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 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  33                                        struct x86_exception *fault)
  34 {
  35         struct vcpu_svm *svm = to_svm(vcpu);
  36
  37         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  38                 /*
  39                  * TODO: track the cause of the nested page fault, and
  40                  * correctly fill in the high bits of exit_info_1.
  41                  */
  42                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  43                 svm->vmcb->control.exit_code_hi = 0;
  44                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
  45                 svm->vmcb->control.exit_info_2 = fault->address;
  46         }
  47
  48         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  49         svm->vmcb->control.exit_info_1 |= fault->error_code;
  50
  51         nested_svm_vmexit(svm);
  52 }
  53
  54 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  55 {
  56         struct vcpu_svm *svm = to_svm(vcpu);
  57         u64 cr3 = svm->nested.ctl.nested_cr3;
  58         u64 pdpte;
  59         int ret;
  60
  61         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
  62                                        offset_in_page(cr3) + index * 8, 8);
  63         if (ret)
  64                 return 0;
  65         return pdpte;
  66 }
  67
  68 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  69 {
  70         struct vcpu_svm *svm = to_svm(vcpu);
  71
  72         return svm->nested.ctl.nested_cr3;
  73 }
  74
  75 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  76 {
  77         struct vcpu_svm *svm = to_svm(vcpu);
  78         struct vmcb *hsave = svm->nested.hsave;
  79
  80         WARN_ON(mmu_is_nested(vcpu));
  81
  82         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
  83         kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, hsave->save.cr4, hsave->save.efer,
  84                                 svm->nested.ctl.nested_cr3);
  85         vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
  86         vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
  87         vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
  88         reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
  89         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
  90 }
  91
  92 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
  93 {
  94         vcpu->arch.mmu = &vcpu->arch.root_mmu;
  95         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
  96 }
  97
  98 void recalc_intercepts(struct vcpu_svm *svm)
  99 {
 100         struct vmcb_control_area *c, *h, *g;
 101
 102         vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 103
 104         if (!is_guest_mode(&svm->vcpu))
 105                 return;
 106
 107         c = &svm->vmcb->control;
 108         h = &svm->nested.hsave->control;
 109         g = &svm->nested.ctl;
 110
 111         svm->nested.host_intercept_exceptions = h->intercept_exceptions;
 112
 113         c->intercept_cr = h->intercept_cr;
 114         c->intercept_dr = h->intercept_dr;
 115         c->intercept_exceptions = h->intercept_exceptions;
 116         c->intercept = h->intercept;
 117
 118         if (g->int_ctl & V_INTR_MASKING_MASK) {
 119                 /* We only want the cr8 intercept bits of L1 */
 120                 c->intercept_cr &= ~(1U << INTERCEPT_CR8_READ);
 121                 c->intercept_cr &= ~(1U << INTERCEPT_CR8_WRITE);
 122
 123                 /*
 124                  * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 125                  * affect any interrupt we may want to inject; therefore,
 126                  * interrupt window vmexits are irrelevant to L0.
 127                  */
 128                 c->intercept &= ~(1ULL << INTERCEPT_VINTR);
 129         }
 130
 131         /* We don't want to see VMMCALLs from a nested guest */
 132         c->intercept &= ~(1ULL << INTERCEPT_VMMCALL);
 133
 134         c->intercept_cr |= g->intercept_cr;
 135         c->intercept_dr |= g->intercept_dr;
 136         c->intercept_exceptions |= g->intercept_exceptions;
 137         c->intercept |= g->intercept;
 138 }
 139
 140 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
 141                                    struct vmcb_control_area *from)
 142 {
 143         dst->intercept_cr         = from->intercept_cr;
 144         dst->intercept_dr         = from->intercept_dr;
 145         dst->intercept_exceptions = from->intercept_exceptions;
 146         dst->intercept            = from->intercept;
 147         dst->iopm_base_pa         = from->iopm_base_pa;
 148         dst->msrpm_base_pa        = from->msrpm_base_pa;
 149         dst->tsc_offset           = from->tsc_offset;
 150         /* asid not copied, it is handled manually for svm->vmcb.  */
 151         dst->tlb_ctl              = from->tlb_ctl;
 152         dst->int_ctl              = from->int_ctl;
 153         dst->int_vector           = from->int_vector;
 154         dst->int_state            = from->int_state;
 155         dst->exit_code            = from->exit_code;
 156         dst->exit_code_hi         = from->exit_code_hi;
 157         dst->exit_info_1          = from->exit_info_1;
 158         dst->exit_info_2          = from->exit_info_2;
 159         dst->exit_int_info        = from->exit_int_info;
 160         dst->exit_int_info_err    = from->exit_int_info_err;
