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         if (CC(!(save->efer & EFER_SVME)))
 276                 return false;
 277
 278         if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 279             CC(save->cr0 & ~0xffffffffULL))
 280                 return false;
 281
 282         if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 283                 return false;
 284
 285         if (!nested_vmcb_check_cr3_cr4(vcpu, save))
 286                 return false;
 287
 288         if (CC(!kvm_valid_efer(vcpu, save->efer)))
 289                 return false;
 290
 291         return true;
 292 }
 293
 294 static bool nested_vmcb_checks(struct kvm_vcpu *vcpu, struct vmcb *vmcb12)
 295 {
 296         if (!nested_vmcb_valid_sregs(vcpu, &vmcb12->save))
 297                 return false;
 298
 299         return nested_vmcb_check_controls(&vmcb12->control);
 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         svm_set_efer(&svm->vcpu, vmcb12->save.efer);
 453         svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
 454         svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
 455
 456         svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 457
 458         kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
 459         kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
 460         kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
 461
 462         /* In case we don't even reach vcpu_run, the fields are not updated */
 463         svm->vmcb->save.rax = vmcb12->save.rax;
 464         svm->vmcb->save.rsp = vmcb12->save.rsp;
 465         svm->vmcb->save.rip = vmcb12->save.rip;
 466
 467         /* These bits will be set properly on the first execution when new_vmc12 is true */
 468         if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 469                 svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
 470                 svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
 471                 vmcb_mark_dirty(svm->vmcb, VMCB_DR);
 472         }
 473 }
 474
 475 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
 476 {
 477         const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
 478
 479         /*
 480          * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 481          * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 482          */
 483
 484         /*
 485          * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
 486          * avic_physical_id.
 487          */
 488         WARN_ON(svm->vmcb01.ptr->control.int_ctl & AVIC_ENABLE_MASK);
 489
 490         /* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 491         svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
 492         svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
 493         svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
 494
 495         /* Done at vmrun: asid.  */
 496
 497         /* Also overwritten later if necessary.  */
 498         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 499
 500         /* nested_cr3.  */
 501         if (nested_npt_enabled(svm))
 502                 nested_svm_init_mmu_context(&svm->vcpu);
 503
 504         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 505                 svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
 506
 507         svm->vmcb->control.int_ctl             =
 508                 (svm->nested.ctl.int_ctl & ~mask) |
 509                 (svm->vmcb01.ptr->control.int_ctl & mask);
 510
 511         svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
 512         svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
 513         svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
 514         svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
 515         svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 516
 517         svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
 518         svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
 519
 520         /* Enter Guest-Mode */
 521         enter_guest_mode(&svm->vcpu);
 522
 523         /*
 524          * Merge guest and host intercepts - must be called with vcpu in
 525          * guest-mode to take effect.
 526          */
 527         recalc_intercepts(svm);
 528 }
 529
 530 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 531 {
 532         /*
 533          * Some VMCB state is shared between L1 and L2 and thus has to be
 534          * moved at the time of nested vmrun and vmexit.
 535          *
 536          * VMLOAD/VMSAVE state would also belong in this category, but KVM
 537          * always performs VMLOAD and VMSAVE from the VMCB01.
