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
  23 #include "kvm_emulate.h"
  24 #include "trace.h"
  25 #include "mmu.h"
  26 #include "x86.h"
  27 #include "svm.h"
  28
  29 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  30                                        struct x86_exception *fault)
  31 {
  32         struct vcpu_svm *svm = to_svm(vcpu);
  33
  34         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  35                 /*
  36                  * TODO: track the cause of the nested page fault, and
  37                  * correctly fill in the high bits of exit_info_1.
  38                  */
  39                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  40                 svm->vmcb->control.exit_code_hi = 0;
  41                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
  42                 svm->vmcb->control.exit_info_2 = fault->address;
  43         }
  44
  45         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  46         svm->vmcb->control.exit_info_1 |= fault->error_code;
  47
  48         /*
  49          * The present bit is always zero for page structure faults on real
  50          * hardware.
  51          */
  52         if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
  53                 svm->vmcb->control.exit_info_1 &= ~1;
  54
  55         nested_svm_vmexit(svm);
  56 }
  57
  58 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  59 {
  60         struct vcpu_svm *svm = to_svm(vcpu);
  61         u64 cr3 = svm->nested.nested_cr3;
  62         u64 pdpte;
  63         int ret;
  64
  65         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
  66                                        offset_in_page(cr3) + index * 8, 8);
  67         if (ret)
  68                 return 0;
  69         return pdpte;
  70 }
  71
  72 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  73 {
  74         struct vcpu_svm *svm = to_svm(vcpu);
  75
  76         return svm->nested.nested_cr3;
  77 }
  78
  79 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  80 {
  81         WARN_ON(mmu_is_nested(vcpu));
  82
  83         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
  84         kvm_init_shadow_mmu(vcpu);
  85         vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
  86         vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
  87         vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
  88         vcpu->arch.mmu->shadow_root_level = kvm_x86_ops.get_tdp_level(vcpu);
  89         reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
  90         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
  91 }
  92
  93 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
  94 {
  95         vcpu->arch.mmu = &vcpu->arch.root_mmu;
  96         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
  97 }
  98
  99 void recalc_intercepts(struct vcpu_svm *svm)
 100 {
 101         struct vmcb_control_area *c, *h;
 102         struct nested_state *g;
 103
 104         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 105
 106         if (!is_guest_mode(&svm->vcpu))
 107                 return;
 108
 109         c = &svm->vmcb->control;
 110         h = &svm->nested.hsave->control;
 111         g = &svm->nested;
 112
 113         c->intercept_cr = h->intercept_cr;
 114         c->intercept_dr = h->intercept_dr;
 115         c->intercept_exceptions = h->intercept_exceptions;
 116         c->intercept = h->intercept;
 117
 118         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
 119                 /* We only want the cr8 intercept bits of L1 */
 120                 c->intercept_cr &= ~(1U << INTERCEPT_CR8_READ);
 121                 c->intercept_cr &= ~(1U << INTERCEPT_CR8_WRITE);
 122
 123                 /*
 124                  * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 125                  * affect any interrupt we may want to inject; therefore,
 126                  * interrupt window vmexits are irrelevant to L0.
