arch/x86/kvm/svm/nested.c

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