arch/x86/kvm/svm/nested.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Kernel-based Virtual Machine driver for Linux
   4  *
   5  * AMD SVM support
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *   Avi Kivity   <avi@qumranet.com>
  13  */
  14
  15 #define pr_fmt(fmt) "SVM: " fmt
  16
  17 #include <linux/kvm_types.h>
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20
  21 #include <asm/msr-index.h>
  22 #include <asm/debugreg.h>
  23
  24 #include "kvm_emulate.h"
  25 #include "trace.h"
  26 #include "mmu.h"
  27 #include "x86.h"
  28 #include "cpuid.h"
  29 #include "lapic.h"
  30 #include "svm.h"
  31
  32 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
  33
  34 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  35                                        struct x86_exception *fault)
  36 {
  37         struct vcpu_svm *svm = to_svm(vcpu);
  38
  39         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  40                 /*
  41                  * TODO: track the cause of the nested page fault, and
  42                  * correctly fill in the high bits of exit_info_1.
  43                  */
  44                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  45                 svm->vmcb->control.exit_code_hi = 0;
  46                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
  47                 svm->vmcb->control.exit_info_2 = fault->address;
  48         }
  49
  50         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  51         svm->vmcb->control.exit_info_1 |= fault->error_code;
  52
  53         nested_svm_vmexit(svm);
  54 }
  55
  56 static void svm_inject_page_fault_nested(struct kvm_vcpu *vcpu, struct x86_exception *fault)
  57 {
  58        struct vcpu_svm *svm = to_svm(vcpu);
  59        WARN_ON(!is_guest_mode(vcpu));
  60
  61        if (vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
  62            !svm->nested.nested_run_pending) {
  63                svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
  64                svm->vmcb->control.exit_code_hi = 0;
  65                svm->vmcb->control.exit_info_1 = fault->error_code;
  66                svm->vmcb->control.exit_info_2 = fault->address;
  67                nested_svm_vmexit(svm);
  68        } else {
  69                kvm_inject_page_fault(vcpu, fault);
  70        }
  71 }
  72
  73 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  74 {
  75         struct vcpu_svm *svm = to_svm(vcpu);
  76         u64 cr3 = svm->nested.ctl.nested_cr3;
  77         u64 pdpte;
  78         int ret;
  79
  80         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
  81                                        offset_in_page(cr3) + index * 8, 8);
  82         if (ret)
  83                 return 0;
  84         return pdpte;
  85 }
  86
  87 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  88 {
  89         struct vcpu_svm *svm = to_svm(vcpu);
  90
  91         return svm->nested.ctl.nested_cr3;
  92 }
  93
  94 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  95 {
  96         struct vcpu_svm *svm = to_svm(vcpu);
  97
  98         WARN_ON(mmu_is_nested(vcpu));
  99
 100         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
 101         kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
 102                                 svm->vmcb01.ptr->save.efer,
 103                                 svm->nested.ctl.nested_cr3);
 104         vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
 105         vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
 106         vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
 107         reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
 108         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 109 }
 110
 111 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 112 {
 113         vcpu->arch.mmu = &vcpu->arch.root_mmu;
 114         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 115 }
 116
 117 void recalc_intercepts(struct vcpu_svm *svm)
 118 {
 119         struct vmcb_control_area *c, *h, *g;
 120         unsigned int i;
 121
 122         vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 123
 124         if (!is_guest_mode(&svm->vcpu))
 125                 return;
 126
 127         c = &svm->vmcb->control;
 128         h = &svm->vmcb01.ptr->control;
 129         g = &svm->nested.ctl;
 130
 131         for (i = 0; i < MAX_INTERCEPT; i++)
 132                 c->intercepts[i] = h->intercepts[i];
 133
 134         if (g->int_ctl & V_INTR_MASKING_MASK) {
 135                 /* We only want the cr8 intercept bits of L1 */
 136                 vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
 137                 vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
 138
 139                 /*
 140                  * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 141                  * affect any interrupt we may want to inject; therefore,
 142                  * interrupt window vmexits are irrelevant to L0.
