->vmcb01.ptr->control;
 156         g = &svm->nested.ctl;
 157
 158         for (i = 0; i < MAX_INTERCEPT; i++)
 159                 c->intercepts[i] = h->intercepts[i];
 160
 161         if (g->int_ctl & V_INTR_MASKING_MASK) {
 162                 /* We only want the cr8 intercept bits of L1 */
 163                 vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
 164                 vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
 165
 166                 /*
 167                  * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 168                  * affect any interrupt we may want to inject; therefore,
 169                  * interrupt window vmexits are irrelevant to L0.
 170                  */
 171                 vmcb_clr_intercept(c, INTERCEPT_VINTR);
 172         }
 173
 174         /* We don't want to see VMMCALLs from a nested guest */
 175         vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
 176
 177         for (i = 0; i < MAX_INTERCEPT; i++)
 178                 c->intercepts[i] |= g->intercepts[i];
 179
 180         /* If SMI is not intercepted, ignore guest SMI intercept as well  */
 181         if (!intercept_smi)
 182                 vmcb_clr_intercept(c, INTERCEPT_SMI);
 183
 184         if (nested_vmcb_needs_vls_intercept(svm)) {
 185                 /*
 186                  * If the virtual VMLOAD/VMSAVE is not enabled for the L2,
 187                  * we must intercept these instructions to correctly
 188                  * emulate them in case L1 doesn't intercept them.
 189                  */
 190                 vmcb_set_intercept(c, INTERCEPT_VMLOAD);
 191                 vmcb_set_intercept(c, INTERCEPT_VMSAVE);
 192         } else {
 193                 WARN_ON(!(c->virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK));
 194         }
 195 }
 196
 197 /*
 198  * Merge L0's (KVM) and L1's (Nested VMCB) MSR permission bitmaps. The function
 199  * is optimized in that it only merges the parts where KVM MSR permission bitmap
 200  * may contain zero bits.
 201  */
 202 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 203 {
 204         struct hv_enlightenments *hve =
 205                 (struct hv_enlightenments *)svm->nested.ctl.reserved_sw;
 206         int i;
 207
 208         /*
 209          * MSR bitmap update can be skipped when:
 210          * - MSR bitmap for L1 hasn't changed.
 211          * - Nested hypervisor (L1) is attempting to launch the same L2 as
 212          *   before.
 213          * - Nested hypervisor (L1) is using Hyper-V emulation interface and
 214          * tells KVM (L0) there were no changes in MSR bitmap for L2.
 215          */
 216         if (!svm->nested.force_msr_bitmap_recalc &&
 217             kvm_hv_hypercall_enabled(&svm->vcpu) &&
 218             hve->hv_enlightenments_control.msr_bitmap &&
 219             (svm->nested.ctl.clean & BIT(VMCB_HV_NESTED_ENLIGHTENMENTS)))
 220                 goto set_msrpm_base_pa;
 221
 222         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 223                 return true;
 224
 225         for (i = 0; i < MSRPM_OFFSETS; i++) {
 226                 u32 value, p;
 227                 u64 offset;
 228
 229                 if (msrpm_offsets[i] == 0xffffffff)
 230                         break;
 231
 232                 p      = msrpm_offsets[i];
 233                 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 234
 235                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 236                         return false;
 237
 238                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 239         }
 240
 241         svm->nested.force_msr_bitmap_recalc = false;
 242
 243 set_msrpm_base_pa:
 244         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 245
 246         return true;
 247 }
 248
 249 /*
 250  * Bits 11:0 of bitmap address are ignored by hardware
 251  */
 252 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
 253 {
 254         u64 addr = PAGE_ALIGN(pa);
 255
 256         return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
 257             kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
 258 }
 259
 260 static bool nested_svm_check_tlb_ctl(struct kvm_vcpu *vcpu, u8 tlb_ctl)
 261 {
 262         /* Nested FLUSHBYASID is not supported yet.  */
 263         switch(tlb_ctl) {
 264                 case TLB_CONTROL_DO_NOTHING:
 265                 case TLB_CONTROL_FLUSH_ALL_ASID:
 266                         return true;
 267                 default:
 268                         return false;
 269         }
 270 }
 271
 272 static bool __nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
 273                                          struct vmcb_ctrl_area_cached *control)
 274 {
 275         if (CC(!vmcb12_is_intercept(control, INTERCEPT_VMRUN)))
 276                 return false;
 277
 278         if (CC(control->asid == 0))
 279                 return false;
 280
 281         if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
 282                 return false;
 283
 284         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
 285                                            MSRPM_SIZE)))
 286                 return false;
 287         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
 288                                            IOPM_SIZE)))
 289                 return false;
 290
 291         if (CC(!nested_svm_check_tlb_ctl(vcpu, control->tlb_ctl)))
 292                 return false;
 293
 294         return true;
 295 }
 296
 297 /* Common checks that apply to both L1 and L2 state.  */
 298 static bool __nested_vmcb_check_save(struct kvm_vcpu *vcpu,
 299                                      struct vmcb_save_area_cached *save)
 300 {
 301         if (CC(!(save->efer & EFER_SVME)))
 302                 return false;
 303
 304         if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 305             CC(save->cr0 & ~0xffffffffULL))
 306                 return false;
 307
 308         if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 309                 return false;
 310
 311         /*
 312          * These checks are also performed by KVM_SET_SREGS,
 313          * except that EFER.LMA is not checked by SVM against
 314          * CR0.PG && EFER.LME.
