arch/x86/kvm/svm/nested.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Kernel-based Virtual Machine driver for Linux
   4  *
   5  * AMD SVM support
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *   Avi Kivity   <avi@qumranet.com>
  13  */
  14
  15 #define pr_fmt(fmt) "SVM: " fmt
  16
  17 #include <linux/kvm_types.h>
  18 #include <linux/kvm_host.h>
  19 #include <linux/kernel.h>
  20
  21 #include <asm/msr-index.h>
  22 #include <asm/debugreg.h>
  23
  24 #include "kvm_emulate.h"
  25 #include "trace.h"
  26 #include "mmu.h"
  27 #include "x86.h"
  28 #include "cpuid.h"
  29 #include "lapic.h"
  30 #include "svm.h"
  31
  32 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
  33
  34 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  35                                        struct x86_exception *fault)
  36 {
  37         struct vcpu_svm *svm = to_svm(vcpu);
  38
  39         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  40                 /*
  41                  * TODO: track the cause of the nested page fault, and
  42                  * correctly fill in the high bits of exit_info_1.
  43                  */
  44                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  45                 svm->vmcb->control.exit_code_hi = 0;
  46                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
  47                 svm->vmcb->control.exit_info_2 = fault->address;
  48         }
  49
  50         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  51         svm->vmcb->control.exit_info_1 |= fault->error_code;
  52
  53         nested_svm_vmexit(svm);
  54 }
  55
  56 static void svm_inject_page_fault_nested(struct kvm_vcpu *vcpu, struct x86_exception *fault)
  57 {
  58        struct vcpu_svm *svm = to_svm(vcpu);
  59        WARN_ON(!is_guest_mode(vcpu));
  60
  61        if (vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
  62            !svm->nested.nested_run_pending) {
  63                svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
  64                svm->vmcb->control.exit_code_hi = 0;
  65                svm->vmcb->control.exit_info_1 = fault->error_code;
  66                svm->vmcb->control.exit_info_2 = fault->address;
  67                nested_svm_vmexit(svm);
  68        } else {
  69                kvm_inject_page_fault(vcpu, fault);
  70        }
  71 }
  72
  73 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  74 {
  75         struct vcpu_svm *svm = to_svm(vcpu);
  76         u64 cr3 = svm->nested.ctl.nested_cr3;
  77         u64 pdpte;
  78         int ret;
  79
  80         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
  81                                        offset_in_page(cr3) + index * 8, 8);
  82         if (ret)
  83                 return 0;
  84         return pdpte;
  85 }
  86
  87 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  88 {
  89         struct vcpu_svm *svm = to_svm(vcpu);
  90
  91         return svm->nested.ctl.nested_cr3;
  92 }
  93
  94 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  95 {
  96         struct vcpu_svm *svm = to_svm(vcpu);
  97
  98         WARN_ON(mmu_is_nested(vcpu));
  99
 100         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
 101         kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
 102                                 svm->vmcb01.ptr->save.efer,
 103                                 svm->nested.ctl.nested_cr3);
 104         vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
 105         vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
 106         vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
 107         reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
 108         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 109 }
 110
 111 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 112 {
 113         vcpu->arch.mmu = &vcpu->arch.root_mmu;
 114         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 115 }
 116
 117 void recalc_intercepts(struct vcpu_svm *svm)
 118 {
 119         struct vmcb_control_area *c, *h, *g;
 120         unsigned int i;
 121
 122         vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 123
 124         if (!is_guest_mode(&svm->vcpu))
 125                 return;
 126
 127         c = &svm->vmcb->control;
 128         h = &svm->vmcb01.ptr->control;
 129         g = &svm->nested.ctl;
 130
 131         for (i = 0; i < MAX_INTERCEPT; i++)
 132                 c->intercepts[i] = h->intercepts[i];
 133
 134         if (g->int_ctl & V_INTR_MASKING_MASK) {
 135                 /* We only want the cr8 intercept bits of L1 */
 136                 vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
 137                 vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
 138
 139                 /*
 140                  * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 141                  * affect any interrupt we may want to inject; therefore,
 142                  * interrupt window vmexits are irrelevant to L0.