 143                  */
 144                 vmcb_clr_intercept(c, INTERCEPT_VINTR);
 145         }
 146
 147         /* We don't want to see VMMCALLs from a nested guest */
 148         vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
 149
 150         for (i = 0; i < MAX_INTERCEPT; i++)
 151                 c->intercepts[i] |= g->intercepts[i];
 152 }
 153
 154 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
 155                                    struct vmcb_control_area *from)
 156 {
 157         unsigned int i;
 158
 159         for (i = 0; i < MAX_INTERCEPT; i++)
 160                 dst->intercepts[i] = from->intercepts[i];
 161
 162         dst->iopm_base_pa         = from->iopm_base_pa;
 163         dst->msrpm_base_pa        = from->msrpm_base_pa;
 164         dst->tsc_offset           = from->tsc_offset;
 165         /* asid not copied, it is handled manually for svm->vmcb.  */
 166         dst->tlb_ctl              = from->tlb_ctl;
 167         dst->int_ctl              = from->int_ctl;
 168         dst->int_vector           = from->int_vector;
 169         dst->int_state            = from->int_state;
 170         dst->exit_code            = from->exit_code;
 171         dst->exit_code_hi         = from->exit_code_hi;
 172         dst->exit_info_1          = from->exit_info_1;
 173         dst->exit_info_2          = from->exit_info_2;
 174         dst->exit_int_info        = from->exit_int_info;
 175         dst->exit_int_info_err    = from->exit_int_info_err;
 176         dst->nested_ctl           = from->nested_ctl;
 177         dst->event_inj            = from->event_inj;
 178         dst->event_inj_err        = from->event_inj_err;
 179         dst->nested_cr3           = from->nested_cr3;
 180         dst->virt_ext              = from->virt_ext;
 181         dst->pause_filter_count   = from->pause_filter_count;
 182         dst->pause_filter_thresh  = from->pause_filter_thresh;
 183 }
 184
 185 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 186 {
 187         /*
 188          * This function merges the msr permission bitmaps of kvm and the
 189          * nested vmcb. It is optimized in that it only merges the parts where
 190          * the kvm msr permission bitmap may contain zero bits
 191          */
 192         int i;
 193
 194         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 195                 return true;
 196
 197         for (i = 0; i < MSRPM_OFFSETS; i++) {
 198                 u32 value, p;
 199                 u64 offset;
 200
 201                 if (msrpm_offsets[i] == 0xffffffff)
 202                         break;
 203
 204                 p      = msrpm_offsets[i];
 205                 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 206
 207                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 208                         return false;
 209
 210                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 211         }
 212
 213         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 214
 215         return true;
 216 }
 217
 218 /*
 219  * Bits 11:0 of bitmap address are ignored by hardware
 220  */
 221 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
 222 {
 223         u64 addr = PAGE_ALIGN(pa);
 224
 225         return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
 226             kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
 227 }
 228
 229 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
 230                                        struct vmcb_control_area *control)
 231 {
 232         if (CC(!vmcb_is_intercept(control, INTERCEPT_VMRUN)))
 233                 return false;
 234
 235         if (CC(control->asid == 0))
 236                 return false;
 237
 238         if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
 239                 return false;
 240
 241         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
 242                                            MSRPM_SIZE)))
 243                 return false;
 244         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
 245                                            IOPM_SIZE)))
 246                 return false;
 247
 248         return true;
 249 }
 250
 251 static bool nested_vmcb_check_cr3_cr4(struct kvm_vcpu *vcpu,
 252                                       struct vmcb_save_area *save)
 253 {
 254         /*
 255          * These checks are also performed by KVM_SET_SREGS,
 256          * except that EFER.LMA is not checked by SVM against
 257          * CR0.PG && EFER.LME.
 258          */
 259         if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
 260                 if (CC(!(save->cr4 & X86_CR4_PAE)) ||
 261                     CC(!(save->cr0 & X86_CR0_PE)) ||
 262                     CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
 263                         return false;
 264         }
 265
 266         if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
 267                 return false;
 268
 269         return true;
 270 }
 271
 272 /* Common checks that apply to both L1 and L2 state.  */
 273 static bool nested_vmcb_valid_sregs(struct kvm_vcpu *vcpu,
 274                                     struct vmcb_save_area *save)
 275 {
 276         /*
 277          * FIXME: these should be done after copying the fields,
 278          * to avoid TOC/TOU races.  For these save area checks
 279          * the possible damage is limited since kvm_set_cr0 and
 280          * kvm_set_cr4 handle failure; EFER_SVME is an exception
 281          * so it is force-set later in nested_prepare_vmcb_save.
 282          */
 283         if (CC(!(save->efer & EFER_SVME)))
 284                 return false;
 285
 286         if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 287             CC(save->cr0 & ~0xffffffffULL))
 288                 return false;
 289
 290         if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 291                 return false;
 292
 293         if (!nested_vmcb_check_cr3_cr4(vcpu, save))
 294                 return false;
 295
 296         if (CC(!kvm_valid_efer(vcpu, save->efer)))
 297                 return false;
 298
 299         return true;
 300 }
 301
 302 static void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
 303                                             struct vmcb_control_area *control)
 304 {
 305         copy_vmcb_control_area(&svm->nested.ctl, control);
 306
 307         /* Copy it here because nested_svm_check_controls will check it.  */
 308         svm->nested.ctl.asid           = control->asid;
 309         svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
 310         svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
 311 }
 312
 313 /*
 314  * Synchronize fields that are written by the processor, so that
 315  * they can be copied back into the vmcb12.
 316  */
 317 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
 318 {
 319         u32 mask;
 320         svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 321         svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 322
 323         /* Only a few fields of int_ctl are written by the processor.  */
 324         mask = V_IRQ_MASK | V_TPR_MASK;
 325         if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
 326             svm_is_intercept(svm, INTERCEPT_VINTR)) {
 327                 /*
 328                  * In order to request an interrupt window, L0 is usurping
 329                  * svm->vmcb->control.int_ctl and possibly setting V_IRQ
 330                  * even if it was clear in L1's VMCB.  Restoring it would be
 331                  * wrong.  However, in this case V_IRQ will remain true until
 332                  * interrupt_window_interception calls svm_clear_vintr and
 333                  * restores int_ctl.  We can just leave it aside.
 334                  */
 335                 mask &= ~V_IRQ_MASK;
 336         }
 337         svm->nested.ctl.int_ctl        &= ~mask;
 338         svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 339 }
 340
 341 /*
 342  * Transfer any event that L0 or L1 wanted to inject into L2 to
 343  * EXIT_INT_INFO.