 315          */
 316         if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
 317                 if (CC(!(save->cr4 & X86_CR4_PAE)) ||
 318                     CC(!(save->cr0 & X86_CR0_PE)) ||
 319                     CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
 320                         return false;
 321         }
 322
 323         if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
 324                 return false;
 325
 326         if (CC(!kvm_valid_efer(vcpu, save->efer)))
 327                 return false;
 328
 329         return true;
 330 }
 331
 332 static bool nested_vmcb_check_save(struct kvm_vcpu *vcpu)
 333 {
 334         struct vcpu_svm *svm = to_svm(vcpu);
 335         struct vmcb_save_area_cached *save = &svm->nested.save;
 336
 337         return __nested_vmcb_check_save(vcpu, save);
 338 }
 339
 340 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu)
 341 {
 342         struct vcpu_svm *svm = to_svm(vcpu);
 343         struct vmcb_ctrl_area_cached *ctl = &svm->nested.ctl;
 344
 345         return __nested_vmcb_check_controls(vcpu, ctl);
 346 }
 347
 348 static
 349 void __nested_copy_vmcb_control_to_cache(struct kvm_vcpu *vcpu,
 350                                          struct vmcb_ctrl_area_cached *to,
 351                                          struct vmcb_control_area *from)
 352 {
 353         unsigned int i;
 354
 355         for (i = 0; i < MAX_INTERCEPT; i++)
 356                 to->intercepts[i] = from->intercepts[i];
 357
 358         to->iopm_base_pa        = from->iopm_base_pa;
 359         to->msrpm_base_pa       = from->msrpm_base_pa;
 360         to->tsc_offset          = from->tsc_offset;
 361         to->tlb_ctl             = from->tlb_ctl;
 362         to->int_ctl             = from->int_ctl;
 363         to->int_vector          = from->int_vector;
 364         to->int_state           = from->int_state;
 365         to->exit_code           = from->exit_code;
 366         to->exit_code_hi        = from->exit_code_hi;
 367         to->exit_info_1         = from->exit_info_1;
 368         to->exit_info_2         = from->exit_info_2;
 369         to->exit_int_info       = from->exit_int_info;
 370         to->exit_int_info_err   = from->exit_int_info_err;
 371         to->nested_ctl          = from->nested_ctl;
 372         to->event_inj           = from->event_inj;
 373         to->event_inj_err       = from->event_inj_err;
 374         to->nested_cr3          = from->nested_cr3;
 375         to->virt_ext            = from->virt_ext;
 376         to->pause_filter_count  = from->pause_filter_count;
 377         to->pause_filter_thresh = from->pause_filter_thresh;
 378
 379         /* Copy asid here because nested_vmcb_check_controls will check it.  */
 380         to->asid           = from->asid;
 381         to->msrpm_base_pa &= ~0x0fffULL;
 382         to->iopm_base_pa  &= ~0x0fffULL;
 383
 384         /* Hyper-V extensions (Enlightened VMCB) */
 385         if (kvm_hv_hypercall_enabled(vcpu)) {
 386                 to->clean = from->clean;
 387                 memcpy(to->reserved_sw, from->reserved_sw,
 388                        sizeof(struct hv_enlightenments));
 389         }
 390 }
 391
 392 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
 393                                        struct vmcb_control_area *control)
 394 {
 395         __nested_copy_vmcb_control_to_cache(&svm->vcpu, &svm->nested.ctl, control);
 396 }
 397
 398 static void __nested_copy_vmcb_save_to_cache(struct vmcb_save_area_cached *to,
 399                                              struct vmcb_save_area *from)
 400 {
 401         /*
 402          * Copy only fields that are validated, as we need them
 403          * to avoid TOC/TOU races.
 404          */
 405         to->efer = from->efer;
 406         to->cr0 = from->cr0;
 407         to->cr3 = from->cr3;
 408         to->cr4 = from->cr4;
 409
 410         to->dr6 = from->dr6;
 411         to->dr7 = from->dr7;
 412 }
 413
 414 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
 415                                     struct vmcb_save_area *save)
 416 {
 417         __nested_copy_vmcb_save_to_cache(&svm->nested.save, save);
 418 }
 419
 420 /*
 421  * Synchronize fields that are written by the processor, so that
 422  * they can be copied back into the vmcb12.
 423  */
 424 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
 425 {
 426         u32 mask;
 427         svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 428         svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 429
 430         /* Only a few fields of int_ctl are written by the processor.  */
 431         mask = V_IRQ_MASK | V_TPR_MASK;
 432         if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
 433             svm_is_intercept(svm, INTERCEPT_VINTR)) {
 434                 /*
 435                  * In order to request an interrupt window, L0 is usurping
 436                  * svm->vmcb->control.int_ctl and possibly setting V_IRQ
 437                  * even if it was clear in L1's VMCB.  Restoring it would be
 438                  * wrong.  However, in this case V_IRQ will remain true until
 439                  * interrupt_window_interception calls svm_clear_vintr and
 440                  * restores int_ctl.  We can just leave it aside.
 441                  */
 442                 mask &= ~V_IRQ_MASK;
 443         }
 444
 445         if (nested_vgif_enabled(svm))
 446                 mask |= V_GIF_MASK;
 447
 448         svm->nested.ctl.int_ctl        &= ~mask;
 449         svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 450 }
 451
 452 /*
 453  * Transfer any event that L0 or L1 wanted to inject into L2 to
 454  * EXIT_INT_INFO.
 455  */
 456 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
 457                                                 struct vmcb *vmcb12)
 458 {
 459         struct kvm_vcpu *vcpu = &svm->vcpu;
 460         u32 exit_int_info = 0;
 461         unsigned int nr;
 462
 463         if (vcpu->arch.exception.injected) {
 464                 nr = vcpu->arch.exception.nr;
 465                 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 466
 467                 if (vcpu->arch.exception.has_error_code) {
 468                         exit_int_info |= SVM_EVTINJ_VALID_ERR;
 469                         vmcb12->control.exit_int_info_err =
 470                                 vcpu->arch.exception.error_code;
 471                 }
 472
 473         } else if (vcpu->arch.nmi_injected) {
 474                 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 475
 476         } else if (vcpu->arch.interrupt.injected) {
 477                 nr = vcpu->arch.interrupt.nr;
 478                 exit_int_info = nr | SVM_EVTINJ_VALID;
 479
 480                 if (vcpu->arch.interrupt.soft)
 481                         exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 482                 else
 483                         exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 484         }
 485
 486         vmcb12->control.exit_int_info = exit_int_info;
 487 }
 488
 489 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
 490 {
 491         /*
 492          * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
 493          * things to fix before this can be conditional:
 494          *
 495          *  - Flush TLBs for both L1 and L2 remote TLB flush
 496          *  - Honor L1's request to flush an ASID on nested VMRUN
 497          *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
 498          *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
 499          *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
 500          *
 501          * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
 502          *     NPT guest-physical mappings on VMRUN.
 503          */
 504         kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
 505         kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
 506 }
 507
 508 /*
 509  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 510  * if we are emulating VM-Entry into a guest with NPT enabled.