 161         dst->nested_ctl           = from->nested_ctl;
 162         dst->event_inj            = from->event_inj;
 163         dst->event_inj_err        = from->event_inj_err;
 164         dst->nested_cr3           = from->nested_cr3;
 165         dst->virt_ext              = from->virt_ext;
 166         dst->pause_filter_count   = from->pause_filter_count;
 167         dst->pause_filter_thresh  = from->pause_filter_thresh;
 168 }
 169
 170 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 171 {
 172         /*
 173          * This function merges the msr permission bitmaps of kvm and the
 174          * nested vmcb. It is optimized in that it only merges the parts where
 175          * the kvm msr permission bitmap may contain zero bits
 176          */
 177         int i;
 178
 179         if (!(svm->nested.ctl.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 180                 return true;
 181
 182         for (i = 0; i < MSRPM_OFFSETS; i++) {
 183                 u32 value, p;
 184                 u64 offset;
 185
 186                 if (msrpm_offsets[i] == 0xffffffff)
 187                         break;
 188
 189                 p      = msrpm_offsets[i];
 190                 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 191
 192                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 193                         return false;
 194
 195                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 196         }
 197
 198         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 199
 200         return true;
 201 }
 202
 203 static bool nested_vmcb_check_controls(struct vmcb_control_area *control)
 204 {
 205         if ((control->intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
 206                 return false;
 207
 208         if (control->asid == 0)
 209                 return false;
 210
 211         if ((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
 212             !npt_enabled)
 213                 return false;
 214
 215         return true;
 216 }
 217
 218 static bool nested_vmcb_checks(struct vcpu_svm *svm, struct vmcb *vmcb)
 219 {
 220         bool nested_vmcb_lma;
 221         if ((vmcb->save.efer & EFER_SVME) == 0)
 222                 return false;
 223
 224         if (((vmcb->save.cr0 & X86_CR0_CD) == 0) &&
 225             (vmcb->save.cr0 & X86_CR0_NW))
 226                 return false;
 227
 228         if (!kvm_dr6_valid(vmcb->save.dr6) || !kvm_dr7_valid(vmcb->save.dr7))
 229                 return false;
 230
 231         nested_vmcb_lma =
 232                 (vmcb->save.efer & EFER_LME) &&
 233                 (vmcb->save.cr0 & X86_CR0_PG);
 234
 235         if (!nested_vmcb_lma) {
 236                 if (vmcb->save.cr4 & X86_CR4_PAE) {
 237                         if (vmcb->save.cr3 & MSR_CR3_LEGACY_PAE_RESERVED_MASK)
 238                                 return false;
 239                 } else {
 240                         if (vmcb->save.cr3 & MSR_CR3_LEGACY_RESERVED_MASK)
 241                                 return false;
 242                 }
 243         } else {
 244                 if (!(vmcb->save.cr4 & X86_CR4_PAE) ||
 245                     !(vmcb->save.cr0 & X86_CR0_PE) ||
 246                     (vmcb->save.cr3 & MSR_CR3_LONG_RESERVED_MASK))
 247                         return false;
 248         }
 249         if (kvm_valid_cr4(&svm->vcpu, vmcb->save.cr4))
 250                 return false;
 251
 252         return nested_vmcb_check_controls(&vmcb->control);
 253 }
 254
 255 static void load_nested_vmcb_control(struct vcpu_svm *svm,
 256                                      struct vmcb_control_area *control)
 257 {
 258         copy_vmcb_control_area(&svm->nested.ctl, control);
 259
 260         /* Copy it here because nested_svm_check_controls will check it.  */
 261         svm->nested.ctl.asid           = control->asid;
 262         svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
 263         svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
 264 }
 265
 266 /*
 267  * Synchronize fields that are written by the processor, so that
 268  * they can be copied back into the nested_vmcb.
 269  */
 270 void sync_nested_vmcb_control(struct vcpu_svm *svm)
 271 {
 272         u32 mask;
 273         svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 274         svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 275
 276         /* Only a few fields of int_ctl are written by the processor.  */
 277         mask = V_IRQ_MASK | V_TPR_MASK;
 278         if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
 279             svm_is_intercept(svm, INTERCEPT_VINTR)) {
 280                 /*
 281                  * In order to request an interrupt window, L0 is usurping
 282                  * svm->vmcb->control.int_ctl and possibly setting V_IRQ
 283                  * even if it was clear in L1's VMCB.  Restoring it would be
 284                  * wrong.  However, in this case V_IRQ will remain true until
 285                  * interrupt_window_interception calls svm_clear_vintr and
 286                  * restores int_ctl.  We can just leave it aside.
 287                  */
 288                 mask &= ~V_IRQ_MASK;
 289         }
 290         svm->nested.ctl.int_ctl        &= ~mask;
 291         svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 292 }
 293
 294 /*
 295  * Transfer any event that L0 or L1 wanted to inject into L2 to
 296  * EXIT_INT_INFO.