 127                  */
 128                 c->intercept &= ~(1ULL << INTERCEPT_VINTR);
 129         }
 130
 131         /* We don't want to see VMMCALLs from a nested guest */
 132         c->intercept &= ~(1ULL << INTERCEPT_VMMCALL);
 133
 134         c->intercept_cr |= g->intercept_cr;
 135         c->intercept_dr |= g->intercept_dr;
 136         c->intercept_exceptions |= g->intercept_exceptions;
 137         c->intercept |= g->intercept;
 138 }
 139
 140 static void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
 141 {
 142         struct vmcb_control_area *dst  = &dst_vmcb->control;
 143         struct vmcb_control_area *from = &from_vmcb->control;
 144
 145         dst->intercept_cr         = from->intercept_cr;
 146         dst->intercept_dr         = from->intercept_dr;
 147         dst->intercept_exceptions = from->intercept_exceptions;
 148         dst->intercept            = from->intercept;
 149         dst->iopm_base_pa         = from->iopm_base_pa;
 150         dst->msrpm_base_pa        = from->msrpm_base_pa;
 151         dst->tsc_offset           = from->tsc_offset;
 152         dst->asid                 = from->asid;
 153         dst->tlb_ctl              = from->tlb_ctl;
 154         dst->int_ctl              = from->int_ctl;
 155         dst->int_vector           = from->int_vector;
 156         dst->int_state            = from->int_state;
 157         dst->exit_code            = from->exit_code;
 158         dst->exit_code_hi         = from->exit_code_hi;
 159         dst->exit_info_1          = from->exit_info_1;
 160         dst->exit_info_2          = from->exit_info_2;
 161         dst->exit_int_info        = from->exit_int_info;
 162         dst->exit_int_info_err    = from->exit_int_info_err;
 163         dst->nested_ctl           = from->nested_ctl;
 164         dst->event_inj            = from->event_inj;
 165         dst->event_inj_err        = from->event_inj_err;
 166         dst->nested_cr3           = from->nested_cr3;
 167         dst->virt_ext              = from->virt_ext;
 168         dst->pause_filter_count   = from->pause_filter_count;
 169         dst->pause_filter_thresh  = from->pause_filter_thresh;
 170 }
 171
 172 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 173 {
 174         /*
 175          * This function merges the msr permission bitmaps of kvm and the
 176          * nested vmcb. It is optimized in that it only merges the parts where
 177          * the kvm msr permission bitmap may contain zero bits
 178          */
 179         int i;
 180
 181         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 182                 return true;
 183
 184         for (i = 0; i < MSRPM_OFFSETS; i++) {
 185                 u32 value, p;
 186                 u64 offset;
 187
 188                 if (msrpm_offsets[i] == 0xffffffff)
 189                         break;
 190
 191                 p      = msrpm_offsets[i];
 192                 offset = svm->nested.vmcb_msrpm + (p * 4);
 193
 194                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 195                         return false;
 196
 197                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 198         }
 199
 200         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 201
 202         return true;
 203 }
 204
 205 static bool nested_vmcb_checks(struct vmcb *vmcb)
 206 {
 207         if ((vmcb->save.efer & EFER_SVME) == 0)
 208                 return false;
 209
 210         if (((vmcb->save.cr0 & X86_CR0_CD) == 0) &&
 211             (vmcb->save.cr0 & X86_CR0_NW))
 212                 return false;
 213
 214         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
 215                 return false;
 216
 217         if (vmcb->control.asid == 0)
 218                 return false;
 219
 220         if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
 221             !npt_enabled)
 222                 return false;
 223
 224         return true;
 225 }
 226
 227 void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
 228                           struct vmcb *nested_vmcb, struct kvm_host_map *map)
 229 {
 230         bool evaluate_pending_interrupts =
 231                 is_intercept(svm, INTERCEPT_VINTR) ||
 232                 is_intercept(svm, INTERCEPT_IRET);
 233
 234         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
 235                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
 236         else
 237                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
 238
 239         if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
 240                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
 241                 nested_svm_init_mmu_context(&svm->vcpu);
 242         }
 243
 244         /* Load the nested guest state */
 245         svm->vmcb->save.es = nested_vmcb->save.es;
 246         svm->vmcb->save.cs = nested_vmcb->save.cs;
 247         svm->vmcb->save.ss = nested_vmcb->save.ss;
 248         svm->vmcb->save.ds = nested_vmcb->save.ds;
 249         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
 250         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
 251         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
 252         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
 253         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
 254         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
 255         if (npt_enabled) {
 256                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
 257                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
 258         } else