 538          */
 539         to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 540 }
 541
 542 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 543                          struct vmcb *vmcb12)
 544 {
 545         struct vcpu_svm *svm = to_svm(vcpu);
 546         int ret;
 547
 548         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
 549                                vmcb12->save.rip,
 550                                vmcb12->control.int_ctl,
 551                                vmcb12->control.event_inj,
 552                                vmcb12->control.nested_ctl);
 553
 554         trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 555                                     vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 556                                     vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 557                                     vmcb12->control.intercepts[INTERCEPT_WORD3],
 558                                     vmcb12->control.intercepts[INTERCEPT_WORD4],
 559                                     vmcb12->control.intercepts[INTERCEPT_WORD5]);
 560
 561
 562         svm->nested.vmcb12_gpa = vmcb12_gpa;
 563
 564         WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 565
 566         nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 567         nested_load_control_from_vmcb12(svm, &vmcb12->control);
 568
 569         svm_switch_vmcb(svm, &svm->nested.vmcb02);
 570         nested_vmcb02_prepare_control(svm);
 571         nested_vmcb02_prepare_save(svm, vmcb12);
 572
 573         ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
 574                                   nested_npt_enabled(svm));
 575         if (ret)
 576                 return ret;
 577
 578         if (!npt_enabled)
 579                 vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
 580
 581         svm_set_gif(svm, true);
 582
 583         return 0;
 584 }
 585
 586 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 587 {
 588         struct vcpu_svm *svm = to_svm(vcpu);
 589         int ret;
 590         struct vmcb *vmcb12;
 591         struct kvm_host_map map;
 592         u64 vmcb12_gpa;
 593
 594         ++vcpu->stat.nested_run;
 595
 596         if (is_smm(vcpu)) {
 597                 kvm_queue_exception(vcpu, UD_VECTOR);
 598                 return 1;
 599         }
 600
 601         vmcb12_gpa = svm->vmcb->save.rax;
 602         ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 603         if (ret == -EINVAL) {
 604                 kvm_inject_gp(vcpu, 0);
 605                 return 1;
 606         } else if (ret) {
 607                 return kvm_skip_emulated_instruction(vcpu);
 608         }
 609
 610         ret = kvm_skip_emulated_instruction(vcpu);
 611
 612         vmcb12 = map.hva;
 613
 614         if (WARN_ON_ONCE(!svm->nested.initialized))
 615                 return -EINVAL;
 616
 617         if (!nested_vmcb_checks(vcpu, vmcb12)) {
 618                 vmcb12->control.exit_code    = SVM_EXIT_ERR;
 619                 vmcb12->control.exit_code_hi = 0;
 620                 vmcb12->control.exit_info_1  = 0;
 621                 vmcb12->control.exit_info_2  = 0;
 622                 goto out;
 623         }
 624
 625
 626         /* Clear internal status */
 627         kvm_clear_exception_queue(vcpu);
 628         kvm_clear_interrupt_queue(vcpu);
 629
 630         /*
 631          * Since vmcb01 is not in use, we can use it to store some of the L1
 632          * state.
 633          */
 634         svm->vmcb01.ptr->save.efer   = vcpu->arch.efer;
 635         svm->vmcb01.ptr->save.cr0    = kvm_read_cr0(vcpu);
 636         svm->vmcb01.ptr->save.cr4    = vcpu->arch.cr4;
 637         svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
 638         svm->vmcb01.ptr->save.rip    = kvm_rip_read(vcpu);
 639
 640         if (!npt_enabled)
 641                 svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
 642
 643         svm->nested.nested_run_pending = 1;
 644
 645         if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12))
 646                 goto out_exit_err;
 647
 648         if (nested_svm_vmrun_msrpm(svm))
 649                 goto out;
 650
 651 out_exit_err:
 652         svm->nested.nested_run_pending = 0;
 653
 654         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 655         svm->vmcb->control.exit_code_hi = 0;
 656         svm->vmcb->control.exit_info_1  = 0;
 657         svm->vmcb->control.exit_info_2  = 0;
 658
 659         nested_svm_vmexit(svm);
 660
 661 out:
 662         kvm_vcpu_unmap(vcpu, &map, true);
 663
 664         return ret;
 665 }
 666
 667 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 668 {
 669         to_vmcb->save.fs = from_vmcb->save.fs;
 670         to_vmcb->save.gs = from_vmcb->save.gs;
 671         to_vmcb->save.tr = from_vmcb->save.tr;
 672         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 673         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 674         to_vmcb->save.star = from_vmcb->save.star;
 675         to_vmcb->save.lstar = from_vmcb->save.lstar;
 676         to_vmcb->save.cstar = from_vmcb->save.cstar;
 677         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 678         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 679         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 680         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 681 }
