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         unsigned int i;
 102
 103         vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 104
 105         if (!is_guest_mode(&svm->vcpu))
 106                 return;
 107
 108         c = &svm->vmcb->control;
 109         h = &svm->nested.hsave->control;
 110         g = &svm->nested.ctl;
 111
 112         for (i = 0; i < MAX_INTERCEPT; i++)
 113                 c->intercepts[i] = h->intercepts[i];
 114
 115         if (g->int_ctl & V_INTR_MASKING_MASK) {
 116                 /* We only want the cr8 intercept bits of L1 */
 117                 vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
 118                 vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
 119
 120                 /*
 121                  * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 122                  * affect any interrupt we may want to inject; therefore,
 123                  * interrupt window vmexits are irrelevant to L0.
 124                  */
 125                 vmcb_clr_intercept(c, INTERCEPT_VINTR);
 126         }
 127
 128         /* We don't want to see VMMCALLs from a nested guest */
 129         vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
 130
 131         for (i = 0; i < MAX_INTERCEPT; i++)
 132                 c->intercepts[i] |= g->intercepts[i];
 133 }
 134
 135 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
 136                                    struct vmcb_control_area *from)
 137 {
 138         unsigned int i;
 139
 140         for (i = 0; i < MAX_INTERCEPT; i++)
 141                 dst->intercepts[i] = from->intercepts[i];
 142
 143         dst->iopm_base_pa         = from->iopm_base_pa;
 144         dst->msrpm_base_pa        = from->msrpm_base_pa;
 145         dst->tsc_offset           = from->tsc_offset;
 146         /* asid not copied, it is handled manually for svm->vmcb.  */
 147         dst->tlb_ctl              = from->tlb_ctl;
 148         dst->int_ctl              = from->int_ctl;
 149         dst->int_vector           = from->int_vector;
 150         dst->int_state            = from->int_state;
 151         dst->exit_code            = from->exit_code;
 152         dst->exit_code_hi         = from->exit_code_hi;
 153         dst->exit_info_1          = from->exit_info_1;
 154         dst->exit_info_2          = from->exit_info_2;
 155         dst->exit_int_info        = from->exit_int_info;
 156         dst->exit_int_info_err    = from->exit_int_info_err;
 157         dst->nested_ctl           = from->nested_ctl;
 158         dst->event_inj            = from->event_inj;
 159         dst->event_inj_err        = from->event_inj_err;
 160         dst->nested_cr3           = from->nested_cr3;
 161         dst->virt_ext              = from->virt_ext;
 162         dst->pause_filter_count   = from->pause_filter_count;
 163         dst->pause_filter_thresh  = from->pause_filter_thresh;
 164 }
 165
 166 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 167 {
 168         /*
 169          * This function merges the msr permission bitmaps of kvm and the
 170          * nested vmcb. It is optimized in that it only merges the parts where
 171          * the kvm msr permission bitmap may contain zero bits
 172          */
 173         int i;
 174
 175         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 176                 return true;
 177
 178         for (i = 0; i < MSRPM_OFFSETS; i++) {
 179                 u32 value, p;
 180                 u64 offset;
 181
 182                 if (msrpm_offsets[i] == 0xffffffff)
 183                         break;
 184
 185                 p      = msrpm_offsets[i];
 186                 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 187
 188                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 189                         return false;
 190
 191                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 192         }
 193
 194         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 195
 196         return true;
 197 }
 198
 199 static bool nested_vmcb_check_controls(struct vmcb_control_area *control)
 200 {
 201         if ((vmcb_is_intercept(control, INTERCEPT_VMRUN)) == 0)
 202                 return false;
 203
 204         if (control->asid == 0)
 205                 return false;
 206
 207         if ((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
 208             !npt_enabled)
 209                 return false;
 210
 211         return true;
 212 }
 213
 214 static bool nested_vmcb_checks(struct vcpu_svm *svm, struct vmcb *vmcb12)
 215 {
 216         bool vmcb12_lma;
 217
 218         if ((vmcb12->save.efer & EFER_SVME) == 0)
 219                 return false;
 220
 221         if (((vmcb12->save.cr0 & X86_CR0_CD) == 0) && (vmcb12->save.cr0 & X86_CR0_NW))
 222                 return false;
 223
 224         if (!kvm_dr6_valid(vmcb12->save.dr6) || !kvm_dr7_valid(vmcb12->save.dr7))
 225                 return false;
 226
 227         vmcb12_lma = (vmcb12->save.efer & EFER_LME) && (vmcb12->save.cr0 & X86_CR0_PG);
 228