 511  */
 512 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 513                                bool nested_npt, bool reload_pdptrs)
 514 {
 515         if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
 516                 return -EINVAL;
 517
 518         if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
 519             CC(!load_pdptrs(vcpu, cr3)))
 520                 return -EINVAL;
 521
 522         vcpu->arch.cr3 = cr3;
 523
 524         /* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
 525         kvm_init_mmu(vcpu);
 526
 527         if (!nested_npt)
 528                 kvm_mmu_new_pgd(vcpu, cr3);
 529
 530         return 0;
 531 }
 532
 533 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
 534 {
 535         if (!svm->nested.vmcb02.ptr)
 536                 return;
 537
 538         /* FIXME: merge g_pat from vmcb01 and vmcb12.  */
 539         svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
 540 }
 541
 542 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 543 {
 544         bool new_vmcb12 = false;
 545         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 546         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 547
 548         nested_vmcb02_compute_g_pat(svm);
 549
 550         /* Load the nested guest state */
 551         if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
 552                 new_vmcb12 = true;
 553                 svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
 554                 svm->nested.force_msr_bitmap_recalc = true;
 555         }
 556
 557         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
 558                 vmcb02->save.es = vmcb12->save.es;
 559                 vmcb02->save.cs = vmcb12->save.cs;
 560                 vmcb02->save.ss = vmcb12->save.ss;
 561                 vmcb02->save.ds = vmcb12->save.ds;
 562                 vmcb02->save.cpl = vmcb12->save.cpl;
 563                 vmcb_mark_dirty(vmcb02, VMCB_SEG);
 564         }
 565
 566         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
 567                 vmcb02->save.gdtr = vmcb12->save.gdtr;
 568                 vmcb02->save.idtr = vmcb12->save.idtr;
 569                 vmcb_mark_dirty(vmcb02, VMCB_DT);
 570         }
 571
 572         kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
 573
 574         svm_set_efer(&svm->vcpu, svm->nested.save.efer);
 575
 576         svm_set_cr0(&svm->vcpu, svm->nested.save.cr0);
 577         svm_set_cr4(&svm->vcpu, svm->nested.save.cr4);
 578
 579         svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 580
 581         kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
 582         kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
 583         kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
 584
 585         /* In case we don't even reach vcpu_run, the fields are not updated */
 586         vmcb02->save.rax = vmcb12->save.rax;
 587         vmcb02->save.rsp = vmcb12->save.rsp;
 588         vmcb02->save.rip = vmcb12->save.rip;
 589
 590         /* These bits will be set properly on the first execution when new_vmc12 is true */
 591         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 592                 vmcb02->save.dr7 = svm->nested.save.dr7 | DR7_FIXED_1;
 593                 svm->vcpu.arch.dr6  = svm->nested.save.dr6 | DR6_ACTIVE_LOW;
 594                 vmcb_mark_dirty(vmcb02, VMCB_DR);
 595         }
 596
 597         if (unlikely(svm->lbrv_enabled && (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
 598                 /*
 599                  * Reserved bits of DEBUGCTL are ignored.  Be consistent with
 600                  * svm_set_msr's definition of reserved bits.
 601                  */
 602                 svm_copy_lbrs(vmcb02, vmcb12);
 603                 vmcb02->save.dbgctl &= ~DEBUGCTL_RESERVED_BITS;
 604                 svm_update_lbrv(&svm->vcpu);
 605
 606         } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
 607                 svm_copy_lbrs(vmcb02, vmcb01);
 608         }
 609 }
 610
 611 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
 612 {
 613         u32 int_ctl_vmcb01_bits = V_INTR_MASKING_MASK;
 614         u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
 615
 616         struct kvm_vcpu *vcpu = &svm->vcpu;
 617         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 618         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 619
 620         /*
 621          * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 622          * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 623          */
 624
 625         if (svm->vgif_enabled && (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK))
 626                 int_ctl_vmcb12_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 627         else
 628                 int_ctl_vmcb01_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 629
 630         /* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 631         vmcb02->control.nested_ctl = vmcb01->control.nested_ctl;
 632         vmcb02->control.iopm_base_pa = vmcb01->control.iopm_base_pa;
 633         vmcb02->control.msrpm_base_pa = vmcb01->control.msrpm_base_pa;
 634
 635         /* Done at vmrun: asid.  */
 636
 637         /* Also overwritten later if necessary.  */
 638         vmcb02->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 639
 640         /* nested_cr3.  */
 641         if (nested_npt_enabled(svm))
 642                 nested_svm_init_mmu_context(vcpu);
 643
 644         vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
 645                         vcpu->arch.l1_tsc_offset,
 646                         svm->nested.ctl.tsc_offset,
 647                         svm->tsc_ratio_msr);
 648
 649         vmcb02->control.tsc_offset = vcpu->arch.tsc_offset;
 650
 651         if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
 652                 WARN_ON(!svm->tsc_scaling_enabled);
 653                 nested_svm_update_tsc_ratio_msr(vcpu);
 654         }
 655
 656         vmcb02->control.int_ctl             =
 657                 (svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
 658                 (vmcb01->control.int_ctl & int_ctl_vmcb01_bits);
 659
 660         vmcb02->control.int_vector          = svm->nested.ctl.int_vector;
 661         vmcb02->control.int_state           = svm->nested.ctl.int_state;
 662         vmcb02->control.event_inj           = svm->nested.ctl.event_inj;
 663         vmcb02->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 664
 665         vmcb02->control.virt_ext            = vmcb01->control.virt_ext &
 666                                               LBR_CTL_ENABLE_MASK;
 667         if (svm->lbrv_enabled)
 668                 vmcb02->control.virt_ext  |=
 669                         (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK);
 670
 671         if (!nested_vmcb_needs_vls_intercept(svm))
 672                 vmcb02->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
 673
 674         if (kvm_pause_in_guest(svm->vcpu.kvm)) {
 675                 /* use guest values since host doesn't use them */
 676                 vmcb02->control.pause_filter_count =
 677                                 svm->pause_filter_enabled ?
 678                                 svm->nested.ctl.pause_filter_count : 0;
 679
 680                 vmcb02->control.pause_filter_thresh =
 681                                 svm->pause_threshold_enabled ?
 682                                 svm->nested.ctl.pause_filter_thresh : 0;
 683
 684         } else if (!vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_PAUSE)) {
 685                 /* use host values when guest doesn't use them */
 686                 vmcb02->control.pause_filter_count = vmcb01->control.pause_filter_count;
 687                 vmcb02->control.pause_filter_thresh = vmcb01->control.pause_filter_thresh;
 688         } else {
 689                 /*
 690                  * Intercept every PAUSE otherwise and
 691                  * ignore both host and guest values
 692                  */
 693                 vmcb02->control.pause_filter_count = 0;
 694                 vmcb02->control.pause_filter_thresh = 0;
 695         }
 696
 697         nested_svm_transition_tlb_flush(vcpu);
 698
 699         /* Enter Guest-Mode */
 700         enter_guest_mode(vcpu);
 701
 702         /*
 703          * Merge guest and host intercepts - must be called with vcpu in
 704          * guest-mode to take effect.