 682
 683 int nested_svm_vmexit(struct vcpu_svm *svm)
 684 {
 685         struct kvm_vcpu *vcpu = &svm->vcpu;
 686         struct vmcb *vmcb12;
 687         struct vmcb *vmcb = svm->vmcb;
 688         struct kvm_host_map map;
 689         int rc;
 690
 691         /* Triple faults in L2 should never escape. */
 692         WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
 693
 694         rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 695         if (rc) {
 696                 if (rc == -EINVAL)
 697                         kvm_inject_gp(vcpu, 0);
 698                 return 1;
 699         }
 700
 701         vmcb12 = map.hva;
 702
 703         /* Exit Guest-Mode */
 704         leave_guest_mode(vcpu);
 705         svm->nested.vmcb12_gpa = 0;
 706         WARN_ON_ONCE(svm->nested.nested_run_pending);
 707
 708         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 709
 710         /* in case we halted in L2 */
 711         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 712
 713         /* Give the current vmcb to the guest */
 714
 715         vmcb12->save.es     = vmcb->save.es;
 716         vmcb12->save.cs     = vmcb->save.cs;
 717         vmcb12->save.ss     = vmcb->save.ss;
 718         vmcb12->save.ds     = vmcb->save.ds;
 719         vmcb12->save.gdtr   = vmcb->save.gdtr;
 720         vmcb12->save.idtr   = vmcb->save.idtr;
 721         vmcb12->save.efer   = svm->vcpu.arch.efer;
 722         vmcb12->save.cr0    = kvm_read_cr0(vcpu);
 723         vmcb12->save.cr3    = kvm_read_cr3(vcpu);
 724         vmcb12->save.cr2    = vmcb->save.cr2;
 725         vmcb12->save.cr4    = svm->vcpu.arch.cr4;
 726         vmcb12->save.rflags = kvm_get_rflags(vcpu);
 727         vmcb12->save.rip    = kvm_rip_read(vcpu);
 728         vmcb12->save.rsp    = kvm_rsp_read(vcpu);
 729         vmcb12->save.rax    = kvm_rax_read(vcpu);
 730         vmcb12->save.dr7    = vmcb->save.dr7;
 731         vmcb12->save.dr6    = svm->vcpu.arch.dr6;
 732         vmcb12->save.cpl    = vmcb->save.cpl;
 733
 734         vmcb12->control.int_state         = vmcb->control.int_state;
 735         vmcb12->control.exit_code         = vmcb->control.exit_code;
 736         vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
 737         vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
 738         vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
 739
 740         if (vmcb12->control.exit_code != SVM_EXIT_ERR)
 741                 nested_save_pending_event_to_vmcb12(svm, vmcb12);
 742
 743         if (svm->nrips_enabled)
 744                 vmcb12->control.next_rip  = vmcb->control.next_rip;
 745
 746         vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
 747         vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 748         vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
 749         vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 750
 751         vmcb12->control.pause_filter_count =
 752                 svm->vmcb->control.pause_filter_count;
 753         vmcb12->control.pause_filter_thresh =
 754                 svm->vmcb->control.pause_filter_thresh;
 755
 756         nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
 757
 758         svm_switch_vmcb(svm, &svm->vmcb01);
 759
 760         /*
 761          * On vmexit the  GIF is set to false and
 762          * no event can be injected in L1.
 763          */
 764         svm_set_gif(svm, false);
 765         svm->vmcb->control.exit_int_info = 0;
 766
 767         svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
 768         if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
 769                 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 770                 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 771         }
 772
 773         svm->nested.ctl.nested_cr3 = 0;
 774
 775         /*
 776          * Restore processor state that had been saved in vmcb01
 777          */
 778         kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
 779         svm_set_efer(vcpu, svm->vmcb->save.efer);
 780         svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
 781         svm_set_cr4(vcpu, svm->vmcb->save.cr4);
 782         kvm_rax_write(vcpu, svm->vmcb->save.rax);
 783         kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
 784         kvm_rip_write(vcpu, svm->vmcb->save.rip);
 785
 786         svm->vcpu.arch.dr7 = DR7_FIXED_1;
 787         kvm_update_dr7(&svm->vcpu);
 788
 789         trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
 790                                        vmcb12->control.exit_info_1,
 791                                        vmcb12->control.exit_info_2,
 792                                        vmcb12->control.exit_int_info,
 793                                        vmcb12->control.exit_int_info_err,
 794                                        KVM_ISA_SVM);
 795
 796         kvm_vcpu_unmap(vcpu, &map, true);
 797
 798         nested_svm_uninit_mmu_context(vcpu);
 799
 800         rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false);
 801         if (rc)
 802                 return 1;
 803
 804         /*
 805          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 806          * doesn't end up in L1.