 259                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
 260
 261         /* Guest paging mode is active - reset mmu */
 262         kvm_mmu_reset_context(&svm->vcpu);
 263
 264         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
 265         kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax);
 266         kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp);
 267         kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip);
 268
 269         /* In case we don't even reach vcpu_run, the fields are not updated */
 270         svm->vmcb->save.rax = nested_vmcb->save.rax;
 271         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
 272         svm->vmcb->save.rip = nested_vmcb->save.rip;
 273         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
 274         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
 275         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
 276
 277         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
 278         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
 279
 280         /* cache intercepts */
 281         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
 282         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
 283         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
 284         svm->nested.intercept            = nested_vmcb->control.intercept;
 285
 286         svm_flush_tlb(&svm->vcpu);
 287         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
 288         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
 289                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
 290         else
 291                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
 292
 293         svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
 294         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 295
 296         svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
 297         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
 298         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
 299         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
 300         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
 301
 302         svm->vmcb->control.pause_filter_count =
 303                 nested_vmcb->control.pause_filter_count;
 304         svm->vmcb->control.pause_filter_thresh =
 305                 nested_vmcb->control.pause_filter_thresh;
 306
 307         kvm_vcpu_unmap(&svm->vcpu, map, true);
 308
 309         /* Enter Guest-Mode */
 310         enter_guest_mode(&svm->vcpu);
 311
 312         /*
 313          * Merge guest and host intercepts - must be called  with vcpu in
 314          * guest-mode to take affect here
 315          */
 316         recalc_intercepts(svm);
 317
 318         svm->nested.vmcb = vmcb_gpa;
 319
 320         /*
 321          * If L1 had a pending IRQ/NMI before executing VMRUN,
 322          * which wasn't delivered because it was disallowed (e.g.
 323          * interrupts disabled), L0 needs to evaluate if this pending
 324          * event should cause an exit from L2 to L1 or be delivered
 325          * directly to L2.
 326          *
 327          * Usually this would be handled by the processor noticing an
 328          * IRQ/NMI window request.  However, VMRUN can unblock interrupts
 329          * by implicitly setting GIF, so force L0 to perform pending event
 330          * evaluation by requesting a KVM_REQ_EVENT.
 331          */
 332         enable_gif(svm);
 333         if (unlikely(evaluate_pending_interrupts))
 334                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
 335
 336         mark_all_dirty(svm->vmcb);
 337 }
 338
 339 int nested_svm_vmrun(struct vcpu_svm *svm)
 340 {
 341         int ret;
 342         struct vmcb *nested_vmcb;
 343         struct vmcb *hsave = svm->nested.hsave;
 344         struct vmcb *vmcb = svm->vmcb;
 345         struct kvm_host_map map;
 346         u64 vmcb_gpa;
 347
 348         vmcb_gpa = svm->vmcb->save.rax;
 349
 350         ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map);
 351         if (ret == -EINVAL) {
 352                 kvm_inject_gp(&svm->vcpu, 0);
 353                 return 1;
 354         } else if (ret) {
 355                 return kvm_skip_emulated_instruction(&svm->vcpu);
 356         }
 357
 358         ret = kvm_skip_emulated_instruction(&svm->vcpu);
 359
 360         nested_vmcb = map.hva;
 361
 362         if (!nested_vmcb_checks(nested_vmcb)) {
 363                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
 364                 nested_vmcb->control.exit_code_hi = 0;
 365                 nested_vmcb->control.exit_info_1  = 0;
 366                 nested_vmcb->control.exit_info_2  = 0;
 367
 368                 kvm_vcpu_unmap(&svm->vcpu, &map, true);
 369
 370                 return ret;
 371         }
 372
 373         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
 374                                nested_vmcb->save.rip,
 375                                nested_vmcb->control.int_ctl,
 376                                nested_vmcb->control.event_inj,
 377                                nested_vmcb->control.nested_ctl);
 378
 379         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
 380                                     nested_vmcb->control.intercept_cr >> 16,
 381                                     nested_vmcb->control.intercept_exceptions,
 382                                     nested_vmcb->control.intercept);
 383
 384         /* Clear internal status */
 385         kvm_clear_exception_queue(&svm->vcpu);