 344  */
 345 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
 346                                                 struct vmcb *vmcb12)
 347 {
 348         struct kvm_vcpu *vcpu = &svm->vcpu;
 349         u32 exit_int_info = 0;
 350         unsigned int nr;
 351
 352         if (vcpu->arch.exception.injected) {
 353                 nr = vcpu->arch.exception.nr;
 354                 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 355
 356                 if (vcpu->arch.exception.has_error_code) {
 357                         exit_int_info |= SVM_EVTINJ_VALID_ERR;
 358                         vmcb12->control.exit_int_info_err =
 359                                 vcpu->arch.exception.error_code;
 360                 }
 361
 362         } else if (vcpu->arch.nmi_injected) {
 363                 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 364
 365         } else if (vcpu->arch.interrupt.injected) {
 366                 nr = vcpu->arch.interrupt.nr;
 367                 exit_int_info = nr | SVM_EVTINJ_VALID;
 368
 369                 if (vcpu->arch.interrupt.soft)
 370                         exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 371                 else
 372                         exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 373         }
 374
 375         vmcb12->control.exit_int_info = exit_int_info;
 376 }
 377
 378 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
 379 {
 380         return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
 381 }
 382
 383 /*
 384  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 385  * if we are emulating VM-Entry into a guest with NPT enabled.
 386  */
 387 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 388                                bool nested_npt)
 389 {
 390         if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
 391                 return -EINVAL;
 392
 393         if (!nested_npt && is_pae_paging(vcpu) &&
 394             (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
 395                 if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)))
 396                         return -EINVAL;
 397         }
 398
 399         /*
 400          * TODO: optimize unconditional TLB flush/MMU sync here and in
 401          * kvm_init_shadow_npt_mmu().
 402          */
 403         if (!nested_npt)
 404                 kvm_mmu_new_pgd(vcpu, cr3, false, false);
 405
 406         vcpu->arch.cr3 = cr3;
 407         kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
 408
 409         kvm_init_mmu(vcpu, false);
 410
 411         return 0;
 412 }
 413
 414 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
 415 {
 416         if (!svm->nested.vmcb02.ptr)
 417                 return;
 418
 419         /* FIXME: merge g_pat from vmcb01 and vmcb12.  */
 420         svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
 421 }
 422
 423 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 424 {
 425         bool new_vmcb12 = false;
 426
 427         nested_vmcb02_compute_g_pat(svm);
 428
 429         /* Load the nested guest state */
 430         if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
 431                 new_vmcb12 = true;
 432                 svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
 433         }
 434
 435         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
 436                 svm->vmcb->save.es = vmcb12->save.es;
 437                 svm->vmcb->save.cs = vmcb12->save.cs;
 438                 svm->vmcb->save.ss = vmcb12->save.ss;
 439                 svm->vmcb->save.ds = vmcb12->save.ds;
 440                 svm->vmcb->save.cpl = vmcb12->save.cpl;
 441                 vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
 442         }
 443
 444         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
 445                 svm->vmcb->save.gdtr = vmcb12->save.gdtr;
 446                 svm->vmcb->save.idtr = vmcb12->save.idtr;
 447                 vmcb_mark_dirty(svm->vmcb, VMCB_DT);
 448         }
 449
 450         kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
 451
 452         /*
 453          * Force-set EFER_SVME even though it is checked earlier on the
 454          * VMCB12, because the guest can flip the bit between the check
 455          * and now.  Clearing EFER_SVME would call svm_free_nested.
 456          */
 457         svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
 458
 459         svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
 460         svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
 461
 462         svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 463
 464         kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
 465         kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
 466         kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
 467
 468         /* In case we don't even reach vcpu_run, the fields are not updated */
 469         svm->vmcb->save.rax = vmcb12->save.rax;
 470         svm->vmcb->save.rsp = vmcb12->save.rsp;
 471         svm->vmcb->save.rip = vmcb12->save.rip;
 472
 473         /* These bits will be set properly on the first execution when new_vmc12 is true */
 474         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 475                 svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
 476                 svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
 477                 vmcb_mark_dirty(svm->vmcb, VMCB_DR);
 478         }
 479 }
 480
 481 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
 482 {
 483         const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
 484
 485         /*
 486          * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 487          * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 488          */
 489
 490         /*
 491          * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
 492          * avic_physical_id.
 493          */
 494         WARN_ON(svm->vmcb01.ptr->control.int_ctl & AVIC_ENABLE_MASK);
 495
 496         /* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 497         svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
 498         svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
 499         svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
 500
 501         /* Done at vmrun: asid.  */
 502
 503         /* Also overwritten later if necessary.  */
 504         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 505
 506         /* nested_cr3.  */
 507         if (nested_npt_enabled(svm))
 508                 nested_svm_init_mmu_context(&svm->vcpu);
 509
 510         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 511                 svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
 512
 513         svm->vmcb->control.int_ctl             =
 514                 (svm->nested.ctl.int_ctl & ~mask) |
 515                 (svm->vmcb01.ptr->control.int_ctl & mask);
 516
 517         svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
 518         svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
 519         svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
 520         svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
 521         svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 522
 523         svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
 524         svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
 525
 526         /* Enter Guest-Mode */
 527         enter_guest_mode(&svm->vcpu);
 528
 529         /*
 530          * Merge guest and host intercepts - must be called with vcpu in
 531          * guest-mode to take effect.