 807          */
 808         svm->vcpu.arch.nmi_injected = false;
 809         kvm_clear_exception_queue(vcpu);
 810         kvm_clear_interrupt_queue(vcpu);
 811
 812         return 0;
 813 }
 814
 815 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
 816 {
 817         nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
 818 }
 819
 820 int svm_allocate_nested(struct vcpu_svm *svm)
 821 {
 822         struct page *vmcb02_page;
 823
 824         if (svm->nested.initialized)
 825                 return 0;
 826
 827         vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 828         if (!vmcb02_page)
 829                 return -ENOMEM;
 830         svm->nested.vmcb02.ptr = page_address(vmcb02_page);
 831         svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
 832
 833         svm->nested.msrpm = svm_vcpu_alloc_msrpm();
 834         if (!svm->nested.msrpm)
 835                 goto err_free_vmcb02;
 836         svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
 837
 838         svm->nested.initialized = true;
 839         return 0;
 840
 841 err_free_vmcb02:
 842         __free_page(vmcb02_page);
 843         return -ENOMEM;
 844 }
 845
 846 void svm_free_nested(struct vcpu_svm *svm)
 847 {
 848         if (!svm->nested.initialized)
 849                 return;
 850
 851         svm_vcpu_free_msrpm(svm->nested.msrpm);
 852         svm->nested.msrpm = NULL;
 853
 854         __free_page(virt_to_page(svm->nested.vmcb02.ptr));
 855         svm->nested.vmcb02.ptr = NULL;
 856
 857         svm->nested.initialized = false;
 858 }
 859
 860 /*
 861  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
 862  */
 863 void svm_leave_nested(struct vcpu_svm *svm)
 864 {
 865         struct kvm_vcpu *vcpu = &svm->vcpu;
 866
 867         if (is_guest_mode(vcpu)) {
 868                 svm->nested.nested_run_pending = 0;
 869                 leave_guest_mode(vcpu);
 870
 871                 svm_switch_vmcb(svm, &svm->nested.vmcb02);
 872
 873                 nested_svm_uninit_mmu_context(vcpu);
 874                 vmcb_mark_all_dirty(svm->vmcb);
 875         }
 876
 877         kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 878 }
 879
 880 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 881 {
 882         u32 offset, msr, value;
 883         int write, mask;
 884
 885         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 886                 return NESTED_EXIT_HOST;
 887
 888         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 889         offset = svm_msrpm_offset(msr);
 890         write  = svm->vmcb->control.exit_info_1 & 1;
 891         mask   = 1 << ((2 * (msr & 0xf)) + write);
 892
 893         if (offset == MSR_INVALID)
 894                 return NESTED_EXIT_DONE;
 895
 896         /* Offset is in 32 bit units but need in 8 bit units */
 897         offset *= 4;
 898
 899         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
 900                 return NESTED_EXIT_DONE;
 901
 902         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 903 }
 904
 905 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 906 {
 907         unsigned port, size, iopm_len;
 908         u16 val, mask;
 909         u8 start_bit;
 910         u64 gpa;
 911
 912         if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
 913                 return NESTED_EXIT_HOST;
 914
 915         port = svm->vmcb->control.exit_info_1 >> 16;
 916         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 917                 SVM_IOIO_SIZE_SHIFT;
 918         gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
 919         start_bit = port % 8;
 920         iopm_len = (start_bit + size > 8) ? 2 : 1;
 921         mask = (0xf >> (4 - size)) << start_bit;
 922         val = 0;
 923
 924         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 925                 return NESTED_EXIT_DONE;
 926
 927         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 928 }
 929
 930 static int nested_svm_intercept(struct vcpu_svm *svm)
 931 {
 932         u32 exit_code = svm->vmcb->control.exit_code;
 933         int vmexit = NESTED_EXIT_HOST;
 934
 935         switch (exit_code) {
 936         case SVM_EXIT_MSR:
 937                 vmexit = nested_svm_exit_handled_msr(svm);
 938                 break;
 939         case SVM_EXIT_IOIO:
 940                 vmexit = nested_svm_intercept_ioio(svm);
 941                 break;
 942         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 943                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 944                         vmexit = NESTED_EXIT_DONE;
 945                 break;
 946         }
 947         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 948                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 949                         vmexit = NESTED_EXIT_DONE;
 950                 break;
 951         }
 952         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 953                 /*
 954                  * Host-intercepted exceptions have been checked already in
 955                  * nested_svm_exit_special.  There is nothing to do here,
 956                  * the vmexit is injected by svm_check_nested_events.