 297  */
 298 static void nested_vmcb_save_pending_event(struct vcpu_svm *svm,
 299                                            struct vmcb *nested_vmcb)
 300 {
 301         struct kvm_vcpu *vcpu = &svm->vcpu;
 302         u32 exit_int_info = 0;
 303         unsigned int nr;
 304
 305         if (vcpu->arch.exception.injected) {
 306                 nr = vcpu->arch.exception.nr;
 307                 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 308
 309                 if (vcpu->arch.exception.has_error_code) {
 310                         exit_int_info |= SVM_EVTINJ_VALID_ERR;
 311                         nested_vmcb->control.exit_int_info_err =
 312                                 vcpu->arch.exception.error_code;
 313                 }
 314
 315         } else if (vcpu->arch.nmi_injected) {
 316                 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 317
 318         } else if (vcpu->arch.interrupt.injected) {
 319                 nr = vcpu->arch.interrupt.nr;
 320                 exit_int_info = nr | SVM_EVTINJ_VALID;
 321
 322                 if (vcpu->arch.interrupt.soft)
 323                         exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 324                 else
 325                         exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 326         }
 327
 328         nested_vmcb->control.exit_int_info = exit_int_info;
 329 }
 330
 331 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
 332 {
 333         return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
 334 }
 335
 336 /*
 337  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 338  * if we are emulating VM-Entry into a guest with NPT enabled.
 339  */
 340 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 341                                bool nested_npt)
 342 {
 343         if (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))
 344                 return -EINVAL;
 345
 346         if (!nested_npt && is_pae_paging(vcpu) &&
 347             (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
 348                 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 349                         return -EINVAL;
 350         }
 351
 352         /*
 353          * TODO: optimize unconditional TLB flush/MMU sync here and in
 354          * kvm_init_shadow_npt_mmu().
 355          */
 356         if (!nested_npt)
 357                 kvm_mmu_new_pgd(vcpu, cr3, false, false);
 358
 359         vcpu->arch.cr3 = cr3;
 360         kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
 361
 362         kvm_init_mmu(vcpu, false);
 363
 364         return 0;
 365 }
 366
 367 static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *nested_vmcb)
 368 {
 369         /* Load the nested guest state */
 370         svm->vmcb->save.es = nested_vmcb->save.es;
 371         svm->vmcb->save.cs = nested_vmcb->save.cs;
 372         svm->vmcb->save.ss = nested_vmcb->save.ss;
 373         svm->vmcb->save.ds = nested_vmcb->save.ds;
 374         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
 375         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
 376         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
 377         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
 378         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
 379         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
 380         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
 381         kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax);
 382         kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp);
 383         kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip);
 384
 385         /* In case we don't even reach vcpu_run, the fields are not updated */
 386         svm->vmcb->save.rax = nested_vmcb->save.rax;
 387         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
 388         svm->vmcb->save.rip = nested_vmcb->save.rip;
 389         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
 390         svm->vcpu.arch.dr6  = nested_vmcb->save.dr6;
 391         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
 392 }
 393
 394 static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
 395 {
 396         const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
 397
 398         if (nested_npt_enabled(svm))
 399                 nested_svm_init_mmu_context(&svm->vcpu);
 400
 401         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 402                 svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
 403
 404         svm->vmcb->control.int_ctl             =
 405                 (svm->nested.ctl.int_ctl & ~mask) |
 406                 (svm->nested.hsave->control.int_ctl & mask);
 407
 408         svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
 409         svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
 410         svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
 411         svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
 412         svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 413
 414         svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
 415         svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
 416
 417         /* Enter Guest-Mode */
 418         enter_guest_mode(&svm->vcpu);
 419
 420         /*
 421          * Merge guest and host intercepts - must be called  with vcpu in
 422          * guest-mode to take affect here
 423          */