 532          */
 533         recalc_intercepts(svm);
 534 }
 535
 536 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 537 {
 538         /*
 539          * Some VMCB state is shared between L1 and L2 and thus has to be
 540          * moved at the time of nested vmrun and vmexit.
 541          *
 542          * VMLOAD/VMSAVE state would also belong in this category, but KVM
 543          * always performs VMLOAD and VMSAVE from the VMCB01.
 544          */
 545         to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 546 }
 547
 548 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 549                          struct vmcb *vmcb12)
 550 {
 551         struct vcpu_svm *svm = to_svm(vcpu);
 552         int ret;
 553
 554         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
 555                                vmcb12->save.rip,
 556                                vmcb12->control.int_ctl,
 557                                vmcb12->control.event_inj,
 558                                vmcb12->control.nested_ctl);
 559
 560         trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 561                                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 562                                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 563                                     vmcb12->control.intercepts[INTERCEPT_WORD3],
 564                                     vmcb12->control.intercepts[INTERCEPT_WORD4],
 565                                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
 566
 567
 568         svm->nested.vmcb12_gpa = vmcb12_gpa;
 569
 570         WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 571
 572         nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 573
 574         svm_switch_vmcb(svm, &svm->nested.vmcb02);
 575         nested_vmcb02_prepare_control(svm);
 576         nested_vmcb02_prepare_save(svm, vmcb12);
 577
 578         ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
 579                                   nested_npt_enabled(svm));
 580         if (ret)
 581                 return ret;
 582
 583         if (!npt_enabled)
 584                 vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
 585
 586         svm_set_gif(svm, true);
 587
 588         return 0;
 589 }
 590
 591 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 592 {
 593         struct vcpu_svm *svm = to_svm(vcpu);
 594         int ret;
 595         struct vmcb *vmcb12;
 596         struct kvm_host_map map;
 597         u64 vmcb12_gpa;
 598
 599         ++vcpu->stat.nested_run;
 600
 601         if (is_smm(vcpu)) {
 602                 kvm_queue_exception(vcpu, UD_VECTOR);
 603                 return 1;
 604         }
 605
 606         vmcb12_gpa = svm->vmcb->save.rax;
 607         ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 608         if (ret == -EINVAL) {
 609                 kvm_inject_gp(vcpu, 0);
 610                 return 1;
 611         } else if (ret) {
 612                 return kvm_skip_emulated_instruction(vcpu);
 613         }
 614
 615         ret = kvm_skip_emulated_instruction(vcpu);
 616
 617         vmcb12 = map.hva;
 618
 619         if (WARN_ON_ONCE(!svm->nested.initialized))
 620                 return -EINVAL;
 621
 622         nested_load_control_from_vmcb12(svm, &vmcb12->control);
 623
 624         if (!nested_vmcb_valid_sregs(vcpu, &vmcb12->save) ||
 625             !nested_vmcb_check_controls(vcpu, &svm->nested.ctl)) {
 626                 vmcb12->control.exit_code    = SVM_EXIT_ERR;
 627                 vmcb12->control.exit_code_hi = 0;
 628                 vmcb12->control.exit_info_1  = 0;
 629                 vmcb12->control.exit_info_2  = 0;
 630                 goto out;
 631         }
 632
 633
 634         /* Clear internal status */
 635         kvm_clear_exception_queue(vcpu);
 636         kvm_clear_interrupt_queue(vcpu);
 637
 638         /*
 639          * Since vmcb01 is not in use, we can use it to store some of the L1
 640          * state.
 641          */
 642         svm->vmcb01.ptr->save.efer   = vcpu->arch.efer;
 643         svm->vmcb01.ptr->save.cr0    = kvm_read_cr0(vcpu);
 644         svm->vmcb01.ptr->save.cr4    = vcpu->arch.cr4;
 645         svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
 646         svm->vmcb01.ptr->save.rip    = kvm_rip_read(vcpu);
 647
 648         if (!npt_enabled)
 649                 svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
 650
 651         svm->nested.nested_run_pending = 1;
 652
 653         if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12))
 654                 goto out_exit_err;
 655
 656         if (nested_svm_vmrun_msrpm(svm))
 657                 goto out;
 658
 659 out_exit_err:
 660         svm->nested.nested_run_pending = 0;
 661
 662         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 663         svm->vmcb->control.exit_code_hi = 0;
 664         svm->vmcb->control.exit_info_1  = 0;
 665         svm->vmcb->control.exit_info_2  = 0;
 666
 667         nested_svm_vmexit(svm);
 668
 669 out:
 670         kvm_vcpu_unmap(vcpu, &map, true);
 671
 672         return ret;
 673 }
 674
 675 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 676 {
 677         to_vmcb->save.fs = from_vmcb->save.fs;
 678         to_vmcb->save.gs = from_vmcb->save.gs;
 679         to_vmcb->save.tr = from_vmcb->save.tr;
 680         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 681         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 682         to_vmcb->save.star = from_vmcb->save.star;
 683         to_vmcb->save.lstar = from_vmcb->save.lstar;
 684         to_vmcb->save.cstar = from_vmcb->save.cstar;
 685         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 686         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 687         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 688         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 689 }