 229         if (!vmcb12_lma) {
 230                 if (vmcb12->save.cr4 & X86_CR4_PAE) {
 231                         if (vmcb12->save.cr3 & MSR_CR3_LEGACY_PAE_RESERVED_MASK)
 232                                 return false;
 233                 } else {
 234                         if (vmcb12->save.cr3 & MSR_CR3_LEGACY_RESERVED_MASK)
 235                                 return false;
 236                 }
 237         } else {
 238                 if (!(vmcb12->save.cr4 & X86_CR4_PAE) ||
 239                     !(vmcb12->save.cr0 & X86_CR0_PE) ||
 240                     (vmcb12->save.cr3 & MSR_CR3_LONG_RESERVED_MASK))
 241                         return false;
 242         }
 243         if (kvm_valid_cr4(&svm->vcpu, vmcb12->save.cr4))
 244                 return false;
 245
 246         return nested_vmcb_check_controls(&vmcb12->control);
 247 }
 248
 249 static void load_nested_vmcb_control(struct vcpu_svm *svm,
 250                                      struct vmcb_control_area *control)
 251 {
 252         copy_vmcb_control_area(&svm->nested.ctl, control);
 253
 254         /* Copy it here because nested_svm_check_controls will check it.  */
 255         svm->nested.ctl.asid           = control->asid;
 256         svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
 257         svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
 258 }
 259
 260 /*
 261  * Synchronize fields that are written by the processor, so that
 262  * they can be copied back into the nested_vmcb.
 263  */
 264 void sync_nested_vmcb_control(struct vcpu_svm *svm)
 265 {
 266         u32 mask;
 267         svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 268         svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 269
 270         /* Only a few fields of int_ctl are written by the processor.  */
 271         mask = V_IRQ_MASK | V_TPR_MASK;
 272         if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
 273             svm_is_intercept(svm, INTERCEPT_VINTR)) {
 274                 /*
 275                  * In order to request an interrupt window, L0 is usurping
 276                  * svm->vmcb->control.int_ctl and possibly setting V_IRQ
 277                  * even if it was clear in L1's VMCB.  Restoring it would be
 278                  * wrong.  However, in this case V_IRQ will remain true until
 279                  * interrupt_window_interception calls svm_clear_vintr and
 280                  * restores int_ctl.  We can just leave it aside.
 281                  */
 282                 mask &= ~V_IRQ_MASK;
 283         }
 284         svm->nested.ctl.int_ctl        &= ~mask;
 285         svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 286 }
 287
 288 /*
 289  * Transfer any event that L0 or L1 wanted to inject into L2 to
 290  * EXIT_INT_INFO.
 291  */
 292 static void nested_vmcb_save_pending_event(struct vcpu_svm *svm,
 293                                            struct vmcb *vmcb12)
 294 {
 295         struct kvm_vcpu *vcpu = &svm->vcpu;
 296         u32 exit_int_info = 0;
 297         unsigned int nr;
 298
 299         if (vcpu->arch.exception.injected) {
 300                 nr = vcpu->arch.exception.nr;
 301                 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 302
 303                 if (vcpu->arch.exception.has_error_code) {
 304                         exit_int_info |= SVM_EVTINJ_VALID_ERR;
 305                         vmcb12->control.exit_int_info_err =
 306                                 vcpu->arch.exception.error_code;
 307                 }
 308
 309         } else if (vcpu->arch.nmi_injected) {
 310                 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 311
 312         } else if (vcpu->arch.interrupt.injected) {
 313                 nr = vcpu->arch.interrupt.nr;
 314                 exit_int_info = nr | SVM_EVTINJ_VALID;
 315
 316                 if (vcpu->arch.interrupt.soft)
 317                         exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 318                 else
 319                         exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 320         }
 321
 322         vmcb12->control.exit_int_info = exit_int_info;
 323 }
 324
 325 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
 326 {
 327         return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
 328 }
 329
 330 /*
 331  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 332  * if we are emulating VM-Entry into a guest with NPT enabled.
 333  */
 334 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 335                                bool nested_npt)
 336 {
 337         if (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))
 338                 return -EINVAL;
 339
 340         if (!nested_npt && is_pae_paging(vcpu) &&
 341             (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
 342                 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 343                         return -EINVAL;
 344         }
 345
 346         /*
 347          * TODO: optimize unconditional TLB flush/MMU sync here and in
 348          * kvm_init_shadow_npt_mmu().