 705          */
 706         recalc_intercepts(svm);
 707 }
 708
 709 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 710 {
 711         /*
 712          * Some VMCB state is shared between L1 and L2 and thus has to be
 713          * moved at the time of nested vmrun and vmexit.
 714          *
 715          * VMLOAD/VMSAVE state would also belong in this category, but KVM
 716          * always performs VMLOAD and VMSAVE from the VMCB01.
 717          */
 718         to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 719 }
 720
 721 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 722                          struct vmcb *vmcb12, bool from_vmrun)
 723 {
 724         struct vcpu_svm *svm = to_svm(vcpu);
 725         int ret;
 726
 727         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
 728                                vmcb12->save.rip,
 729                                vmcb12->control.int_ctl,
 730                                vmcb12->control.event_inj,
 731                                vmcb12->control.nested_ctl);
 732
 733         trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 734                                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 735                                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 736                                     vmcb12->control.intercepts[INTERCEPT_WORD3],
 737                                     vmcb12->control.intercepts[INTERCEPT_WORD4],
 738                                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
 739
 740
 741         svm->nested.vmcb12_gpa = vmcb12_gpa;
 742
 743         WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 744
 745         nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 746
 747         svm_switch_vmcb(svm, &svm->nested.vmcb02);
 748         nested_vmcb02_prepare_control(svm);
 749         nested_vmcb02_prepare_save(svm, vmcb12);
 750
 751         ret = nested_svm_load_cr3(&svm->vcpu, svm->nested.save.cr3,
 752                                   nested_npt_enabled(svm), from_vmrun);
 753         if (ret)
 754                 return ret;
 755
 756         if (!from_vmrun)
 757                 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 758
 759         svm_set_gif(svm, true);
 760
 761         if (kvm_vcpu_apicv_active(vcpu))
 762                 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
 763
 764         return 0;
 765 }
 766
 767 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 768 {
 769         struct vcpu_svm *svm = to_svm(vcpu);
 770         int ret;
 771         struct vmcb *vmcb12;
 772         struct kvm_host_map map;
 773         u64 vmcb12_gpa;
 774         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 775
 776         if (!svm->nested.hsave_msr) {
 777                 kvm_inject_gp(vcpu, 0);
 778                 return 1;
 779         }
 780
 781         if (is_smm(vcpu)) {
 782                 kvm_queue_exception(vcpu, UD_VECTOR);
 783                 return 1;
 784         }
 785
 786         vmcb12_gpa = svm->vmcb->save.rax;
 787         ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 788         if (ret == -EINVAL) {
 789                 kvm_inject_gp(vcpu, 0);
 790                 return 1;
 791         } else if (ret) {
 792                 return kvm_skip_emulated_instruction(vcpu);
 793         }
 794
 795         ret = kvm_skip_emulated_instruction(vcpu);
 796
 797         vmcb12 = map.hva;
 798
 799         if (WARN_ON_ONCE(!svm->nested.initialized))
 800                 return -EINVAL;
 801
 802         nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
 803         nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
 804
 805         if (!nested_vmcb_check_save(vcpu) ||
 806             !nested_vmcb_check_controls(vcpu)) {
 807                 vmcb12->control.exit_code    = SVM_EXIT_ERR;
 808                 vmcb12->control.exit_code_hi = 0;
 809                 vmcb12->control.exit_info_1  = 0;
 810                 vmcb12->control.exit_info_2  = 0;
 811                 goto out;
 812         }
 813
 814         /*
 815          * Since vmcb01 is not in use, we can use it to store some of the L1
 816          * state.
 817          */
 818         vmcb01->save.efer   = vcpu->arch.efer;
 819         vmcb01->save.cr0    = kvm_read_cr0(vcpu);
 820         vmcb01->save.cr4    = vcpu->arch.cr4;
 821         vmcb01->save.rflags = kvm_get_rflags(vcpu);
 822         vmcb01->save.rip    = kvm_rip_read(vcpu);
 823
 824         if (!npt_enabled)
 825                 vmcb01->save.cr3 = kvm_read_cr3(vcpu);
 826
 827         svm->nested.nested_run_pending = 1;
 828
 829         if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
 830                 goto out_exit_err;
 831
 832         if (nested_svm_vmrun_msrpm(svm))
 833                 goto out;
 834
 835 out_exit_err:
 836         svm->nested.nested_run_pending = 0;
 837
 838         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 839         svm->vmcb->control.exit_code_hi = 0;
 840         svm->vmcb->control.exit_info_1  = 0;
 841         svm->vmcb->control.exit_info_2  = 0;
 842
 843         nested_svm_vmexit(svm);
 844
 845 out:
 846         kvm_vcpu_unmap(vcpu, &map, true);
 847
 848         return ret;
 849 }
 850
 851 /* Copy state save area fields which are handled by VMRUN */
 852 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
 853                           struct vmcb_save_area *from_save)
 854 {
 855         to_save->es = from_save->es;
 856         to_save->cs = from_save->cs;
 857         to_save->ss = from_save->ss;
 858         to_save->ds = from_save->ds;
 859         to_save->gdtr = from_save->gdtr;
 860         to_save->idtr = from_save->idtr;
 861         to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
 862         to_save->efer = from_save->efer;
 863         to_save->cr0 = from_save->cr0;
 864         to_save->cr3 = from_save->cr3;
 865         to_save->cr4 = from_save->cr4;
 866         to_save->rax = from_save->rax;
 867         to_save->rsp = from_save->rsp;
 868         to_save->rip = from_save->rip;
 869         to_save->cpl = 0;
 870 }
 871
 872 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
 873 {
 874         to_vmcb->save.fs = from_vmcb->save.fs;
 875         to_vmcb->save.gs = from_vmcb->save.gs;
 876         to_vmcb->save.tr = from_vmcb->save.tr;
 877         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 878         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 879         to_vmcb->save.star = from_vmcb->save.star;
 880         to_vmcb->save.lstar = from_vmcb->save.lstar;
 881         to_vmcb->save.cstar = from_vmcb->save.cstar;
