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) KBUILD_MODNAME ": " 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 "smm.h"
  29 #include "cpuid.h"
  30 #include "lapic.h"
  31 #include "svm.h"
  32 #include "hyperv.h"
  33
  34 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
  35
  36 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  37                                        struct x86_exception *fault)
  38 {
  39         struct vcpu_svm *svm = to_svm(vcpu);
  40         struct vmcb *vmcb = svm->vmcb;
  41
  42         if (vmcb->control.exit_code != SVM_EXIT_NPF) {
  43                 /*
  44                  * TODO: track the cause of the nested page fault, and
  45                  * correctly fill in the high bits of exit_info_1.
  46                  */
  47                 vmcb->control.exit_code = SVM_EXIT_NPF;
  48                 vmcb->control.exit_code_hi = 0;
  49                 vmcb->control.exit_info_1 = (1ULL << 32);
  50                 vmcb->control.exit_info_2 = fault->address;
  51         }
  52
  53         vmcb->control.exit_info_1 &= ~0xffffffffULL;
  54         vmcb->control.exit_info_1 |= fault->error_code;
  55
  56         nested_svm_vmexit(svm);
  57 }
  58
  59 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  60 {
  61         struct vcpu_svm *svm = to_svm(vcpu);
  62         u64 cr3 = svm->nested.ctl.nested_cr3;
  63         u64 pdpte;
  64         int ret;
  65
  66         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
  67                                        offset_in_page(cr3) + index * 8, 8);
  68         if (ret)
  69                 return 0;
  70         return pdpte;
  71 }
  72
  73 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  74 {
  75         struct vcpu_svm *svm = to_svm(vcpu);
  76
  77         return svm->nested.ctl.nested_cr3;
  78 }
  79
  80 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  81 {
  82         struct vcpu_svm *svm = to_svm(vcpu);
  83
  84         WARN_ON(mmu_is_nested(vcpu));
  85
  86         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
  87
  88         /*
  89          * The NPT format depends on L1's CR4 and EFER, which is in vmcb01.  Note,
  90          * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
  91          * vCPU state.  CR0.WP is explicitly ignored, while CR0.PG is required.
  92          */
  93         kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
  94                                 svm->vmcb01.ptr->save.efer,
  95                                 svm->nested.ctl.nested_cr3);
  96         vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
  97         vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
  98         vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
  99         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 100 }
 101
 102 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 103 {
 104         vcpu->arch.mmu = &vcpu->arch.root_mmu;
 105         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 106 }
 107
 108 static bool nested_vmcb_needs_vls_intercept(struct vcpu_svm *svm)
 109 {
 110         if (!guest_can_use(&svm->vcpu, X86_FEATURE_V_VMSAVE_VMLOAD))
 111                 return true;
 112
 113         if (!nested_npt_enabled(svm))
 114                 return true;
 115
 116         if (!(svm->nested.ctl.virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK))
 117                 return true;
 118
 119         return false;
 120 }
 121
 122 void recalc_intercepts(struct vcpu_svm *svm)
 123 {
 124         struct vmcb_control_area *c, *h;
 125         struct vmcb_ctrl_area_cached *g;
 126         unsigned int i;
 127
 128         vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 129
 130         if (!is_guest_mode(&svm->vcpu))
 131                 return;
 132
 133         c = &svm->vmcb->control;
 134         h = &svm->vmcb01.ptr->control;
 135         g = &svm->nested.ctl;
 136
 137         for (i = 0; i < MAX_INTERCEPT; i++)
 138                 c->intercepts[i] = h->intercepts[i];
 139
 140         if (g->int_ctl & V_INTR_MASKING_MASK) {
 141                 /*
 142                  * If L2 is active and V_INTR_MASKING is enabled in vmcb12,
 143                  * disable intercept of CR8 writes as L2's CR8 does not affect
 144                  * any interrupt KVM may want to inject.
 145                  *
 146                  * Similarly, disable intercept of virtual interrupts (used to
 147                  * detect interrupt windows) if the saved RFLAGS.IF is '0', as
 148                  * the effective RFLAGS.IF for L1 interrupts will never be set
 149                  * while L2 is running (L2's RFLAGS.IF doesn't affect L1 IRQs).
 150                  */
 151                 vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
 152                 if (!(svm->vmcb01.ptr->save.rflags & X86_EFLAGS_IF))
 153                         vmcb_clr_intercept(c, INTERCEPT_VINTR);
 154         }
 155
 156         /*
 157          * We want to see VMMCALLs from a nested guest only when Hyper-V L2 TLB
 158          * flush feature is enabled.
 159          */
 160         if (!nested_svm_l2_tlb_flush_enabled(&svm->vcpu))
 161                 vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
 162
 163         for (i = 0; i < MAX_INTERCEPT; i++)
 164                 c->intercepts[i] |= g->intercepts[i];
 165
 166         /* If SMI is not intercepted, ignore guest SMI intercept as well  */
 167         if (!intercept_smi)
 168                 vmcb_clr_intercept(c, INTERCEPT_SMI);
 169
 170         if (nested_vmcb_needs_vls_intercept(svm)) {
 171                 /*
 172                  * If the virtual VMLOAD/VMSAVE is not enabled for the L2,
 173                  * we must intercept these instructions to correctly
 174                  * emulate them in case L1 doesn't intercept them.
 175                  */
 176                 vmcb_set_intercept(c, INTERCEPT_VMLOAD);
 177                 vmcb_set_intercept(c, INTERCEPT_VMSAVE);
 178         } else {
 179                 WARN_ON(!(c->virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK));
 180         }
 181 }
 182
 183 /*
 184  * Merge L0's (KVM) and L1's (Nested VMCB) MSR permission bitmaps. The function
 185  * is optimized in that it only merges the parts where KVM MSR permission bitmap
 186  * may contain zero bits.
 187  */
 188 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 189 {
 190         int i;
 191
 192         /*
 193          * MSR bitmap update can be skipped when:
 194          * - MSR bitmap for L1 hasn't changed.
 195          * - Nested hypervisor (L1) is attempting to launch the same L2 as
 196          *   before.
 197          * - Nested hypervisor (L1) is using Hyper-V emulation interface and
 198          * tells KVM (L0) there were no changes in MSR bitmap for L2.
 199          */
 200 #ifdef CONFIG_KVM_HYPERV
 201         if (!svm->nested.force_msr_bitmap_recalc) {
 202                 struct hv_vmcb_enlightenments *hve = &svm->nested.ctl.hv_enlightenments;
 203
 204                 if (kvm_hv_hypercall_enabled(&svm->vcpu) &&
 205                     hve->hv_enlightenments_control.msr_bitmap &&
 206                     (svm->nested.ctl.clean & BIT(HV_VMCB_NESTED_ENLIGHTENMENTS)))
 207                         goto set_msrpm_base_pa;
 208         }
 209 #endif
 210
 211         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 212                 return true;
 213
 214         for (i = 0; i < MSRPM_OFFSETS; i++) {
 215                 u32 value, p;
 216                 u64 offset;
 217
 218                 if (msrpm_offsets[i] == 0xffffffff)
 219                         break;
 220
 221                 p      = msrpm_offsets[i];
 222
 223                 /* x2apic msrs are intercepted always for the nested guest */
 224                 if (is_x2apic_msrpm_offset(p))
 225                         continue;
 226
 227                 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 228
 229                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 230                         return false;
 231
 232                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 233         }
 234
 235         svm->nested.force_msr_bitmap_recalc = false;
 236
 237 #ifdef CONFIG_KVM_HYPERV
 238 set_msrpm_base_pa:
 239 #endif
 240         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 241
 242         return true;
 243 }
 244
 245 /*
 246  * Bits 11:0 of bitmap address are ignored by hardware
 247  */
 248 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
 249 {
 250         u64 addr = PAGE_ALIGN(pa);
 251
 252         return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
 253             kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
 254 }
 255
 256 static bool __nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
 257                                          struct vmcb_ctrl_area_cached *control)
 258 {
 259         if (CC(!vmcb12_is_intercept(control, INTERCEPT_VMRUN)))
 260                 return false;
 261
 262         if (CC(control->asid == 0))
 263                 return false;
 264
 265         if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
 266                 return false;
 267
 268         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
 269                                            MSRPM_SIZE)))
 270                 return false;
 271         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
 272                                            IOPM_SIZE)))
 273                 return false;
 274
 275         if (CC((control->int_ctl & V_NMI_ENABLE_MASK) &&
 276                !vmcb12_is_intercept(control, INTERCEPT_NMI))) {
 277                 return false;
 278         }
 279
 280         return true;
 281 }
 282
 283 /* Common checks that apply to both L1 and L2 state.  */
 284 static bool __nested_vmcb_check_save(struct kvm_vcpu *vcpu,
 285                                      struct vmcb_save_area_cached *save)
 286 {
 287         if (CC(!(save->efer & EFER_SVME)))
 288                 return false;
 289
 290         if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 291             CC(save->cr0 & ~0xffffffffULL))
 292                 return false;
 293
 294         if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 295                 return false;
 296
 297         /*
 298          * These checks are also performed by KVM_SET_SREGS,
 299          * except that EFER.LMA is not checked by SVM against
 300          * CR0.PG && EFER.LME.