 349          */
 350         if (!nested_npt)
 351                 kvm_mmu_new_pgd(vcpu, cr3, false, false);
 352
 353         vcpu->arch.cr3 = cr3;
 354         kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
 355
 356         kvm_init_mmu(vcpu, false);
 357
 358         return 0;
 359 }
 360
 361 static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 362 {
 363         /* Load the nested guest state */
 364         svm->vmcb->save.es = vmcb12->save.es;
 365         svm->vmcb->save.cs = vmcb12->save.cs;
 366         svm->vmcb->save.ss = vmcb12->save.ss;
 367         svm->vmcb->save.ds = vmcb12->save.ds;
 368         svm->vmcb->save.gdtr = vmcb12->save.gdtr;
 369         svm->vmcb->save.idtr = vmcb12->save.idtr;
 370         kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags);
 371         svm_set_efer(&svm->vcpu, vmcb12->save.efer);
 372         svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
 373         svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
 374         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 375         kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
 376         kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
 377         kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
 378
 379         /* In case we don't even reach vcpu_run, the fields are not updated */
 380         svm->vmcb->save.rax = vmcb12->save.rax;
 381         svm->vmcb->save.rsp = vmcb12->save.rsp;
 382         svm->vmcb->save.rip = vmcb12->save.rip;
 383         svm->vmcb->save.dr7 = vmcb12->save.dr7;
 384         svm->vcpu.arch.dr6  = vmcb12->save.dr6;
 385         svm->vmcb->save.cpl = vmcb12->save.cpl;
 386 }
 387
 388 static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
 389 {
 390         const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
 391
 392         if (nested_npt_enabled(svm))
 393                 nested_svm_init_mmu_context(&svm->vcpu);
 394
 395         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 396                 svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
 397
 398         svm->vmcb->control.int_ctl             =
 399                 (svm->nested.ctl.int_ctl & ~mask) |
 400                 (svm->nested.hsave->control.int_ctl & mask);
 401
 402         svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
 403         svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
 404         svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
 405         svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
 406         svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 407
 408         svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
 409         svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
 410
 411         /* Enter Guest-Mode */
 412         enter_guest_mode(&svm->vcpu);
 413
 414         /*
 415          * Merge guest and host intercepts - must be called  with vcpu in
 416          * guest-mode to take affect here
 417          */
 418         recalc_intercepts(svm);
 419
 420         vmcb_mark_all_dirty(svm->vmcb);
 421 }
 422
 423 int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb12_gpa,
 424                          struct vmcb *vmcb12)
 425 {
 426         int ret;
 427
 428         svm->nested.vmcb12_gpa = vmcb12_gpa;
 429         load_nested_vmcb_control(svm, &vmcb12->control);
 430         nested_prepare_vmcb_save(svm, vmcb12);
 431         nested_prepare_vmcb_control(svm);
 432
 433         ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
 434                                   nested_npt_enabled(svm));
 435         if (ret)
 436                 return ret;
 437
 438         svm_set_gif(svm, true);
 439
 440         return 0;
 441 }
 442
 443 int nested_svm_vmrun(struct vcpu_svm *svm)
 444 {
 445         int ret;
 446         struct vmcb *vmcb12;
 447         struct vmcb *hsave = svm->nested.hsave;
 448         struct vmcb *vmcb = svm->vmcb;
 449         struct kvm_host_map map;
 450         u64 vmcb12_gpa;
 451
 452         if (is_smm(&svm->vcpu)) {
 453                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 454                 return 1;
 455         }
 456
 457         vmcb12_gpa = svm->vmcb->save.rax;
 458         ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 459         if (ret == -EINVAL) {
 460                 kvm_inject_gp(&svm->vcpu, 0);
 461                 return 1;
 462         } else if (ret) {
 463                 return kvm_skip_emulated_instruction(&svm->vcpu);
 464         }
 465
 466         ret = kvm_skip_emulated_instruction(&svm->vcpu);
 467
