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         struct hv_vmcb_enlightenments *hve = &svm->nested.ctl.hv_enlightenments;
 191         int i;
 192
 193         /*
 194          * MSR bitmap update can be skipped when:
 195          * - MSR bitmap for L1 hasn't changed.
 196          * - Nested hypervisor (L1) is attempting to launch the same L2 as
 197          *   before.
 198          * - Nested hypervisor (L1) is using Hyper-V emulation interface and
 199          * tells KVM (L0) there were no changes in MSR bitmap for L2.
 200          */
 201         if (!svm->nested.force_msr_bitmap_recalc &&
 202             kvm_hv_hypercall_enabled(&svm->vcpu) &&
 203             hve->hv_enlightenments_control.msr_bitmap &&
 204             (svm->nested.ctl.clean & BIT(HV_VMCB_NESTED_ENLIGHTENMENTS)))
 205                 goto set_msrpm_base_pa;
 206
 207         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 208                 return true;
 209
 210         for (i = 0; i < MSRPM_OFFSETS; i++) {
 211                 u32 value, p;
 212                 u64 offset;
 213
 214                 if (msrpm_offsets[i] == 0xffffffff)
 215                         break;
 216
 217                 p      = msrpm_offsets[i];
 218
 219                 /* x2apic msrs are intercepted always for the nested guest */
 220                 if (is_x2apic_msrpm_offset(p))
 221                         continue;
 222
 223                 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 224
 225                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 226                         return false;
 227
 228                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 229         }
 230
 231         svm->nested.force_msr_bitmap_recalc = false;
 232
 233 set_msrpm_base_pa:
 234         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 235
 236         return true;
 237 }
 238
 239 /*
 240  * Bits 11:0 of bitmap address are ignored by hardware
 241  */
 242 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
 243 {
 244         u64 addr = PAGE_ALIGN(pa);
 245
 246         return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
 247             kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
 248 }
 249
 250 static bool nested_svm_check_tlb_ctl(struct kvm_vcpu *vcpu, u8 tlb_ctl)
 251 {
 252         /* Nested FLUSHBYASID is not supported yet.  */
 253         switch(tlb_ctl) {
 254                 case TLB_CONTROL_DO_NOTHING:
 255                 case TLB_CONTROL_FLUSH_ALL_ASID:
 256                         return true;
 257                 default:
 258                         return false;
 259         }
 260 }
 261
 262 static bool __nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
 263                                          struct vmcb_ctrl_area_cached *control)
 264 {
 265         if (CC(!vmcb12_is_intercept(control, INTERCEPT_VMRUN)))
 266                 return false;
 267
 268         if (CC(control->asid == 0))
 269                 return false;
 270
 271         if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
 272                 return false;
 273
 274         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
 275                                            MSRPM_SIZE)))
 276                 return false;
 277         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
 278                                            IOPM_SIZE)))
 279                 return false;
 280
 281         if (CC(!nested_svm_check_tlb_ctl(vcpu, control->tlb_ctl)))
 282                 return false;
 283
 284         if (CC((control->int_ctl & V_NMI_ENABLE_MASK) &&
 285                !vmcb12_is_intercept(control, INTERCEPT_NMI))) {
 286                 return false;
 287         }
 288
 289         return true;
 290 }
 291
 292 /* Common checks that apply to both L1 and L2 state.  */
 293 static bool __nested_vmcb_check_save(struct kvm_vcpu *vcpu,
 294                                      struct vmcb_save_area_cached *save)
 295 {
 296         if (CC(!(save->efer & EFER_SVME)))
 297                 return false;
 298
 299         if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 300             CC(save->cr0 & ~0xffffffffULL))
 301                 return false;
 302
 303         if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 304                 return false;
 305
 306         /*
 307          * These checks are also performed by KVM_SET_SREGS,
 308          * except that EFER.LMA is not checked by SVM against
 309          * CR0.PG && EFER.LME.
 310          */
 311         if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
 312                 if (CC(!(save->cr4 & X86_CR4_PAE)) ||
 313                     CC(!(save->cr0 & X86_CR0_PE)) ||
 314                     CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
 315                         return false;
 316         }
 317
 318         /* Note, SVM doesn't have any additional restrictions on CR4. */
 319         if (CC(!__kvm_is_valid_cr4(vcpu, save->cr4)))
 320                 return false;
 321
 322         if (CC(!kvm_valid_efer(vcpu, save->efer)))
 323                 return false;
 324
 325         return true;
 326 }
 327
 328 static bool nested_vmcb_check_save(struct kvm_vcpu *vcpu)
 329 {
 330         struct vcpu_svm *svm = to_svm(vcpu);
 331         struct vmcb_save_area_cached *save = &svm->nested.save;
 332
 333         return __nested_vmcb_check_save(vcpu, save);
 334 }
 335
 336 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu)
 337 {
 338         struct vcpu_svm *svm = to_svm(vcpu);
 339         struct vmcb_ctrl_area_cached *ctl = &svm->nested.ctl;
 340
 341         return __nested_vmcb_check_controls(vcpu, ctl);
 342 }
 343
 344 static
 345 void __nested_copy_vmcb_control_to_cache(struct kvm_vcpu *vcpu,
 346                                          struct vmcb_ctrl_area_cached *to,
 347                                          struct vmcb_control_area *from)
 348 {
 349         unsigned int i;
 350
 351         for (i = 0; i < MAX_INTERCEPT; i++)
 352                 to->intercepts[i] = from->intercepts[i];
 353
 354         to->iopm_base_pa        = from->iopm_base_pa;
 355         to->msrpm_base_pa       = from->msrpm_base_pa;
 356         to->tsc_offset          = from->tsc_offset;
 357         to->tlb_ctl             = from->tlb_ctl;
 358         to->int_ctl             = from->int_ctl;
 359         to->int_vector          = from->int_vector;
 360         to->int_state           = from->int_state;
 361         to->exit_code           = from->exit_code;
 362         to->exit_code_hi        = from->exit_code_hi;
 363         to->exit_info_1         = from->exit_info_1;
 364         to->exit_info_2         = from->exit_info_2;
 365         to->exit_int_info       = from->exit_int_info;
 366         to->exit_int_info_err   = from->exit_int_info_err;
 367         to->nested_ctl          = from->nested_ctl;
 368         to->event_inj           = from->event_inj;
 369         to->event_inj_err       = from->event_inj_err;
 370         to->next_rip            = from->next_rip;
 371         to->nested_cr3          = from->nested_cr3;
 372         to->virt_ext            = from->virt_ext;
 373         to->pause_filter_count  = from->pause_filter_count;
 374         to->pause_filter_thresh = from->pause_filter_thresh;
 375
 376         /* Copy asid here because nested_vmcb_check_controls will check it.  */
 377         to->asid           = from->asid;
 378         to->msrpm_base_pa &= ~0x0fffULL;
 379         to->iopm_base_pa  &= ~0x0fffULL;
 380
 381         /* Hyper-V extensions (Enlightened VMCB) */
 382         if (kvm_hv_hypercall_enabled(vcpu)) {
 383                 to->clean = from->clean;
 384                 memcpy(&to->hv_enlightenments, &from->hv_enlightenments,
 385                        sizeof(to->hv_enlightenments));
 386         }
 387 }
 388
 389 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
 390                                        struct vmcb_control_area *control)
 391 {
 392         __nested_copy_vmcb_control_to_cache(&svm->vcpu, &svm->nested.ctl, control);
 393 }
 394
 395 static void __nested_copy_vmcb_save_to_cache(struct vmcb_save_area_cached *to,
 396                                              struct vmcb_save_area *from)
 397 {
 398         /*
 399          * Copy only fields that are validated, as we need them
 400          * to avoid TOC/TOU races.