 424         recalc_intercepts(svm);
 425
 426         vmcb_mark_all_dirty(svm->vmcb);
 427 }
 428
 429 int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
 430                           struct vmcb *nested_vmcb)
 431 {
 432         int ret;
 433
 434         svm->nested.vmcb = vmcb_gpa;
 435         load_nested_vmcb_control(svm, &nested_vmcb->control);
 436         nested_prepare_vmcb_save(svm, nested_vmcb);
 437         nested_prepare_vmcb_control(svm);
 438
 439         ret = nested_svm_load_cr3(&svm->vcpu, nested_vmcb->save.cr3,
 440                                   nested_npt_enabled(svm));
 441         if (ret)
 442                 return ret;
 443
 444         svm_set_gif(svm, true);
 445
 446         return 0;
 447 }
 448
 449 int nested_svm_vmrun(struct vcpu_svm *svm)
 450 {
 451         int ret;
 452         struct vmcb *nested_vmcb;
 453         struct vmcb *hsave = svm->nested.hsave;
 454         struct vmcb *vmcb = svm->vmcb;
 455         struct kvm_host_map map;
 456         u64 vmcb_gpa;
 457
 458         if (is_smm(&svm->vcpu)) {
 459                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 460                 return 1;
 461         }
 462
 463         vmcb_gpa = svm->vmcb->save.rax;
 464         ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map);
 465         if (ret == -EINVAL) {
 466                 kvm_inject_gp(&svm->vcpu, 0);
 467                 return 1;
 468         } else if (ret) {
 469                 return kvm_skip_emulated_instruction(&svm->vcpu);
 470         }
 471
 472         ret = kvm_skip_emulated_instruction(&svm->vcpu);
 473
 474         nested_vmcb = map.hva;
 475
 476         if (!nested_vmcb_checks(svm, nested_vmcb)) {
 477                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
 478                 nested_vmcb->control.exit_code_hi = 0;
 479                 nested_vmcb->control.exit_info_1  = 0;
 480                 nested_vmcb->control.exit_info_2  = 0;
 481                 goto out;
 482         }
 483
 484         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
 485                                nested_vmcb->save.rip,
 486                                nested_vmcb->control.int_ctl,
 487                                nested_vmcb->control.event_inj,
 488                                nested_vmcb->control.nested_ctl);
 489
 490         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
 491                                     nested_vmcb->control.intercept_cr >> 16,
 492                                     nested_vmcb->control.intercept_exceptions,
 493                                     nested_vmcb->control.intercept);
 494
 495         /* Clear internal status */
 496         kvm_clear_exception_queue(&svm->vcpu);
 497         kvm_clear_interrupt_queue(&svm->vcpu);
 498
 499         /*
 500          * Save the old vmcb, so we don't need to pick what we save, but can
 501          * restore everything when a VMEXIT occurs
 502          */
 503         hsave->save.es     = vmcb->save.es;
 504         hsave->save.cs     = vmcb->save.cs;
 505         hsave->save.ss     = vmcb->save.ss;
 506         hsave->save.ds     = vmcb->save.ds;
 507         hsave->save.gdtr   = vmcb->save.gdtr;
 508         hsave->save.idtr   = vmcb->save.idtr;
 509         hsave->save.efer   = svm->vcpu.arch.efer;
 510         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
 511         hsave->save.cr4    = svm->vcpu.arch.cr4;
 512         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
 513         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
 514         hsave->save.rsp    = vmcb->save.rsp;
 515         hsave->save.rax    = vmcb->save.rax;
 516         if (npt_enabled)
 517                 hsave->save.cr3    = vmcb->save.cr3;
 518         else
 519                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
 520
 521         copy_vmcb_control_area(&hsave->control, &vmcb->control);
 522
 523         svm->nested.nested_run_pending = 1;
 524
 525         if (enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb))
 526                 goto out_exit_err;
 527
 528         if (nested_svm_vmrun_msrpm(svm))
 529                 goto out;
 530
 531 out_exit_err:
 532         svm->nested.nested_run_pending = 0;
 533
 534         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 535         svm->vmcb->control.exit_code_hi = 0;
 536         svm->vmcb->control.exit_info_1  = 0;
 537         svm->vmcb->control.exit_info_2  = 0;
 538
 539         nested_svm_vmexit(svm);
 540
 541 out:
 542         kvm_vcpu_unmap(&svm->vcpu, &map, true);
 543
 544         return ret;
 545 }
 546
 547 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 548 {
 549         to_vmcb->save.fs = from_vmcb->save.fs;
 550         to_vmcb->save.gs = from_vmcb->save.gs;
 551         to_vmcb->save.tr = from_vmcb->save.tr;
 552         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 553         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 554         to_vmcb->save.star = from_vmcb->save.star;
 555         to_vmcb->save.lstar = from_vmcb->save.lstar;
 556         to_vmcb->save.cstar = from_vmcb->save.cstar;
 557         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 558         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 559         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 560         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 561 }
 562