 301          */
 302         if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
 303                 if (CC(!(save->cr4 & X86_CR4_PAE)) ||
 304                     CC(!(save->cr0 & X86_CR0_PE)) ||
 305                     CC(!kvm_vcpu_is_legal_cr3(vcpu, save->cr3)))
 306                         return false;
 307         }
 308
 309         /* Note, SVM doesn't have any additional restrictions on CR4. */
 310         if (CC(!__kvm_is_valid_cr4(vcpu, save->cr4)))
 311                 return false;
 312
 313         if (CC(!kvm_valid_efer(vcpu, save->efer)))
 314                 return false;
 315
 316         return true;
 317 }
 318
 319 static bool nested_vmcb_check_save(struct kvm_vcpu *vcpu)
 320 {
 321         struct vcpu_svm *svm = to_svm(vcpu);
 322         struct vmcb_save_area_cached *save = &svm->nested.save;
 323
 324         return __nested_vmcb_check_save(vcpu, save);
 325 }
 326
 327 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu)
 328 {
 329         struct vcpu_svm *svm = to_svm(vcpu);
 330         struct vmcb_ctrl_area_cached *ctl = &svm->nested.ctl;
 331
 332         return __nested_vmcb_check_controls(vcpu, ctl);
 333 }
 334
 335 static
 336 void __nested_copy_vmcb_control_to_cache(struct kvm_vcpu *vcpu,
 337                                          struct vmcb_ctrl_area_cached *to,
 338                                          struct vmcb_control_area *from)
 339 {
 340         unsigned int i;
 341
 342         for (i = 0; i < MAX_INTERCEPT; i++)
 343                 to->intercepts[i] = from->intercepts[i];
 344
 345         to->iopm_base_pa        = from->iopm_base_pa;
 346         to->msrpm_base_pa       = from->msrpm_base_pa;
 347         to->tsc_offset          = from->tsc_offset;
 348         to->tlb_ctl             = from->tlb_ctl;
 349         to->int_ctl             = from->int_ctl;
 350         to->int_vector          = from->int_vector;
 351         to->int_state           = from->int_state;
 352         to->exit_code           = from->exit_code;
 353         to->exit_code_hi        = from->exit_code_hi;
 354         to->exit_info_1         = from->exit_info_1;
 355         to->exit_info_2         = from->exit_info_2;
 356         to->exit_int_info       = from->exit_int_info;
 357         to->exit_int_info_err   = from->exit_int_info_err;
 358         to->nested_ctl          = from->nested_ctl;
 359         to->event_inj           = from->event_inj;
 360         to->event_inj_err       = from->event_inj_err;
 361         to->next_rip            = from->next_rip;
 362         to->nested_cr3          = from->nested_cr3;
 363         to->virt_ext            = from->virt_ext;
 364         to->pause_filter_count  = from->pause_filter_count;
 365         to->pause_filter_thresh = from->pause_filter_thresh;
 366
 367         /* Copy asid here because nested_vmcb_check_controls will check it.  */
 368         to->asid           = from->asid;
 369         to->msrpm_base_pa &= ~0x0fffULL;
 370         to->iopm_base_pa  &= ~0x0fffULL;
 371
 372 #ifdef CONFIG_KVM_HYPERV
 373         /* Hyper-V extensions (Enlightened VMCB) */
 374         if (kvm_hv_hypercall_enabled(vcpu)) {
 375                 to->clean = from->clean;
 376                 memcpy(&to->hv_enlightenments, &from->hv_enlightenments,
 377                        sizeof(to->hv_enlightenments));
 378         }
 379 #endif
 380 }
 381
 382 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
 383                                        struct vmcb_control_area *control)
 384 {
 385         __nested_copy_vmcb_control_to_cache(&svm->vcpu, &svm->nested.ctl, control);
 386 }
 387
 388 static void __nested_copy_vmcb_save_to_cache(struct vmcb_save_area_cached *to,
 389                                              struct vmcb_save_area *from)
 390 {
 391         /*
 392          * Copy only fields that are validated, as we need them
 393          * to avoid TOC/TOU races.
 394          */
 395         to->efer = from->efer;
 396         to->cr0 = from->cr0;
 397         to->cr3 = from->cr3;
 398         to->cr4 = from->cr4;
 399
 400         to->dr6 = from->dr6;
 401         to->dr7 = from->dr7;
 402 }
 403
 404 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
 405                                     struct vmcb_save_area *save)
 406 {
 407         __nested_copy_vmcb_save_to_cache(&svm->nested.save, save);
 408 }
 409
 410 /*
 411  * Synchronize fields that are written by the processor, so that
 412  * they can be copied back into the vmcb12.
 413  */
 414 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
 415 {
 416         u32 mask;
 417         svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 418         svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 419
 420         /* Only a few fields of int_ctl are written by the processor.  */
 421         mask = V_IRQ_MASK | V_TPR_MASK;
 422         /*
 423          * Don't sync vmcb02 V_IRQ back to vmcb12 if KVM (L0) is intercepting
 424          * virtual interrupts in order to request an interrupt window, as KVM
 425          * has usurped vmcb02's int_ctl.  If an interrupt window opens before
 426          * the next VM-Exit, svm_clear_vintr() will restore vmcb12's int_ctl.
 427          * If no window opens, V_IRQ will be correctly preserved in vmcb12's
 428          * int_ctl (because it was never recognized while L2 was running).
 429          */
 430         if (svm_is_intercept(svm, INTERCEPT_VINTR) &&
 431             !test_bit(INTERCEPT_VINTR, (unsigned long *)svm->nested.ctl.intercepts))
 432                 mask &= ~V_IRQ_MASK;
 433
 434         if (nested_vgif_enabled(svm))
 435                 mask |= V_GIF_MASK;
 436
 437         if (nested_vnmi_enabled(svm))
 438                 mask |= V_NMI_BLOCKING_MASK | V_NMI_PENDING_MASK;
 439
 440         svm->nested.ctl.int_ctl        &= ~mask;
 441         svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 442 }
 443
 444 /*
 445  * Transfer any event that L0 or L1 wanted to inject into L2 to
 446  * EXIT_INT_INFO.