 401          */
 402         to->efer = from->efer;
 403         to->cr0 = from->cr0;
 404         to->cr3 = from->cr3;
 405         to->cr4 = from->cr4;
 406
 407         to->dr6 = from->dr6;
 408         to->dr7 = from->dr7;
 409 }
 410
 411 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
 412                                     struct vmcb_save_area *save)
 413 {
 414         __nested_copy_vmcb_save_to_cache(&svm->nested.save, save);
 415 }
 416
 417 /*
 418  * Synchronize fields that are written by the processor, so that
 419  * they can be copied back into the vmcb12.
 420  */
 421 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
 422 {
 423         u32 mask;
 424         svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 425         svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 426
 427         /* Only a few fields of int_ctl are written by the processor.  */
 428         mask = V_IRQ_MASK | V_TPR_MASK;
 429         /*
 430          * Don't sync vmcb02 V_IRQ back to vmcb12 if KVM (L0) is intercepting
 431          * virtual interrupts in order to request an interrupt window, as KVM
 432          * has usurped vmcb02's int_ctl.  If an interrupt window opens before
 433          * the next VM-Exit, svm_clear_vintr() will restore vmcb12's int_ctl.
 434          * If no window opens, V_IRQ will be correctly preserved in vmcb12's
 435          * int_ctl (because it was never recognized while L2 was running).
 436          */
 437         if (svm_is_intercept(svm, INTERCEPT_VINTR) &&
 438             !test_bit(INTERCEPT_VINTR, (unsigned long *)svm->nested.ctl.intercepts))
 439                 mask &= ~V_IRQ_MASK;
 440
 441         if (nested_vgif_enabled(svm))
 442                 mask |= V_GIF_MASK;
 443
 444         if (nested_vnmi_enabled(svm))
 445                 mask |= V_NMI_BLOCKING_MASK | V_NMI_PENDING_MASK;
 446
 447         svm->nested.ctl.int_ctl        &= ~mask;
 448         svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 449 }
 450
 451 /*
 452  * Transfer any event that L0 or L1 wanted to inject into L2 to
 453  * EXIT_INT_INFO.
 454  */
 455 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
 456                                                 struct vmcb *vmcb12)
 457 {
 458         struct kvm_vcpu *vcpu = &svm->vcpu;
 459         u32 exit_int_info = 0;
 460         unsigned int nr;
 461
 462         if (vcpu->arch.exception.injected) {
 463                 nr = vcpu->arch.exception.vector;
 464                 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 465
 466                 if (vcpu->arch.exception.has_error_code) {
 467                         exit_int_info |= SVM_EVTINJ_VALID_ERR;
 468                         vmcb12->control.exit_int_info_err =
 469                                 vcpu->arch.exception.error_code;
 470                 }
 471
 472         } else if (vcpu->arch.nmi_injected) {
 473                 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 474
 475         } else if (vcpu->arch.interrupt.injected) {
 476                 nr = vcpu->arch.interrupt.nr;
 477                 exit_int_info = nr | SVM_EVTINJ_VALID;
 478
 479                 if (vcpu->arch.interrupt.soft)
 480                         exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 481                 else
 482                         exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 483         }
 484
 485         vmcb12->control.exit_int_info = exit_int_info;
 486 }
 487
 488 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
 489 {
 490         /*
 491          * KVM_REQ_HV_TLB_FLUSH flushes entries from either L1's VP_ID or
 492          * L2's VP_ID upon request from the guest. Make sure we check for
 493          * pending entries in the right FIFO upon L1/L2 transition as these
 494          * requests are put by other vCPUs asynchronously.
 495          */
 496         if (to_hv_vcpu(vcpu) && npt_enabled)
 497                 kvm_make_request(KVM_REQ_HV_TLB_FLUSH, vcpu);
 498
 499         /*
 500          * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
 501          * things to fix before this can be conditional:
 502          *
 503          *  - Flush TLBs for both L1 and L2 remote TLB flush
 504          *  - Honor L1's request to flush an ASID on nested VMRUN
 505          *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
 506          *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
 507          *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
 508          *
 509          * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
 510          *     NPT guest-physical mappings on VMRUN.
 511          */
 512         kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
 513         kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
 514 }
 515
 516 /*
 517  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 518  * if we are emulating VM-Entry into a guest with NPT enabled.
 519  */
 520 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 521                                bool nested_npt, bool reload_pdptrs)
 522 {
 523         if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
 524                 return -EINVAL;
 525
 526         if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
 527             CC(!load_pdptrs(vcpu, cr3)))
 528                 return -EINVAL;
 529
 530         vcpu->arch.cr3 = cr3;
 531
 532         /* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
 533         kvm_init_mmu(vcpu);
 534
 535         if (!nested_npt)
 536                 kvm_mmu_new_pgd(vcpu, cr3);
 537
 538         return 0;
 539 }
 540
 541 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
 542 {
 543         if (!svm->nested.vmcb02.ptr)
 544                 return;
 545
 546         /* FIXME: merge g_pat from vmcb01 and vmcb12.  */
 547         svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
 548 }
 549
 550 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 551 {
 552         bool new_vmcb12 = false;
 553         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 554         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 555         struct kvm_vcpu *vcpu = &svm->vcpu;
 556
 557         nested_vmcb02_compute_g_pat(svm);
 558
 559         /* Load the nested guest state */
 560         if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
 561                 new_vmcb12 = true;
 562                 svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
 563                 svm->nested.force_msr_bitmap_recalc = true;
 564         }
 565
 566         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
 567                 vmcb02->save.es = vmcb12->save.es;
 568                 vmcb02->save.cs = vmcb12->save.cs;
 569                 vmcb02->save.ss = vmcb12->save.ss;
 570                 vmcb02->save.ds = vmcb12->save.ds;
 571                 vmcb02->save.cpl = vmcb12->save.cpl;
 572                 vmcb_mark_dirty(vmcb02, VMCB_SEG);
 573         }
 574
 575         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
 576                 vmcb02->save.gdtr = vmcb12->save.gdtr;
 577                 vmcb02->save.idtr = vmcb12->save.idtr;
 578                 vmcb_mark_dirty(vmcb02, VMCB_DT);
 579         }
 580
 581         kvm_set_rflags(vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
 582
 583         svm_set_efer(vcpu, svm->nested.save.efer);
 584
 585         svm_set_cr0(vcpu, svm->nested.save.cr0);
 586         svm_set_cr4(vcpu, svm->nested.save.cr4);
 587
 588         svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 589
 590         kvm_rax_write(vcpu, vmcb12->save.rax);
 591         kvm_rsp_write(vcpu, vmcb12->save.rsp);
 592         kvm_rip_write(vcpu, vmcb12->save.rip);
 593
 594         /* In case we don't even reach vcpu_run, the fields are not updated */
 595         vmcb02->save.rax = vmcb12->save.rax;
 596         vmcb02->save.rsp = vmcb12->save.rsp;
 597         vmcb02->save.rip = vmcb12->save.rip;
 598
 599         /* These bits will be set properly on the first execution when new_vmc12 is true */
 600         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 601                 vmcb02->save.dr7 = svm->nested.save.dr7 | DR7_FIXED_1;
 602                 svm->vcpu.arch.dr6  = svm->nested.save.dr6 | DR6_ACTIVE_LOW;
 603                 vmcb_mark_dirty(vmcb02, VMCB_DR);
 604         }
 605
 606         if (unlikely(guest_can_use(vcpu, X86_FEATURE_LBRV) &&
 607                      (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
 608                 /*
 609                  * Reserved bits of DEBUGCTL are ignored.  Be consistent with
 610                  * svm_set_msr's definition of reserved bits.