 563 int nested_svm_vmexit(struct vcpu_svm *svm)
 564 {
 565         int rc;
 566         struct vmcb *nested_vmcb;
 567         struct vmcb *hsave = svm->nested.hsave;
 568         struct vmcb *vmcb = svm->vmcb;
 569         struct kvm_host_map map;
 570
 571         rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map);
 572         if (rc) {
 573                 if (rc == -EINVAL)
 574                         kvm_inject_gp(&svm->vcpu, 0);
 575                 return 1;
 576         }
 577
 578         nested_vmcb = map.hva;
 579
 580         /* Exit Guest-Mode */
 581         leave_guest_mode(&svm->vcpu);
 582         svm->nested.vmcb = 0;
 583         WARN_ON_ONCE(svm->nested.nested_run_pending);
 584
 585         /* in case we halted in L2 */
 586         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 587
 588         /* Give the current vmcb to the guest */
 589
 590         nested_vmcb->save.es     = vmcb->save.es;
 591         nested_vmcb->save.cs     = vmcb->save.cs;
 592         nested_vmcb->save.ss     = vmcb->save.ss;
 593         nested_vmcb->save.ds     = vmcb->save.ds;
 594         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
 595         nested_vmcb->save.idtr   = vmcb->save.idtr;
 596         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
 597         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
 598         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
 599         nested_vmcb->save.cr2    = vmcb->save.cr2;
 600         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 601         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
 602         nested_vmcb->save.rip    = kvm_rip_read(&svm->vcpu);
 603         nested_vmcb->save.rsp    = kvm_rsp_read(&svm->vcpu);
 604         nested_vmcb->save.rax    = kvm_rax_read(&svm->vcpu);
 605         nested_vmcb->save.dr7    = vmcb->save.dr7;
 606         nested_vmcb->save.dr6    = svm->vcpu.arch.dr6;
 607         nested_vmcb->save.cpl    = vmcb->save.cpl;
 608
 609         nested_vmcb->control.int_state         = vmcb->control.int_state;
 610         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
 611         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
 612         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
 613         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
 614
 615         if (nested_vmcb->control.exit_code != SVM_EXIT_ERR)
 616                 nested_vmcb_save_pending_event(svm, nested_vmcb);
 617
 618         if (svm->nrips_enabled)
 619                 nested_vmcb->control.next_rip  = vmcb->control.next_rip;
 620
 621         nested_vmcb->control.int_ctl           = svm->nested.ctl.int_ctl;
 622         nested_vmcb->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 623         nested_vmcb->control.event_inj         = svm->nested.ctl.event_inj;
 624         nested_vmcb->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 625
 626         nested_vmcb->control.pause_filter_count =
 627                 svm->vmcb->control.pause_filter_count;
 628         nested_vmcb->control.pause_filter_thresh =
 629                 svm->vmcb->control.pause_filter_thresh;
 630
 631         /* Restore the original control entries */
 632         copy_vmcb_control_area(&vmcb->control, &hsave->control);
 633
 634         /* On vmexit the  GIF is set to false */
 635         svm_set_gif(svm, false);
 636
 637         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 638                 svm->vcpu.arch.l1_tsc_offset;
 639
 640         svm->nested.ctl.nested_cr3 = 0;
 641
 642         /* Restore selected save entries */
 643         svm->vmcb->save.es = hsave->save.es;
 644         svm->vmcb->save.cs = hsave->save.cs;
 645         svm->vmcb->save.ss = hsave->save.ss;
 646         svm->vmcb->save.ds = hsave->save.ds;
 647         svm->vmcb->save.gdtr = hsave->save.gdtr;
 648         svm->vmcb->save.idtr = hsave->save.idtr;
 649         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
 650         svm_set_efer(&svm->vcpu, hsave->save.efer);
 651         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
 652         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
 653         kvm_rax_write(&svm->vcpu, hsave->save.rax);
 654         kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
 655         kvm_rip_write(&svm->vcpu, hsave->save.rip);
 656         svm->vmcb->save.dr7 = 0;
 657         svm->vmcb->save.cpl = 0;
 658         svm->vmcb->control.exit_int_info = 0;
 659
 660         vmcb_mark_all_dirty(svm->vmcb);
 661
 662         trace_kvm_nested_vmexit_inject(nested_vmcb->control.exit_code,
 663                                        nested_vmcb->control.exit_info_1,
 664                                        nested_vmcb->control.exit_info_2,
 665                                        nested_vmcb->control.exit_int_info,
 666                                        nested_vmcb->control.exit_int_info_err,
 667                                        KVM_ISA_SVM);
 668
 669         kvm_vcpu_unmap(&svm->vcpu, &map, true);
 670
 671         nested_svm_uninit_mmu_context(&svm->vcpu);
 672
 673         rc = nested_svm_load_cr3(&svm->vcpu, hsave->save.cr3, false);
 674         if (rc)
 675                 return 1;
 676
 677         if (npt_enabled)
 678                 svm->vmcb->save.cr3 = hsave->save.cr3;
 679
 680         /*
 681          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 682          * doesn't end up in L1.