 957                  */
 958                 vmexit = NESTED_EXIT_DONE;
 959                 break;
 960         }
 961         case SVM_EXIT_ERR: {
 962                 vmexit = NESTED_EXIT_DONE;
 963                 break;
 964         }
 965         default: {
 966                 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
 967                         vmexit = NESTED_EXIT_DONE;
 968         }
 969         }
 970
 971         return vmexit;
 972 }
 973
 974 int nested_svm_exit_handled(struct vcpu_svm *svm)
 975 {
 976         int vmexit;
 977
 978         vmexit = nested_svm_intercept(svm);
 979
 980         if (vmexit == NESTED_EXIT_DONE)
 981                 nested_svm_vmexit(svm);
 982
 983         return vmexit;
 984 }
 985
 986 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
 987 {
 988         if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
 989                 kvm_queue_exception(vcpu, UD_VECTOR);
 990                 return 1;
 991         }
 992
 993         if (to_svm(vcpu)->vmcb->save.cpl) {
 994                 kvm_inject_gp(vcpu, 0);
 995                 return 1;
 996         }
 997
 998         return 0;
 999 }
1000
1001 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1002 {
1003         unsigned int nr = svm->vcpu.arch.exception.nr;
1004
1005         return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1006 }
1007
1008 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1009 {
1010         unsigned int nr = svm->vcpu.arch.exception.nr;
1011
1012         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1013         svm->vmcb->control.exit_code_hi = 0;
1014
1015         if (svm->vcpu.arch.exception.has_error_code)
1016                 svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1017
1018         /*
1019          * EXITINFO2 is undefined for all exception intercepts other
1020          * than #PF.
1021          */
1022         if (nr == PF_VECTOR) {
1023                 if (svm->vcpu.arch.exception.nested_apf)
1024                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1025                 else if (svm->vcpu.arch.exception.has_payload)
1026                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1027                 else
1028                         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1029         } else if (nr == DB_VECTOR) {
1030                 /* See inject_pending_event.  */
1031                 kvm_deliver_exception_payload(&svm->vcpu);
1032                 if (svm->vcpu.arch.dr7 & DR7_GD) {
1033                         svm->vcpu.arch.dr7 &= ~DR7_GD;
1034                         kvm_update_dr7(&svm->vcpu);
1035                 }
1036         } else
1037                 WARN_ON(svm->vcpu.arch.exception.has_payload);
1038
1039         nested_svm_vmexit(svm);
1040 }
1041
1042 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1043 {
1044         return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1045 }
1046
1047 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1048 {
1049         struct vcpu_svm *svm = to_svm(vcpu);
1050         bool block_nested_events =
1051                 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1052         struct kvm_lapic *apic = vcpu->arch.apic;
1053
1054         if (lapic_in_kernel(vcpu) &&
1055             test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1056                 if (block_nested_events)
1057                         return -EBUSY;
1058                 if (!nested_exit_on_init(svm))
1059                         return 0;
1060                 nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1061                 return 0;
1062         }
1063
1064         if (vcpu->arch.exception.pending) {
1065                 if (block_nested_events)
1066                         return -EBUSY;
1067                 if (!nested_exit_on_exception(svm))
1068                         return 0;
1069                 nested_svm_inject_exception_vmexit(svm);
1070                 return 0;