 882         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 883         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 884         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 885         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 886 }
 887
 888 int nested_svm_vmexit(struct vcpu_svm *svm)
 889 {
 890         struct kvm_vcpu *vcpu = &svm->vcpu;
 891         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 892         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 893         struct vmcb *vmcb12;
 894         struct kvm_host_map map;
 895         int rc;
 896
 897         rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 898         if (rc) {
 899                 if (rc == -EINVAL)
 900                         kvm_inject_gp(vcpu, 0);
 901                 return 1;
 902         }
 903
 904         vmcb12 = map.hva;
 905
 906         /* Exit Guest-Mode */
 907         leave_guest_mode(vcpu);
 908         svm->nested.vmcb12_gpa = 0;
 909         WARN_ON_ONCE(svm->nested.nested_run_pending);
 910
 911         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 912
 913         /* in case we halted in L2 */
 914         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 915
 916         /* Give the current vmcb to the guest */
 917
 918         vmcb12->save.es     = vmcb02->save.es;
 919         vmcb12->save.cs     = vmcb02->save.cs;
 920         vmcb12->save.ss     = vmcb02->save.ss;
 921         vmcb12->save.ds     = vmcb02->save.ds;
 922         vmcb12->save.gdtr   = vmcb02->save.gdtr;
 923         vmcb12->save.idtr   = vmcb02->save.idtr;
 924         vmcb12->save.efer   = svm->vcpu.arch.efer;
 925         vmcb12->save.cr0    = kvm_read_cr0(vcpu);
 926         vmcb12->save.cr3    = kvm_read_cr3(vcpu);
 927         vmcb12->save.cr2    = vmcb02->save.cr2;
 928         vmcb12->save.cr4    = svm->vcpu.arch.cr4;
 929         vmcb12->save.rflags = kvm_get_rflags(vcpu);
 930         vmcb12->save.rip    = kvm_rip_read(vcpu);
 931         vmcb12->save.rsp    = kvm_rsp_read(vcpu);
 932         vmcb12->save.rax    = kvm_rax_read(vcpu);
 933         vmcb12->save.dr7    = vmcb02->save.dr7;
 934         vmcb12->save.dr6    = svm->vcpu.arch.dr6;
 935         vmcb12->save.cpl    = vmcb02->save.cpl;
 936
 937         vmcb12->control.int_state         = vmcb02->control.int_state;
 938         vmcb12->control.exit_code         = vmcb02->control.exit_code;
 939         vmcb12->control.exit_code_hi      = vmcb02->control.exit_code_hi;
 940         vmcb12->control.exit_info_1       = vmcb02->control.exit_info_1;
 941         vmcb12->control.exit_info_2       = vmcb02->control.exit_info_2;
 942
 943         if (vmcb12->control.exit_code != SVM_EXIT_ERR)
 944                 nested_save_pending_event_to_vmcb12(svm, vmcb12);
 945
 946         if (svm->nrips_enabled)
 947                 vmcb12->control.next_rip  = vmcb02->control.next_rip;
 948
 949         vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
 950         vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 951         vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
 952         vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 953
 954         if (!kvm_pause_in_guest(vcpu->kvm) && vmcb02->control.pause_filter_count)
 955                 vmcb01->control.pause_filter_count = vmcb02->control.pause_filter_count;
 956
 957         nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
 958
 959         svm_switch_vmcb(svm, &svm->vmcb01);
 960
 961         if (unlikely(svm->lbrv_enabled && (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
 962                 svm_copy_lbrs(vmcb12, vmcb02);
 963                 svm_update_lbrv(vcpu);
 964         } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
 965                 svm_copy_lbrs(vmcb01, vmcb02);
 966                 svm_update_lbrv(vcpu);
 967         }
 968
 969         /*
 970          * On vmexit the  GIF is set to false and
 971          * no event can be injected in L1.
 972          */
 973         svm_set_gif(svm, false);
 974         vmcb01->control.exit_int_info = 0;
 975
 976         svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
 977         if (vmcb01->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
 978                 vmcb01->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 979                 vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
 980         }
 981
 982         if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
 983                 WARN_ON(!svm->tsc_scaling_enabled);
 984                 vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
 985                 __svm_write_tsc_multiplier(vcpu->arch.tsc_scaling_ratio);
 986         }
 987
 988         svm->nested.ctl.nested_cr3 = 0;
 989
 990         /*
 991          * Restore processor state that had been saved in vmcb01
 992          */
 993         kvm_set_rflags(vcpu, vmcb01->save.rflags);
 994         svm_set_efer(vcpu, vmcb01->save.efer);
 995         svm_set_cr0(vcpu, vmcb01->save.cr0 | X86_CR0_PE);
 996         svm_set_cr4(vcpu, vmcb01->save.cr4);
 997         kvm_rax_write(vcpu, vmcb01->save.rax);
 998         kvm_rsp_write(vcpu, vmcb01->save.rsp);
 999         kvm_rip_write(vcpu, vmcb01->save.rip);
1000
1001         svm->vcpu.arch.dr7 = DR7_FIXED_1;
1002         kvm_update_dr7(&svm->vcpu);
1003
1004         trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
1005                                        vmcb12->control.exit_info_1,
1006                                        vmcb12->control.exit_info_2,
1007                                        vmcb12->control.exit_int_info,
1008                                        vmcb12->control.exit_int_info_err,
1009                                        KVM_ISA_SVM);
1010
1011         kvm_vcpu_unmap(vcpu, &map, true);
1012
1013         nested_svm_transition_tlb_flush(vcpu);
1014
1015         nested_svm_uninit_mmu_context(vcpu);
1016
1017         rc = nested_svm_load_cr3(vcpu, vmcb01->save.cr3, false, true);
1018         if (rc)
1019                 return 1;
1020
1021         /*
1022          * Drop what we picked up for L2 via svm_complete_interrupts() so it
1023          * doesn't end up in L1.
1024          */
1025         svm->vcpu.arch.nmi_injected = false;
1026         kvm_clear_exception_queue(vcpu);
1027         kvm_clear_interrupt_queue(vcpu);
1028
1029         /*
1030          * If we are here following the completion of a VMRUN that
1031          * is being single-stepped, queue the pending #DB intercept
1032          * right now so that it an be accounted for before we execute
1033          * L1's next instruction.
1034          */
1035         if (unlikely(vmcb01->save.rflags & X86_EFLAGS_TF))
1036                 kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
1037
1038         /*
1039          * Un-inhibit the AVIC right away, so that other vCPUs can start
1040          * to benefit from it right away.