 386         kvm_clear_interrupt_queue(&svm->vcpu);
 387
 388         /*
 389          * Save the old vmcb, so we don't need to pick what we save, but can
 390          * restore everything when a VMEXIT occurs
 391          */
 392         hsave->save.es     = vmcb->save.es;
 393         hsave->save.cs     = vmcb->save.cs;
 394         hsave->save.ss     = vmcb->save.ss;
 395         hsave->save.ds     = vmcb->save.ds;
 396         hsave->save.gdtr   = vmcb->save.gdtr;
 397         hsave->save.idtr   = vmcb->save.idtr;
 398         hsave->save.efer   = svm->vcpu.arch.efer;
 399         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
 400         hsave->save.cr4    = svm->vcpu.arch.cr4;
 401         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
 402         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
 403         hsave->save.rsp    = vmcb->save.rsp;
 404         hsave->save.rax    = vmcb->save.rax;
 405         if (npt_enabled)
 406                 hsave->save.cr3    = vmcb->save.cr3;
 407         else
 408                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
 409
 410         copy_vmcb_control_area(hsave, vmcb);
 411
 412         enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, &map);
 413
 414         if (!nested_svm_vmrun_msrpm(svm)) {
 415                 svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 416                 svm->vmcb->control.exit_code_hi = 0;
 417                 svm->vmcb->control.exit_info_1  = 0;
 418                 svm->vmcb->control.exit_info_2  = 0;
 419
 420                 nested_svm_vmexit(svm);
 421         }
 422
 423         return ret;
 424 }
 425
 426 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 427 {
 428         to_vmcb->save.fs = from_vmcb->save.fs;
 429         to_vmcb->save.gs = from_vmcb->save.gs;
 430         to_vmcb->save.tr = from_vmcb->save.tr;
 431         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 432         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 433         to_vmcb->save.star = from_vmcb->save.star;
 434         to_vmcb->save.lstar = from_vmcb->save.lstar;
 435         to_vmcb->save.cstar = from_vmcb->save.cstar;
 436         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 437         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 438         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 439         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 440 }
 441
 442 int nested_svm_vmexit(struct vcpu_svm *svm)
 443 {
 444         int rc;
 445         struct vmcb *nested_vmcb;
 446         struct vmcb *hsave = svm->nested.hsave;
 447         struct vmcb *vmcb = svm->vmcb;
 448         struct kvm_host_map map;
 449
 450         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
 451                                        vmcb->control.exit_info_1,
 452                                        vmcb->control.exit_info_2,
 453                                        vmcb->control.exit_int_info,
 454                                        vmcb->control.exit_int_info_err,
 455                                        KVM_ISA_SVM);
 456
 457         rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map);
 458         if (rc) {
 459                 if (rc == -EINVAL)
 460                         kvm_inject_gp(&svm->vcpu, 0);
 461                 return 1;
 462         }
 463
 464         nested_vmcb = map.hva;
 465
 466         /* Exit Guest-Mode */
 467         leave_guest_mode(&svm->vcpu);
 468         svm->nested.vmcb = 0;
 469
 470         /* Give the current vmcb to the guest */
 471         disable_gif(svm);
 472
 473         nested_vmcb->save.es     = vmcb->save.es;
 474         nested_vmcb->save.cs     = vmcb->save.cs;
 475         nested_vmcb->save.ss     = vmcb->save.ss;
 476         nested_vmcb->save.ds     = vmcb->save.ds;
 477         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
 478         nested_vmcb->save.idtr   = vmcb->save.idtr;
 479         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
 480         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
 481         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
 482         nested_vmcb->save.cr2    = vmcb->save.cr2;
 483         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 484         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
 485         nested_vmcb->save.rip    = vmcb->save.rip;
 486         nested_vmcb->save.rsp    = vmcb->save.rsp;
 487         nested_vmcb->save.rax    = vmcb->save.rax;
 488         nested_vmcb->save.dr7    = vmcb->save.dr7;
 489         nested_vmcb->save.dr6    = vmcb->save.dr6;
 490         nested_vmcb->save.cpl    = vmcb->save.cpl;
 491
 492         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
 493         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
 494         nested_vmcb->control.int_state         = vmcb->control.int_state;
 495         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
 496         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
 497         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
 498         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
 499         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
 500         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
 501
 502         if (svm->nrips_enabled)
 503                 nested_vmcb->control.next_rip  = vmcb->control.next_rip;
 504
 505         /*
 506          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
 507          * to make sure that we do not lose injected events. So check event_inj
 508          * here and copy it to exit_int_info if it is valid.