 447  */
 448 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
 449                                                 struct vmcb *vmcb12)
 450 {
 451         struct kvm_vcpu *vcpu = &svm->vcpu;
 452         u32 exit_int_info = 0;
 453         unsigned int nr;
 454
 455         if (vcpu->arch.exception.injected) {
 456                 nr = vcpu->arch.exception.vector;
 457                 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 458
 459                 if (vcpu->arch.exception.has_error_code) {
 460                         exit_int_info |= SVM_EVTINJ_VALID_ERR;
 461                         vmcb12->control.exit_int_info_err =
 462                                 vcpu->arch.exception.error_code;
 463                 }
 464
 465         } else if (vcpu->arch.nmi_injected) {
 466                 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 467
 468         } else if (vcpu->arch.interrupt.injected) {
 469                 nr = vcpu->arch.interrupt.nr;
 470                 exit_int_info = nr | SVM_EVTINJ_VALID;
 471
 472                 if (vcpu->arch.interrupt.soft)
 473                         exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 474                 else
 475                         exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 476         }
 477
 478         vmcb12->control.exit_int_info = exit_int_info;
 479 }
 480
 481 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
 482 {
 483         /* Handle pending Hyper-V TLB flush requests */
 484         kvm_hv_nested_transtion_tlb_flush(vcpu, npt_enabled);
 485
 486         /*
 487          * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
 488          * things to fix before this can be conditional:
 489          *
 490          *  - Flush TLBs for both L1 and L2 remote TLB flush
 491          *  - Honor L1's request to flush an ASID on nested VMRUN
 492          *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
 493          *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
 494          *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
 495          *
 496          * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
 497          *     NPT guest-physical mappings on VMRUN.
 498          */
 499         kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
 500         kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
 501 }
 502
 503 /*
 504  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 505  * if we are emulating VM-Entry into a guest with NPT enabled.
 506  */
 507 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 508                                bool nested_npt, bool reload_pdptrs)
 509 {
 510         if (CC(!kvm_vcpu_is_legal_cr3(vcpu, cr3)))
 511                 return -EINVAL;
 512
 513         if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
 514             CC(!load_pdptrs(vcpu, cr3)))
 515                 return -EINVAL;
 516
 517         vcpu->arch.cr3 = cr3;
 518
 519         /* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
 520         kvm_init_mmu(vcpu);
 521
 522         if (!nested_npt)
 523                 kvm_mmu_new_pgd(vcpu, cr3);
 524
 525         return 0;
 526 }
 527
 528 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
 529 {
 530         if (!svm->nested.vmcb02.ptr)
 531                 return;
 532
 533         /* FIXME: merge g_pat from vmcb01 and vmcb12.  */
 534         svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
 535 }
 536
 537 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 538 {
 539         bool new_vmcb12 = false;
 540         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 541         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 542         struct kvm_vcpu *vcpu = &svm->vcpu;
 543
 544         nested_vmcb02_compute_g_pat(svm);
 545
 546         /* Load the nested guest state */
 547         if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
 548                 new_vmcb12 = true;
 549                 svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
 550                 svm->nested.force_msr_bitmap_recalc = true;
 551         }
 552
 553         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
 554                 vmcb02->save.es = vmcb12->save.es;
 555                 vmcb02->save.cs = vmcb12->save.cs;
 556                 vmcb02->save.ss = vmcb12->save.ss;
 557                 vmcb02->save.ds = vmcb12->save.ds;
 558                 vmcb02->save.cpl = vmcb12->save.cpl;
 559                 vmcb_mark_dirty(vmcb02, VMCB_SEG);
 560         }
 561
 562         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
 563                 vmcb02->save.gdtr = vmcb12->save.gdtr;
 564                 vmcb02->save.idtr = vmcb12->save.idtr;
 565                 vmcb_mark_dirty(vmcb02, VMCB_DT);
 566         }
 567
 568         kvm_set_rflags(vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
 569
 570         svm_set_efer(vcpu, svm->nested.save.efer);
 571
 572         svm_set_cr0(vcpu, svm->nested.save.cr0);
 573         svm_set_cr4(vcpu, svm->nested.save.cr4);
 574
 575         svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 576
 577         kvm_rax_write(vcpu, vmcb12->save.rax);
 578         kvm_rsp_write(vcpu, vmcb12->save.rsp);
 579         kvm_rip_write(vcpu, vmcb12->save.rip);
 580
 581         /* In case we don't even reach vcpu_run, the fields are not updated */
 582         vmcb02->save.rax = vmcb12->save.rax;
 583         vmcb02->save.rsp = vmcb12->save.rsp;
 584         vmcb02->save.rip = vmcb12->save.rip;
 585
 586         /* These bits will be set properly on the first execution when new_vmc12 is true */
 587         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 588                 vmcb02->save.dr7 = svm->nested.save.dr7 | DR7_FIXED_1;
 589                 svm->vcpu.arch.dr6  = svm->nested.save.dr6 | DR6_ACTIVE_LOW;
 590                 vmcb_mark_dirty(vmcb02, VMCB_DR);
 591         }
 592
 593         if (unlikely(guest_can_use(vcpu, X86_FEATURE_LBRV) &&
 594                      (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
 595                 /*
 596                  * Reserved bits of DEBUGCTL are ignored.  Be consistent with
 597                  * svm_set_msr's definition of reserved bits.
 598                  */
 599                 svm_copy_lbrs(vmcb02, vmcb12);
 600                 vmcb02->save.dbgctl &= ~DEBUGCTL_RESERVED_BITS;
 601                 svm_update_lbrv(&svm->vcpu);
 602
 603         } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
 604                 svm_copy_lbrs(vmcb02, vmcb01);
 605         }
 606 }
 607
 608 static inline bool is_evtinj_soft(u32 evtinj)
 609 {
 610         u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
 611         u8 vector = evtinj & SVM_EVTINJ_VEC_MASK;
 612
 613         if (!(evtinj & SVM_EVTINJ_VALID))
 614                 return false;
 615
 616         if (type == SVM_EVTINJ_TYPE_SOFT)
 617                 return true;
 618
 619         return type == SVM_EVTINJ_TYPE_EXEPT && kvm_exception_is_soft(vector);
 620 }
 621
 622 static bool is_evtinj_nmi(u32 evtinj)
 623 {
 624         u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
 625
 626         if (!(evtinj & SVM_EVTINJ_VALID))
 627                 return false;
 628
 629         return type == SVM_EVTINJ_TYPE_NMI;
 630 }
 631
 632 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm,
 633                                           unsigned long vmcb12_rip,
 634                                           unsigned long vmcb12_csbase)
 635 {
 636         u32 int_ctl_vmcb01_bits = V_INTR_MASKING_MASK;
 637         u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
 638
 639         struct kvm_vcpu *vcpu = &svm->vcpu;
 640         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 641         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 642         u32 pause_count12;
 643         u32 pause_thresh12;
 644
 645         /*
 646          * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 647          * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 648          */
 649
 650         if (guest_can_use(vcpu, X86_FEATURE_VGIF) &&
 651             (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK))
 652                 int_ctl_vmcb12_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 653         else
 654                 int_ctl_vmcb01_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 655
 656         if (vnmi) {
 657                 if (vmcb01->control.int_ctl & V_NMI_PENDING_MASK) {
 658                         svm->vcpu.arch.nmi_pending++;
 659                         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
 660                 }
 661                 if (nested_vnmi_enabled(svm))
 662                         int_ctl_vmcb12_bits |= (V_NMI_PENDING_MASK |
 663                                                 V_NMI_ENABLE_MASK |
 664                                                 V_NMI_BLOCKING_MASK);
 665         }
 666
 667         /* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 668         vmcb02->control.nested_ctl = vmcb01->control.nested_ctl;
 669         vmcb02->control.iopm_base_pa = vmcb01->control.iopm_base_pa;
 670         vmcb02->control.msrpm_base_pa = vmcb01->control.msrpm_base_pa;
 671
 672         /* Done at vmrun: asid.  */
 673
 674         /* Also overwritten later if necessary.  */
 675         vmcb02->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 676
 677         /* nested_cr3.  */
 678         if (nested_npt_enabled(svm))
 679                 nested_svm_init_mmu_context(vcpu);
 680
 681         vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
 682                         vcpu->arch.l1_tsc_offset,
 683                         svm->nested.ctl.tsc_offset,
 684                         svm->tsc_ratio_msr);
 685
 686         vmcb02->control.tsc_offset = vcpu->arch.tsc_offset;
 687
 688         if (guest_can_use(vcpu, X86_FEATURE_TSCRATEMSR) &&
 689             svm->tsc_ratio_msr != kvm_caps.default_tsc_scaling_ratio)
 690                 nested_svm_update_tsc_ratio_msr(vcpu);
 691
 692         vmcb02->control.int_ctl             =
 693                 (svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
 694                 (vmcb01->control.int_ctl & int_ctl_vmcb01_bits);
 695
 696         vmcb02->control.int_vector          = svm->nested.ctl.int_vector;
 697         vmcb02->control.int_state           = svm->nested.ctl.int_state;
 698         vmcb02->control.event_inj           = svm->nested.ctl.event_inj;
 699         vmcb02->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 700
 701         /*
 702          * next_rip is consumed on VMRUN as the return address pushed on the
 703          * stack for injected soft exceptions/interrupts.  If nrips is exposed
 704          * to L1, take it verbatim from vmcb12.  If nrips is supported in
 705          * hardware but not exposed to L1, stuff the actual L2 RIP to emulate
 706          * what a nrips=0 CPU would do (L1 is responsible for advancing RIP
 707          * prior to injecting the event).