1041          */
1042         if (kvm_apicv_activated(vcpu->kvm))
1043                 kvm_vcpu_update_apicv(vcpu);
1044
1045         return 0;
1046 }
1047
1048 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
1049 {
1050         nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
1051 }
1052
1053 int svm_allocate_nested(struct vcpu_svm *svm)
1054 {
1055         struct page *vmcb02_page;
1056
1057         if (svm->nested.initialized)
1058                 return 0;
1059
1060         vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1061         if (!vmcb02_page)
1062                 return -ENOMEM;
1063         svm->nested.vmcb02.ptr = page_address(vmcb02_page);
1064         svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
1065
1066         svm->nested.msrpm = svm_vcpu_alloc_msrpm();
1067         if (!svm->nested.msrpm)
1068                 goto err_free_vmcb02;
1069         svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
1070
1071         svm->nested.initialized = true;
1072         return 0;
1073
1074 err_free_vmcb02:
1075         __free_page(vmcb02_page);
1076         return -ENOMEM;
1077 }
1078
1079 void svm_free_nested(struct vcpu_svm *svm)
1080 {
1081         if (!svm->nested.initialized)
1082                 return;
1083
1084         svm_vcpu_free_msrpm(svm->nested.msrpm);
1085         svm->nested.msrpm = NULL;
1086
1087         __free_page(virt_to_page(svm->nested.vmcb02.ptr));
1088         svm->nested.vmcb02.ptr = NULL;
1089
1090         /*
1091          * When last_vmcb12_gpa matches the current vmcb12 gpa,
1092          * some vmcb12 fields are not loaded if they are marked clean
1093          * in the vmcb12, since in this case they are up to date already.
1094          *
1095          * When the vmcb02 is freed, this optimization becomes invalid.
1096          */
1097         svm->nested.last_vmcb12_gpa = INVALID_GPA;
1098
1099         svm->nested.initialized = false;
1100 }
1101
1102 /*
1103  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
1104  */
1105 void svm_leave_nested(struct kvm_vcpu *vcpu)
1106 {
1107         struct vcpu_svm *svm = to_svm(vcpu);
1108
1109         if (is_guest_mode(vcpu)) {
1110                 svm->nested.nested_run_pending = 0;
1111                 svm->nested.vmcb12_gpa = INVALID_GPA;
1112
1113                 leave_guest_mode(vcpu);
1114
1115                 svm_switch_vmcb(svm, &svm->vmcb01);
1116
1117                 nested_svm_uninit_mmu_context(vcpu);
1118                 vmcb_mark_all_dirty(svm->vmcb);
1119         }
1120
1121         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1122 }
1123
1124 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1125 {
1126         u32 offset, msr, value;
1127         int write, mask;
1128
1129         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
1130                 return NESTED_EXIT_HOST;
1131
1132         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1133         offset = svm_msrpm_offset(msr);
1134         write  = svm->vmcb->control.exit_info_1 & 1;
1135         mask   = 1 << ((2 * (msr & 0xf)) + write);
1136
1137         if (offset == MSR_INVALID)
1138                 return NESTED_EXIT_DONE;
1139
1140         /* Offset is in 32 bit units but need in 8 bit units */
1141         offset *= 4;
1142
1143         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
1144                 return NESTED_EXIT_DONE;
1145
1146         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1147 }
1148
1149 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1150 {
1151         unsigned port, size, iopm_len;
1152         u16 val, mask;
1153         u8 start_bit;
1154         u64 gpa;
1155
1156         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
1157                 return NESTED_EXIT_HOST;
1158
1159         port = svm->vmcb->control.exit_info_1 >> 16;
1160         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1161                 SVM_IOIO_SIZE_SHIFT;
1162         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1163         start_bit = port % 8;
1164         iopm_len = (start_bit + size > 8) ? 2 : 1;
1165         mask = (0xf >> (4 - size)) << start_bit;
1166         val = 0;
1167
1168         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1169                 return NESTED_EXIT_DONE;
1170
1171         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1172 }
1173
1174 static int nested_svm_intercept(struct vcpu_svm *svm)
1175 {
1176         u32 exit_code = svm->vmcb->control.exit_code;
1177         int vmexit = NESTED_EXIT_HOST;
1178
1179         switch (exit_code) {
1180         case SVM_EXIT_MSR:
1181                 vmexit = nested_svm_exit_handled_msr(svm);
1182                 break;
1183         case SVM_EXIT_IOIO:
1184                 vmexit = nested_svm_intercept_ioio(svm);
1185                 break;
1186         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1187                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1188                         vmexit = NESTED_EXIT_DONE;
1189                 break;
1190         }
1191         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1192                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1193                         vmexit = NESTED_EXIT_DONE;
1194                 break;
1195         }
1196         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1197                 /*
1198                  * Host-intercepted exceptions have been checked already in
1199                  * nested_svm_exit_special.  There is nothing to do here,
1200                  * the vmexit is injected by svm_check_nested_events.
1201                  */
1202                 vmexit = NESTED_EXIT_DONE;
1203                 break;
1204         }
1205         case SVM_EXIT_ERR: {
1206                 vmexit = NESTED_EXIT_DONE;
1207                 break;
1208         }
1209         default: {
1210                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1211                         vmexit = NESTED_EXIT_DONE;
1212         }
1213         }
1214
1215         return vmexit;
1216 }
1217
1218 int nested_svm_exit_handled(struct vcpu_svm *svm)
1219 {
1220         int vmexit;
1221
1222         vmexit = nested_svm_intercept(svm);
1223
1224         if (vmexit == NESTED_EXIT_DONE)
1225                 nested_svm_vmexit(svm);
1226
1227         return vmexit;
1228 }
1229
1230 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1231 {
1232         if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1233                 kvm_queue_exception(vcpu, UD_VECTOR);
1234                 return 1;
1235         }
1236
1237         if (to_svm(vcpu)->vmcb->save.cpl) {
1238                 kvm_inject_gp(vcpu, 0);
1239                 return 1;
1240         }
1241
1242         return 0;
1243 }
1244
1245 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1246 {
1247         unsigned int nr = svm->vcpu.arch.exception.nr;
1248
1249         return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1250 }
1251
1252 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1253 {
1254         unsigned int nr = svm->vcpu.arch.exception.nr;
1255         struct vmcb *vmcb = svm->vmcb;
1256
1257         vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1258         vmcb->control.exit_code_hi = 0;
1259
1260         if (svm->vcpu.arch.exception.has_error_code)
1261                 vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1262
1263         /*
1264          * EXITINFO2 is undefined for all exception intercepts other
1265          * than #PF.