 468         vmcb12 = map.hva;
 469
 470         if (!nested_vmcb_checks(svm, vmcb12)) {
 471                 vmcb12->control.exit_code    = SVM_EXIT_ERR;
 472                 vmcb12->control.exit_code_hi = 0;
 473                 vmcb12->control.exit_info_1  = 0;
 474                 vmcb12->control.exit_info_2  = 0;
 475                 goto out;
 476         }
 477
 478         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
 479                                vmcb12->save.rip,
 480                                vmcb12->control.int_ctl,
 481                                vmcb12->control.event_inj,
 482                                vmcb12->control.nested_ctl);
 483
 484         trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 485                                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 486                                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 487                                     vmcb12->control.intercepts[INTERCEPT_WORD3],
 488                                     vmcb12->control.intercepts[INTERCEPT_WORD4],
 489                                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
 490
 491         /* Clear internal status */
 492         kvm_clear_exception_queue(&svm->vcpu);
 493         kvm_clear_interrupt_queue(&svm->vcpu);
 494
 495         /*
 496          * Save the old vmcb, so we don't need to pick what we save, but can
 497          * restore everything when a VMEXIT occurs
 498          */
 499         hsave->save.es     = vmcb->save.es;
 500         hsave->save.cs     = vmcb->save.cs;
 501         hsave->save.ss     = vmcb->save.ss;
 502         hsave->save.ds     = vmcb->save.ds;
 503         hsave->save.gdtr   = vmcb->save.gdtr;
 504         hsave->save.idtr   = vmcb->save.idtr;
 505         hsave->save.efer   = svm->vcpu.arch.efer;
 506         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
 507         hsave->save.cr4    = svm->vcpu.arch.cr4;
 508         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
 509         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
 510         hsave->save.rsp    = vmcb->save.rsp;
 511         hsave->save.rax    = vmcb->save.rax;
 512         if (npt_enabled)
 513                 hsave->save.cr3    = vmcb->save.cr3;
 514         else
 515                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
 516
 517         copy_vmcb_control_area(&hsave->control, &vmcb->control);
 518
 519         svm->nested.nested_run_pending = 1;
 520
 521         if (enter_svm_guest_mode(svm, vmcb12_gpa, vmcb12))
 522                 goto out_exit_err;
 523
 524         if (nested_svm_vmrun_msrpm(svm))
 525                 goto out;
 526
 527 out_exit_err:
 528         svm->nested.nested_run_pending = 0;
 529
 530         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 531         svm->vmcb->control.exit_code_hi = 0;
 532         svm->vmcb->control.exit_info_1  = 0;
 533         svm->vmcb->control.exit_info_2  = 0;
 534
 535         nested_svm_vmexit(svm);
 536
 537 out:
 538         kvm_vcpu_unmap(&svm->vcpu, &map, true);
 539
 540         return ret;
 541 }
 542
 543 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 544 {
 545         to_vmcb->save.fs = from_vmcb->save.fs;
 546         to_vmcb->save.gs = from_vmcb->save.gs;
 547         to_vmcb->save.tr = from_vmcb->save.tr;
 548         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 549         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 550         to_vmcb->save.star = from_vmcb->save.star;
 551         to_vmcb->save.lstar = from_vmcb->save.lstar;
 552         to_vmcb->save.cstar = from_vmcb->save.cstar;
 553         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 554         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 555         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 556         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 557 }
 558
 559 int nested_svm_vmexit(struct vcpu_svm *svm)
 560 {
 561         int rc;
 562         struct vmcb *vmcb12;
 563         struct vmcb *hsave = svm->nested.hsave;
 564         struct vmcb *vmcb = svm->vmcb;
 565         struct kvm_host_map map;
 566
 567         rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 568         if (rc) {
 569                 if (rc == -EINVAL)
 570                         kvm_inject_gp(&svm->vcpu, 0);
 571                 return 1;
 572         }
 573
 574         vmcb12 = map.hva;
 575
 576         /* Exit Guest-Mode */
 577         leave_guest_mode(&svm->vcpu);
 578         svm->nested.vmcb12_gpa = 0;