 690
 691 int nested_svm_vmexit(struct vcpu_svm *svm)
 692 {
 693         struct kvm_vcpu *vcpu = &svm->vcpu;
 694         struct vmcb *vmcb12;
 695         struct vmcb *vmcb = svm->vmcb;
 696         struct kvm_host_map map;
 697         int rc;
 698
 699         /* Triple faults in L2 should never escape. */
 700         WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
 701
 702         rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 703         if (rc) {
 704                 if (rc == -EINVAL)
 705                         kvm_inject_gp(vcpu, 0);
 706                 return 1;
 707         }
 708
 709         vmcb12 = map.hva;
 710
 711         /* Exit Guest-Mode */
 712         leave_guest_mode(vcpu);
 713         svm->nested.vmcb12_gpa = 0;
 714         WARN_ON_ONCE(svm->nested.nested_run_pending);
 715
 716         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 717
 718         /* in case we halted in L2 */
 719         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 720
 721         /* Give the current vmcb to the guest */
 722
 723         vmcb12->save.es     = vmcb->save.es;
 724         vmcb12->save.cs     = vmcb->save.cs;
 725         vmcb12->save.ss     = vmcb->save.ss;
 726         vmcb12->save.ds     = vmcb->save.ds;
 727         vmcb12->save.gdtr   = vmcb->save.gdtr;
 728         vmcb12->save.idtr   = vmcb->save.idtr;
 729         vmcb12->save.efer   = svm->vcpu.arch.efer;
 730         vmcb12->save.cr0    = kvm_read_cr0(vcpu);
 731         vmcb12->save.cr3    = kvm_read_cr3(vcpu);
 732         vmcb12->save.cr2    = vmcb->save.cr2;
 733         vmcb12->save.cr4    = svm->vcpu.arch.cr4;
 734         vmcb12->save.rflags = kvm_get_rflags(vcpu);
 735         vmcb12->save.rip    = kvm_rip_read(vcpu);
 736         vmcb12->save.rsp    = kvm_rsp_read(vcpu);
 737         vmcb12->save.rax    = kvm_rax_read(vcpu);
 738         vmcb12->save.dr7    = vmcb->save.dr7;
 739         vmcb12->save.dr6    = svm->vcpu.arch.dr6;
 740         vmcb12->save.cpl    = vmcb->save.cpl;
 741
 742         vmcb12->control.int_state         = vmcb->control.int_state;
 743         vmcb12->control.exit_code         = vmcb->control.exit_code;
 744         vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
 745         vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
 746         vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
 747
 748         if (vmcb12->control.exit_code != SVM_EXIT_ERR)
 749                 nested_save_pending_event_to_vmcb12(svm, vmcb12);
 750
 751         if (svm->nrips_enabled)
 752                 vmcb12->control.next_rip  = vmcb->control.next_rip;
 753
 754         vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
 755         vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 756         vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
 757         vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 758
 759         vmcb12->control.pause_filter_count =
 760                 svm->vmcb->control.pause_filter_count;
 761         vmcb12->control.pause_filter_thresh =
 762                 svm->vmcb->control.pause_filter_thresh;
 763
 764         nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
 765
 766         svm_switch_vmcb(svm, &svm->vmcb01);
 767         WARN_ON_ONCE(svm->vmcb->control.exit_code != SVM_EXIT_VMRUN);
 768
 769         /*
 770          * On vmexit the  GIF is set to false and
 771          * no event can be injected in L1.
 772          */
 773         svm_set_gif(svm, false);
 774         svm->vmcb->control.exit_int_info = 0;
 775
 776         svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
 777         if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
 778                 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 779                 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 780         }
 781
 782         svm->nested.ctl.nested_cr3 = 0;
 783
 784         /*
 785          * Restore processor state that had been saved in vmcb01
 786          */
 787         kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
 788         svm_set_efer(vcpu, svm->vmcb->save.efer);
 789         svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
 790         svm_set_cr4(vcpu, svm->vmcb->save.cr4);
 791         kvm_rax_write(vcpu, svm->vmcb->save.rax);
 792         kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
 793         kvm_rip_write(vcpu, svm->vmcb->save.rip);
 794
 795         svm->vcpu.arch.dr7 = DR7_FIXED_1;
 796         kvm_update_dr7(&svm->vcpu);
 797
 798         trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
 799                                        vmcb12->control.exit_info_1,
 800                                        vmcb12->control.exit_info_2,
 801                                        vmcb12->control.exit_int_info,
 802                                        vmcb12->control.exit_int_info_err,
 803                                        KVM_ISA_SVM);
 804
 805         kvm_vcpu_unmap(vcpu, &map, true);
 806
 807         nested_svm_uninit_mmu_context(vcpu);
 808
 809         rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false);
 810         if (rc)
 811                 return 1;
 812
 813         /*
 814          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 815          * doesn't end up in L1.
 816          */
 817         svm->vcpu.arch.nmi_injected = false;
 818         kvm_clear_exception_queue(vcpu);
 819         kvm_clear_interrupt_queue(vcpu);
 820
 821         /*
 822          * If we are here following the completion of a VMRUN that
 823          * is being single-stepped, queue the pending #DB intercept
 824          * right now so that it an be accounted for before we execute
 825          * L1's next instruction.