 708          */
 709         if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
 710                 vmcb02->control.next_rip    = svm->nested.ctl.next_rip;
 711         else if (boot_cpu_has(X86_FEATURE_NRIPS))
 712                 vmcb02->control.next_rip    = vmcb12_rip;
 713
 714         svm->nmi_l1_to_l2 = is_evtinj_nmi(vmcb02->control.event_inj);
 715         if (is_evtinj_soft(vmcb02->control.event_inj)) {
 716                 svm->soft_int_injected = true;
 717                 svm->soft_int_csbase = vmcb12_csbase;
 718                 svm->soft_int_old_rip = vmcb12_rip;
 719                 if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
 720                         svm->soft_int_next_rip = svm->nested.ctl.next_rip;
 721                 else
 722                         svm->soft_int_next_rip = vmcb12_rip;
 723         }
 724
 725         vmcb02->control.virt_ext            = vmcb01->control.virt_ext &
 726                                               LBR_CTL_ENABLE_MASK;
 727         if (guest_can_use(vcpu, X86_FEATURE_LBRV))
 728                 vmcb02->control.virt_ext  |=
 729                         (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK);
 730
 731         if (!nested_vmcb_needs_vls_intercept(svm))
 732                 vmcb02->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
 733
 734         if (guest_can_use(vcpu, X86_FEATURE_PAUSEFILTER))
 735                 pause_count12 = svm->nested.ctl.pause_filter_count;
 736         else
 737                 pause_count12 = 0;
 738         if (guest_can_use(vcpu, X86_FEATURE_PFTHRESHOLD))
 739                 pause_thresh12 = svm->nested.ctl.pause_filter_thresh;
 740         else
 741                 pause_thresh12 = 0;
 742         if (kvm_pause_in_guest(svm->vcpu.kvm)) {
 743                 /* use guest values since host doesn't intercept PAUSE */
 744                 vmcb02->control.pause_filter_count = pause_count12;
 745                 vmcb02->control.pause_filter_thresh = pause_thresh12;
 746
 747         } else {
 748                 /* start from host values otherwise */
 749                 vmcb02->control.pause_filter_count = vmcb01->control.pause_filter_count;
 750                 vmcb02->control.pause_filter_thresh = vmcb01->control.pause_filter_thresh;
 751
 752                 /* ... but ensure filtering is disabled if so requested.  */
 753                 if (vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_PAUSE)) {
 754                         if (!pause_count12)
 755                                 vmcb02->control.pause_filter_count = 0;
 756                         if (!pause_thresh12)
 757                                 vmcb02->control.pause_filter_thresh = 0;
 758                 }
 759         }
 760
 761         nested_svm_transition_tlb_flush(vcpu);
 762
 763         /* Enter Guest-Mode */
 764         enter_guest_mode(vcpu);
 765
 766         /*
 767          * Merge guest and host intercepts - must be called with vcpu in
 768          * guest-mode to take effect.
 769          */
 770         recalc_intercepts(svm);
 771 }
 772
 773 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 774 {
 775         /*
 776          * Some VMCB state is shared between L1 and L2 and thus has to be
 777          * moved at the time of nested vmrun and vmexit.
 778          *
 779          * VMLOAD/VMSAVE state would also belong in this category, but KVM
 780          * always performs VMLOAD and VMSAVE from the VMCB01.
 781          */
 782         to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 783 }
 784
 785 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 786                          struct vmcb *vmcb12, bool from_vmrun)
 787 {
 788         struct vcpu_svm *svm = to_svm(vcpu);
 789         int ret;
 790
 791         trace_kvm_nested_vmenter(svm->vmcb->save.rip,
 792                                  vmcb12_gpa,
 793                                  vmcb12->save.rip,
 794                                  vmcb12->control.int_ctl,
 795                                  vmcb12->control.event_inj,
 796                                  vmcb12->control.nested_ctl,
 797                                  vmcb12->control.nested_cr3,
 798                                  vmcb12->save.cr3,
 799                                  KVM_ISA_SVM);
 800
 801         trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 802                                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 803                                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 804                                     vmcb12->control.intercepts[INTERCEPT_WORD3],
 805                                     vmcb12->control.intercepts[INTERCEPT_WORD4],
 806                                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
 807
 808
 809         svm->nested.vmcb12_gpa = vmcb12_gpa;
 810
 811         WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 812
 813         nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 814
 815         svm_switch_vmcb(svm, &svm->nested.vmcb02);
 816         nested_vmcb02_prepare_control(svm, vmcb12->save.rip, vmcb12->save.cs.base);
 817         nested_vmcb02_prepare_save(svm, vmcb12);
 818
 819         ret = nested_svm_load_cr3(&svm->vcpu, svm->nested.save.cr3,
 820                                   nested_npt_enabled(svm), from_vmrun);
 821         if (ret)
 822                 return ret;
 823
 824         if (!from_vmrun)
 825                 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 826
 827         svm_set_gif(svm, true);
 828
 829         if (kvm_vcpu_apicv_active(vcpu))
 830                 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
 831
 832         nested_svm_hv_update_vm_vp_ids(vcpu);
 833
 834         return 0;
 835 }
 836
 837 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 838 {
 839         struct vcpu_svm *svm = to_svm(vcpu);
 840         int ret;
 841         struct vmcb *vmcb12;
 842         struct kvm_host_map map;
 843         u64 vmcb12_gpa;
 844         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 845
 846         if (!svm->nested.hsave_msr) {
 847                 kvm_inject_gp(vcpu, 0);
 848                 return 1;
 849         }
 850
 851         if (is_smm(vcpu)) {
 852                 kvm_queue_exception(vcpu, UD_VECTOR);
 853                 return 1;
 854         }
 855
 856         /* This fails when VP assist page is enabled but the supplied GPA is bogus */
 857         ret = kvm_hv_verify_vp_assist(vcpu);
 858         if (ret) {
 859                 kvm_inject_gp(vcpu, 0);
 860                 return ret;
 861         }
 862
 863         vmcb12_gpa = svm->vmcb->save.rax;
 864         ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 865         if (ret == -EINVAL) {
 866                 kvm_inject_gp(vcpu, 0);
 867                 return 1;
 868         } else if (ret) {
 869                 return kvm_skip_emulated_instruction(vcpu);
 870         }
 871
 872         ret = kvm_skip_emulated_instruction(vcpu);
 873
 874         vmcb12 = map.hva;
 875
 876         if (WARN_ON_ONCE(!svm->nested.initialized))
 877                 return -EINVAL;
 878
 879         nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
 880         nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
 881
 882         if (!nested_vmcb_check_save(vcpu) ||
 883             !nested_vmcb_check_controls(vcpu)) {
 884                 vmcb12->control.exit_code    = SVM_EXIT_ERR;
 885                 vmcb12->control.exit_code_hi = 0;
 886                 vmcb12->control.exit_info_1  = 0;
 887                 vmcb12->control.exit_info_2  = 0;
 888                 goto out;
 889         }
 890
 891         /*
 892          * Since vmcb01 is not in use, we can use it to store some of the L1
 893          * state.