1071         }
1072
1073         if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1074                 if (block_nested_events)
1075                         return -EBUSY;
1076                 if (!nested_exit_on_smi(svm))
1077                         return 0;
1078                 nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1079                 return 0;
1080         }
1081
1082         if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1083                 if (block_nested_events)
1084                         return -EBUSY;
1085                 if (!nested_exit_on_nmi(svm))
1086                         return 0;
1087                 nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1088                 return 0;
1089         }
1090
1091         if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1092                 if (block_nested_events)
1093                         return -EBUSY;
1094                 if (!nested_exit_on_intr(svm))
1095                         return 0;
1096                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1097                 nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1098                 return 0;
1099         }
1100
1101         return 0;
1102 }
1103
1104 int nested_svm_exit_special(struct vcpu_svm *svm)
1105 {
1106         u32 exit_code = svm->vmcb->control.exit_code;
1107
1108         switch (exit_code) {
1109         case SVM_EXIT_INTR:
1110         case SVM_EXIT_NMI:
1111         case SVM_EXIT_NPF:
1112                 return NESTED_EXIT_HOST;
1113         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1114                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1115
1116                 if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1117                     excp_bits)
1118                         return NESTED_EXIT_HOST;
1119                 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1120                          svm->vcpu.arch.apf.host_apf_flags)
1121                         /* Trap async PF even if not shadowing */
1122                         return NESTED_EXIT_HOST;
1123                 break;
1124         }
1125         default:
1126                 break;
1127         }
1128
1129         return NESTED_EXIT_CONTINUE;
1130 }
1131
1132 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1133                                 struct kvm_nested_state __user *user_kvm_nested_state,
1134                                 u32 user_data_size)
1135 {
1136         struct vcpu_svm *svm;
1137         struct kvm_nested_state kvm_state = {
1138                 .flags = 0,
1139                 .format = KVM_STATE_NESTED_FORMAT_SVM,
1140                 .size = sizeof(kvm_state),
1141         };
1142         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1143                 &user_kvm_nested_state->data.svm[0];
1144
1145         if (!vcpu)
1146                 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1147
1148         svm = to_svm(vcpu);
1149
1150         if (user_data_size < kvm_state.size)
1151                 goto out;
1152
1153         /* First fill in the header and copy it out.  */
1154         if (is_guest_mode(vcpu)) {
1155                 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1156                 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1157                 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1158
1159                 if (svm->nested.nested_run_pending)
1160                         kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1161         }
1162
1163         if (gif_set(svm))
1164                 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1165
1166         if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1167                 return -EFAULT;
1168
1169         if (!is_guest_mode(vcpu))
1170                 goto out;
1171
1172         /*
1173          * Copy over the full size of the VMCB rather than just the size
1174          * of the structs.