 143                  */
 144                 vmcb_clr_intercept(c, INTERCEPT_VINTR);
 145         }
 146
 147         /* We don't want to see VMMCALLs from a nested guest */
 148         vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
 149
 150         for (i = 0; i < MAX_INTERCEPT; i++)
 151                 c->intercepts[i] |= g->intercepts[i];
 152 }
 153
 154 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
 155                                    struct vmcb_control_area *from)
 156 {
 157         unsigned int i;
 158
 159         for (i = 0; i < MAX_INTERCEPT; i++)
 160                 dst->intercepts[i] = from->intercepts[i];
 161
 162         dst->iopm_base_pa         = from->iopm_base_pa;
 163         dst->msrpm_base_pa        = from->msrpm_base_pa;
 164         dst->tsc_offset           = from->tsc_offset;
 165         /* asid not copied, it is handled manually for svm->vmcb.  */
 166         dst->tlb_ctl              = from->tlb_ctl;
 167         dst->int_ctl              = from->int_ctl;
 168         dst->int_vector           = from->int_vector;
 169         dst->int_state            = from->int_state;
 170         dst->exit_code            = from->exit_code;
 171         dst->exit_code_hi         = from->exit_code_hi;
 172         dst->exit_info_1          = from->exit_info_1;
 173         dst->exit_info_2          = from->exit_info_2;
 174         dst->exit_int_info        = from->exit_int_info;
 175         dst->exit_int_info_err    = from->exit_int_info_err;
 176         dst->nested_ctl           = from->nested_ctl;
 177         dst->event_inj            = from->event_inj;
 178         dst->event_inj_err        = from->event_inj_err;
 179         dst->nested_cr3           = from->nested_cr3;
 180         dst->virt_ext              = from->virt_ext;
 181         dst->pause_filter_count   = from->pause_filter_count;
 182         dst->pause_filter_thresh  = from->pause_filter_thresh;
 183 }
 184
 185 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 186 {
 187         /*
 188          * This function merges the msr permission bitmaps of kvm and the
 189          * nested vmcb. It is optimized in that it only merges the parts where
 190          * the kvm msr permission bitmap may contain zero bits
 191          */
 192         int i;
 193
 194         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 195                 return true;
 196
 197         for (i = 0; i < MSRPM_OFFSETS; i++) {
 198                 u32 value, p;
 199                 u64 offset;
 200
 201                 if (msrpm_offsets[i] == 0xffffffff)
 202                         break;
 203
 204                 p      = msrpm_offsets[i];
 205                 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 206
 207                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 208                         return false;
 209
 210                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 211         }
 212
 213         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 214
 215         return true;
 216 }
 217
 218 /*
 219  * Bits 11:0 of bitmap address are ignored by hardware
 220  */
 221 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
 222 {
 223         u64 addr = PAGE_ALIGN(pa);
 224
 225         return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
 226             kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
 227 }
 228
 229 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
 230                                        struct vmcb_control_area *control)
 231 {
 232         if (CC(!vmcb_is_intercept(control, INTERCEPT_VMRUN)))
 233                 return false;
 234
 235         if (CC(control->asid == 0))
 236                 return false;
 237
 238         if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
 239                 return false;
 240
 241         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
 242                                            MSRPM_SIZE)))
 243                 return false;
 244         if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
 245                                            IOPM_SIZE)))
 246                 return false;
 247
 248         return true;
 249 }
 250
 251 static bool nested_vmcb_check_cr3_cr4(struct kvm_vcpu *vcpu,
 252                                       struct vmcb_save_area *save)
 253 {
 254         /*
 255          * These checks are also performed by KVM_SET_SREGS,
 256          * except that EFER.LMA is not checked by SVM against
 257          * CR0.PG && EFER.LME.
 258          */
 259         if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
 260                 if (CC(!(save->cr4 & X86_CR4_PAE)) ||
 261                     CC(!(save->cr0 & X86_CR0_PE)) ||
 262                     CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
 263                         return false;
 264         }
 265
 266         if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
 267                 return false;
 268
 269         return true;
 270 }
 271
 272 /* Common checks that apply to both L1 and L2 state.  */
 273 static bool nested_vmcb_valid_sregs(struct kvm_vcpu *vcpu,
 274                                     struct vmcb_save_area *save)
 275 {
 276         /*
 277          * FIXME: these should be done after copying the fields,
 278          * to avoid TOC/TOU races.  For these save area checks
 279          * the possible damage is limited since kvm_set_cr0 and
 280          * kvm_set_cr4 handle failure; EFER_SVME is an exception
 281          * so it is force-set later in nested_prepare_vmcb_save.
 282          */
 283         if (CC(!(save->efer & EFER_SVME)))
 284                 return false;
 285
 286         if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 287             CC(save->cr0 & ~0xffffffffULL))
 288                 return false;
 289
 290         if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 291                 return false;
 292
 293         if (!nested_vmcb_check_cr3_cr4(vcpu, save))
 294                 return false;
 295
 296         if (CC(!kvm_valid_efer(vcpu, save->efer)))
 297                 return false;
 298
 299         return true;
 300 }
 301
 302 static void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
 303                                             struct vmcb_control_area *control)
 304 {
 305         copy_vmcb_control_area(&svm->nested.ctl, control);
 306
 307         /* Copy it here because nested_svm_check_controls will check it.  */
 308         svm->nested.ctl.asid           = control->asid;
 309         svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
 310         svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
 311 }
 312
 313 /*
 314  * Synchronize fields that are written by the processor, so that
 315  * they can be copied back into the vmcb12.
 316  */
 317 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
 318 {
 319         u32 mask;
 320         svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 321         svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 322
 323         /* Only a few fields of int_ctl are written by the processor.  */
 324         mask = V_IRQ_MASK | V_TPR_MASK;
 325         if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
 326             svm_is_intercept(svm, INTERCEPT_VINTR)) {
 327                 /*
 328                  * In order to request an interrupt window, L0 is usurping
 329                  * svm->vmcb->control.int_ctl and possibly setting V_IRQ
 330                  * even if it was clear in L1's VMCB.  Restoring it would be
 331                  * wrong.  However, in this case V_IRQ will remain true until
 332                  * interrupt_window_interception calls svm_clear_vintr and
 333                  * restores int_ctl.  We can just leave it aside.
 334                  */
 335                 mask &= ~V_IRQ_MASK;
 336         }
 337         svm->nested.ctl.int_ctl        &= ~mask;
 338         svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 339 }
 340
 341 /*
 342  * Transfer any event that L0 or L1 wanted to inject into L2 to
 343  * EXIT_INT_INFO.