 894          */
 895         vmcb01->save.efer   = vcpu->arch.efer;
 896         vmcb01->save.cr0    = kvm_read_cr0(vcpu);
 897         vmcb01->save.cr4    = vcpu->arch.cr4;
 898         vmcb01->save.rflags = kvm_get_rflags(vcpu);
 899         vmcb01->save.rip    = kvm_rip_read(vcpu);
 900
 901         if (!npt_enabled)
 902                 vmcb01->save.cr3 = kvm_read_cr3(vcpu);
 903
 904         svm->nested.nested_run_pending = 1;
 905
 906         if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
 907                 goto out_exit_err;
 908
 909         if (nested_svm_vmrun_msrpm(svm))
 910                 goto out;
 911
 912 out_exit_err:
 913         svm->nested.nested_run_pending = 0;
 914         svm->nmi_l1_to_l2 = false;
 915         svm->soft_int_injected = false;
 916
 917         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 918         svm->vmcb->control.exit_code_hi = 0;
 919         svm->vmcb->control.exit_info_1  = 0;
 920         svm->vmcb->control.exit_info_2  = 0;
 921
 922         nested_svm_vmexit(svm);
 923
 924 out:
 925         kvm_vcpu_unmap(vcpu, &map, true);
 926
 927         return ret;
 928 }
 929
 930 /* Copy state save area fields which are handled by VMRUN */
 931 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
 932                           struct vmcb_save_area *from_save)
 933 {
 934         to_save->es = from_save->es;
 935         to_save->cs = from_save->cs;
 936         to_save->ss = from_save->ss;
 937         to_save->ds = from_save->ds;
 938         to_save->gdtr = from_save->gdtr;
 939         to_save->idtr = from_save->idtr;
 940         to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
 941         to_save->efer = from_save->efer;
 942         to_save->cr0 = from_save->cr0;
 943         to_save->cr3 = from_save->cr3;
 944         to_save->cr4 = from_save->cr4;
 945         to_save->rax = from_save->rax;
 946         to_save->rsp = from_save->rsp;
 947         to_save->rip = from_save->rip;
 948         to_save->cpl = 0;
 949 }
 950
 951 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
 952 {
 953         to_vmcb->save.fs = from_vmcb->save.fs;
 954         to_vmcb->save.gs = from_vmcb->save.gs;
 955         to_vmcb->save.tr = from_vmcb->save.tr;
 956         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 957         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 958         to_vmcb->save.star = from_vmcb->save.star;
 959         to_vmcb->save.lstar = from_vmcb->save.lstar;
 960         to_vmcb->save.cstar = from_vmcb->save.cstar;
 961         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 962         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 963         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 964         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 965 }
 966
 967 int nested_svm_vmexit(struct vcpu_svm *svm)
 968 {
 969         struct kvm_vcpu *vcpu = &svm->vcpu;
 970         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 971         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 972         struct vmcb *vmcb12;
 973         struct kvm_host_map map;
 974         int rc;
 975
 976         rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 977         if (rc) {
 978                 if (rc == -EINVAL)
 979                         kvm_inject_gp(vcpu, 0);
 980                 return 1;
 981         }
 982
 983         vmcb12 = map.hva;
 984
 985         /* Exit Guest-Mode */
 986         leave_guest_mode(vcpu);
 987         svm->nested.vmcb12_gpa = 0;
 988         WARN_ON_ONCE(svm->nested.nested_run_pending);
 989
 990         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 991
 992         /* in case we halted in L2 */
 993         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 994
 995         /* Give the current vmcb to the guest */
 996
 997         vmcb12->save.es     = vmcb02->save.es;
 998         vmcb12->save.cs     = vmcb02->save.cs;
 999         vmcb12->save.ss     = vmcb02->save.ss;
1000         vmcb12->save.ds     = vmcb02->save.ds;
1001         vmcb12->save.gdtr   = vmcb02->save.gdtr;
1002         vmcb12->save.idtr   = vmcb02->save.idtr;
1003         vmcb12->save.efer   = svm->vcpu.arch.efer;
1004         vmcb12->save.cr0    = kvm_read_cr0(vcpu);
1005         vmcb12->save.cr3    = kvm_read_cr3(vcpu);
1006         vmcb12->save.cr2    = vmcb02->save.cr2;
1007         vmcb12->save.cr4    = svm->vcpu.arch.cr4;
1008         vmcb12->save.rflags = kvm_get_rflags(vcpu);
1009         vmcb12->save.rip    = kvm_rip_read(vcpu);
1010         vmcb12->save.rsp    = kvm_rsp_read(vcpu);
1011         vmcb12->save.rax    = kvm_rax_read(vcpu);
1012         vmcb12->save.dr7    = vmcb02->save.dr7;
1013         vmcb12->save.dr6    = svm->vcpu.arch.dr6;
1014         vmcb12->save.cpl    = vmcb02->save.cpl;
1015
1016         vmcb12->control.int_state         = vmcb02->control.int_state;
1017         vmcb12->control.exit_code         = vmcb02->control.exit_code;
1018         vmcb12->control.exit_code_hi      = vmcb02->control.exit_code_hi;
1019         vmcb12->control.exit_info_1       = vmcb02->control.exit_info_1;
1020         vmcb12->control.exit_info_2       = vmcb02->control.exit_info_2;
1021
1022         if (vmcb12->control.exit_code != SVM_EXIT_ERR)
1023                 nested_save_pending_event_to_vmcb12(svm, vmcb12);
1024
1025         if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
1026                 vmcb12->control.next_rip  = vmcb02->control.next_rip;
1027
1028         vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
1029         vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
1030         vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
1031
1032         if (!kvm_pause_in_guest(vcpu->kvm)) {
1033                 vmcb01->control.pause_filter_count = vmcb02->control.pause_filter_count;
1034                 vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1035
1036         }
1037
1038         nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
1039
1040         svm_switch_vmcb(svm, &svm->vmcb01);
1041
1042         /*
1043          * Rules for synchronizing int_ctl bits from vmcb02 to vmcb01:
1044          *
1045          * V_IRQ, V_IRQ_VECTOR, V_INTR_PRIO_MASK, V_IGN_TPR:  If L1 doesn't
1046          * intercept interrupts, then KVM will use vmcb02's V_IRQ (and related
1047          * flags) to detect interrupt windows for L1 IRQs (even if L1 uses
1048          * virtual interrupt masking).  Raise KVM_REQ_EVENT to ensure that
1049          * KVM re-requests an interrupt window if necessary, which implicitly
1050          * copies this bits from vmcb02 to vmcb01.
1051          *
1052          * V_TPR: If L1 doesn't use virtual interrupt masking, then L1's vTPR
1053          * is stored in vmcb02, but its value doesn't need to be copied from/to
1054          * vmcb01 because it is copied from/to the virtual APIC's TPR register
1055          * on each VM entry/exit.
1056          *
1057          * V_GIF: If nested vGIF is not used, KVM uses vmcb02's V_GIF for L1's
1058          * V_GIF.  However, GIF is architecturally clear on each VM exit, thus
1059          * there is no need to copy V_GIF from vmcb02 to vmcb01.
1060          */
1061         if (!nested_exit_on_intr(svm))
1062                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
1063
1064         if (unlikely(guest_can_use(vcpu, X86_FEATURE_LBRV) &&
1065                      (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
1066                 svm_copy_lbrs(vmcb12, vmcb02);
1067                 svm_update_lbrv(vcpu);
1068         } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
1069                 svm_copy_lbrs(vmcb01, vmcb02);
1070                 svm_update_lbrv(vcpu);
1071         }
1072
1073         if (vnmi) {
1074                 if (vmcb02->control.int_ctl & V_NMI_BLOCKING_MASK)
1075                         vmcb01->control.int_ctl |= V_NMI_BLOCKING_MASK;
1076                 else
1077                         vmcb01->control.int_ctl &= ~V_NMI_BLOCKING_MASK;
1078
1079                 if (vcpu->arch.nmi_pending) {
1080                         vcpu->arch.nmi_pending--;
1081                         vmcb01->control.int_ctl |= V_NMI_PENDING_MASK;
1082                 } else {
1083                         vmcb01->control.int_ctl &= ~V_NMI_PENDING_MASK;
1084                 }
1085         }
1086
1087         /*
1088          * On vmexit the  GIF is set to false and
1089          * no event can be injected in L1.