 683          */
 684         svm->vcpu.arch.nmi_injected = false;
 685         kvm_clear_exception_queue(&svm->vcpu);
 686         kvm_clear_interrupt_queue(&svm->vcpu);
 687
 688         return 0;
 689 }
 690
 691 /*
 692  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
 693  */
 694 void svm_leave_nested(struct vcpu_svm *svm)
 695 {
 696         if (is_guest_mode(&svm->vcpu)) {
 697                 struct vmcb *hsave = svm->nested.hsave;
 698                 struct vmcb *vmcb = svm->vmcb;
 699
 700                 svm->nested.nested_run_pending = 0;
 701                 leave_guest_mode(&svm->vcpu);
 702                 copy_vmcb_control_area(&vmcb->control, &hsave->control);
 703                 nested_svm_uninit_mmu_context(&svm->vcpu);
 704         }
 705 }
 706
 707 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 708 {
 709         u32 offset, msr, value;
 710         int write, mask;
 711
 712         if (!(svm->nested.ctl.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 713                 return NESTED_EXIT_HOST;
 714
 715         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 716         offset = svm_msrpm_offset(msr);
 717         write  = svm->vmcb->control.exit_info_1 & 1;
 718         mask   = 1 << ((2 * (msr & 0xf)) + write);
 719
 720         if (offset == MSR_INVALID)
 721                 return NESTED_EXIT_DONE;
 722
 723         /* Offset is in 32 bit units but need in 8 bit units */
 724         offset *= 4;
 725
 726         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
 727                 return NESTED_EXIT_DONE;
 728
 729         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 730 }
 731
 732 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 733 {
 734         unsigned port, size, iopm_len;
 735         u16 val, mask;
 736         u8 start_bit;
 737         u64 gpa;
 738
 739         if (!(svm->nested.ctl.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
 740                 return NESTED_EXIT_HOST;
 741
 742         port = svm->vmcb->control.exit_info_1 >> 16;
 743         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 744                 SVM_IOIO_SIZE_SHIFT;
 745         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
 746         start_bit = port % 8;
 747         iopm_len = (start_bit + size > 8) ? 2 : 1;
 748         mask = (0xf >> (4 - size)) << start_bit;
 749         val = 0;
 750
 751         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 752                 return NESTED_EXIT_DONE;
 753
 754         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 755 }
 756
 757 static int nested_svm_intercept(struct vcpu_svm *svm)
 758 {
 759         u32 exit_code = svm->vmcb->control.exit_code;
 760         int vmexit = NESTED_EXIT_HOST;
 761
 762         switch (exit_code) {
 763         case SVM_EXIT_MSR:
 764                 vmexit = nested_svm_exit_handled_msr(svm);
 765                 break;
 766         case SVM_EXIT_IOIO:
 767                 vmexit = nested_svm_intercept_ioio(svm);
 768                 break;
 769         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 770                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
 771                 if (svm->nested.ctl.intercept_cr & bit)
 772                         vmexit = NESTED_EXIT_DONE;
 773                 break;
 774         }
 775         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 776                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
 777                 if (svm->nested.ctl.intercept_dr & bit)
 778                         vmexit = NESTED_EXIT_DONE;
 779                 break;
 780         }
 781         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 782                 /*
 783                  * Host-intercepted exceptions have been checked already in
 784                  * nested_svm_exit_special.  There is nothing to do here,
 785                  * the vmexit is injected by svm_check_nested_events.
 786                  */
 787                 vmexit = NESTED_EXIT_DONE;
 788                 break;
 789         }
 790         case SVM_EXIT_ERR: {
 791                 vmexit = NESTED_EXIT_DONE;
 792                 break;
 793         }
 794         default: {
 795                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
 796                 if (svm->nested.ctl.intercept & exit_bits)
 797                         vmexit = NESTED_EXIT_DONE;
 798         }
 799         }
 800
 801         return vmexit;
 802 }
 803
 804 int nested_svm_exit_handled(struct vcpu_svm *svm)
 805 {
 806         int vmexit;
 807
 808         vmexit = nested_svm_intercept(svm);
 809
 810         if (vmexit == NESTED_EXIT_DONE)
 811                 nested_svm_vmexit(svm);
 812
 813         return vmexit;
 814 }
 815
 816 int nested_svm_check_permissions(struct vcpu_svm *svm)
 817 {
 818         if (!(svm->vcpu.arch.efer & EFER_SVME) ||
 819             !is_paging(&svm->vcpu)) {
 820                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 821                 return 1;
 822         }
 823
 824         if (svm->vmcb->save.cpl) {
 825                 kvm_inject_gp(&svm->vcpu, 0);
 826                 return 1;
 827         }
 828
 829         return 0;
 830 }
 831
 832 static bool nested_exit_on_exception(struct vcpu_svm *svm)
 833 {
 834         unsigned int nr = svm->vcpu.arch.exception.nr;
 835
 836         return (svm->nested.ctl.intercept_exceptions & (1 << nr));
 837 }
 838
 839 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
 840 {
 841         unsigned int nr = svm->vcpu.arch.exception.nr;
 842
 843         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
 844         svm->vmcb->control.exit_code_hi = 0;
 845
 846         if (svm->vcpu.arch.exception.has_error_code)
 847                 svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
 848
 849         /*
 850          * EXITINFO2 is undefined for all exception intercepts other
 851          * than #PF.