 344  */
 345 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
 346                                                 struct vmcb *vmcb12)
 347 {
 348         struct kvm_vcpu *vcpu = &svm->vcpu;
 349         u32 exit_int_info = 0;
 350         unsigned int nr;
 351
 352         if (vcpu->arch.exception.injected) {
 353                 nr = vcpu->arch.exception.nr;
 354                 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 355
 356                 if (vcpu->arch.exception.has_error_code) {
 357                         exit_int_info |= SVM_EVTINJ_VALID_ERR;
 358                         vmcb12->control.exit_int_info_err =
 359                                 vcpu->arch.exception.error_code;
 360                 }
 361
 362         } else if (vcpu->arch.nmi_injected) {
 363                 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 364
 365         } else if (vcpu->arch.interrupt.injected) {
 366                 nr = vcpu->arch.interrupt.nr;
 367                 exit_int_info = nr | SVM_EVTINJ_VALID;
 368
 369                 if (vcpu->arch.interrupt.soft)
 370                         exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 371                 else
 372                         exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 373         }
 374
 375         vmcb12->control.exit_int_info = exit_int_info;
 376 }
 377
 378 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
 379 {
 380         return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
 381 }
 382
 383 /*
 384  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 385  * if we are emulating VM-Entry into a guest with NPT enabled.
 386  */
 387 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 388                                bool nested_npt)
 389 {
 390         if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
 391                 return -EINVAL;
 392
 393         if (!nested_npt && is_pae_paging(vcpu) &&
 394             (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
 395                 if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)))
 396                         return -EINVAL;
 397         }
 398
 399         /*
 400          * TODO: optimize unconditional TLB flush/MMU sync here and in
 401          * kvm_init_shadow_npt_mmu().
 402          */
 403         if (!nested_npt)
 404                 kvm_mmu_new_pgd(vcpu, cr3, false, false);
 405
 406         vcpu->arch.cr3 = cr3;
 407         kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
 408
 409         kvm_init_mmu(vcpu, false);
 410
 411         return 0;
 412 }
 413
 414 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
 415 {
 416         if (!svm->nested.vmcb02.ptr)
 417                 return;
 418
 419         /* FIXME: merge g_pat from vmcb01 and vmcb12.  */
 420         svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
 421 }
 422
 423 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 424 {
 425         bool new_vmcb12 = false;
 426
 427         nested_vmcb02_compute_g_pat(svm);
 428
 429         /* Load the nested guest state */
 430
 431         if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
 432                 new_vmcb12 = true;
 433                 svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
 434         }
 435
 436         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
 437                 svm->vmcb->save.es = vmcb12->save.es;
 438                 svm->vmcb->save.cs = vmcb12->save.cs;
 439                 svm->vmcb->save.ss = vmcb12->save.ss;
 440                 svm->vmcb->save.ds = vmcb12->save.ds;
 441                 svm->vmcb->save.cpl = vmcb12->save.cpl;
 442                 vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
 443         }
 444
 445         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
 446                 svm->vmcb->save.gdtr = vmcb12->save.gdtr;
 447                 svm->vmcb->save.idtr = vmcb12->save.idtr;
 448                 vmcb_mark_dirty(svm->vmcb, VMCB_DT);
 449         }
 450
 451         kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
 452
 453         /*
 454          * Force-set EFER_SVME even though it is checked earlier on the
 455          * VMCB12, because the guest can flip the bit between the check
 456          * and now.  Clearing EFER_SVME would call svm_free_nested.
 457          */
 458         svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
 459
 460         svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
 461         svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
 462
 463         svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 464
 465         kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
 466         kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
 467         kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
 468
 469         /* In case we don't even reach vcpu_run, the fields are not updated */
 470         svm->vmcb->save.rax = vmcb12->save.rax;
 471         svm->vmcb->save.rsp = vmcb12->save.rsp;
 472         svm->vmcb->save.rip = vmcb12->save.rip;
 473
 474         /* These bits will be set properly on the first execution when new_vmc12 is true */
 475         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 476                 svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
 477                 svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
 478                 vmcb_mark_dirty(svm->vmcb, VMCB_DR);
 479         }
 480 }
 481
 482 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
 483 {
 484         const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
 485
 486         /*
 487          * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 488          * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 489          */
 490
 491         /*
 492          * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
 493          * avic_physical_id.
 494          */
 495         WARN_ON(svm->vmcb01.ptr->control.int_ctl & AVIC_ENABLE_MASK);
 496
 497         /* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 498         svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
 499         svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
 500         svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
 501
 502         /* Done at vmrun: asid.  */
 503
 504         /* Also overwritten later if necessary.  */
 505         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 506
 507         /* nested_cr3.  */
 508         if (nested_npt_enabled(svm))
 509                 nested_svm_init_mmu_context(&svm->vcpu);
 510
 511         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 512                 svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
 513
 514         svm->vmcb->control.int_ctl             =
 515                 (svm->nested.ctl.int_ctl & ~mask) |
 516                 (svm->vmcb01.ptr->control.int_ctl & mask);
 517
 518         svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
 519         svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
 520         svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
 521         svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
 522         svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 523
 524         svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
 525         svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
 526
 527         /* Enter Guest-Mode */
 528         enter_guest_mode(&svm->vcpu);
 529
 530         /*
 531          * Merge guest and host intercepts - must be called with vcpu in
 532          * guest-mode to take effect.