1090          */
1091         svm_set_gif(svm, false);
1092         vmcb01->control.exit_int_info = 0;
1093
1094         svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
1095         if (vmcb01->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
1096                 vmcb01->control.tsc_offset = svm->vcpu.arch.tsc_offset;
1097                 vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1098         }
1099
1100         if (kvm_caps.has_tsc_control &&
1101             vcpu->arch.tsc_scaling_ratio != vcpu->arch.l1_tsc_scaling_ratio) {
1102                 vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
1103                 svm_write_tsc_multiplier(vcpu);
1104         }
1105
1106         svm->nested.ctl.nested_cr3 = 0;
1107
1108         /*
1109          * Restore processor state that had been saved in vmcb01
1110          */
1111         kvm_set_rflags(vcpu, vmcb01->save.rflags);
1112         svm_set_efer(vcpu, vmcb01->save.efer);
1113         svm_set_cr0(vcpu, vmcb01->save.cr0 | X86_CR0_PE);
1114         svm_set_cr4(vcpu, vmcb01->save.cr4);
1115         kvm_rax_write(vcpu, vmcb01->save.rax);
1116         kvm_rsp_write(vcpu, vmcb01->save.rsp);
1117         kvm_rip_write(vcpu, vmcb01->save.rip);
1118
1119         svm->vcpu.arch.dr7 = DR7_FIXED_1;
1120         kvm_update_dr7(&svm->vcpu);
1121
1122         trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
1123                                        vmcb12->control.exit_info_1,
1124                                        vmcb12->control.exit_info_2,
1125                                        vmcb12->control.exit_int_info,
1126                                        vmcb12->control.exit_int_info_err,
1127                                        KVM_ISA_SVM);
1128
1129         kvm_vcpu_unmap(vcpu, &map, true);
1130
1131         nested_svm_transition_tlb_flush(vcpu);
1132
1133         nested_svm_uninit_mmu_context(vcpu);
1134
1135         rc = nested_svm_load_cr3(vcpu, vmcb01->save.cr3, false, true);
1136         if (rc)
1137                 return 1;
1138
1139         /*
1140          * Drop what we picked up for L2 via svm_complete_interrupts() so it
1141          * doesn't end up in L1.
1142          */
1143         svm->vcpu.arch.nmi_injected = false;
1144         kvm_clear_exception_queue(vcpu);
1145         kvm_clear_interrupt_queue(vcpu);
1146
1147         /*
1148          * If we are here following the completion of a VMRUN that
1149          * is being single-stepped, queue the pending #DB intercept
1150          * right now so that it an be accounted for before we execute
1151          * L1's next instruction.
1152          */
1153         if (unlikely(vmcb01->save.rflags & X86_EFLAGS_TF))
1154                 kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
1155
1156         /*
1157          * Un-inhibit the AVIC right away, so that other vCPUs can start
1158          * to benefit from it right away.
1159          */
1160         if (kvm_apicv_activated(vcpu->kvm))
1161                 __kvm_vcpu_update_apicv(vcpu);
1162
1163         return 0;
1164 }
1165
1166 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
1167 {
1168         struct vcpu_svm *svm = to_svm(vcpu);
1169
1170         if (!vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SHUTDOWN))
1171                 return;
1172
1173         kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1174         nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
1175 }
1176
1177 int svm_allocate_nested(struct vcpu_svm *svm)
1178 {
1179         struct page *vmcb02_page;
1180
1181         if (svm->nested.initialized)
1182                 return 0;
1183
1184         vmcb02_page = snp_safe_alloc_page(&svm->vcpu);
1185         if (!vmcb02_page)
1186                 return -ENOMEM;
1187         svm->nested.vmcb02.ptr = page_address(vmcb02_page);
1188         svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
1189
1190         svm->nested.msrpm = svm_vcpu_alloc_msrpm();
1191         if (!svm->nested.msrpm)
1192                 goto err_free_vmcb02;
1193         svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
1194
1195         svm->nested.initialized = true;
1196         return 0;
1197
1198 err_free_vmcb02:
1199         __free_page(vmcb02_page);
1200         return -ENOMEM;
1201 }
1202
1203 void svm_free_nested(struct vcpu_svm *svm)
1204 {
1205         if (!svm->nested.initialized)
1206                 return;
1207
1208         if (WARN_ON_ONCE(svm->vmcb != svm->vmcb01.ptr))
1209                 svm_switch_vmcb(svm, &svm->vmcb01);
1210
1211         svm_vcpu_free_msrpm(svm->nested.msrpm);
1212         svm->nested.msrpm = NULL;
1213
1214         __free_page(virt_to_page(svm->nested.vmcb02.ptr));
1215         svm->nested.vmcb02.ptr = NULL;
1216
1217         /*
1218          * When last_vmcb12_gpa matches the current vmcb12 gpa,
1219          * some vmcb12 fields are not loaded if they are marked clean
1220          * in the vmcb12, since in this case they are up to date already.
1221          *
1222          * When the vmcb02 is freed, this optimization becomes invalid.
1223          */
1224         svm->nested.last_vmcb12_gpa = INVALID_GPA;
1225
1226         svm->nested.initialized = false;
1227 }
1228
1229 void svm_leave_nested(struct kvm_vcpu *vcpu)
1230 {
1231         struct vcpu_svm *svm = to_svm(vcpu);
1232
1233         if (is_guest_mode(vcpu)) {
1234                 svm->nested.nested_run_pending = 0;
1235                 svm->nested.vmcb12_gpa = INVALID_GPA;
1236
1237                 leave_guest_mode(vcpu);
1238
1239                 svm_switch_vmcb(svm, &svm->vmcb01);
1240
1241                 nested_svm_uninit_mmu_context(vcpu);
1242                 vmcb_mark_all_dirty(svm->vmcb);
1243
1244                 if (kvm_apicv_activated(vcpu->kvm))
1245                         kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
1246         }
1247
1248         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1249 }
1250
1251 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1252 {
1253         u32 offset, msr, value;
1254         int write, mask;
1255
1256         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
1257                 return NESTED_EXIT_HOST;
1258
1259         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1260         offset = svm_msrpm_offset(msr);
1261         write  = svm->vmcb->control.exit_info_1 & 1;
1262         mask   = 1 << ((2 * (msr & 0xf)) + write);
1263
1264         if (offset == MSR_INVALID)
1265                 return NESTED_EXIT_DONE;
1266
1267         /* Offset is in 32 bit units but need in 8 bit units */
1268         offset *= 4;
1269
1270         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
1271                 return NESTED_EXIT_DONE;
1272
1273         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1274 }
1275
1276 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1277 {
1278         unsigned port, size, iopm_len;
1279         u16 val, mask;
1280         u8 start_bit;
1281         u64 gpa;
1282
1283         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
1284                 return NESTED_EXIT_HOST;
1285
1286         port = svm->vmcb->control.exit_info_1 >> 16;
1287         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1288                 SVM_IOIO_SIZE_SHIFT;
1289         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1290         start_bit = port % 8;
1291         iopm_len = (start_bit + size > 8) ? 2 : 1;
1292         mask = (0xf >> (4 - size)) << start_bit;
1293         val = 0;
1294
1295         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1296                 return NESTED_EXIT_DONE;
1297
1298         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1299 }
1300
1301 static int nested_svm_intercept(struct vcpu_svm *svm)
1302 {
1303         u32 exit_code = svm->vmcb->control.exit_code;
1304         int vmexit = NESTED_EXIT_HOST;
1305
1306         switch (exit_code) {
1307         case SVM_EXIT_MSR:
1308                 vmexit = nested_svm_exit_handled_msr(svm);
1309                 break;
1310         case SVM_EXIT_IOIO:
1311                 vmexit = nested_svm_intercept_ioio(svm);
1312                 break;
1313         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1314                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1315                         vmexit = NESTED_EXIT_DONE;
1316                 break;
1317         }
1318         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1319                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1320                         vmexit = NESTED_EXIT_DONE;
1321                 break;
1322         }
1323         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1324                 /*
1325                  * Host-intercepted exceptions have been checked already in
1326                  * nested_svm_exit_special.  There is nothing to do here,
1327                  * the vmexit is injected by svm_check_nested_events.