1175          */
1176         if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1177                 return -EFAULT;
1178         if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1179                          sizeof(user_vmcb->control)))
1180                 return -EFAULT;
1181         if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1182                          sizeof(user_vmcb->save)))
1183                 return -EFAULT;
1184 out:
1185         return kvm_state.size;
1186 }
1187
1188 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1189                                 struct kvm_nested_state __user *user_kvm_nested_state,
1190                                 struct kvm_nested_state *kvm_state)
1191 {
1192         struct vcpu_svm *svm = to_svm(vcpu);
1193         struct vmcb __user *user_vmcb = (struct vmcb __user *)
1194                 &user_kvm_nested_state->data.svm[0];
1195         struct vmcb_control_area *ctl;
1196         struct vmcb_save_area *save;
1197         int ret;
1198         u32 cr0;
1199
1200         BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1201                      KVM_STATE_NESTED_SVM_VMCB_SIZE);
1202
1203         if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1204                 return -EINVAL;
1205
1206         if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1207                                  KVM_STATE_NESTED_RUN_PENDING |
1208                                  KVM_STATE_NESTED_GIF_SET))
1209                 return -EINVAL;
1210
1211         /*
1212          * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1213          * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1214          */
1215         if (!(vcpu->arch.efer & EFER_SVME)) {
1216                 /* GIF=1 and no guest mode are required if SVME=0.  */
1217                 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1218                         return -EINVAL;
1219         }
1220
1221         /* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1222         if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1223                 return -EINVAL;
1224
1225         if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1226                 svm_leave_nested(svm);
1227                 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1228                 return 0;
1229         }
1230
1231         if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1232                 return -EINVAL;
1233         if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1234                 return -EINVAL;
1235
1236         ret  = -ENOMEM;
1237         ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
1238         save = kzalloc(sizeof(*save), GFP_KERNEL);
1239         if (!ctl || !save)
1240                 goto out_free;
1241
1242         ret = -EFAULT;
1243         if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1244                 goto out_free;
1245         if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1246                 goto out_free;
1247
1248         ret = -EINVAL;
1249         if (!nested_vmcb_check_controls(ctl))
1250                 goto out_free;
1251
1252         /*
1253          * Processor state contains L2 state.  Check that it is
1254          * valid for guest mode (see nested_vmcb_checks).
1255          */
1256         cr0 = kvm_read_cr0(vcpu);
1257         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1258                 goto out_free;
1259
1260         /*
1261          * Validate host state saved from before VMRUN (see
1262          * nested_svm_check_permissions).
1263          */
1264         if (!(save->cr0 & X86_CR0_PG) ||
1265             !(save->cr0 & X86_CR0_PE) ||
1266             (save->rflags & X86_EFLAGS_VM) ||
1267             !nested_vmcb_valid_sregs(vcpu, save))
1268                 goto out_free;
1269
1270         /*
1271          * All checks done, we can enter guest mode. Userspace provides
1272          * vmcb12.control, which will be combined with L1 and stored into
1273          * vmcb02, and the L1 save state which we store in vmcb01.
1274          * L2 registers if needed are moved from the current VMCB to VMCB02.
1275          */
1276
1277         svm->nested.nested_run_pending =
1278                 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1279
1280         svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1281         if (svm->current_vmcb == &svm->vmcb01)
1282                 svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1283
1284         svm->vmcb01.ptr->save.es = save->es;
1285         svm->vmcb01.ptr->save.cs = save->cs;
1286         svm->vmcb01.ptr->save.ss = save->ss;
1287         svm->vmcb01.ptr->save.ds = save->ds;
1288         svm->vmcb01.ptr->save.gdtr = save->gdtr;
1289         svm->vmcb01.ptr->save.idtr = save->idtr;
1290         svm->vmcb01.ptr->save.rflags = save->rflags | X86_EFLAGS_FIXED;
1291         svm->vmcb01.ptr->save.efer = save->efer;
1292         svm->vmcb01.ptr->save.cr0 = save->cr0;
1293         svm->vmcb01.ptr->save.cr3 = save->cr3;
1294         svm->vmcb01.ptr->save.cr4 = save->cr4;
1295         svm->vmcb01.ptr->save.rax = save->rax;
1296         svm->vmcb01.ptr->save.rsp = save->rsp;
1297         svm->vmcb01.ptr->save.rip = save->rip;
1298         svm->vmcb01.ptr->save.cpl = 0;
1299
1300         nested_load_control_from_vmcb12(svm, ctl);
1301
1302         svm_switch_vmcb(svm, &svm->nested.vmcb02);
1303
1304         nested_vmcb02_prepare_control(svm);
1305
1306         kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1307         ret = 0;
1308 out_free:
1309         kfree(save);
1310         kfree(ctl);
1311
1312         return ret;
1313 }
1314
1315 struct kvm_x86_nested_ops svm_nested_ops = {
1316         .check_events = svm_check_nested_events,
1317         .triple_fault = nested_svm_triple_fault,
1318         .get_nested_state_pages = svm_get_nested_state_pages,
1319         .get_state = svm_get_nested_state,
1320         .set_state = svm_set_nested_state,
1321 };