1266          */
1267         if (nr == PF_VECTOR) {
1268                 if (svm->vcpu.arch.exception.nested_apf)
1269                         vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1270                 else if (svm->vcpu.arch.exception.has_payload)
1271                         vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1272                 else
1273                         vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1274         } else if (nr == DB_VECTOR) {
1275                 /* See inject_pending_event.  */
1276                 kvm_deliver_exception_payload(&svm->vcpu);
1277                 if (svm->vcpu.arch.dr7 & DR7_GD) {
1278                         svm->vcpu.arch.dr7 &= ~DR7_GD;
1279                         kvm_update_dr7(&svm->vcpu);
1280                 }
1281         } else
1282                 WARN_ON(svm->vcpu.arch.exception.has_payload);
1283
1284         nested_svm_vmexit(svm);
1285 }
1286
1287 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1288 {
1289         return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1290 }
1291
1292 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1293 {
1294         struct vcpu_svm *svm = to_svm(vcpu);
1295         bool block_nested_events =
1296                 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1297         struct kvm_lapic *apic = vcpu->arch.apic;
1298
1299         if (lapic_in_kernel(vcpu) &&
1300             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1301                 if (block_nested_events)
1302                         return -EBUSY;
1303                 if (!nested_exit_on_init(svm))
1304                         return 0;
1305                 nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1306                 return 0;
1307         }
1308
1309         if (vcpu->arch.exception.pending) {
1310                 /*
1311                  * Only a pending nested run can block a pending exception.
1312                  * Otherwise an injected NMI/interrupt should either be
1313                  * lost or delivered to the nested hypervisor in the EXITINTINFO
1314                  * vmcb field, while delivering the pending exception.
1315                  */
1316                 if (svm->nested.nested_run_pending)
1317                         return -EBUSY;
1318                 if (!nested_exit_on_exception(svm))
1319                         return 0;
1320                 nested_svm_inject_exception_vmexit(svm);
1321                 return 0;
1322         }
1323
1324         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1325                 if (block_nested_events)
1326                         return -EBUSY;
1327                 if (!nested_exit_on_smi(svm))
1328                         return 0;
1329                 nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1330                 return 0;
1331         }
1332
1333         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1334                 if (block_nested_events)
1335                         return -EBUSY;
1336                 if (!nested_exit_on_nmi(svm))
1337                         return 0;
1338                 nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1339                 return 0;
1340         }
1341
1342         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1343                 if (block_nested_events)
1344                         return -EBUSY;
1345                 if (!nested_exit_on_intr(svm))
1346                         return 0;
1347                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1348                 nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1349                 return 0;
1350         }
1351
1352         return 0;
1353 }
1354
1355 int nested_svm_exit_special(struct vcpu_svm *svm)
1356 {
1357         u32 exit_code = svm->vmcb->control.exit_code;
1358
1359         switch (exit_code) {
1360         case SVM_EXIT_INTR:
1361         case SVM_EXIT_NMI:
1362         case SVM_EXIT_NPF:
1363                 return NESTED_EXIT_HOST;
1364         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1365                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1366
1367                 if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1368                     excp_bits)
1369                         return NESTED_EXIT_HOST;
1370                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1371                          svm->vcpu.arch.apf.host_apf_flags)
1372                         /* Trap async PF even if not shadowing */
1373                         return NESTED_EXIT_HOST;
1374                 break;
1375         }
1376         default:
1377                 break;
1378         }
1379
1380         return NESTED_EXIT_CONTINUE;
1381 }
1382
1383 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
1384 {
1385         struct vcpu_svm *svm = to_svm(vcpu);
1386
1387         vcpu->arch.tsc_scaling_ratio =
1388                 kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
1389                                                svm->tsc_ratio_msr);
1390         __svm_write_tsc_multiplier(vcpu->arch.tsc_scaling_ratio);
1391 }
1392
1393 /* Inverse operation of nested_copy_vmcb_control_to_cache(). asid is copied too. */
1394 static void nested_copy_vmcb_cache_to_control(struct vmcb_control_area *dst,
1395                                               struct vmcb_ctrl_area_cached *from)
1396 {
1397         unsigned int i;
1398
1399         memset(dst, 0, sizeof(struct vmcb_control_area));
1400
1401         for (i = 0; i < MAX_INTERCEPT; i++)
1402                 dst->intercepts[i] = from->intercepts[i];
1403
1404         dst->iopm_base_pa         = from->iopm_base_pa;
1405         dst->msrpm_base_pa        = from->msrpm_base_pa;
1406         dst->tsc_offset           = from->tsc_offset;
1407         dst->asid                 = from->asid;
1408         dst->tlb_ctl              = from->tlb_ctl;
1409         dst->int_ctl              = from->int_ctl;
1410         dst->int_vector           = from->int_vector;
1411         dst->int_state            = from->int_state;
1412         dst->exit_code            = from->exit_code;
1413         dst->exit_code_hi         = from->exit_code_hi;
1414         dst->exit_info_1          = from->exit_info_1;
1415         dst->exit_info_2          = from->exit_info_2;
1416         dst->exit_int_info        = from->exit_int_info;
1417         dst->exit_int_info_err    = from->exit_int_info_err;
1418         dst->nested_ctl           = from->nested_ctl;
1419         dst->event_inj            = from->event_inj;
1420         dst->event_inj_err        = from->event_inj_err;
1421         dst->nested_cr3           = from->nested_cr3;
1422         dst->virt_ext              = from->virt_ext;
1423         dst->pause_filter_count   = from->pause_filter_count;
1424         dst->pause_filter_thresh  = from->pause_filter_thresh;
1425         /* 'clean' and 'reserved_sw' are not changed by KVM */
1426 }
1427
1428 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1429                                 struct kvm_nested_state __user *user_kvm_nested_state,
1430                                 u32 user_data_size)
1431 {
1432         struct vcpu_svm *svm;
1433         struct vmcb_control_area *ctl;
1434         unsigned long r;
1435         struct kvm_nested_state kvm_state = {
1436                 .flags = 0,
1437                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1438                 .size = sizeof(kvm_state),
1439         };
1440         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1441                 &user_kvm_nested_state->data.svm[0];
1442
1443         if (!vcpu)
1444                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1445
1446         svm = to_svm(vcpu);
1447
1448         if (user_data_size < kvm_state.size)
1449                 goto out;
1450
1451         /* First fill in the header and copy it out.  */
1452         if (is_guest_mode(vcpu)) {
1453                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1454                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1455                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1456
1457                 if (svm->nested.nested_run_pending)
1458                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1459         }
1460
1461         if (gif_set(svm))
1462                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1463
1464         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1465                 return -EFAULT;
1466
1467         if (!is_guest_mode(vcpu))
1468                 goto out;
1469
1470         /*
1471          * Copy over the full size of the VMCB rather than just the size
1472          * of the structs.