 579         WARN_ON_ONCE(svm->nested.nested_run_pending);
 580
 581         /* in case we halted in L2 */
 582         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 583
 584         /* Give the current vmcb to the guest */
 585
 586         vmcb12->save.es     = vmcb->save.es;
 587         vmcb12->save.cs     = vmcb->save.cs;
 588         vmcb12->save.ss     = vmcb->save.ss;
 589         vmcb12->save.ds     = vmcb->save.ds;
 590         vmcb12->save.gdtr   = vmcb->save.gdtr;
 591         vmcb12->save.idtr   = vmcb->save.idtr;
 592         vmcb12->save.efer   = svm->vcpu.arch.efer;
 593         vmcb12->save.cr0    = kvm_read_cr0(&svm->vcpu);
 594         vmcb12->save.cr3    = kvm_read_cr3(&svm->vcpu);
 595         vmcb12->save.cr2    = vmcb->save.cr2;
 596         vmcb12->save.cr4    = svm->vcpu.arch.cr4;
 597         vmcb12->save.rflags = kvm_get_rflags(&svm->vcpu);
 598         vmcb12->save.rip    = kvm_rip_read(&svm->vcpu);
 599         vmcb12->save.rsp    = kvm_rsp_read(&svm->vcpu);
 600         vmcb12->save.rax    = kvm_rax_read(&svm->vcpu);
 601         vmcb12->save.dr7    = vmcb->save.dr7;
 602         vmcb12->save.dr6    = svm->vcpu.arch.dr6;
 603         vmcb12->save.cpl    = vmcb->save.cpl;
 604
 605         vmcb12->control.int_state         = vmcb->control.int_state;
 606         vmcb12->control.exit_code         = vmcb->control.exit_code;
 607         vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
 608         vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
 609         vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
 610
 611         if (vmcb12->control.exit_code != SVM_EXIT_ERR)
 612                 nested_vmcb_save_pending_event(svm, vmcb12);
 613
 614         if (svm->nrips_enabled)
 615                 vmcb12->control.next_rip  = vmcb->control.next_rip;
 616
 617         vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
 618         vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 619         vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
 620         vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 621
 622         vmcb12->control.pause_filter_count =
 623                 svm->vmcb->control.pause_filter_count;
 624         vmcb12->control.pause_filter_thresh =
 625                 svm->vmcb->control.pause_filter_thresh;
 626
 627         /* Restore the original control entries */
 628         copy_vmcb_control_area(&vmcb->control, &hsave->control);
 629
 630         /* On vmexit the  GIF is set to false */
 631         svm_set_gif(svm, false);
 632
 633         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 634                 svm->vcpu.arch.l1_tsc_offset;
 635
 636         svm->nested.ctl.nested_cr3 = 0;
 637
 638         /* Restore selected save entries */
 639         svm->vmcb->save.es = hsave->save.es;
 640         svm->vmcb->save.cs = hsave->save.cs;
 641         svm->vmcb->save.ss = hsave->save.ss;
 642         svm->vmcb->save.ds = hsave->save.ds;
 643         svm->vmcb->save.gdtr = hsave->save.gdtr;
 644         svm->vmcb->save.idtr = hsave->save.idtr;
 645         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
 646         svm_set_efer(&svm->vcpu, hsave->save.efer);
 647         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
 648         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
 649         kvm_rax_write(&svm->vcpu, hsave->save.rax);
 650         kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
 651         kvm_rip_write(&svm->vcpu, hsave->save.rip);
 652         svm->vmcb->save.dr7 = 0;
 653         svm->vmcb->save.cpl = 0;
 654         svm->vmcb->control.exit_int_info = 0;
 655
 656         vmcb_mark_all_dirty(svm->vmcb);
 657
 658         trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
 659                                        vmcb12->control.exit_info_1,
 660                                        vmcb12->control.exit_info_2,
 661                                        vmcb12->control.exit_int_info,
 662                                        vmcb12->control.exit_int_info_err,
 663                                        KVM_ISA_SVM);
 664
 665         kvm_vcpu_unmap(&svm->vcpu, &map, true);
 666
 667         nested_svm_uninit_mmu_context(&svm->vcpu);
 668
 669         rc = nested_svm_load_cr3(&svm->vcpu, hsave->save.cr3, false);
 670         if (rc)
 671                 return 1;
 672
 673         if (npt_enabled)
 674                 svm->vmcb->save.cr3 = hsave->save.cr3;
 675
 676         /*
 677          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 678          * doesn't end up in L1.