 826          */
 827         if (unlikely(svm->vmcb->save.rflags & X86_EFLAGS_TF))
 828                 kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
 829
 830         return 0;
 831 }
 832
 833 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
 834 {
 835         nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
 836 }
 837
 838 int svm_allocate_nested(struct vcpu_svm *svm)
 839 {
 840         struct page *vmcb02_page;
 841
 842         if (svm->nested.initialized)
 843                 return 0;
 844
 845         vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 846         if (!vmcb02_page)
 847                 return -ENOMEM;
 848         svm->nested.vmcb02.ptr = page_address(vmcb02_page);
 849         svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
 850
 851         svm->nested.msrpm = svm_vcpu_alloc_msrpm();
 852         if (!svm->nested.msrpm)
 853                 goto err_free_vmcb02;
 854         svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
 855
 856         svm->nested.initialized = true;
 857         return 0;
 858
 859 err_free_vmcb02:
 860         __free_page(vmcb02_page);
 861         return -ENOMEM;
 862 }
 863
 864 void svm_free_nested(struct vcpu_svm *svm)
 865 {
 866         if (!svm->nested.initialized)
 867                 return;
 868
 869         svm_vcpu_free_msrpm(svm->nested.msrpm);
 870         svm->nested.msrpm = NULL;
 871
 872         __free_page(virt_to_page(svm->nested.vmcb02.ptr));
 873         svm->nested.vmcb02.ptr = NULL;
 874
 875         svm->nested.initialized = false;
 876 }
 877
 878 /*
 879  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
 880  */
 881 void svm_leave_nested(struct vcpu_svm *svm)
 882 {
 883         struct kvm_vcpu *vcpu = &svm->vcpu;
 884
 885         if (is_guest_mode(vcpu)) {
 886                 svm->nested.nested_run_pending = 0;
 887                 leave_guest_mode(vcpu);
 888
 889                 svm_switch_vmcb(svm, &svm->nested.vmcb02);
 890
 891                 nested_svm_uninit_mmu_context(vcpu);
 892                 vmcb_mark_all_dirty(svm->vmcb);
 893         }
 894
 895         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 896 }
 897
 898 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 899 {
 900         u32 offset, msr, value;
 901         int write, mask;
 902
 903         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 904                 return NESTED_EXIT_HOST;
 905
 906         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 907         offset = svm_msrpm_offset(msr);
 908         write  = svm->vmcb->control.exit_info_1 & 1;
 909         mask   = 1 << ((2 * (msr & 0xf)) + write);
 910
 911         if (offset == MSR_INVALID)
 912                 return NESTED_EXIT_DONE;
 913
 914         /* Offset is in 32 bit units but need in 8 bit units */
 915         offset *= 4;
 916
 917         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
 918                 return NESTED_EXIT_DONE;
 919
 920         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 921 }
 922
 923 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 924 {
 925         unsigned port, size, iopm_len;
 926         u16 val, mask;
 927         u8 start_bit;
 928         u64 gpa;
 929
 930         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
 931                 return NESTED_EXIT_HOST;
 932
 933         port = svm->vmcb->control.exit_info_1 >> 16;
 934         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 935                 SVM_IOIO_SIZE_SHIFT;
 936         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
 937         start_bit = port % 8;
 938         iopm_len = (start_bit + size > 8) ? 2 : 1;
 939         mask = (0xf >> (4 - size)) << start_bit;
 940         val = 0;
 941
 942         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 943                 return NESTED_EXIT_DONE;
 944
 945         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 946 }
 947
 948 static int nested_svm_intercept(struct vcpu_svm *svm)
 949 {
 950         u32 exit_code = svm->vmcb->control.exit_code;
 951         int vmexit = NESTED_EXIT_HOST;
 952
 953         switch (exit_code) {
 954         case SVM_EXIT_MSR:
 955                 vmexit = nested_svm_exit_handled_msr(svm);
 956                 break;
 957         case SVM_EXIT_IOIO:
 958                 vmexit = nested_svm_intercept_ioio(svm);
 959                 break;
 960         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 961                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 962                         vmexit = NESTED_EXIT_DONE;
 963                 break;
 964         }
 965         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 966                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 967                         vmexit = NESTED_EXIT_DONE;
 968                 break;
 969         }
 970         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 971                 /*
 972                  * Host-intercepted exceptions have been checked already in
 973                  * nested_svm_exit_special.  There is nothing to do here,
 974                  * the vmexit is injected by svm_check_nested_events.
 975                  */
 976                 vmexit = NESTED_EXIT_DONE;
 977                 break;
 978         }
 979         case SVM_EXIT_ERR: {
 980                 vmexit = NESTED_EXIT_DONE;
 981                 break;
 982         }
 983         default: {
 984                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 985                         vmexit = NESTED_EXIT_DONE;
 986         }
 987         }
 988
 989         return vmexit;
 990 }
 991
 992 int nested_svm_exit_handled(struct vcpu_svm *svm)
 993 {
 994         int vmexit;
 995
 996         vmexit = nested_svm_intercept(svm);
 997
 998         if (vmexit == NESTED_EXIT_DONE)
 999                 nested_svm_vmexit(svm);
1000
1001         return vmexit;
1002 }
1003
1004 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1005 {
1006         if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1007                 kvm_queue_exception(vcpu, UD_VECTOR);
1008                 return 1;
1009         }
1010
1011         if (to_svm(vcpu)->vmcb->save.cpl) {
1012                 kvm_inject_gp(vcpu, 0);
1013                 return 1;
1014         }
1015
1016         return 0;
1017 }
1018
1019 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1020 {
1021         unsigned int nr = svm->vcpu.arch.exception.nr;
1022
1023         return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1024 }
1025
1026 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1027 {
1028         unsigned int nr = svm->vcpu.arch.exception.nr;
1029
1030         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1031         svm->vmcb->control.exit_code_hi = 0;
1032
1033         if (svm->vcpu.arch.exception.has_error_code)
1034                 svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1035
1036         /*
1037          * EXITINFO2 is undefined for all exception intercepts other
1038          * than #PF.