 509          * Exit_int_info and event_inj can't be both valid because the case
 510          * below only happens on a VMRUN instruction intercept which has
 511          * no valid exit_int_info set.
 512          */
 513         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
 514                 struct vmcb_control_area *nc = &nested_vmcb->control;
 515
 516                 nc->exit_int_info     = vmcb->control.event_inj;
 517                 nc->exit_int_info_err = vmcb->control.event_inj_err;
 518         }
 519
 520         nested_vmcb->control.tlb_ctl           = 0;
 521         nested_vmcb->control.event_inj         = 0;
 522         nested_vmcb->control.event_inj_err     = 0;
 523
 524         nested_vmcb->control.pause_filter_count =
 525                 svm->vmcb->control.pause_filter_count;
 526         nested_vmcb->control.pause_filter_thresh =
 527                 svm->vmcb->control.pause_filter_thresh;
 528
 529         /* We always set V_INTR_MASKING and remember the old value in hflags */
 530         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
 531                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
 532
 533         /* Restore the original control entries */
 534         copy_vmcb_control_area(vmcb, hsave);
 535
 536         svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
 537         kvm_clear_exception_queue(&svm->vcpu);
 538         kvm_clear_interrupt_queue(&svm->vcpu);
 539
 540         svm->nested.nested_cr3 = 0;
 541
 542         /* Restore selected save entries */
 543         svm->vmcb->save.es = hsave->save.es;
 544         svm->vmcb->save.cs = hsave->save.cs;
 545         svm->vmcb->save.ss = hsave->save.ss;
 546         svm->vmcb->save.ds = hsave->save.ds;
 547         svm->vmcb->save.gdtr = hsave->save.gdtr;
 548         svm->vmcb->save.idtr = hsave->save.idtr;
 549         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
 550         svm_set_efer(&svm->vcpu, hsave->save.efer);
 551         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
 552         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
 553         if (npt_enabled) {
 554                 svm->vmcb->save.cr3 = hsave->save.cr3;
 555                 svm->vcpu.arch.cr3 = hsave->save.cr3;
 556         } else {
 557                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
 558         }
 559         kvm_rax_write(&svm->vcpu, hsave->save.rax);
 560         kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
 561         kvm_rip_write(&svm->vcpu, hsave->save.rip);
 562         svm->vmcb->save.dr7 = 0;
 563         svm->vmcb->save.cpl = 0;
 564         svm->vmcb->control.exit_int_info = 0;
 565
 566         mark_all_dirty(svm->vmcb);
 567
 568         kvm_vcpu_unmap(&svm->vcpu, &map, true);
 569
 570         nested_svm_uninit_mmu_context(&svm->vcpu);
 571         kvm_mmu_reset_context(&svm->vcpu);
 572         kvm_mmu_load(&svm->vcpu);
 573
 574         /*
 575          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 576          * doesn't end up in L1.