 679          */
 680         svm->vcpu.arch.nmi_injected = false;
 681         kvm_clear_exception_queue(&svm->vcpu);
 682         kvm_clear_interrupt_queue(&svm->vcpu);
 683
 684         return 0;
 685 }
 686
 687 /*
 688  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
 689  */
 690 void svm_leave_nested(struct vcpu_svm *svm)
 691 {
 692         if (is_guest_mode(&svm->vcpu)) {
 693                 struct vmcb *hsave = svm->nested.hsave;
 694                 struct vmcb *vmcb = svm->vmcb;
 695
 696                 svm->nested.nested_run_pending = 0;
 697                 leave_guest_mode(&svm->vcpu);
 698                 copy_vmcb_control_area(&vmcb->control, &hsave->control);
 699                 nested_svm_uninit_mmu_context(&svm->vcpu);
 700         }
 701 }
 702
 703 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 704 {
 705         u32 offset, msr, value;
 706         int write, mask;
 707
 708         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 709                 return NESTED_EXIT_HOST;
 710
 711         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 712         offset = svm_msrpm_offset(msr);
 713         write  = svm->vmcb->control.exit_info_1 & 1;
 714         mask   = 1 << ((2 * (msr & 0xf)) + write);
 715
 716         if (offset == MSR_INVALID)
 717                 return NESTED_EXIT_DONE;
 718
 719         /* Offset is in 32 bit units but need in 8 bit units */
 720         offset *= 4;
 721
 722         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
 723                 return NESTED_EXIT_DONE;
 724
 725         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 726 }
 727
 728 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 729 {
 730         unsigned port, size, iopm_len;
 731         u16 val, mask;
 732         u8 start_bit;
 733         u64 gpa;
 734
 735         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
 736                 return NESTED_EXIT_HOST;
 737
 738         port = svm->vmcb->control.exit_info_1 >> 16;
 739         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 740                 SVM_IOIO_SIZE_SHIFT;
 741         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
 742         start_bit = port % 8;
 743         iopm_len = (start_bit + size > 8) ? 2 : 1;
 744         mask = (0xf >> (4 - size)) << start_bit;
 745         val = 0;
 746
 747         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 748                 return NESTED_EXIT_DONE;
 749
 750         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 751 }
 752
 753 static int nested_svm_intercept(struct vcpu_svm *svm)
 754 {
 755         u32 exit_code = svm->vmcb->control.exit_code;
 756         int vmexit = NESTED_EXIT_HOST;
 757
 758         switch (exit_code) {
 759         case SVM_EXIT_MSR:
 760                 vmexit = nested_svm_exit_handled_msr(svm);
 761                 break;
 762         case SVM_EXIT_IOIO:
 763                 vmexit = nested_svm_intercept_ioio(svm);
 764                 break;
 765         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 766                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 767                         vmexit = NESTED_EXIT_DONE;
 768                 break;
 769         }
 770         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 771                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 772                         vmexit = NESTED_EXIT_DONE;
 773                 break;
 774         }
 775         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 776                 /*
 777                  * Host-intercepted exceptions have been checked already in
 778                  * nested_svm_exit_special.  There is nothing to do here,
 779                  * the vmexit is injected by svm_check_nested_events.
 780                  */
 781                 vmexit = NESTED_EXIT_DONE;
 782                 break;
 783         }
 784         case SVM_EXIT_ERR: {
 785                 vmexit = NESTED_EXIT_DONE;
 786                 break;
 787         }
 788         default: {
 789                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 790                         vmexit = NESTED_EXIT_DONE;
 791         }
 792         }
 793
 794         return vmexit;
 795 }
 796
 797 int nested_svm_exit_handled(struct vcpu_svm *svm)
 798 {
 799         int vmexit;
 800
 801         vmexit = nested_svm_intercept(svm);
 802
 803         if (vmexit == NESTED_EXIT_DONE)
 804                 nested_svm_vmexit(svm);
 805
 806         return vmexit;
 807 }
 808
 809 int nested_svm_check_permissions(struct vcpu_svm *svm)
 810 {
 811         if (!(svm->vcpu.arch.efer & EFER_SVME) ||
 812             !is_paging(&svm->vcpu)) {
 813                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 814                 return 1;
 815         }
 816
 817         if (svm->vmcb->save.cpl) {
 818                 kvm_inject_gp(&svm->vcpu, 0);
 819                 return 1;
 820         }
 821
 822         return 0;
 823 }
 824
 825 static bool nested_exit_on_exception(struct vcpu_svm *svm)
 826 {
 827         unsigned int nr = svm->vcpu.arch.exception.nr;
 828
 829         return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
 830 }
 831
 832 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
 833 {
 834         unsigned int nr = svm->vcpu.arch.exception.nr;
 835
 836         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
 837         svm->vmcb->control.exit_code_hi = 0;
 838
 839         if (svm->vcpu.arch.exception.has_error_code)
 840                 svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
 841
 842         /*
 843          * EXITINFO2 is undefined for all exception intercepts other
 844          * than #PF.