 611                  */
 612                 svm_copy_lbrs(vmcb02, vmcb12);
 613                 vmcb02->save.dbgctl &= ~DEBUGCTL_RESERVED_BITS;
 614                 svm_update_lbrv(&svm->vcpu);
 615
 616         } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
 617                 svm_copy_lbrs(vmcb02, vmcb01);
 618         }
 619 }
 620
 621 static inline bool is_evtinj_soft(u32 evtinj)
 622 {
 623         u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
 624         u8 vector = evtinj & SVM_EVTINJ_VEC_MASK;
 625
 626         if (!(evtinj & SVM_EVTINJ_VALID))
 627                 return false;
 628
 629         if (type == SVM_EVTINJ_TYPE_SOFT)
 630                 return true;
 631
 632         return type == SVM_EVTINJ_TYPE_EXEPT && kvm_exception_is_soft(vector);
 633 }
 634
 635 static bool is_evtinj_nmi(u32 evtinj)
 636 {
 637         u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
 638
 639         if (!(evtinj & SVM_EVTINJ_VALID))
 640                 return false;
 641
 642         return type == SVM_EVTINJ_TYPE_NMI;
 643 }
 644
 645 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm,
 646                                           unsigned long vmcb12_rip,
 647                                           unsigned long vmcb12_csbase)
 648 {
 649         u32 int_ctl_vmcb01_bits = V_INTR_MASKING_MASK;
 650         u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
 651
 652         struct kvm_vcpu *vcpu = &svm->vcpu;
 653         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 654         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 655         u32 pause_count12;
 656         u32 pause_thresh12;
 657
 658         /*
 659          * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 660          * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 661          */
 662
 663         if (guest_can_use(vcpu, X86_FEATURE_VGIF) &&
 664             (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK))
 665                 int_ctl_vmcb12_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 666         else
 667                 int_ctl_vmcb01_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 668
 669         if (vnmi) {
 670                 if (vmcb01->control.int_ctl & V_NMI_PENDING_MASK) {
 671                         svm->vcpu.arch.nmi_pending++;
 672                         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
 673                 }
 674                 if (nested_vnmi_enabled(svm))
 675                         int_ctl_vmcb12_bits |= (V_NMI_PENDING_MASK |
 676                                                 V_NMI_ENABLE_MASK |
 677                                                 V_NMI_BLOCKING_MASK);
 678         }
 679
 680         /* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 681         vmcb02->control.nested_ctl = vmcb01->control.nested_ctl;
 682         vmcb02->control.iopm_base_pa = vmcb01->control.iopm_base_pa;
 683         vmcb02->control.msrpm_base_pa = vmcb01->control.msrpm_base_pa;
 684
 685         /* Done at vmrun: asid.  */
 686
 687         /* Also overwritten later if necessary.  */
 688         vmcb02->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 689
 690         /* nested_cr3.  */
 691         if (nested_npt_enabled(svm))
 692                 nested_svm_init_mmu_context(vcpu);
 693
 694         vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
 695                         vcpu->arch.l1_tsc_offset,
 696                         svm->nested.ctl.tsc_offset,
 697                         svm->tsc_ratio_msr);
 698
 699         vmcb02->control.tsc_offset = vcpu->arch.tsc_offset;
 700
 701         if (guest_can_use(vcpu, X86_FEATURE_TSCRATEMSR) &&
 702             svm->tsc_ratio_msr != kvm_caps.default_tsc_scaling_ratio)
 703                 nested_svm_update_tsc_ratio_msr(vcpu);
 704
 705         vmcb02->control.int_ctl             =
 706                 (svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
 707                 (vmcb01->control.int_ctl & int_ctl_vmcb01_bits);
 708
 709         vmcb02->control.int_vector          = svm->nested.ctl.int_vector;
 710         vmcb02->control.int_state           = svm->nested.ctl.int_state;
 711         vmcb02->control.event_inj           = svm->nested.ctl.event_inj;
 712         vmcb02->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 713
 714         /*
 715          * next_rip is consumed on VMRUN as the return address pushed on the
 716          * stack for injected soft exceptions/interrupts.  If nrips is exposed
 717          * to L1, take it verbatim from vmcb12.  If nrips is supported in
 718          * hardware but not exposed to L1, stuff the actual L2 RIP to emulate
 719          * what a nrips=0 CPU would do (L1 is responsible for advancing RIP
 720          * prior to injecting the event).
 721          */
 722         if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
 723                 vmcb02->control.next_rip    = svm->nested.ctl.next_rip;
 724         else if (boot_cpu_has(X86_FEATURE_NRIPS))
 725                 vmcb02->control.next_rip    = vmcb12_rip;
 726
 727         svm->nmi_l1_to_l2 = is_evtinj_nmi(vmcb02->control.event_inj);
 728         if (is_evtinj_soft(vmcb02->control.event_inj)) {
 729                 svm->soft_int_injected = true;
 730                 svm->soft_int_csbase = vmcb12_csbase;
 731                 svm->soft_int_old_rip = vmcb12_rip;
 732                 if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
 733                         svm->soft_int_next_rip = svm->nested.ctl.next_rip;
 734                 else
 735                         svm->soft_int_next_rip = vmcb12_rip;
 736         }
 737
 738         vmcb02->control.virt_ext            = vmcb01->control.virt_ext &
 739                                               LBR_CTL_ENABLE_MASK;
 740         if (guest_can_use(vcpu, X86_FEATURE_LBRV))
 741                 vmcb02->control.virt_ext  |=
 742                         (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK);
 743
 744         if (!nested_vmcb_needs_vls_intercept(svm))
 745                 vmcb02->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
 746
 747         if (guest_can_use(vcpu, X86_FEATURE_PAUSEFILTER))
 748                 pause_count12 = svm->nested.ctl.pause_filter_count;
 749         else
 750                 pause_count12 = 0;
 751         if (guest_can_use(vcpu, X86_FEATURE_PFTHRESHOLD))
 752                 pause_thresh12 = svm->nested.ctl.pause_filter_thresh;
 753         else
 754                 pause_thresh12 = 0;
 755         if (kvm_pause_in_guest(svm->vcpu.kvm)) {
 756                 /* use guest values since host doesn't intercept PAUSE */
 757                 vmcb02->control.pause_filter_count = pause_count12;
 758                 vmcb02->control.pause_filter_thresh = pause_thresh12;
 759
 760         } else {
 761                 /* start from host values otherwise */
 762                 vmcb02->control.pause_filter_count = vmcb01->control.pause_filter_count;
 763                 vmcb02->control.pause_filter_thresh = vmcb01->control.pause_filter_thresh;
 764
 765                 /* ... but ensure filtering is disabled if so requested.  */
 766                 if (vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_PAUSE)) {
 767                         if (!pause_count12)
 768                                 vmcb02->control.pause_filter_count = 0;
 769                         if (!pause_thresh12)
 770                                 vmcb02->control.pause_filter_thresh = 0;
 771                 }
 772         }
 773
 774         nested_svm_transition_tlb_flush(vcpu);
 775
 776         /* Enter Guest-Mode */
 777         enter_guest_mode(vcpu);
 778
 779         /*
 780          * Merge guest and host intercepts - must be called with vcpu in
 781          * guest-mode to take effect.