 533          */
 534         recalc_intercepts(svm);
 535 }
 536
 537 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 538 {
 539         /*
 540          * Some VMCB state is shared between L1 and L2 and thus has to be
 541          * moved at the time of nested vmrun and vmexit.
 542          *
 543          * VMLOAD/VMSAVE state would also belong in this category, but KVM
 544          * always performs VMLOAD and VMSAVE from the VMCB01.
 545          */
 546         to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 547 }
 548
 549 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 550                          struct vmcb *vmcb12)
 551 {
 552         struct vcpu_svm *svm = to_svm(vcpu);
 553         int ret;
 554
 555         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
 556                                vmcb12->save.rip,
 557                                vmcb12->control.int_ctl,
 558                                vmcb12->control.event_inj,
 559                                vmcb12->control.nested_ctl);
 560
 561         trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 562                                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 563                                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 564                                     vmcb12->control.intercepts[INTERCEPT_WORD3],
 565                                     vmcb12->control.intercepts[INTERCEPT_WORD4],
 566                                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
 567
 568
 569         svm->nested.vmcb12_gpa = vmcb12_gpa;
 570
 571         WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 572
 573         nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 574
 575         svm_switch_vmcb(svm, &svm->nested.vmcb02);
 576         nested_vmcb02_prepare_control(svm);
 577         nested_vmcb02_prepare_save(svm, vmcb12);
 578
 579         ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
 580                                   nested_npt_enabled(svm));
 581         if (ret)
 582                 return ret;
 583
 584         if (!npt_enabled)
 585                 vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
 586
 587         svm_set_gif(svm, true);
 588
 589         return 0;
 590 }
 591
 592 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 593 {
 594         struct vcpu_svm *svm = to_svm(vcpu);
 595         int ret;
 596         struct vmcb *vmcb12;
 597         struct kvm_host_map map;
 598         u64 vmcb12_gpa;
 599
 600         ++vcpu->stat.nested_run;
 601
 602         if (is_smm(vcpu)) {
 603                 kvm_queue_exception(vcpu, UD_VECTOR);
 604                 return 1;
 605         }
 606
 607         vmcb12_gpa = svm->vmcb->save.rax;
 608         ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 609         if (ret == -EINVAL) {
 610                 kvm_inject_gp(vcpu, 0);
 611                 return 1;
 612         } else if (ret) {
 613                 return kvm_skip_emulated_instruction(vcpu);
 614         }
 615
 616         ret = kvm_skip_emulated_instruction(vcpu);
 617
 618         vmcb12 = map.hva;
 619
 620         if (WARN_ON_ONCE(!svm->nested.initialized))
 621                 return -EINVAL;
 622
 623         nested_load_control_from_vmcb12(svm, &vmcb12->control);
 624
 625         if (!nested_vmcb_valid_sregs(vcpu, &vmcb12->save) ||
 626             !nested_vmcb_check_controls(vcpu, &svm->nested.ctl)) {
 627                 vmcb12->control.exit_code    = SVM_EXIT_ERR;
 628                 vmcb12->control.exit_code_hi = 0;
 629                 vmcb12->control.exit_info_1  = 0;
 630                 vmcb12->control.exit_info_2  = 0;
 631                 goto out;
 632         }
 633
 634
 635         /* Clear internal status */
 636         kvm_clear_exception_queue(vcpu);
 637         kvm_clear_interrupt_queue(vcpu);
 638
 639         /*
 640          * Since vmcb01 is not in use, we can use it to store some of the L1
 641          * state.
 642          */
 643         svm->vmcb01.ptr->save.efer   = vcpu->arch.efer;
 644         svm->vmcb01.ptr->save.cr0    = kvm_read_cr0(vcpu);
 645         svm->vmcb01.ptr->save.cr4    = vcpu->arch.cr4;
 646         svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
 647         svm->vmcb01.ptr->save.rip    = kvm_rip_read(vcpu);
 648
 649         if (!npt_enabled)
 650                 svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
 651
 652         svm->nested.nested_run_pending = 1;
 653
 654         if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12))
 655                 goto out_exit_err;
 656
 657         if (nested_svm_vmrun_msrpm(svm))
 658                 goto out;
 659
 660 out_exit_err:
 661         svm->nested.nested_run_pending = 0;
 662
 663         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 664         svm->vmcb->control.exit_code_hi = 0;
 665         svm->vmcb->control.exit_info_1  = 0;
 666         svm->vmcb->control.exit_info_2  = 0;
 667
 668         nested_svm_vmexit(svm);
 669
 670 out:
 671         kvm_vcpu_unmap(vcpu, &map, true);
 672
 673         return ret;
 674 }
 675
 676 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 677 {
 678         to_vmcb->save.fs = from_vmcb->save.fs;
 679         to_vmcb->save.gs = from_vmcb->save.gs;
 680         to_vmcb->save.tr = from_vmcb->save.tr;
 681         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 682         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 683         to_vmcb->save.star = from_vmcb->save.star;
 684         to_vmcb->save.lstar = from_vmcb->save.lstar;
 685         to_vmcb->save.cstar = from_vmcb->save.cstar;
 686         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 687         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 688         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 689         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 690 }