 845          */
 846         if (nr == PF_VECTOR) {
 847                 if (svm->vcpu.arch.exception.nested_apf)
 848                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
 849                 else if (svm->vcpu.arch.exception.has_payload)
 850                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
 851                 else
 852                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
 853         } else if (nr == DB_VECTOR) {
 854                 /* See inject_pending_event.  */
 855                 kvm_deliver_exception_payload(&svm->vcpu);
 856                 if (svm->vcpu.arch.dr7 & DR7_GD) {
 857                         svm->vcpu.arch.dr7 &= ~DR7_GD;
 858                         kvm_update_dr7(&svm->vcpu);
 859                 }
 860         } else
 861                 WARN_ON(svm->vcpu.arch.exception.has_payload);
 862
 863         nested_svm_vmexit(svm);
 864 }
 865
 866 static void nested_svm_smi(struct vcpu_svm *svm)
 867 {
 868         svm->vmcb->control.exit_code = SVM_EXIT_SMI;
 869         svm->vmcb->control.exit_info_1 = 0;
 870         svm->vmcb->control.exit_info_2 = 0;
 871
 872         nested_svm_vmexit(svm);
 873 }
 874
 875 static void nested_svm_nmi(struct vcpu_svm *svm)
 876 {
 877         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
 878         svm->vmcb->control.exit_info_1 = 0;
 879         svm->vmcb->control.exit_info_2 = 0;
 880
 881         nested_svm_vmexit(svm);
 882 }
 883
 884 static void nested_svm_intr(struct vcpu_svm *svm)
 885 {
 886         trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
 887
 888         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
 889         svm->vmcb->control.exit_info_1 = 0;
 890         svm->vmcb->control.exit_info_2 = 0;
 891
 892         nested_svm_vmexit(svm);
 893 }
 894
 895 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
 896 {
 897         return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
 898 }
 899
 900 static void nested_svm_init(struct vcpu_svm *svm)
 901 {
 902         svm->vmcb->control.exit_code   = SVM_EXIT_INIT;
 903         svm->vmcb->control.exit_info_1 = 0;
 904         svm->vmcb->control.exit_info_2 = 0;
 905
 906         nested_svm_vmexit(svm);
 907 }
 908
 909
 910 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
 911 {
 912         struct vcpu_svm *svm = to_svm(vcpu);
 913         bool block_nested_events =
 914                 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
 915         struct kvm_lapic *apic = vcpu->arch.apic;
 916
 917         if (lapic_in_kernel(vcpu) &&
 918             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
 919                 if (block_nested_events)
 920                         return -EBUSY;
 921                 if (!nested_exit_on_init(svm))
 922                         return 0;
 923                 nested_svm_init(svm);
 924                 return 0;
 925         }
 926
 927         if (vcpu->arch.exception.pending) {
 928                 if (block_nested_events)
 929                         return -EBUSY;
 930                 if (!nested_exit_on_exception(svm))
 931                         return 0;
 932                 nested_svm_inject_exception_vmexit(svm);
 933                 return 0;
 934         }
 935
 936         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
 937                 if (block_nested_events)
 938                         return -EBUSY;
 939                 if (!nested_exit_on_smi(svm))
 940                         return 0;
 941                 nested_svm_smi(svm);
 942                 return 0;
 943         }
 944
 945         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
 946                 if (block_nested_events)
 947                         return -EBUSY;
 948                 if (!nested_exit_on_nmi(svm))
 949                         return 0;
 950                 nested_svm_nmi(svm);
 951                 return 0;
 952         }
 953
 954         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
 955                 if (block_nested_events)
 956                         return -EBUSY;
 957                 if (!nested_exit_on_intr(svm))
 958                         return 0;
 959                 nested_svm_intr(svm);
 960                 return 0;
 961         }
 962
 963         return 0;
 964 }
 965
 966 int nested_svm_exit_special(struct vcpu_svm *svm)
 967 {
 968         u32 exit_code = svm->vmcb->control.exit_code;
 969
 970         switch (exit_code) {
 971         case SVM_EXIT_INTR:
 972         case SVM_EXIT_NMI:
 973         case SVM_EXIT_NPF:
 974                 return NESTED_EXIT_HOST;
 975         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 976                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
 977
 978                 if (get_host_vmcb(svm)->control.intercepts[INTERCEPT_EXCEPTION] &
 979                                 excp_bits)
 980                         return NESTED_EXIT_HOST;
 981                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
 982                          svm->vcpu.arch.apf.host_apf_flags)
 983                         /* Trap async PF even if not shadowing */
 984                         return NESTED_EXIT_HOST;
 985                 break;
 986         }
 987         default:
 988                 break;
 989         }
 990
 991         return NESTED_EXIT_CONTINUE;
 992 }
 993
 994 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
 995                                 struct kvm_nested_state __user *user_kvm_nested_state,
 996                                 u32 user_data_size)
 997 {
 998         struct vcpu_svm *svm;
 999         struct kvm_nested_state kvm_state = {
1000                 .flags = 0,
1001                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1002                 .size = sizeof(kvm_state),
1003         };
1004         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1005                 &user_kvm_nested_state->data.svm[0];
1006
1007         if (!vcpu)
1008                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1009
1010         svm = to_svm(vcpu);
1011
1012         if (user_data_size < kvm_state.size)
1013                 goto out;
1014
1015         /* First fill in the header and copy it out.  */
1016         if (is_guest_mode(vcpu)) {
1017                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1018                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1019                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1020
1021                 if (svm->nested.nested_run_pending)
1022                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1023         }
1024
1025         if (gif_set(svm))
1026                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1027
1028         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1029                 return -EFAULT;
1030
1031         if (!is_guest_mode(vcpu))
1032                 goto out;
1033
1034         /*
1035          * Copy over the full size of the VMCB rather than just the size
1036          * of the structs.