 577          */
 578         svm->vcpu.arch.nmi_injected = false;
 579         kvm_clear_exception_queue(&svm->vcpu);
 580         kvm_clear_interrupt_queue(&svm->vcpu);
 581
 582         return 0;
 583 }
 584
 585 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 586 {
 587         u32 offset, msr, value;
 588         int write, mask;
 589
 590         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 591                 return NESTED_EXIT_HOST;
 592
 593         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 594         offset = svm_msrpm_offset(msr);
 595         write  = svm->vmcb->control.exit_info_1 & 1;
 596         mask   = 1 << ((2 * (msr & 0xf)) + write);
 597
 598         if (offset == MSR_INVALID)
 599                 return NESTED_EXIT_DONE;
 600
 601         /* Offset is in 32 bit units but need in 8 bit units */
 602         offset *= 4;
 603
 604         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
 605                 return NESTED_EXIT_DONE;
 606
 607         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 608 }
 609
 610 /* DB exceptions for our internal use must not cause vmexit */
 611 static int nested_svm_intercept_db(struct vcpu_svm *svm)
 612 {
 613         unsigned long dr6;
 614
 615         /* if we're not singlestepping, it's not ours */
 616         if (!svm->nmi_singlestep)
 617                 return NESTED_EXIT_DONE;
 618
 619         /* if it's not a singlestep exception, it's not ours */
 620         if (kvm_get_dr(&svm->vcpu, 6, &dr6))
 621                 return NESTED_EXIT_DONE;
 622         if (!(dr6 & DR6_BS))
 623                 return NESTED_EXIT_DONE;
 624
 625         /* if the guest is singlestepping, it should get the vmexit */
 626         if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
 627                 disable_nmi_singlestep(svm);
 628                 return NESTED_EXIT_DONE;
 629         }
 630
 631         /* it's ours, the nested hypervisor must not see this one */
 632         return NESTED_EXIT_HOST;
 633 }
 634
 635 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 636 {
 637         unsigned port, size, iopm_len;
 638         u16 val, mask;
 639         u8 start_bit;
 640         u64 gpa;
 641
 642         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
 643                 return NESTED_EXIT_HOST;
 644
 645         port = svm->vmcb->control.exit_info_1 >> 16;
 646         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 647                 SVM_IOIO_SIZE_SHIFT;
 648         gpa  = svm->nested.vmcb_iopm + (port / 8);
 649         start_bit = port % 8;
 650         iopm_len = (start_bit + size > 8) ? 2 : 1;
 651         mask = (0xf >> (4 - size)) << start_bit;
 652         val = 0;
 653
 654         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 655                 return NESTED_EXIT_DONE;
 656
 657         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 658 }
 659
 660 static int nested_svm_intercept(struct vcpu_svm *svm)
 661 {
 662         u32 exit_code = svm->vmcb->control.exit_code;
 663         int vmexit = NESTED_EXIT_HOST;
 664
 665         switch (exit_code) {
 666         case SVM_EXIT_MSR:
 667                 vmexit = nested_svm_exit_handled_msr(svm);
 668                 break;
 669         case SVM_EXIT_IOIO:
 670                 vmexit = nested_svm_intercept_ioio(svm);
 671                 break;
 672         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 673                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
 674                 if (svm->nested.intercept_cr & bit)
 675                         vmexit = NESTED_EXIT_DONE;
 676                 break;
 677         }
 678         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 679                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
 680                 if (svm->nested.intercept_dr & bit)
 681                         vmexit = NESTED_EXIT_DONE;
 682                 break;
 683         }
 684         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 685                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
 686                 if (svm->nested.intercept_exceptions & excp_bits) {
 687                         if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
 688                                 vmexit = nested_svm_intercept_db(svm);
 689                         else
 690                                 vmexit = NESTED_EXIT_DONE;
 691                 }
 692                 /* async page fault always cause vmexit */
 693                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