 691
 692 int nested_svm_vmexit(struct vcpu_svm *svm)
 693 {
 694         struct kvm_vcpu *vcpu = &svm->vcpu;
 695         struct vmcb *vmcb12;
 696         struct vmcb *vmcb = svm->vmcb;
 697         struct kvm_host_map map;
 698         int rc;
 699
 700         /* Triple faults in L2 should never escape. */
 701         WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
 702
 703         rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 704         if (rc) {
 705                 if (rc == -EINVAL)
 706                         kvm_inject_gp(vcpu, 0);
 707                 return 1;
 708         }
 709
 710         vmcb12 = map.hva;
 711
 712         /* Exit Guest-Mode */
 713         leave_guest_mode(vcpu);
 714         svm->nested.vmcb12_gpa = 0;
 715         WARN_ON_ONCE(svm->nested.nested_run_pending);
 716
 717         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 718
 719         /* in case we halted in L2 */
 720         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 721
 722         /* Give the current vmcb to the guest */
 723
 724         vmcb12->save.es     = vmcb->save.es;
 725         vmcb12->save.cs     = vmcb->save.cs;
 726         vmcb12->save.ss     = vmcb->save.ss;
 727         vmcb12->save.ds     = vmcb->save.ds;
 728         vmcb12->save.gdtr   = vmcb->save.gdtr;
 729         vmcb12->save.idtr   = vmcb->save.idtr;
 730         vmcb12->save.efer   = svm->vcpu.arch.efer;
 731         vmcb12->save.cr0    = kvm_read_cr0(vcpu);
 732         vmcb12->save.cr3    = kvm_read_cr3(vcpu);
 733         vmcb12->save.cr2    = vmcb->save.cr2;
 734         vmcb12->save.cr4    = svm->vcpu.arch.cr4;
 735         vmcb12->save.rflags = kvm_get_rflags(vcpu);
 736         vmcb12->save.rip    = kvm_rip_read(vcpu);
 737         vmcb12->save.rsp    = kvm_rsp_read(vcpu);
 738         vmcb12->save.rax    = kvm_rax_read(vcpu);
 739         vmcb12->save.dr7    = vmcb->save.dr7;
 740         vmcb12->save.dr6    = svm->vcpu.arch.dr6;
 741         vmcb12->save.cpl    = vmcb->save.cpl;
 742
 743         vmcb12->control.int_state         = vmcb->control.int_state;
 744         vmcb12->control.exit_code         = vmcb->control.exit_code;
 745         vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
 746         vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
 747         vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
 748
 749         if (vmcb12->control.exit_code != SVM_EXIT_ERR)
 750                 nested_save_pending_event_to_vmcb12(svm, vmcb12);
 751
 752         if (svm->nrips_enabled)
 753                 vmcb12->control.next_rip  = vmcb->control.next_rip;
 754
 755         vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
 756         vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 757         vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
 758         vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 759
 760         vmcb12->control.pause_filter_count =
 761                 svm->vmcb->control.pause_filter_count;
 762         vmcb12->control.pause_filter_thresh =
 763                 svm->vmcb->control.pause_filter_thresh;
 764
 765         nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
 766
 767         svm_switch_vmcb(svm, &svm->vmcb01);
 768         WARN_ON_ONCE(svm->vmcb->control.exit_code != SVM_EXIT_VMRUN);
 769
 770         /*
 771          * On vmexit the  GIF is set to false and
 772          * no event can be injected in L1.
 773          */
 774         svm_set_gif(svm, false);
 775         svm->vmcb->control.exit_int_info = 0;
 776
 777         svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
 778         if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
 779                 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 780                 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 781         }
 782
 783         svm->nested.ctl.nested_cr3 = 0;
 784
 785         /*
 786          * Restore processor state that had been saved in vmcb01
 787          */
 788         kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
 789         svm_set_efer(vcpu, svm->vmcb->save.efer);
 790         svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
 791         svm_set_cr4(vcpu, svm->vmcb->save.cr4);
 792         kvm_rax_write(vcpu, svm->vmcb->save.rax);
 793         kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
 794         kvm_rip_write(vcpu, svm->vmcb->save.rip);
 795
 796         svm->vcpu.arch.dr7 = DR7_FIXED_1;
 797         kvm_update_dr7(&svm->vcpu);
 798
 799         trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
 800                                        vmcb12->control.exit_info_1,
 801                                        vmcb12->control.exit_info_2,
 802                                        vmcb12->control.exit_int_info,
 803                                        vmcb12->control.exit_int_info_err,
 804                                        KVM_ISA_SVM);
 805
 806         kvm_vcpu_unmap(vcpu, &map, true);
 807
 808         nested_svm_uninit_mmu_context(vcpu);
 809
 810         rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false);
 811         if (rc)
 812                 return 1;
 813
 814         /*
 815          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 816          * doesn't end up in L1.
 817          */
 818         svm->vcpu.arch.nmi_injected = false;
 819         kvm_clear_exception_queue(vcpu);
 820         kvm_clear_interrupt_queue(vcpu);
 821
 822         /*
 823          * If we are here following the completion of a VMRUN that
 824          * is being single-stepped, queue the pending #DB intercept
 825          * right now so that it an be accounted for before we execute
 826          * L1's next instruction.