 782          */
 783         recalc_intercepts(svm);
 784 }
 785
 786 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 787 {
 788         /*
 789          * Some VMCB state is shared between L1 and L2 and thus has to be
 790          * moved at the time of nested vmrun and vmexit.
 791          *
 792          * VMLOAD/VMSAVE state would also belong in this category, but KVM
 793          * always performs VMLOAD and VMSAVE from the VMCB01.
 794          */
 795         to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 796 }
 797
 798 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 799                          struct vmcb *vmcb12, bool from_vmrun)
 800 {
 801         struct vcpu_svm *svm = to_svm(vcpu);
 802         int ret;
 803
 804         trace_kvm_nested_vmenter(svm->vmcb->save.rip,
 805                                  vmcb12_gpa,
 806                                  vmcb12->save.rip,
 807                                  vmcb12->control.int_ctl,
 808                                  vmcb12->control.event_inj,
 809                                  vmcb12->control.nested_ctl,
 810                                  vmcb12->control.nested_cr3,
 811                                  vmcb12->save.cr3,
 812                                  KVM_ISA_SVM);
 813
 814         trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 815                                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 816                                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 817                                     vmcb12->control.intercepts[INTERCEPT_WORD3],
 818                                     vmcb12->control.intercepts[INTERCEPT_WORD4],
 819                                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
 820
 821
 822         svm->nested.vmcb12_gpa = vmcb12_gpa;
 823
 824         WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 825
 826         nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 827
 828         svm_switch_vmcb(svm, &svm->nested.vmcb02);
 829         nested_vmcb02_prepare_control(svm, vmcb12->save.rip, vmcb12->save.cs.base);
 830         nested_vmcb02_prepare_save(svm, vmcb12);
 831
 832         ret = nested_svm_load_cr3(&svm->vcpu, svm->nested.save.cr3,
 833                                   nested_npt_enabled(svm), from_vmrun);
 834         if (ret)
 835                 return ret;
 836
 837         if (!from_vmrun)
 838                 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 839
 840         svm_set_gif(svm, true);
 841
 842         if (kvm_vcpu_apicv_active(vcpu))
 843                 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
 844
 845         nested_svm_hv_update_vm_vp_ids(vcpu);
 846
 847         return 0;
 848 }
 849
 850 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 851 {
 852         struct vcpu_svm *svm = to_svm(vcpu);
 853         int ret;
 854         struct vmcb *vmcb12;
 855         struct kvm_host_map map;
 856         u64 vmcb12_gpa;
 857         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 858
 859         if (!svm->nested.hsave_msr) {
 860                 kvm_inject_gp(vcpu, 0);
 861                 return 1;
 862         }
 863
 864         if (is_smm(vcpu)) {
 865                 kvm_queue_exception(vcpu, UD_VECTOR);
 866                 return 1;
 867         }
 868
 869         /* This fails when VP assist page is enabled but the supplied GPA is bogus */
 870         ret = kvm_hv_verify_vp_assist(vcpu);
 871         if (ret) {
 872                 kvm_inject_gp(vcpu, 0);
 873                 return ret;
 874         }
 875
 876         vmcb12_gpa = svm->vmcb->save.rax;
 877         ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 878         if (ret == -EINVAL) {
 879                 kvm_inject_gp(vcpu, 0);
 880                 return 1;
 881         } else if (ret) {
 882                 return kvm_skip_emulated_instruction(vcpu);
 883         }
 884
 885         ret = kvm_skip_emulated_instruction(vcpu);
 886
 887         vmcb12 = map.hva;
 888
 889         if (WARN_ON_ONCE(!svm->nested.initialized))
 890                 return -EINVAL;
 891
 892         nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
 893         nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
 894
 895         if (!nested_vmcb_check_save(vcpu) ||
 896             !nested_vmcb_check_controls(vcpu)) {
 897                 vmcb12->control.exit_code    = SVM_EXIT_ERR;
 898                 vmcb12->control.exit_code_hi = 0;
 899                 vmcb12->control.exit_info_1  = 0;
 900                 vmcb12->control.exit_info_2  = 0;
 901                 goto out;
 902         }
 903
 904         /*
 905          * Since vmcb01 is not in use, we can use it to store some of the L1
 906          * state.
 907          */
 908         vmcb01->save.efer   = vcpu->arch.efer;
 909         vmcb01->save.cr0    = kvm_read_cr0(vcpu);
 910         vmcb01->save.cr4    = vcpu->arch.cr4;
 911         vmcb01->save.rflags = kvm_get_rflags(vcpu);
 912         vmcb01->save.rip    = kvm_rip_read(vcpu);
 913
 914         if (!npt_enabled)
 915                 vmcb01->save.cr3 = kvm_read_cr3(vcpu);
 916
 917         svm->nested.nested_run_pending = 1;
 918
 919         if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
 920                 goto out_exit_err;
 921
 922         if (nested_svm_vmrun_msrpm(svm))
 923                 goto out;
 924
 925 out_exit_err:
 926         svm->nested.nested_run_pending = 0;
 927         svm->nmi_l1_to_l2 = false;
 928         svm->soft_int_injected = false;
 929
 930         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 931         svm->vmcb->control.exit_code_hi = 0;
 932         svm->vmcb->control.exit_info_1  = 0;
 933         svm->vmcb->control.exit_info_2  = 0;
 934
 935         nested_svm_vmexit(svm);
 936
 937 out:
 938         kvm_vcpu_unmap(vcpu, &map, true);
 939
 940         return ret;
 941 }
 942
 943 /* Copy state save area fields which are handled by VMRUN */
 944 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
 945                           struct vmcb_save_area *from_save)
 946 {
 947         to_save->es = from_save->es;
 948         to_save->cs = from_save->cs;
 949         to_save->ss = from_save->ss;
 950         to_save->ds = from_save->ds;
 951         to_save->gdtr = from_save->gdtr;
 952         to_save->idtr = from_save->idtr;
 953         to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
 954         to_save->efer = from_save->efer;
 955         to_save->cr0 = from_save->cr0;
 956         to_save->cr3 = from_save->cr3;
 957         to_save->cr4 = from_save->cr4;
 958         to_save->rax = from_save->rax;
 959         to_save->rsp = from_save->rsp;
 960         to_save->rip = from_save->rip;
 961         to_save->cpl = 0;
 962 }
 963
 964 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
 965 {
 966         to_vmcb->save.fs = from_vmcb->save.fs;
 967         to_vmcb->save.gs = from_vmcb->save.gs;
 968         to_vmcb->save.tr = from_vmcb->save.tr;
 969         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 970         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 971         to_vmcb->save.star = from_vmcb->save.star;
 972         to_vmcb->save.lstar = from_vmcb->save.lstar;
 973         to_vmcb->save.cstar = from_vmcb->save.cstar;
 974         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 975         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 976         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 977         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 978 }
 979
 980 int nested_svm_vmexit(struct vcpu_svm *svm)
 981 {
 982         struct kvm_vcpu *vcpu = &svm->vcpu;