 852          */
 853         if (nr == PF_VECTOR) {
 854                 if (svm->vcpu.arch.exception.nested_apf)
 855                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
 856                 else if (svm->vcpu.arch.exception.has_payload)
 857                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
 858                 else
 859                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
 860         } else if (nr == DB_VECTOR) {
 861                 /* See inject_pending_event.  */
 862                 kvm_deliver_exception_payload(&svm->vcpu);
 863                 if (svm->vcpu.arch.dr7 & DR7_GD) {
 864                         svm->vcpu.arch.dr7 &= ~DR7_GD;
 865                         kvm_update_dr7(&svm->vcpu);
 866                 }
 867         } else
 868                 WARN_ON(svm->vcpu.arch.exception.has_payload);
 869
 870         nested_svm_vmexit(svm);
 871 }
 872
 873 static void nested_svm_smi(struct vcpu_svm *svm)
 874 {
 875         svm->vmcb->control.exit_code = SVM_EXIT_SMI;
 876         svm->vmcb->control.exit_info_1 = 0;
 877         svm->vmcb->control.exit_info_2 = 0;
 878
 879         nested_svm_vmexit(svm);
 880 }
 881
 882 static void nested_svm_nmi(struct vcpu_svm *svm)
 883 {
 884         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
 885         svm->vmcb->control.exit_info_1 = 0;
 886         svm->vmcb->control.exit_info_2 = 0;
 887
 888         nested_svm_vmexit(svm);
 889 }
 890
 891 static void nested_svm_intr(struct vcpu_svm *svm)
 892 {
 893         trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
 894
 895         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
 896         svm->vmcb->control.exit_info_1 = 0;
 897         svm->vmcb->control.exit_info_2 = 0;
 898
 899         nested_svm_vmexit(svm);
 900 }
 901
 902 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
 903 {
 904         return (svm->nested.ctl.intercept & (1ULL << INTERCEPT_INIT));
 905 }
 906
 907 static void nested_svm_init(struct vcpu_svm *svm)
 908 {
 909         svm->vmcb->control.exit_code   = SVM_EXIT_INIT;
 910         svm->vmcb->control.exit_info_1 = 0;
 911         svm->vmcb->control.exit_info_2 = 0;
 912
 913         nested_svm_vmexit(svm);
 914 }
 915
 916
 917 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
 918 {
 919         struct vcpu_svm *svm = to_svm(vcpu);
 920         bool block_nested_events =
 921                 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
 922         struct kvm_lapic *apic = vcpu->arch.apic;
 923
 924         if (lapic_in_kernel(vcpu) &&
 925             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
 926                 if (block_nested_events)
 927                         return -EBUSY;
 928                 if (!nested_exit_on_init(svm))
 929                         return 0;
 930                 nested_svm_init(svm);
 931                 return 0;
 932         }
 933
 934         if (vcpu->arch.exception.pending) {
 935                 if (block_nested_events)
 936                         return -EBUSY;
 937                 if (!nested_exit_on_exception(svm))
 938                         return 0;
 939                 nested_svm_inject_exception_vmexit(svm);
 940                 return 0;
 941         }
 942
 943         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
 944                 if (block_nested_events)
 945                         return -EBUSY;
 946                 if (!nested_exit_on_smi(svm))
 947                         return 0;
 948                 nested_svm_smi(svm);
 949                 return 0;
 950         }
 951
 952         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
 953                 if (block_nested_events)
 954                         return -EBUSY;
 955                 if (!nested_exit_on_nmi(svm))
 956                         return 0;
 957                 nested_svm_nmi(svm);
 958                 return 0;
 959         }
 960
 961         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
 962                 if (block_nested_events)
 963                         return -EBUSY;
 964                 if (!nested_exit_on_intr(svm))
 965                         return 0;
 966                 nested_svm_intr(svm);
 967                 return 0;
 968         }
 969
 970         return 0;
 971 }
 972
 973 int nested_svm_exit_special(struct vcpu_svm *svm)
 974 {
 975         u32 exit_code = svm->vmcb->control.exit_code;
 976
 977         switch (exit_code) {
 978         case SVM_EXIT_INTR:
 979         case SVM_EXIT_NMI:
 980         case SVM_EXIT_NPF:
 981                 return NESTED_EXIT_HOST;
 982         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 983                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
 984
 985                 if (get_host_vmcb(svm)->control.intercept_exceptions & excp_bits)
 986                         return NESTED_EXIT_HOST;
 987                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
 988                          svm->vcpu.arch.apf.host_apf_flags)
 989                         /* Trap async PF even if not shadowing */
 990                         return NESTED_EXIT_HOST;
 991                 break;
 992         }
 993         default:
 994                 break;
 995         }
 996
 997         return NESTED_EXIT_CONTINUE;
 998 }
 999
1000 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1001                                 struct kvm_nested_state __user *user_kvm_nested_state,
1002                                 u32 user_data_size)
1003 {
1004         struct vcpu_svm *svm;
1005         struct kvm_nested_state kvm_state = {
1006                 .flags = 0,
1007                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1008                 .size = sizeof(kvm_state),
1009         };
1010         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1011                 &user_kvm_nested_state->data.svm[0];
1012
1013         if (!vcpu)
1014                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1015
1016         svm = to_svm(vcpu);
1017
1018         if (user_data_size < kvm_state.size)
1019                 goto out;
1020
1021         /* First fill in the header and copy it out.  */
1022         if (is_guest_mode(vcpu)) {
1023                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb;
1024                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1025                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1026
1027                 if (svm->nested.nested_run_pending)
1028                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1029         }
1030
1031         if (gif_set(svm))
1032                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1033
1034         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1035                 return -EFAULT;
1036
1037         if (!is_guest_mode(vcpu))
1038                 goto out;
1039
1040         /*
1041          * Copy over the full size of the VMCB rather than just the size
1042          * of the structs.