 827          */
 828         if (unlikely(svm->vmcb->save.rflags & X86_EFLAGS_TF))
 829                 kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
 830
 831         return 0;
 832 }
 833
 834 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
 835 {
 836         nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
 837 }
 838
 839 int svm_allocate_nested(struct vcpu_svm *svm)
 840 {
 841         struct page *vmcb02_page;
 842
 843         if (svm->nested.initialized)
 844                 return 0;
 845
 846         vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 847         if (!vmcb02_page)
 848                 return -ENOMEM;
 849         svm->nested.vmcb02.ptr = page_address(vmcb02_page);
 850         svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
 851
 852         svm->nested.msrpm = svm_vcpu_alloc_msrpm();
 853         if (!svm->nested.msrpm)
 854                 goto err_free_vmcb02;
 855         svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
 856
 857         svm->nested.initialized = true;
 858         return 0;
 859
 860 err_free_vmcb02:
 861         __free_page(vmcb02_page);
 862         return -ENOMEM;
 863 }
 864
 865 void svm_free_nested(struct vcpu_svm *svm)
 866 {
 867         if (!svm->nested.initialized)
 868                 return;
 869
 870         svm_vcpu_free_msrpm(svm->nested.msrpm);
 871         svm->nested.msrpm = NULL;
 872
 873         __free_page(virt_to_page(svm->nested.vmcb02.ptr));
 874         svm->nested.vmcb02.ptr = NULL;
 875
 876         svm->nested.initialized = false;
 877 }
 878
 879 /*
 880  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
 881  */
 882 void svm_leave_nested(struct vcpu_svm *svm)
 883 {
 884         struct kvm_vcpu *vcpu = &svm->vcpu;
 885
 886         if (is_guest_mode(vcpu)) {
 887                 svm->nested.nested_run_pending = 0;
 888                 leave_guest_mode(vcpu);
 889
 890                 svm_switch_vmcb(svm, &svm->nested.vmcb02);
 891
 892                 nested_svm_uninit_mmu_context(vcpu);
 893                 vmcb_mark_all_dirty(svm->vmcb);
 894         }
 895
 896         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 897 }
 898
 899 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 900 {
 901         u32 offset, msr, value;
 902         int write, mask;
 903
 904         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 905                 return NESTED_EXIT_HOST;
 906
 907         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 908         offset = svm_msrpm_offset(msr);
 909         write  = svm->vmcb->control.exit_info_1 & 1;
 910         mask   = 1 << ((2 * (msr & 0xf)) + write);
 911
 912         if (offset == MSR_INVALID)
 913                 return NESTED_EXIT_DONE;
 914
 915         /* Offset is in 32 bit units but need in 8 bit units */
 916         offset *= 4;
 917
 918         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
 919                 return NESTED_EXIT_DONE;
 920
 921         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 922 }
 923
 924 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 925 {
 926         unsigned port, size, iopm_len;
 927         u16 val, mask;
 928         u8 start_bit;
 929         u64 gpa;
 930
 931         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
 932                 return NESTED_EXIT_HOST;
 933
 934         port = svm->vmcb->control.exit_info_1 >> 16;
 935         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 936                 SVM_IOIO_SIZE_SHIFT;
 937         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
 938         start_bit = port % 8;
 939         iopm_len = (start_bit + size > 8) ? 2 : 1;
 940         mask = (0xf >> (4 - size)) << start_bit;
 941         val = 0;
 942
 943         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 944                 return NESTED_EXIT_DONE;
 945
 946         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 947 }
 948
 949 static int nested_svm_intercept(struct vcpu_svm *svm)
 950 {
 951         u32 exit_code = svm->vmcb->control.exit_code;
 952         int vmexit = NESTED_EXIT_HOST;
 953
 954         switch (exit_code) {
 955         case SVM_EXIT_MSR:
 956                 vmexit = nested_svm_exit_handled_msr(svm);
 957                 break;
 958         case SVM_EXIT_IOIO:
 959                 vmexit = nested_svm_intercept_ioio(svm);
 960                 break;
 961         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 962                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 963                         vmexit = NESTED_EXIT_DONE;
 964                 break;
 965         }
 966         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 967                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 968                         vmexit = NESTED_EXIT_DONE;
 969                 break;
 970         }
 971         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 972                 /*
 973                  * Host-intercepted exceptions have been checked already in
 974                  * nested_svm_exit_special.  There is nothing to do here,
 975                  * the vmexit is injected by svm_check_nested_events.
 976                  */
 977                 vmexit = NESTED_EXIT_DONE;
 978                 break;
 979         }
 980         case SVM_EXIT_ERR: {
 981                 vmexit = NESTED_EXIT_DONE;
 982                 break;
 983         }
 984         default: {
 985                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 986                         vmexit = NESTED_EXIT_DONE;
 987         }
 988         }
 989
 990         return vmexit;
 991 }
 992
 993 int nested_svm_exit_handled(struct vcpu_svm *svm)
 994 {
 995         int vmexit;
 996
 997         vmexit = nested_svm_intercept(svm);
 998
 999         if (vmexit == NESTED_EXIT_DONE)
1000                 nested_svm_vmexit(svm);
1001
1002         return vmexit;
1003 }
1004
1005 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1006 {
1007         if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1008                 kvm_queue_exception(vcpu, UD_VECTOR);
1009                 return 1;
1010         }
1011
1012         if (to_svm(vcpu)->vmcb->save.cpl) {
1013                 kvm_inject_gp(vcpu, 0);
1014                 return 1;
1015         }
1016
1017         return 0;
1018 }
1019
1020 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1021 {
1022         unsigned int nr = svm->vcpu.arch.exception.nr;
1023
1024         return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1025 }
1026
1027 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1028 {
1029         unsigned int nr = svm->vcpu.arch.exception.nr;
1030
1031         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1032         svm->vmcb->control.exit_code_hi = 0;
1033
1034         if (svm->vcpu.arch.exception.has_error_code)
1035                 svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1036
1037         /*
1038          * EXITINFO2 is undefined for all exception intercepts other
1039          * than #PF.