1039          */
1040         if (nr == PF_VECTOR) {
1041                 if (svm->vcpu.arch.exception.nested_apf)
1042                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1043                 else if (svm->vcpu.arch.exception.has_payload)
1044                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1045                 else
1046                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1047         } else if (nr == DB_VECTOR) {
1048                 /* See inject_pending_event.  */
1049                 kvm_deliver_exception_payload(&svm->vcpu);
1050                 if (svm->vcpu.arch.dr7 & DR7_GD) {
1051                         svm->vcpu.arch.dr7 &= ~DR7_GD;
1052                         kvm_update_dr7(&svm->vcpu);
1053                 }
1054         } else
1055                 WARN_ON(svm->vcpu.arch.exception.has_payload);
1056
1057         nested_svm_vmexit(svm);
1058 }
1059
1060 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1061 {
1062         return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1063 }
1064
1065 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1066 {
1067         struct vcpu_svm *svm = to_svm(vcpu);
1068         bool block_nested_events =
1069                 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1070         struct kvm_lapic *apic = vcpu->arch.apic;
1071
1072         if (lapic_in_kernel(vcpu) &&
1073             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1074                 if (block_nested_events)
1075                         return -EBUSY;
1076                 if (!nested_exit_on_init(svm))
1077                         return 0;
1078                 nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1079                 return 0;
1080         }
1081
1082         if (vcpu->arch.exception.pending) {
1083                 /*
1084                  * Only a pending nested run can block a pending exception.
1085                  * Otherwise an injected NMI/interrupt should either be
1086                  * lost or delivered to the nested hypervisor in the EXITINTINFO
1087                  * vmcb field, while delivering the pending exception.
1088                  */
1089                 if (svm->nested.nested_run_pending)
1090                         return -EBUSY;
1091                 if (!nested_exit_on_exception(svm))
1092                         return 0;
1093                 nested_svm_inject_exception_vmexit(svm);
1094                 return 0;
1095         }
1096
1097         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1098                 if (block_nested_events)
1099                         return -EBUSY;
1100                 if (!nested_exit_on_smi(svm))
1101                         return 0;
1102                 nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1103                 return 0;
1104         }
1105
1106         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1107                 if (block_nested_events)
1108                         return -EBUSY;
1109                 if (!nested_exit_on_nmi(svm))
1110                         return 0;
1111                 nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1112                 return 0;
1113         }
1114
1115         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1116                 if (block_nested_events)
1117                         return -EBUSY;
1118                 if (!nested_exit_on_intr(svm))
1119                         return 0;
1120                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1121                 nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1122                 return 0;
1123         }
1124
1125         return 0;
1126 }
1127
1128 int nested_svm_exit_special(struct vcpu_svm *svm)
1129 {
1130         u32 exit_code = svm->vmcb->control.exit_code;
1131
1132         switch (exit_code) {
1133         case SVM_EXIT_INTR:
1134         case SVM_EXIT_NMI:
1135         case SVM_EXIT_NPF:
1136                 return NESTED_EXIT_HOST;
1137         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1138                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1139
1140                 if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1141                     excp_bits)
1142                         return NESTED_EXIT_HOST;
1143                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1144                          svm->vcpu.arch.apf.host_apf_flags)
1145                         /* Trap async PF even if not shadowing */
1146                         return NESTED_EXIT_HOST;
1147                 break;
1148         }
1149         default:
1150                 break;
1151         }
1152
1153         return NESTED_EXIT_CONTINUE;
1154 }
1155
1156 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1157                                 struct kvm_nested_state __user *user_kvm_nested_state,
1158                                 u32 user_data_size)
1159 {
1160         struct vcpu_svm *svm;
1161         struct kvm_nested_state kvm_state = {
1162                 .flags = 0,
1163                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1164                 .size = sizeof(kvm_state),
1165         };
1166         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1167                 &user_kvm_nested_state->data.svm[0];
1168
1169         if (!vcpu)
1170                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1171
1172         svm = to_svm(vcpu);
1173
1174         if (user_data_size < kvm_state.size)
1175                 goto out;
1176
1177         /* First fill in the header and copy it out.  */
1178         if (is_guest_mode(vcpu)) {
1179                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1180                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1181                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1182
1183                 if (svm->nested.nested_run_pending)
1184                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1185         }
1186
1187         if (gif_set(svm))
1188                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1189
1190         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1191                 return -EFAULT;
1192
1193         if (!is_guest_mode(vcpu))
1194                 goto out;
1195
1196         /*
1197          * Copy over the full size of the VMCB rather than just the size
1198          * of the structs.