 983         struct vmcb *vmcb01 = svm->vmcb01.ptr;
 984         struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 985         struct vmcb *vmcb12;
 986         struct kvm_host_map map;
 987         int rc;
 988
 989         rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 990         if (rc) {
 991                 if (rc == -EINVAL)
 992                         kvm_inject_gp(vcpu, 0);
 993                 return 1;
 994         }
 995
 996         vmcb12 = map.hva;
 997
 998         /* Exit Guest-Mode */
 999         leave_guest_mode(vcpu);
1000         svm->nested.vmcb12_gpa = 0;
1001         WARN_ON_ONCE(svm->nested.nested_run_pending);
1002
1003         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1004
1005         /* in case we halted in L2 */
1006         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
1007
1008         /* Give the current vmcb to the guest */
1009
1010         vmcb12->save.es     = vmcb02->save.es;
1011         vmcb12->save.cs     = vmcb02->save.cs;
1012         vmcb12->save.ss     = vmcb02->save.ss;
1013         vmcb12->save.ds     = vmcb02->save.ds;
1014         vmcb12->save.gdtr   = vmcb02->save.gdtr;
1015         vmcb12->save.idtr   = vmcb02->save.idtr;
1016         vmcb12->save.efer   = svm->vcpu.arch.efer;
1017         vmcb12->save.cr0    = kvm_read_cr0(vcpu);
1018         vmcb12->save.cr3    = kvm_read_cr3(vcpu);
1019         vmcb12->save.cr2    = vmcb02->save.cr2;
1020         vmcb12->save.cr4    = svm->vcpu.arch.cr4;
1021         vmcb12->save.rflags = kvm_get_rflags(vcpu);
1022         vmcb12->save.rip    = kvm_rip_read(vcpu);
1023         vmcb12->save.rsp    = kvm_rsp_read(vcpu);
1024         vmcb12->save.rax    = kvm_rax_read(vcpu);
1025         vmcb12->save.dr7    = vmcb02->save.dr7;
1026         vmcb12->save.dr6    = svm->vcpu.arch.dr6;
1027         vmcb12->save.cpl    = vmcb02->save.cpl;
1028
1029         vmcb12->control.int_state         = vmcb02->control.int_state;
1030         vmcb12->control.exit_code         = vmcb02->control.exit_code;
1031         vmcb12->control.exit_code_hi      = vmcb02->control.exit_code_hi;
1032         vmcb12->control.exit_info_1       = vmcb02->control.exit_info_1;
1033         vmcb12->control.exit_info_2       = vmcb02->control.exit_info_2;
1034
1035         if (vmcb12->control.exit_code != SVM_EXIT_ERR)
1036                 nested_save_pending_event_to_vmcb12(svm, vmcb12);
1037
1038         if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
1039                 vmcb12->control.next_rip  = vmcb02->control.next_rip;
1040
1041         vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
1042         vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
1043         vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
1044
1045         if (!kvm_pause_in_guest(vcpu->kvm)) {
1046                 vmcb01->control.pause_filter_count = vmcb02->control.pause_filter_count;
1047                 vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1048
1049         }
1050
1051         nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
1052
1053         svm_switch_vmcb(svm, &svm->vmcb01);
1054
1055         /*
1056          * Rules for synchronizing int_ctl bits from vmcb02 to vmcb01:
1057          *
1058          * V_IRQ, V_IRQ_VECTOR, V_INTR_PRIO_MASK, V_IGN_TPR:  If L1 doesn't
1059          * intercept interrupts, then KVM will use vmcb02's V_IRQ (and related
1060          * flags) to detect interrupt windows for L1 IRQs (even if L1 uses
1061          * virtual interrupt masking).  Raise KVM_REQ_EVENT to ensure that
1062          * KVM re-requests an interrupt window if necessary, which implicitly
1063          * copies this bits from vmcb02 to vmcb01.
1064          *
1065          * V_TPR: If L1 doesn't use virtual interrupt masking, then L1's vTPR
1066          * is stored in vmcb02, but its value doesn't need to be copied from/to
1067          * vmcb01 because it is copied from/to the virtual APIC's TPR register
1068          * on each VM entry/exit.
1069          *
1070          * V_GIF: If nested vGIF is not used, KVM uses vmcb02's V_GIF for L1's
1071          * V_GIF.  However, GIF is architecturally clear on each VM exit, thus
1072          * there is no need to copy V_GIF from vmcb02 to vmcb01.
1073          */
1074         if (!nested_exit_on_intr(svm))
1075                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
1076
1077         if (unlikely(guest_can_use(vcpu, X86_FEATURE_LBRV) &&
1078                      (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
1079                 svm_copy_lbrs(vmcb12, vmcb02);
1080                 svm_update_lbrv(vcpu);
1081         } else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
1082                 svm_copy_lbrs(vmcb01, vmcb02);
1083                 svm_update_lbrv(vcpu);
1084         }
1085
1086         if (vnmi) {
1087                 if (vmcb02->control.int_ctl & V_NMI_BLOCKING_MASK)
1088                         vmcb01->control.int_ctl |= V_NMI_BLOCKING_MASK;
1089                 else
1090                         vmcb01->control.int_ctl &= ~V_NMI_BLOCKING_MASK;
1091
1092                 if (vcpu->arch.nmi_pending) {
1093                         vcpu->arch.nmi_pending--;
1094                         vmcb01->control.int_ctl |= V_NMI_PENDING_MASK;
1095                 } else {
1096                         vmcb01->control.int_ctl &= ~V_NMI_PENDING_MASK;
1097                 }
1098         }
1099
1100         /*
1101          * On vmexit the  GIF is set to false and
1102          * no event can be injected in L1.
1103          */
1104         svm_set_gif(svm, false);
1105         vmcb01->control.exit_int_info = 0;
1106
1107         svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
1108         if (vmcb01->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
1109                 vmcb01->control.tsc_offset = svm->vcpu.arch.tsc_offset;
1110                 vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1111         }
1112
1113         if (kvm_caps.has_tsc_control &&
1114             vcpu->arch.tsc_scaling_ratio != vcpu->arch.l1_tsc_scaling_ratio) {
1115                 vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
1116                 svm_write_tsc_multiplier(vcpu);
1117         }
1118
1119         svm->nested.ctl.nested_cr3 = 0;
1120
1121         /*
1122          * Restore processor state that had been saved in vmcb01
1123          */
1124         kvm_set_rflags(vcpu, vmcb01->save.rflags);
1125         svm_set_efer(vcpu, vmcb01->save.efer);
1126         svm_set_cr0(vcpu, vmcb01->save.cr0 | X86_CR0_PE);
1127         svm_set_cr4(vcpu, vmcb01->save.cr4);
1128         kvm_rax_write(vcpu, vmcb01->save.rax);
1129         kvm_rsp_write(vcpu, vmcb01->save.rsp);
1130         kvm_rip_write(vcpu, vmcb01->save.rip);
1131
1132         svm->vcpu.arch.dr7 = DR7_FIXED_1;
1133         kvm_update_dr7(&svm->vcpu);
1134
1135         trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
1136                                        vmcb12->control.exit_info_1,
1137                                        vmcb12->control.exit_info_2,
1138                                        vmcb12->control.exit_int_info,
1139                                        vmcb12->control.exit_int_info_err,
1140                                        KVM_ISA_SVM);
1141
1142         kvm_vcpu_unmap(vcpu, &map, true);
1143
1144         nested_svm_transition_tlb_flush(vcpu);
1145
1146         nested_svm_uninit_mmu_context(vcpu);
1147
1148         rc = nested_svm_load_cr3(vcpu, vmcb01->save.cr3, false, true);
1149         if (rc)
1150                 return 1;
1151
1152         /*
1153          * Drop what we picked up for L2 via svm_complete_interrupts() so it
1154          * doesn't end up in L1.