1040          */
1041         if (nr == PF_VECTOR) {
1042                 if (svm->vcpu.arch.exception.nested_apf)
1043                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1044                 else if (svm->vcpu.arch.exception.has_payload)
1045                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1046                 else
1047                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1048         } else if (nr == DB_VECTOR) {
1049                 /* See inject_pending_event.  */
1050                 kvm_deliver_exception_payload(&svm->vcpu);
1051                 if (svm->vcpu.arch.dr7 & DR7_GD) {
1052                         svm->vcpu.arch.dr7 &= ~DR7_GD;
1053                         kvm_update_dr7(&svm->vcpu);
1054                 }
1055         } else
1056                 WARN_ON(svm->vcpu.arch.exception.has_payload);
1057
1058         nested_svm_vmexit(svm);
1059 }
1060
1061 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1062 {
1063         return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1064 }
1065
1066 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1067 {
1068         struct vcpu_svm *svm = to_svm(vcpu);
1069         bool block_nested_events =
1070                 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1071         struct kvm_lapic *apic = vcpu->arch.apic;
1072
1073         if (lapic_in_kernel(vcpu) &&
1074             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1075                 if (block_nested_events)
1076                         return -EBUSY;
1077                 if (!nested_exit_on_init(svm))
1078                         return 0;
1079                 nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1080                 return 0;
1081         }
1082
1083         if (vcpu->arch.exception.pending) {
1084                 /*
1085                  * Only a pending nested run can block a pending exception.
1086                  * Otherwise an injected NMI/interrupt should either be
1087                  * lost or delivered to the nested hypervisor in the EXITINTINFO
1088                  * vmcb field, while delivering the pending exception.
1089                  */
1090                 if (svm->nested.nested_run_pending)
1091                         return -EBUSY;
1092                 if (!nested_exit_on_exception(svm))
1093                         return 0;
1094                 nested_svm_inject_exception_vmexit(svm);
1095                 return 0;
1096         }
1097
1098         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1099                 if (block_nested_events)
1100                         return -EBUSY;
1101                 if (!nested_exit_on_smi(svm))
1102                         return 0;
1103                 nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1104                 return 0;
1105         }
1106
1107         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1108                 if (block_nested_events)
1109                         return -EBUSY;
1110                 if (!nested_exit_on_nmi(svm))
1111                         return 0;
1112                 nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1113                 return 0;
1114         }
1115
1116         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1117                 if (block_nested_events)
1118                         return -EBUSY;
1119                 if (!nested_exit_on_intr(svm))
1120                         return 0;
1121                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1122                 nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1123                 return 0;
1124         }
1125
1126         return 0;
1127 }
1128
1129 int nested_svm_exit_special(struct vcpu_svm *svm)
1130 {
1131         u32 exit_code = svm->vmcb->control.exit_code;
1132
1133         switch (exit_code) {
1134         case SVM_EXIT_INTR:
1135         case SVM_EXIT_NMI:
1136         case SVM_EXIT_NPF:
1137                 return NESTED_EXIT_HOST;
1138         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1139                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1140
1141                 if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1142                     excp_bits)
1143                         return NESTED_EXIT_HOST;
1144                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1145                          svm->vcpu.arch.apf.host_apf_flags)
1146                         /* Trap async PF even if not shadowing */
1147                         return NESTED_EXIT_HOST;
1148                 break;
1149         }
1150         default:
1151                 break;
1152         }
1153
1154         return NESTED_EXIT_CONTINUE;
1155 }
1156
1157 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1158                                 struct kvm_nested_state __user *user_kvm_nested_state,
1159                                 u32 user_data_size)
1160 {
1161         struct vcpu_svm *svm;
1162         struct kvm_nested_state kvm_state = {
1163                 .flags = 0,
1164                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1165                 .size = sizeof(kvm_state),
1166         };
1167         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1168                 &user_kvm_nested_state->data.svm[0];
1169
1170         if (!vcpu)
1171                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1172
1173         svm = to_svm(vcpu);
1174
1175         if (user_data_size < kvm_state.size)
1176                 goto out;
1177
1178         /* First fill in the header and copy it out.  */
1179         if (is_guest_mode(vcpu)) {
1180                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1181                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1182                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1183
1184                 if (svm->nested.nested_run_pending)
1185                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1186         }
1187
1188         if (gif_set(svm))
1189                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1190
1191         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1192                 return -EFAULT;
1193
1194         if (!is_guest_mode(vcpu))
1195                 goto out;
1196
1197         /*
1198          * Copy over the full size of the VMCB rather than just the size
1199          * of the structs.