1199          */
1200         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1201                 return -EFAULT;
1202         if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1203                          sizeof(user_vmcb->control)))
1204                 return -EFAULT;
1205         if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1206                          sizeof(user_vmcb->save)))
1207                 return -EFAULT;
1208 out:
1209         return kvm_state.size;
1210 }
1211
1212 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1213                                 struct kvm_nested_state __user *user_kvm_nested_state,
1214                                 struct kvm_nested_state *kvm_state)
1215 {
1216         struct vcpu_svm *svm = to_svm(vcpu);
1217         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1218                 &user_kvm_nested_state->data.svm[0];
1219         struct vmcb_control_area *ctl;
1220         struct vmcb_save_area *save;
1221         int ret;
1222         u32 cr0;
1223
1224         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1225                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1226
1227         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1228                 return -EINVAL;
1229
1230         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1231                                  KVM_STATE_NESTED_RUN_PENDING |
1232                                  KVM_STATE_NESTED_GIF_SET))
1233                 return -EINVAL;
1234
1235         /*
1236          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1237          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1238          */
1239         if (!(vcpu->arch.efer & EFER_SVME)) {
1240                 /* GIF=1 and no guest mode are required if SVME=0.  */
1241                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1242                         return -EINVAL;
1243         }
1244
1245         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1246         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1247                 return -EINVAL;
1248
1249         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1250                 svm_leave_nested(svm);
1251                 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1252                 return 0;
1253         }
1254
1255         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1256                 return -EINVAL;
1257         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1258                 return -EINVAL;
1259
1260         ret  = -ENOMEM;
1261         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1262         save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1263         if (!ctl || !save)
1264                 goto out_free;
1265
1266         ret = -EFAULT;
1267         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1268                 goto out_free;
1269         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1270                 goto out_free;
1271
1272         ret = -EINVAL;
1273         if (!nested_vmcb_check_controls(vcpu, ctl))
1274                 goto out_free;
1275
1276         /*
1277          * Processor state contains L2 state.  Check that it is
1278          * valid for guest mode (see nested_vmcb_check_save).
1279          */
1280         cr0 = kvm_read_cr0(vcpu);
1281         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1282                 goto out_free;
1283
1284         /*
1285          * Validate host state saved from before VMRUN (see
1286          * nested_svm_check_permissions).
1287          */
1288         if (!(save->cr0 & X86_CR0_PG) ||
1289             !(save->cr0 & X86_CR0_PE) ||
1290             (save->rflags & X86_EFLAGS_VM) ||
1291             !nested_vmcb_valid_sregs(vcpu, save))
1292                 goto out_free;
1293
1294         /*
1295          * All checks done, we can enter guest mode. Userspace provides
1296          * vmcb12.control, which will be combined with L1 and stored into
1297          * vmcb02, and the L1 save state which we store in vmcb01.
1298          * L2 registers if needed are moved from the current VMCB to VMCB02.
1299          */
1300
1301         svm->nested.nested_run_pending =
1302                 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1303
1304         svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1305         if (svm->current_vmcb == &svm->vmcb01)
1306                 svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1307
1308         svm->vmcb01.ptr->save.es = save->es;
1309         svm->vmcb01.ptr->save.cs = save->cs;
1310         svm->vmcb01.ptr->save.ss = save->ss;
1311         svm->vmcb01.ptr->save.ds = save->ds;
1312         svm->vmcb01.ptr->save.gdtr = save->gdtr;
1313         svm->vmcb01.ptr->save.idtr = save->idtr;
1314         svm->vmcb01.ptr->save.rflags = save->rflags | X86_EFLAGS_FIXED;
1315         svm->vmcb01.ptr->save.efer = save->efer;
1316         svm->vmcb01.ptr->save.cr0 = save->cr0;
1317         svm->vmcb01.ptr->save.cr3 = save->cr3;
1318         svm->vmcb01.ptr->save.cr4 = save->cr4;
1319         svm->vmcb01.ptr->save.rax = save->rax;
1320         svm->vmcb01.ptr->save.rsp = save->rsp;
1321         svm->vmcb01.ptr->save.rip = save->rip;
1322         svm->vmcb01.ptr->save.cpl = 0;
1323
1324         nested_load_control_from_vmcb12(svm, ctl);
1325
1326         svm_switch_vmcb(svm, &svm->nested.vmcb02);
1327
1328         nested_vmcb02_prepare_control(svm);
1329
1330         kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1331         ret = 0;
1332 out_free:
1333         kfree(save);
1334         kfree(ctl);
1335
1336         return ret;
1337 }
1338
1339 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1340 {
1341         struct vcpu_svm *svm = to_svm(vcpu);
1342
1343         if (WARN_ON(!is_guest_mode(vcpu)))
1344                 return true;
1345
1346         if (nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1347                                 nested_npt_enabled(svm)))
1348                 return false;
1349
1350         if (!nested_svm_vmrun_msrpm(svm)) {
1351                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1352                 vcpu->run->internal.suberror =
1353                         KVM_INTERNAL_ERROR_EMULATION;
1354                 vcpu->run->internal.ndata = 0;
1355                 return false;
1356         }
1357
1358         return true;
1359 }
1360
1361 struct kvm_x86_nested_ops svm_nested_ops = {
1362         .check_events = svm_check_nested_events,
1363         .triple_fault = nested_svm_triple_fault,
1364         .get_nested_state_pages = svm_get_nested_state_pages,
1365         .get_state = svm_get_nested_state,
1366         .set_state = svm_set_nested_state,
1367 };