1328                  */
1329                 vmexit = NESTED_EXIT_DONE;
1330                 break;
1331         }
1332         case SVM_EXIT_ERR: {
1333                 vmexit = NESTED_EXIT_DONE;
1334                 break;
1335         }
1336         default: {
1337                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1338                         vmexit = NESTED_EXIT_DONE;
1339         }
1340         }
1341
1342         return vmexit;
1343 }
1344
1345 int nested_svm_exit_handled(struct vcpu_svm *svm)
1346 {
1347         int vmexit;
1348
1349         vmexit = nested_svm_intercept(svm);
1350
1351         if (vmexit == NESTED_EXIT_DONE)
1352                 nested_svm_vmexit(svm);
1353
1354         return vmexit;
1355 }
1356
1357 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1358 {
1359         if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1360                 kvm_queue_exception(vcpu, UD_VECTOR);
1361                 return 1;
1362         }
1363
1364         if (to_svm(vcpu)->vmcb->save.cpl) {
1365                 kvm_inject_gp(vcpu, 0);
1366                 return 1;
1367         }
1368
1369         return 0;
1370 }
1371
1372 static bool nested_svm_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
1373                                            u32 error_code)
1374 {
1375         struct vcpu_svm *svm = to_svm(vcpu);
1376
1377         return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(vector));
1378 }
1379
1380 static void nested_svm_inject_exception_vmexit(struct kvm_vcpu *vcpu)
1381 {
1382         struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
1383         struct vcpu_svm *svm = to_svm(vcpu);
1384         struct vmcb *vmcb = svm->vmcb;
1385
1386         vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + ex->vector;
1387         vmcb->control.exit_code_hi = 0;
1388
1389         if (ex->has_error_code)
1390                 vmcb->control.exit_info_1 = ex->error_code;
1391
1392         /*
1393          * EXITINFO2 is undefined for all exception intercepts other
1394          * than #PF.
1395          */
1396         if (ex->vector == PF_VECTOR) {
1397                 if (ex->has_payload)
1398                         vmcb->control.exit_info_2 = ex->payload;
1399                 else
1400                         vmcb->control.exit_info_2 = vcpu->arch.cr2;
1401         } else if (ex->vector == DB_VECTOR) {
1402                 /* See kvm_check_and_inject_events().  */
1403                 kvm_deliver_exception_payload(vcpu, ex);
1404
1405                 if (vcpu->arch.dr7 & DR7_GD) {
1406                         vcpu->arch.dr7 &= ~DR7_GD;
1407                         kvm_update_dr7(vcpu);
1408                 }
1409         } else {
1410                 WARN_ON(ex->has_payload);
1411         }
1412
1413         nested_svm_vmexit(svm);
1414 }
1415
1416 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1417 {
1418         return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1419 }
1420
1421 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1422 {
1423         struct kvm_lapic *apic = vcpu->arch.apic;
1424         struct vcpu_svm *svm = to_svm(vcpu);
1425         /*
1426          * Only a pending nested run blocks a pending exception.  If there is a
1427          * previously injected event, the pending exception occurred while said
1428          * event was being delivered and thus needs to be handled.
1429          */
1430         bool block_nested_exceptions = svm->nested.nested_run_pending;
1431         /*
1432          * New events (not exceptions) are only recognized at instruction
1433          * boundaries.  If an event needs reinjection, then KVM is handling a
1434          * VM-Exit that occurred _during_ instruction execution; new events are
1435          * blocked until the instruction completes.
1436          */
1437         bool block_nested_events = block_nested_exceptions ||
1438                                    kvm_event_needs_reinjection(vcpu);
1439
1440         if (lapic_in_kernel(vcpu) &&
1441             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1442                 if (block_nested_events)
1443                         return -EBUSY;
1444                 if (!nested_exit_on_init(svm))
1445                         return 0;
1446                 nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1447                 return 0;
1448         }
1449
1450         if (vcpu->arch.exception_vmexit.pending) {
1451                 if (block_nested_exceptions)
1452                         return -EBUSY;
1453                 nested_svm_inject_exception_vmexit(vcpu);
1454                 return 0;
1455         }
1456
1457         if (vcpu->arch.exception.pending) {
1458                 if (block_nested_exceptions)
1459                         return -EBUSY;
1460                 return 0;
1461         }
1462
1463 #ifdef CONFIG_KVM_SMM
1464         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1465                 if (block_nested_events)
1466                         return -EBUSY;
1467                 if (!nested_exit_on_smi(svm))
1468                         return 0;
1469                 nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1470                 return 0;
1471         }
1472 #endif
1473
1474         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1475                 if (block_nested_events)
1476                         return -EBUSY;
1477                 if (!nested_exit_on_nmi(svm))
1478                         return 0;
1479                 nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1480                 return 0;
1481         }
1482
1483         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1484                 if (block_nested_events)
1485                         return -EBUSY;
1486                 if (!nested_exit_on_intr(svm))
1487                         return 0;
1488                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1489                 nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1490                 return 0;
1491         }
1492
1493         return 0;
1494 }
1495
1496 int nested_svm_exit_special(struct vcpu_svm *svm)
1497 {
1498         u32 exit_code = svm->vmcb->control.exit_code;
1499         struct kvm_vcpu *vcpu = &svm->vcpu;
1500
1501         switch (exit_code) {
1502         case SVM_EXIT_INTR:
1503         case SVM_EXIT_NMI:
1504         case SVM_EXIT_NPF:
1505                 return NESTED_EXIT_HOST;
1506         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1507                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1508
1509                 if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1510                     excp_bits)
1511                         return NESTED_EXIT_HOST;
1512                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1513                          svm->vcpu.arch.apf.host_apf_flags)
1514                         /* Trap async PF even if not shadowing */
1515                         return NESTED_EXIT_HOST;
1516                 break;
1517         }
1518         case SVM_EXIT_VMMCALL:
1519                 /* Hyper-V L2 TLB flush hypercall is handled by L0 */
1520                 if (guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
1521                     nested_svm_l2_tlb_flush_enabled(vcpu) &&
1522                     kvm_hv_is_tlb_flush_hcall(vcpu))
1523                         return NESTED_EXIT_HOST;
1524                 break;
1525         default:
1526                 break;
1527         }
1528
1529         return NESTED_EXIT_CONTINUE;
1530 }
1531
1532 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
1533 {
1534         struct vcpu_svm *svm = to_svm(vcpu);
1535
1536         vcpu->arch.tsc_scaling_ratio =
1537                 kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
1538                                                svm->tsc_ratio_msr);
1539         svm_write_tsc_multiplier(vcpu);
1540 }
1541
1542 /* Inverse operation of nested_copy_vmcb_control_to_cache(). asid is copied too. */
1543 static void nested_copy_vmcb_cache_to_control(struct vmcb_control_area *dst,
1544                                               struct vmcb_ctrl_area_cached *from)
1545 {
1546         unsigned int i;
1547
1548         memset(dst, 0, sizeof(struct vmcb_control_area));
1549
1550         for (i = 0; i < MAX_INTERCEPT; i++)
1551                 dst->intercepts[i] = from->intercepts[i];
1552
1553         dst->iopm_base_pa         = from->iopm_base_pa;
1554         dst->msrpm_base_pa        = from->msrpm_base_pa;
1555         dst->tsc_offset           = from->tsc_offset;
1556         dst->asid                 = from->asid;
1557         dst->tlb_ctl              = from->tlb_ctl;
1558         dst->int_ctl              = from->int_ctl;
1559         dst->int_vector           = from->int_vector;
1560         dst->int_state            = from->int_state;
1561         dst->exit_code            = from->exit_code;
1562         dst->exit_code_hi         = from->exit_code_hi;
1563         dst->exit_info_1          = from->exit_info_1;
1564         dst->exit_info_2          = from->exit_info_2;
1565         dst->exit_int_info        = from->exit_int_info;
1566         dst->exit_int_info_err    = from->exit_int_info_err;
1567         dst->nested_ctl           = from->nested_ctl;
1568         dst->event_inj            = from->event_inj;
1569         dst->event_inj_err        = from->event_inj_err;
1570         dst->next_rip             = from->next_rip;
1571         dst->nested_cr3           = from->nested_cr3;
1572         dst->virt_ext              = from->virt_ext;
1573         dst->pause_filter_count   = from->pause_filter_count;
1574         dst->pause_filter_thresh  = from->pause_filter_thresh;
1575         /* 'clean' and 'hv_enlightenments' are not changed by KVM */
1576 }
1577
1578 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1579                                 struct kvm_nested_state __user *user_kvm_nested_state,
1580                                 u32 user_data_size)
1581 {
1582         struct vcpu_svm *svm;
1583         struct vmcb_control_area *ctl;
1584         unsigned long r;
1585         struct kvm_nested_state kvm_state = {
1586                 .flags = 0,
1587                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1588                 .size = sizeof(kvm_state),
1589         };
1590         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1591                 &user_kvm_nested_state->data.svm[0];
1592
1593         if (!vcpu)
1594                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1595
1596         svm = to_svm(vcpu);
1597
1598         if (user_data_size < kvm_state.size)
1599                 goto out;
1600
1601         /* First fill in the header and copy it out.  */
1602         if (is_guest_mode(vcpu)) {
1603                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1604                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1605                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1606
1607                 if (svm->nested.nested_run_pending)
1608                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1609         }
1610
1611         if (gif_set(svm))
1612                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1613
1614         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1615                 return -EFAULT;
1616
1617         if (!is_guest_mode(vcpu))
1618                 goto out;
1619
1620         /*
1621          * Copy over the full size of the VMCB rather than just the size
1622          * of the structs.