1155          */
1156         svm->vcpu.arch.nmi_injected = false;
1157         kvm_clear_exception_queue(vcpu);
1158         kvm_clear_interrupt_queue(vcpu);
1159
1160         /*
1161          * If we are here following the completion of a VMRUN that
1162          * is being single-stepped, queue the pending #DB intercept
1163          * right now so that it an be accounted for before we execute
1164          * L1's next instruction.
1165          */
1166         if (unlikely(vmcb01->save.rflags & X86_EFLAGS_TF))
1167                 kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
1168
1169         /*
1170          * Un-inhibit the AVIC right away, so that other vCPUs can start
1171          * to benefit from it right away.
1172          */
1173         if (kvm_apicv_activated(vcpu->kvm))
1174                 __kvm_vcpu_update_apicv(vcpu);
1175
1176         return 0;
1177 }
1178
1179 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
1180 {
1181         struct vcpu_svm *svm = to_svm(vcpu);
1182
1183         if (!vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SHUTDOWN))
1184                 return;
1185
1186         kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1187         nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
1188 }
1189
1190 int svm_allocate_nested(struct vcpu_svm *svm)
1191 {
1192         struct page *vmcb02_page;
1193
1194         if (svm->nested.initialized)
1195                 return 0;
1196
1197         vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1198         if (!vmcb02_page)
1199                 return -ENOMEM;
1200         svm->nested.vmcb02.ptr = page_address(vmcb02_page);
1201         svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
1202
1203         svm->nested.msrpm = svm_vcpu_alloc_msrpm();
1204         if (!svm->nested.msrpm)
1205                 goto err_free_vmcb02;
1206         svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
1207
1208         svm->nested.initialized = true;
1209         return 0;
1210
1211 err_free_vmcb02:
1212         __free_page(vmcb02_page);
1213         return -ENOMEM;
1214 }
1215
1216 void svm_free_nested(struct vcpu_svm *svm)
1217 {
1218         if (!svm->nested.initialized)
1219                 return;
1220
1221         if (WARN_ON_ONCE(svm->vmcb != svm->vmcb01.ptr))
1222                 svm_switch_vmcb(svm, &svm->vmcb01);
1223
1224         svm_vcpu_free_msrpm(svm->nested.msrpm);
1225         svm->nested.msrpm = NULL;
1226
1227         __free_page(virt_to_page(svm->nested.vmcb02.ptr));
1228         svm->nested.vmcb02.ptr = NULL;
1229
1230         /*
1231          * When last_vmcb12_gpa matches the current vmcb12 gpa,
1232          * some vmcb12 fields are not loaded if they are marked clean
1233          * in the vmcb12, since in this case they are up to date already.
1234          *
1235          * When the vmcb02 is freed, this optimization becomes invalid.
1236          */
1237         svm->nested.last_vmcb12_gpa = INVALID_GPA;
1238
1239         svm->nested.initialized = false;
1240 }
1241
1242 void svm_leave_nested(struct kvm_vcpu *vcpu)
1243 {
1244         struct vcpu_svm *svm = to_svm(vcpu);
1245
1246         if (is_guest_mode(vcpu)) {
1247                 svm->nested.nested_run_pending = 0;
1248                 svm->nested.vmcb12_gpa = INVALID_GPA;
1249
1250                 leave_guest_mode(vcpu);
1251
1252                 svm_switch_vmcb(svm, &svm->vmcb01);
1253
1254                 nested_svm_uninit_mmu_context(vcpu);
1255                 vmcb_mark_all_dirty(svm->vmcb);
1256         }
1257
1258         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1259 }
1260
1261 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1262 {
1263         u32 offset, msr, value;
1264         int write, mask;
1265
1266         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
1267                 return NESTED_EXIT_HOST;
1268
1269         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1270         offset = svm_msrpm_offset(msr);
1271         write  = svm->vmcb->control.exit_info_1 & 1;
1272         mask   = 1 << ((2 * (msr & 0xf)) + write);
1273
1274         if (offset == MSR_INVALID)
1275                 return NESTED_EXIT_DONE;
1276
1277         /* Offset is in 32 bit units but need in 8 bit units */
1278         offset *= 4;
1279
1280         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
1281                 return NESTED_EXIT_DONE;
1282
1283         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1284 }
1285
1286 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1287 {
1288         unsigned port, size, iopm_len;
1289         u16 val, mask;
1290         u8 start_bit;
1291         u64 gpa;
1292
1293         if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
1294                 return NESTED_EXIT_HOST;
1295
1296         port = svm->vmcb->control.exit_info_1 >> 16;
1297         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1298                 SVM_IOIO_SIZE_SHIFT;
1299         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1300         start_bit = port % 8;
1301         iopm_len = (start_bit + size > 8) ? 2 : 1;
1302         mask = (0xf >> (4 - size)) << start_bit;
1303         val = 0;
1304
1305         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1306                 return NESTED_EXIT_DONE;
1307
1308         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1309 }
1310
1311 static int nested_svm_intercept(struct vcpu_svm *svm)
1312 {
1313         u32 exit_code = svm->vmcb->control.exit_code;
1314         int vmexit = NESTED_EXIT_HOST;
1315
1316         switch (exit_code) {
1317         case SVM_EXIT_MSR:
1318                 vmexit = nested_svm_exit_handled_msr(svm);
1319                 break;
1320         case SVM_EXIT_IOIO:
1321                 vmexit = nested_svm_intercept_ioio(svm);
1322                 break;
1323         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1324                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1325                         vmexit = NESTED_EXIT_DONE;
1326                 break;
1327         }
1328         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1329                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1330                         vmexit = NESTED_EXIT_DONE;
1331                 break;
1332         }
1333         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1334                 /*
1335                  * Host-intercepted exceptions have been checked already in
1336                  * nested_svm_exit_special.  There is nothing to do here,
1337                  * the vmexit is injected by svm_check_nested_events.
1338                  */
1339                 vmexit = NESTED_EXIT_DONE;
1340                 break;
1341         }
1342         case SVM_EXIT_ERR: {
1343                 vmexit = NESTED_EXIT_DONE;
1344                 break;
1345         }
1346         default: {
1347                 if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1348                         vmexit = NESTED_EXIT_DONE;
1349         }
1350         }
1351
1352         return vmexit;
1353 }
1354
1355 int nested_svm_exit_handled(struct vcpu_svm *svm)
1356 {
1357         int vmexit;
1358
1359         vmexit = nested_svm_intercept(svm);
1360
1361         if (vmexit == NESTED_EXIT_DONE)
1362                 nested_svm_vmexit(svm);
1363
1364         return vmexit;
1365 }
1366
1367 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1368 {
1369         if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1370                 kvm_queue_exception(vcpu, UD_VECTOR);
1371                 return 1;
1372         }
1373
1374         if (to_svm(vcpu)->vmcb->save.cpl) {
1375                 kvm_inject_gp(vcpu, 0);
1376                 return 1;
1377         }
1378
1379         return 0;
1380 }
1381
1382 static bool nested_svm_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
1383                                            u32 error_code)
1384 {
1385         struct vcpu_svm *svm = to_svm(vcpu);
1386
1387         return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(vector));
1388 }
1389
1390 static void nested_svm_inject_exception_vmexit(struct kvm_vcpu *vcpu)
1391 {
1392         struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
1393         struct vcpu_svm *svm = to_svm(vcpu);
1394         struct vmcb *vmcb = svm->vmcb;
1395
1396         vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + ex->vector;
1397         vmcb->control.exit_code_hi = 0;
1398
1399         if (ex->has_error_code)
1400                 vmcb->control.exit_info_1 = ex->error_code;
1401
1402         /*
1403          * EXITINFO2 is undefined for all exception intercepts other
1404          * than #PF.