 694                          svm->vcpu.arch.exception.nested_apf != 0)
 695                         vmexit = NESTED_EXIT_DONE;
 696                 break;
 697         }
 698         case SVM_EXIT_ERR: {
 699                 vmexit = NESTED_EXIT_DONE;
 700                 break;
 701         }
 702         default: {
 703                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
 704                 if (svm->nested.intercept & exit_bits)
 705                         vmexit = NESTED_EXIT_DONE;
 706         }
 707         }
 708
 709         return vmexit;
 710 }
 711
 712 int nested_svm_exit_handled(struct vcpu_svm *svm)
 713 {
 714         int vmexit;
 715
 716         vmexit = nested_svm_intercept(svm);
 717
 718         if (vmexit == NESTED_EXIT_DONE)
 719                 nested_svm_vmexit(svm);
 720
 721         return vmexit;
 722 }
 723
 724 int nested_svm_check_permissions(struct vcpu_svm *svm)
 725 {
 726         if (!(svm->vcpu.arch.efer & EFER_SVME) ||
 727             !is_paging(&svm->vcpu)) {
 728                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 729                 return 1;
 730         }
 731
 732         if (svm->vmcb->save.cpl) {
 733                 kvm_inject_gp(&svm->vcpu, 0);
 734                 return 1;
 735         }
 736
 737         return 0;
 738 }
 739
 740 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 741                                bool has_error_code, u32 error_code)
 742 {
 743         int vmexit;
 744
 745         if (!is_guest_mode(&svm->vcpu))
 746                 return 0;
 747
 748         vmexit = nested_svm_intercept(svm);
 749         if (vmexit != NESTED_EXIT_DONE)
 750                 return 0;
 751
 752         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
 753         svm->vmcb->control.exit_code_hi = 0;
 754         svm->vmcb->control.exit_info_1 = error_code;
 755
 756         /*
 757          * EXITINFO2 is undefined for all exception intercepts other
 758          * than #PF.
 759          */
 760         if (svm->vcpu.arch.exception.nested_apf)
 761                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
 762         else if (svm->vcpu.arch.exception.has_payload)
 763                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
 764         else
 765                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
 766
 767         svm->nested.exit_required = true;
 768         return vmexit;
 769 }
 770
 771 static void nested_svm_intr(struct vcpu_svm *svm)
 772 {
 773         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
 774         svm->vmcb->control.exit_info_1 = 0;
 775         svm->vmcb->control.exit_info_2 = 0;
 776
 777         /* nested_svm_vmexit this gets called afterwards from handle_exit */
 778         svm->nested.exit_required = true;
 779         trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
 780 }
 781
 782 static bool nested_exit_on_intr(struct vcpu_svm *svm)
 783 {
 784         return (svm->nested.intercept & 1ULL);
 785 }
 786
 787 int svm_check_nested_events(struct kvm_vcpu *vcpu)
 788 {
 789         struct vcpu_svm *svm = to_svm(vcpu);
 790         bool block_nested_events =
 791                 kvm_event_needs_reinjection(vcpu) || svm->nested.exit_required;
 792
 793         if (kvm_cpu_has_interrupt(vcpu) && nested_exit_on_intr(svm)) {
 794                 if (block_nested_events)
 795                         return -EBUSY;
 796                 nested_svm_intr(svm);
 797                 return 0;
 798         }
 799
 800         return 0;
 801 }
 802
 803 int nested_svm_exit_special(struct vcpu_svm *svm)
 804 {
 805         u32 exit_code = svm->vmcb->control.exit_code;
 806
 807         switch (exit_code) {
 808         case SVM_EXIT_INTR:
 809         case SVM_EXIT_NMI:
 810         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
 811                 return NESTED_EXIT_HOST;
 812         case SVM_EXIT_NPF:
 813                 /* For now we are always handling NPFs when using them */
 814                 if (npt_enabled)
 815                         return NESTED_EXIT_HOST;
 816                 break;
 817         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
 818                 /* When we're shadowing, trap PFs, but not async PF */
 819                 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
 820                         return NESTED_EXIT_HOST;
 821                 break;
 822         default:
 823                 break;
 824         }
 825
 826         return NESTED_EXIT_CONTINUE;
 827 }