1037          */
1038         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1039                 return -EFAULT;
1040         if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1041                          sizeof(user_vmcb->control)))
1042                 return -EFAULT;
1043         if (copy_to_user(&user_vmcb->save, &svm->nested.hsave->save,
1044                          sizeof(user_vmcb->save)))
1045                 return -EFAULT;
1046
1047 out:
1048         return kvm_state.size;
1049 }
1050
1051 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1052                                 struct kvm_nested_state __user *user_kvm_nested_state,
1053                                 struct kvm_nested_state *kvm_state)
1054 {
1055         struct vcpu_svm *svm = to_svm(vcpu);
1056         struct vmcb *hsave = svm->nested.hsave;
1057         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1058                 &user_kvm_nested_state->data.svm[0];
1059         struct vmcb_control_area *ctl;
1060         struct vmcb_save_area *save;
1061         int ret;
1062         u32 cr0;
1063
1064         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1065                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1066
1067         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1068                 return -EINVAL;
1069
1070         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1071                                  KVM_STATE_NESTED_RUN_PENDING |
1072                                  KVM_STATE_NESTED_GIF_SET))
1073                 return -EINVAL;
1074
1075         /*
1076          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1077          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1078          */
1079         if (!(vcpu->arch.efer & EFER_SVME)) {
1080                 /* GIF=1 and no guest mode are required if SVME=0.  */
1081                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1082                         return -EINVAL;
1083         }
1084
1085         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1086         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1087                 return -EINVAL;
1088
1089         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1090                 svm_leave_nested(svm);
1091                 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1092                 return 0;
1093         }
1094
1095         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1096                 return -EINVAL;
1097         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1098                 return -EINVAL;
1099
1100         ret  = -ENOMEM;
1101         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
1102         save = kzalloc(sizeof(*save), GFP_KERNEL);
1103         if (!ctl || !save)
1104                 goto out_free;
1105
1106         ret = -EFAULT;
1107         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1108                 goto out_free;
1109         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1110                 goto out_free;
1111
1112         ret = -EINVAL;
1113         if (!nested_vmcb_check_controls(ctl))
1114                 goto out_free;
1115
1116         /*
1117          * Processor state contains L2 state.  Check that it is
1118          * valid for guest mode (see nested_vmcb_checks).
1119          */
1120         cr0 = kvm_read_cr0(vcpu);
1121         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1122                 goto out_free;
1123
1124         /*
1125          * Validate host state saved from before VMRUN (see
1126          * nested_svm_check_permissions).
1127          * TODO: validate reserved bits for all saved state.
1128          */
1129         if (!(save->cr0 & X86_CR0_PG))
1130                 goto out_free;
1131
1132         /*
1133          * All checks done, we can enter guest mode.  L1 control fields
1134          * come from the nested save state.  Guest state is already
1135          * in the registers, the save area of the nested state instead
1136          * contains saved L1 state.
1137          */
1138         copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
1139         hsave->save = *save;
1140
1141         svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1142         load_nested_vmcb_control(svm, ctl);
1143         nested_prepare_vmcb_control(svm);
1144
1145         if (!nested_svm_vmrun_msrpm(svm))
1146                 goto out_free;
1147
1148         ret = 0;
1149 out_free:
1150         kfree(save);
1151         kfree(ctl);
1152
1153         return ret;
1154 }
1155
1156 struct kvm_x86_nested_ops svm_nested_ops = {
1157         .check_events = svm_check_nested_events,
1158         .get_state = svm_get_nested_state,
1159         .set_state = svm_set_nested_state,
1160 };