1623          */
1624         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1625                 return -EFAULT;
1626
1627         ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
1628         if (!ctl)
1629                 return -ENOMEM;
1630
1631         nested_copy_vmcb_cache_to_control(ctl, &svm->nested.ctl);
1632         r = copy_to_user(&user_vmcb->control, ctl,
1633                          sizeof(user_vmcb->control));
1634         kfree(ctl);
1635         if (r)
1636                 return -EFAULT;
1637
1638         if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1639                          sizeof(user_vmcb->save)))
1640                 return -EFAULT;
1641 out:
1642         return kvm_state.size;
1643 }
1644
1645 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1646                                 struct kvm_nested_state __user *user_kvm_nested_state,
1647                                 struct kvm_nested_state *kvm_state)
1648 {
1649         struct vcpu_svm *svm = to_svm(vcpu);
1650         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1651                 &user_kvm_nested_state->data.svm[0];
1652         struct vmcb_control_area *ctl;
1653         struct vmcb_save_area *save;
1654         struct vmcb_save_area_cached save_cached;
1655         struct vmcb_ctrl_area_cached ctl_cached;
1656         unsigned long cr0;
1657         int ret;
1658
1659         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1660                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1661
1662         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1663                 return -EINVAL;
1664
1665         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1666                                  KVM_STATE_NESTED_RUN_PENDING |
1667                                  KVM_STATE_NESTED_GIF_SET))
1668                 return -EINVAL;
1669
1670         /*
1671          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1672          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1673          */
1674         if (!(vcpu->arch.efer & EFER_SVME)) {
1675                 /* GIF=1 and no guest mode are required if SVME=0.  */
1676                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1677                         return -EINVAL;
1678         }
1679
1680         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1681         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1682                 return -EINVAL;
1683
1684         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1685                 svm_leave_nested(vcpu);
1686                 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1687                 return 0;
1688         }
1689
1690         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1691                 return -EINVAL;
1692         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1693                 return -EINVAL;
1694
1695         ret  = -ENOMEM;
1696         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1697         save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1698         if (!ctl || !save)
1699                 goto out_free;
1700
1701         ret = -EFAULT;
1702         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1703                 goto out_free;
1704         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1705                 goto out_free;
1706
1707         ret = -EINVAL;
1708         __nested_copy_vmcb_control_to_cache(vcpu, &ctl_cached, ctl);
1709         if (!__nested_vmcb_check_controls(vcpu, &ctl_cached))
1710                 goto out_free;
1711
1712         /*
1713          * Processor state contains L2 state.  Check that it is
1714          * valid for guest mode (see nested_vmcb_check_save).
1715          */
1716         cr0 = kvm_read_cr0(vcpu);
1717         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1718                 goto out_free;
1719
1720         /*
1721          * Validate host state saved from before VMRUN (see
1722          * nested_svm_check_permissions).
1723          */
1724         __nested_copy_vmcb_save_to_cache(&save_cached, save);
1725         if (!(save->cr0 & X86_CR0_PG) ||
1726             !(save->cr0 & X86_CR0_PE) ||
1727             (save->rflags & X86_EFLAGS_VM) ||
1728             !__nested_vmcb_check_save(vcpu, &save_cached))
1729                 goto out_free;
1730
1731
1732         /*
1733          * All checks done, we can enter guest mode. Userspace provides
1734          * vmcb12.control, which will be combined with L1 and stored into
1735          * vmcb02, and the L1 save state which we store in vmcb01.
1736          * L2 registers if needed are moved from the current VMCB to VMCB02.
1737          */
1738
1739         if (is_guest_mode(vcpu))
1740                 svm_leave_nested(vcpu);
1741         else
1742                 svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1743
1744         svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1745
1746         svm->nested.nested_run_pending =
1747                 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1748
1749         svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1750
1751         svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1752         nested_copy_vmcb_control_to_cache(svm, ctl);
1753
1754         svm_switch_vmcb(svm, &svm->nested.vmcb02);
1755         nested_vmcb02_prepare_control(svm, svm->vmcb->save.rip, svm->vmcb->save.cs.base);
1756
1757         /*
1758          * While the nested guest CR3 is already checked and set by
1759          * KVM_SET_SREGS, it was set when nested state was yet loaded,
1760          * thus MMU might not be initialized correctly.
1761          * Set it again to fix this.
1762          */
1763
1764         ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1765                                   nested_npt_enabled(svm), false);
1766         if (WARN_ON_ONCE(ret))
1767                 goto out_free;
1768
1769         svm->nested.force_msr_bitmap_recalc = true;
1770
1771         kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1772         ret = 0;
1773 out_free:
1774         kfree(save);
1775         kfree(ctl);
1776
1777         return ret;
1778 }
1779
1780 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1781 {
1782         struct vcpu_svm *svm = to_svm(vcpu);
1783
1784         if (WARN_ON(!is_guest_mode(vcpu)))
1785                 return true;
1786
1787         if (!vcpu->arch.pdptrs_from_userspace &&
1788             !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1789                 /*
1790                  * Reload the guest's PDPTRs since after a migration
1791                  * the guest CR3 might be restored prior to setting the nested
1792                  * state which can lead to a load of wrong PDPTRs.
1793                  */
1794                 if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
1795                         return false;
1796
1797         if (!nested_svm_vmrun_msrpm(svm)) {
1798                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1799                 vcpu->run->internal.suberror =
1800                         KVM_INTERNAL_ERROR_EMULATION;
1801                 vcpu->run->internal.ndata = 0;
1802                 return false;
1803         }
1804
1805         if (kvm_hv_verify_vp_assist(vcpu))
1806                 return false;
1807
1808         return true;
1809 }
1810
1811 struct kvm_x86_nested_ops svm_nested_ops = {
1812         .leave_nested = svm_leave_nested,
1813         .is_exception_vmexit = nested_svm_is_exception_vmexit,
1814         .check_events = svm_check_nested_events,
1815         .triple_fault = nested_svm_triple_fault,
1816         .get_nested_state_pages = svm_get_nested_state_pages,
1817         .get_state = svm_get_nested_state,
1818         .set_state = svm_set_nested_state,
1819         .hv_inject_synthetic_vmexit_post_tlb_flush = svm_hv_inject_synthetic_vmexit_post_tlb_flush,
1820 };