1405          */
1406         if (ex->vector == PF_VECTOR) {
1407                 if (ex->has_payload)
1408                         vmcb->control.exit_info_2 = ex->payload;
1409                 else
1410                         vmcb->control.exit_info_2 = vcpu->arch.cr2;
1411         } else if (ex->vector == DB_VECTOR) {
1412                 /* See kvm_check_and_inject_events().  */
1413                 kvm_deliver_exception_payload(vcpu, ex);
1414
1415                 if (vcpu->arch.dr7 & DR7_GD) {
1416                         vcpu->arch.dr7 &= ~DR7_GD;
1417                         kvm_update_dr7(vcpu);
1418                 }
1419         } else {
1420                 WARN_ON(ex->has_payload);
1421         }
1422
1423         nested_svm_vmexit(svm);
1424 }
1425
1426 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1427 {
1428         return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1429 }
1430
1431 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1432 {
1433         struct kvm_lapic *apic = vcpu->arch.apic;
1434         struct vcpu_svm *svm = to_svm(vcpu);
1435         /*
1436          * Only a pending nested run blocks a pending exception.  If there is a
1437          * previously injected event, the pending exception occurred while said
1438          * event was being delivered and thus needs to be handled.
1439          */
1440         bool block_nested_exceptions = svm->nested.nested_run_pending;
1441         /*
1442          * New events (not exceptions) are only recognized at instruction
1443          * boundaries.  If an event needs reinjection, then KVM is handling a
1444          * VM-Exit that occurred _during_ instruction execution; new events are
1445          * blocked until the instruction completes.
1446          */
1447         bool block_nested_events = block_nested_exceptions ||
1448                                    kvm_event_needs_reinjection(vcpu);
1449
1450         if (lapic_in_kernel(vcpu) &&
1451             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1452                 if (block_nested_events)
1453                         return -EBUSY;
1454                 if (!nested_exit_on_init(svm))
1455                         return 0;
1456                 nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1457                 return 0;
1458         }
1459
1460         if (vcpu->arch.exception_vmexit.pending) {
1461                 if (block_nested_exceptions)
1462                         return -EBUSY;
1463                 nested_svm_inject_exception_vmexit(vcpu);
1464                 return 0;
1465         }
1466
1467         if (vcpu->arch.exception.pending) {
1468                 if (block_nested_exceptions)
1469                         return -EBUSY;
1470                 return 0;
1471         }
1472
1473 #ifdef CONFIG_KVM_SMM
1474         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1475                 if (block_nested_events)
1476                         return -EBUSY;
1477                 if (!nested_exit_on_smi(svm))
1478                         return 0;
1479                 nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1480                 return 0;
1481         }
1482 #endif
1483
1484         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1485                 if (block_nested_events)
1486                         return -EBUSY;
1487                 if (!nested_exit_on_nmi(svm))
1488                         return 0;
1489                 nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1490                 return 0;
1491         }
1492
1493         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1494                 if (block_nested_events)
1495                         return -EBUSY;
1496                 if (!nested_exit_on_intr(svm))
1497                         return 0;
1498                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1499                 nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1500                 return 0;
1501         }
1502
1503         return 0;
1504 }
1505
1506 int nested_svm_exit_special(struct vcpu_svm *svm)
1507 {
1508         u32 exit_code = svm->vmcb->control.exit_code;
1509         struct kvm_vcpu *vcpu = &svm->vcpu;
1510
1511         switch (exit_code) {
1512         case SVM_EXIT_INTR:
1513         case SVM_EXIT_NMI:
1514         case SVM_EXIT_NPF:
1515                 return NESTED_EXIT_HOST;
1516         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1517                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1518
1519                 if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1520                     excp_bits)
1521                         return NESTED_EXIT_HOST;
1522                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1523                          svm->vcpu.arch.apf.host_apf_flags)
1524                         /* Trap async PF even if not shadowing */
1525                         return NESTED_EXIT_HOST;
1526                 break;
1527         }
1528         case SVM_EXIT_VMMCALL:
1529                 /* Hyper-V L2 TLB flush hypercall is handled by L0 */
1530                 if (guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
1531                     nested_svm_l2_tlb_flush_enabled(vcpu) &&
1532                     kvm_hv_is_tlb_flush_hcall(vcpu))
1533                         return NESTED_EXIT_HOST;
1534                 break;
1535         default:
1536                 break;
1537         }
1538
1539         return NESTED_EXIT_CONTINUE;
1540 }
1541
1542 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
1543 {
1544         struct vcpu_svm *svm = to_svm(vcpu);
1545
1546         vcpu->arch.tsc_scaling_ratio =
1547                 kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
1548                                                svm->tsc_ratio_msr);
1549         svm_write_tsc_multiplier(vcpu);
1550 }
1551
1552 /* Inverse operation of nested_copy_vmcb_control_to_cache(). asid is copied too. */
1553 static void nested_copy_vmcb_cache_to_control(struct vmcb_control_area *dst,
1554                                               struct vmcb_ctrl_area_cached *from)
1555 {
1556         unsigned int i;
1557
1558         memset(dst, 0, sizeof(struct vmcb_control_area));
1559
1560         for (i = 0; i < MAX_INTERCEPT; i++)
1561                 dst->intercepts[i] = from->intercepts[i];
1562
1563         dst->iopm_base_pa         = from->iopm_base_pa;
1564         dst->msrpm_base_pa        = from->msrpm_base_pa;
1565         dst->tsc_offset           = from->tsc_offset;
1566         dst->asid                 = from->asid;
1567         dst->tlb_ctl              = from->tlb_ctl;
1568         dst->int_ctl              = from->int_ctl;
1569         dst->int_vector           = from->int_vector;
1570         dst->int_state            = from->int_state;
1571         dst->exit_code            = from->exit_code;
1572         dst->exit_code_hi         = from->exit_code_hi;
1573         dst->exit_info_1          = from->exit_info_1;
1574         dst->exit_info_2          = from->exit_info_2;
1575         dst->exit_int_info        = from->exit_int_info;
1576         dst->exit_int_info_err    = from->exit_int_info_err;
1577         dst->nested_ctl           = from->nested_ctl;
1578         dst->event_inj            = from->event_inj;
1579         dst->event_inj_err        = from->event_inj_err;
1580         dst->next_rip             = from->next_rip;
1581         dst->nested_cr3           = from->nested_cr3;
1582         dst->virt_ext              = from->virt_ext;
1583         dst->pause_filter_count   = from->pause_filter_count;
1584         dst->pause_filter_thresh  = from->pause_filter_thresh;
1585         /* 'clean' and 'hv_enlightenments' are not changed by KVM */
1586 }
1587
1588 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1589                                 struct kvm_nested_state __user *user_kvm_nested_state,
1590                                 u32 user_data_size)
1591 {
1592         struct vcpu_svm *svm;
1593         struct vmcb_control_area *ctl;
1594         unsigned long r;
1595         struct kvm_nested_state kvm_state = {
1596                 .flags = 0,
1597                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1598                 .size = sizeof(kvm_state),
1599         };
1600         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1601                 &user_kvm_nested_state->data.svm[0];
1602
1603         if (!vcpu)
1604                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1605
1606         svm = to_svm(vcpu);
1607
1608         if (user_data_size < kvm_state.size)
1609                 goto out;
1610
1611         /* First fill in the header and copy it out.  */
1612         if (is_guest_mode(vcpu)) {
1613                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1614                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1615                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1616
1617                 if (svm->nested.nested_run_pending)
1618                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1619         }
1620
1621         if (gif_set(svm))
1622                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1623
1624         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1625                 return -EFAULT;
1626
1627         if (!is_guest_mode(vcpu))
1628                 goto out;
1629
1630         /*
1631          * Copy over the full size of the VMCB rather than just the size
1632          * of the structs.