1200          */
1201         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1202                 return -EFAULT;
1203         if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1204                          sizeof(user_vmcb->control)))
1205                 return -EFAULT;
1206         if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1207                          sizeof(user_vmcb->save)))
1208                 return -EFAULT;
1209 out:
1210         return kvm_state.size;
1211 }
1212
1213 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1214                                 struct kvm_nested_state __user *user_kvm_nested_state,
1215                                 struct kvm_nested_state *kvm_state)
1216 {
1217         struct vcpu_svm *svm = to_svm(vcpu);
1218         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1219                 &user_kvm_nested_state->data.svm[0];
1220         struct vmcb_control_area *ctl;
1221         struct vmcb_save_area *save;
1222         int ret;
1223         u32 cr0;
1224
1225         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1226                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1227
1228         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1229                 return -EINVAL;
1230
1231         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1232                                  KVM_STATE_NESTED_RUN_PENDING |
1233                                  KVM_STATE_NESTED_GIF_SET))
1234                 return -EINVAL;
1235
1236         /*
1237          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1238          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1239          */
1240         if (!(vcpu->arch.efer & EFER_SVME)) {
1241                 /* GIF=1 and no guest mode are required if SVME=0.  */
1242                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1243                         return -EINVAL;
1244         }
1245
1246         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1247         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1248                 return -EINVAL;
1249
1250         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1251                 svm_leave_nested(svm);
1252                 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1253                 return 0;
1254         }
1255
1256         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1257                 return -EINVAL;
1258         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1259                 return -EINVAL;
1260
1261         ret  = -ENOMEM;
1262         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1263         save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1264         if (!ctl || !save)
1265                 goto out_free;
1266
1267         ret = -EFAULT;
1268         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1269                 goto out_free;
1270         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1271                 goto out_free;
1272
1273         ret = -EINVAL;
1274         if (!nested_vmcb_check_controls(vcpu, ctl))
1275                 goto out_free;
1276
1277         /*
1278          * Processor state contains L2 state.  Check that it is
1279          * valid for guest mode (see nested_vmcb_check_save).
1280          */
1281         cr0 = kvm_read_cr0(vcpu);
1282         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1283                 goto out_free;
1284
1285         /*
1286          * Validate host state saved from before VMRUN (see
1287          * nested_svm_check_permissions).
1288          */
1289         if (!(save->cr0 & X86_CR0_PG) ||
1290             !(save->cr0 & X86_CR0_PE) ||
1291             (save->rflags & X86_EFLAGS_VM) ||
1292             !nested_vmcb_valid_sregs(vcpu, save))
1293                 goto out_free;
1294
1295         /*
1296          * All checks done, we can enter guest mode. Userspace provides
1297          * vmcb12.control, which will be combined with L1 and stored into
1298          * vmcb02, and the L1 save state which we store in vmcb01.
1299          * L2 registers if needed are moved from the current VMCB to VMCB02.
1300          */
1301
1302         svm->nested.nested_run_pending =
1303                 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1304
1305         svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1306         if (svm->current_vmcb == &svm->vmcb01)
1307                 svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1308
1309         svm->vmcb01.ptr->save.es = save->es;
1310         svm->vmcb01.ptr->save.cs = save->cs;
1311         svm->vmcb01.ptr->save.ss = save->ss;
1312         svm->vmcb01.ptr->save.ds = save->ds;
1313         svm->vmcb01.ptr->save.gdtr = save->gdtr;
1314         svm->vmcb01.ptr->save.idtr = save->idtr;
1315         svm->vmcb01.ptr->save.rflags = save->rflags | X86_EFLAGS_FIXED;
1316         svm->vmcb01.ptr->save.efer = save->efer;
1317         svm->vmcb01.ptr->save.cr0 = save->cr0;
1318         svm->vmcb01.ptr->save.cr3 = save->cr3;
1319         svm->vmcb01.ptr->save.cr4 = save->cr4;
1320         svm->vmcb01.ptr->save.rax = save->rax;
1321         svm->vmcb01.ptr->save.rsp = save->rsp;
1322         svm->vmcb01.ptr->save.rip = save->rip;
1323         svm->vmcb01.ptr->save.cpl = 0;
1324
1325         nested_load_control_from_vmcb12(svm, ctl);
1326
1327         svm_switch_vmcb(svm, &svm->nested.vmcb02);
1328
1329         nested_vmcb02_prepare_control(svm);
1330
1331         kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1332         ret = 0;
1333 out_free:
1334         kfree(save);
1335         kfree(ctl);
1336
1337         return ret;
1338 }
1339
1340 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1341 {
1342         struct vcpu_svm *svm = to_svm(vcpu);
1343
1344         if (WARN_ON(!is_guest_mode(vcpu)))
1345                 return true;
1346
1347         if (nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1348                                 nested_npt_enabled(svm)))
1349                 return false;
1350
1351         if (!nested_svm_vmrun_msrpm(svm)) {
1352                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1353                 vcpu->run->internal.suberror =
1354                         KVM_INTERNAL_ERROR_EMULATION;
1355                 vcpu->run->internal.ndata = 0;
1356                 return false;
1357         }
1358
1359         return true;
1360 }
1361
1362 struct kvm_x86_nested_ops svm_nested_ops = {
1363         .check_events = svm_check_nested_events,
1364         .triple_fault = nested_svm_triple_fault,
1365         .get_nested_state_pages = svm_get_nested_state_pages,
1366         .get_state = svm_get_nested_state,
1367         .set_state = svm_set_nested_state,
1368 };