1043          */
1044         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1045                 return -EFAULT;
1046         if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1047                          sizeof(user_vmcb->control)))
1048                 return -EFAULT;
1049         if (copy_to_user(&user_vmcb->save, &svm->nested.hsave->save,
1050                          sizeof(user_vmcb->save)))
1051                 return -EFAULT;
1052
1053 out:
1054         return kvm_state.size;
1055 }
1056
1057 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1058                                 struct kvm_nested_state __user *user_kvm_nested_state,
1059                                 struct kvm_nested_state *kvm_state)
1060 {
1061         struct vcpu_svm *svm = to_svm(vcpu);
1062         struct vmcb *hsave = svm->nested.hsave;
1063         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1064                 &user_kvm_nested_state->data.svm[0];
1065         struct vmcb_control_area *ctl;
1066         struct vmcb_save_area *save;
1067         int ret;
1068         u32 cr0;
1069
1070         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1071                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1072
1073         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1074                 return -EINVAL;
1075
1076         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1077                                  KVM_STATE_NESTED_RUN_PENDING |
1078                                  KVM_STATE_NESTED_GIF_SET))
1079                 return -EINVAL;
1080
1081         /*
1082          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1083          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1084          */
1085         if (!(vcpu->arch.efer & EFER_SVME)) {
1086                 /* GIF=1 and no guest mode are required if SVME=0.  */
1087                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1088                         return -EINVAL;
1089         }
1090
1091         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1092         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1093                 return -EINVAL;
1094
1095         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1096                 svm_leave_nested(svm);
1097                 goto out_set_gif;
1098         }
1099
1100         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1101                 return -EINVAL;
1102         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1103                 return -EINVAL;
1104
1105         ret  = -ENOMEM;
1106         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
1107         save = kzalloc(sizeof(*save), GFP_KERNEL);
1108         if (!ctl || !save)
1109                 goto out_free;
1110
1111         ret = -EFAULT;
1112         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1113                 goto out_free;
1114         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1115                 goto out_free;
1116
1117         ret = -EINVAL;
1118         if (!nested_vmcb_check_controls(ctl))
1119                 goto out_free;
1120
1121         /*
1122          * Processor state contains L2 state.  Check that it is
1123          * valid for guest mode (see nested_vmcb_checks).
1124          */
1125         cr0 = kvm_read_cr0(vcpu);
1126         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1127                 goto out_free;
1128
1129         /*
1130          * Validate host state saved from before VMRUN (see
1131          * nested_svm_check_permissions).
1132          * TODO: validate reserved bits for all saved state.
1133          */
1134         if (!(save->cr0 & X86_CR0_PG))
1135                 goto out_free;
1136
1137         /*
1138          * All checks done, we can enter guest mode.  L1 control fields
1139          * come from the nested save state.  Guest state is already
1140          * in the registers, the save area of the nested state instead
1141          * contains saved L1 state.
1142          */
1143         copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
1144         hsave->save = *save;
1145
1146         svm->nested.vmcb = kvm_state->hdr.svm.vmcb_pa;
1147         load_nested_vmcb_control(svm, ctl);
1148         nested_prepare_vmcb_control(svm);
1149
1150         if (!nested_svm_vmrun_msrpm(svm))
1151                 return -EINVAL;
1152
1153 out_set_gif:
1154         svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1155
1156         ret = 0;
1157 out_free:
1158         kfree(save);
1159         kfree(ctl);
1160
1161         return ret;
1162 }
1163
1164 struct kvm_x86_nested_ops svm_nested_ops = {
1165         .check_events = svm_check_nested_events,
1166         .get_state = svm_get_nested_state,
1167         .set_state = svm_set_nested_state,
1168 };