1633          */
1634         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1635                 return -EFAULT;
1636
1637         ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
1638         if (!ctl)
1639                 return -ENOMEM;
1640
1641         nested_copy_vmcb_cache_to_control(ctl, &svm->nested.ctl);
1642         r = copy_to_user(&user_vmcb->control, ctl,
1643                          sizeof(user_vmcb->control));
1644         kfree(ctl);
1645         if (r)
1646                 return -EFAULT;
1647
1648         if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1649                          sizeof(user_vmcb->save)))
1650                 return -EFAULT;
1651 out:
1652         return kvm_state.size;
1653 }
1654
1655 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1656                                 struct kvm_nested_state __user *user_kvm_nested_state,
1657                                 struct kvm_nested_state *kvm_state)
1658 {
1659         struct vcpu_svm *svm = to_svm(vcpu);
1660         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1661                 &user_kvm_nested_state->data.svm[0];
1662         struct vmcb_control_area *ctl;
1663         struct vmcb_save_area *save;
1664         struct vmcb_save_area_cached save_cached;
1665         struct vmcb_ctrl_area_cached ctl_cached;
1666         unsigned long cr0;
1667         int ret;
1668
1669         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1670                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1671
1672         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1673                 return -EINVAL;
1674
1675         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1676                                  KVM_STATE_NESTED_RUN_PENDING |
1677                                  KVM_STATE_NESTED_GIF_SET))
1678                 return -EINVAL;
1679
1680         /*
1681          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1682          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1683          */
1684         if (!(vcpu->arch.efer & EFER_SVME)) {
1685                 /* GIF=1 and no guest mode are required if SVME=0.  */
1686                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1687                         return -EINVAL;
1688         }
1689
1690         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1691         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1692                 return -EINVAL;
1693
1694         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1695                 svm_leave_nested(vcpu);
1696                 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1697                 return 0;
1698         }
1699
1700         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1701                 return -EINVAL;
1702         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1703                 return -EINVAL;
1704
1705         ret  = -ENOMEM;
1706         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1707         save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1708         if (!ctl || !save)
1709                 goto out_free;
1710
1711         ret = -EFAULT;
1712         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1713                 goto out_free;
1714         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1715                 goto out_free;
1716
1717         ret = -EINVAL;
1718         __nested_copy_vmcb_control_to_cache(vcpu, &ctl_cached, ctl);
1719         if (!__nested_vmcb_check_controls(vcpu, &ctl_cached))
1720                 goto out_free;
1721
1722         /*
1723          * Processor state contains L2 state.  Check that it is
1724          * valid for guest mode (see nested_vmcb_check_save).
1725          */
1726         cr0 = kvm_read_cr0(vcpu);
1727         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1728                 goto out_free;
1729
1730         /*
1731          * Validate host state saved from before VMRUN (see
1732          * nested_svm_check_permissions).
1733          */
1734         __nested_copy_vmcb_save_to_cache(&save_cached, save);
1735         if (!(save->cr0 & X86_CR0_PG) ||
1736             !(save->cr0 & X86_CR0_PE) ||
1737             (save->rflags & X86_EFLAGS_VM) ||
1738             !__nested_vmcb_check_save(vcpu, &save_cached))
1739                 goto out_free;
1740
1741
1742         /*
1743          * All checks done, we can enter guest mode. Userspace provides
1744          * vmcb12.control, which will be combined with L1 and stored into
1745          * vmcb02, and the L1 save state which we store in vmcb01.
1746          * L2 registers if needed are moved from the current VMCB to VMCB02.
1747          */
1748
1749         if (is_guest_mode(vcpu))
1750                 svm_leave_nested(vcpu);
1751         else
1752                 svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1753
1754         svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1755
1756         svm->nested.nested_run_pending =
1757                 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1758
1759         svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1760
1761         svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1762         nested_copy_vmcb_control_to_cache(svm, ctl);
1763
1764         svm_switch_vmcb(svm, &svm->nested.vmcb02);
1765         nested_vmcb02_prepare_control(svm, svm->vmcb->save.rip, svm->vmcb->save.cs.base);
1766
1767         /*
1768          * While the nested guest CR3 is already checked and set by
1769          * KVM_SET_SREGS, it was set when nested state was yet loaded,
1770          * thus MMU might not be initialized correctly.
1771          * Set it again to fix this.
1772          */
1773
1774         ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1775                                   nested_npt_enabled(svm), false);
1776         if (WARN_ON_ONCE(ret))
1777                 goto out_free;
1778
1779         svm->nested.force_msr_bitmap_recalc = true;
1780
1781         kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1782         ret = 0;
1783 out_free:
1784         kfree(save);
1785         kfree(ctl);
1786
1787         return ret;
1788 }
1789
1790 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1791 {
1792         struct vcpu_svm *svm = to_svm(vcpu);
1793
1794         if (WARN_ON(!is_guest_mode(vcpu)))
1795                 return true;
1796
1797         if (!vcpu->arch.pdptrs_from_userspace &&
1798             !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1799                 /*
1800                  * Reload the guest's PDPTRs since after a migration
1801                  * the guest CR3 might be restored prior to setting the nested
1802                  * state which can lead to a load of wrong PDPTRs.
1803                  */
1804                 if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
1805                         return false;
1806
1807         if (!nested_svm_vmrun_msrpm(svm)) {
1808                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1809                 vcpu->run->internal.suberror =
1810                         KVM_INTERNAL_ERROR_EMULATION;
1811                 vcpu->run->internal.ndata = 0;
1812                 return false;
1813         }
1814
1815         if (kvm_hv_verify_vp_assist(vcpu))
1816                 return false;
1817
1818         return true;
1819 }
1820
1821 struct kvm_x86_nested_ops svm_nested_ops = {
1822         .leave_nested = svm_leave_nested,
1823         .is_exception_vmexit = nested_svm_is_exception_vmexit,
1824         .check_events = svm_check_nested_events,
1825         .triple_fault = nested_svm_triple_fault,
1826         .get_nested_state_pages = svm_get_nested_state_pages,
1827         .get_state = svm_get_nested_state,
1828         .set_state = svm_set_nested_state,
1829         .hv_inject_synthetic_vmexit_post_tlb_flush = svm_hv_inject_synthetic_vmexit_post_tlb_flush,
1830 };