1473          */
1474         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1475                 return -EFAULT;
1476
1477         ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
1478         if (!ctl)
1479                 return -ENOMEM;
1480
1481         nested_copy_vmcb_cache_to_control(ctl, &svm->nested.ctl);
1482         r = copy_to_user(&user_vmcb->control, ctl,
1483                          sizeof(user_vmcb->control));
1484         kfree(ctl);
1485         if (r)
1486                 return -EFAULT;
1487
1488         if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1489                          sizeof(user_vmcb->save)))
1490                 return -EFAULT;
1491 out:
1492         return kvm_state.size;
1493 }
1494
1495 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1496                                 struct kvm_nested_state __user *user_kvm_nested_state,
1497                                 struct kvm_nested_state *kvm_state)
1498 {
1499         struct vcpu_svm *svm = to_svm(vcpu);
1500         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1501                 &user_kvm_nested_state->data.svm[0];
1502         struct vmcb_control_area *ctl;
1503         struct vmcb_save_area *save;
1504         struct vmcb_save_area_cached save_cached;
1505         struct vmcb_ctrl_area_cached ctl_cached;
1506         unsigned long cr0;
1507         int ret;
1508
1509         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1510                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1511
1512         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1513                 return -EINVAL;
1514
1515         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1516                                  KVM_STATE_NESTED_RUN_PENDING |
1517                                  KVM_STATE_NESTED_GIF_SET))
1518                 return -EINVAL;
1519
1520         /*
1521          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1522          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1523          */
1524         if (!(vcpu->arch.efer & EFER_SVME)) {
1525                 /* GIF=1 and no guest mode are required if SVME=0.  */
1526                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1527                         return -EINVAL;
1528         }
1529
1530         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1531         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1532                 return -EINVAL;
1533
1534         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1535                 svm_leave_nested(vcpu);
1536                 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1537                 return 0;
1538         }
1539
1540         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1541                 return -EINVAL;
1542         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1543                 return -EINVAL;
1544
1545         ret  = -ENOMEM;
1546         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1547         save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1548         if (!ctl || !save)
1549                 goto out_free;
1550
1551         ret = -EFAULT;
1552         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1553                 goto out_free;
1554         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1555                 goto out_free;
1556
1557         ret = -EINVAL;
1558         __nested_copy_vmcb_control_to_cache(vcpu, &ctl_cached, ctl);
1559         if (!__nested_vmcb_check_controls(vcpu, &ctl_cached))
1560                 goto out_free;
1561
1562         /*
1563          * Processor state contains L2 state.  Check that it is
1564          * valid for guest mode (see nested_vmcb_check_save).
1565          */
1566         cr0 = kvm_read_cr0(vcpu);
1567         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1568                 goto out_free;
1569
1570         /*
1571          * Validate host state saved from before VMRUN (see
1572          * nested_svm_check_permissions).
1573          */
1574         __nested_copy_vmcb_save_to_cache(&save_cached, save);
1575         if (!(save->cr0 & X86_CR0_PG) ||
1576             !(save->cr0 & X86_CR0_PE) ||
1577             (save->rflags & X86_EFLAGS_VM) ||
1578             !__nested_vmcb_check_save(vcpu, &save_cached))
1579                 goto out_free;
1580
1581
1582         /*
1583          * All checks done, we can enter guest mode. Userspace provides
1584          * vmcb12.control, which will be combined with L1 and stored into
1585          * vmcb02, and the L1 save state which we store in vmcb01.
1586          * L2 registers if needed are moved from the current VMCB to VMCB02.
1587          */
1588
1589         if (is_guest_mode(vcpu))
1590                 svm_leave_nested(vcpu);
1591         else
1592                 svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1593
1594         svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1595
1596         svm->nested.nested_run_pending =
1597                 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1598
1599         svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1600
1601         svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1602         nested_copy_vmcb_control_to_cache(svm, ctl);
1603
1604         svm_switch_vmcb(svm, &svm->nested.vmcb02);
1605         nested_vmcb02_prepare_control(svm);
1606
1607         /*
1608          * While the nested guest CR3 is already checked and set by
1609          * KVM_SET_SREGS, it was set when nested state was yet loaded,
1610          * thus MMU might not be initialized correctly.
1611          * Set it again to fix this.
1612          */
1613
1614         ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1615                                   nested_npt_enabled(svm), false);
1616         if (WARN_ON_ONCE(ret))
1617                 goto out_free;
1618
1619         svm->nested.force_msr_bitmap_recalc = true;
1620
1621         kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1622         ret = 0;
1623 out_free:
1624         kfree(save);
1625         kfree(ctl);
1626
1627         return ret;
1628 }
1629
1630 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1631 {
1632         struct vcpu_svm *svm = to_svm(vcpu);
1633
1634         if (WARN_ON(!is_guest_mode(vcpu)))
1635                 return true;
1636
1637         if (!vcpu->arch.pdptrs_from_userspace &&
1638             !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1639                 /*
1640                  * Reload the guest's PDPTRs since after a migration
1641                  * the guest CR3 might be restored prior to setting the nested
1642                  * state which can lead to a load of wrong PDPTRs.
1643                  */
1644                 if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
1645                         return false;
1646
1647         if (!nested_svm_vmrun_msrpm(svm)) {
1648                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1649                 vcpu->run->internal.suberror =
1650                         KVM_INTERNAL_ERROR_EMULATION;
1651                 vcpu->run->internal.ndata = 0;
1652                 return false;
1653         }
1654
1655         return true;
1656 }
1657
1658 struct kvm_x86_nested_ops svm_nested_ops = {
1659         .leave_nested = svm_leave_nested,
1660         .check_events = svm_check_nested_events,
1661         .handle_page_fault_workaround = nested_svm_handle_page_fault_workaround,
1662         .triple_fault = nested_svm_triple_fault,
1663         .get_nested_state_pages = svm_get_nested_state_pages,
1664         .get_state = svm_get_nested_state,
1665         .set_state = svm_set_nested_state,
1666 };